A method for evaluating abnormal states during the operation of a solar cell

Abstract

The present invention relates to the field of data processing, and particularly to a method for evaluating abnormal states during the operation of solar cells. The method includes: collecting all monitoring data during the operation of the solar panels in a power station and performing window partitioning; determining the short-term abnormal degree of a window according to the importance of data in each dimension within the window and performing short-term abnormal judgment; determining a trend analysis factor according to the frequency change of short-term abnormal windows; determining a distance metric evaluation factor for the data window according to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel; determining the distance metric between multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor and completing the clustering process based on the distance metric, and evaluating the abnormal state according to the result of the clustering process, thereby improving the accuracy and effectiveness of abnormal state evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and particularly to a method for evaluating abnormal states during the operation of solar cells. Background Art

[0002] The long-term stability and efficiency of solar panels directly affect the performance and economic benefits of the entire solar power generation system. Over time, the panels may be affected by factors such as environmental changes, equipment aging, and external interference, resulting in a decline in operating efficiency or malfunctions, and even potential safety hazards. By monitoring the panels in real time and evaluating their abnormal states promptly, potential problems can be detected early and effective maintenance measures can be taken to avoid equipment downtime or performance degradation, reduce economic losses, extend the service life of the panels, and improve the safety and reliability of the system. Therefore, the evaluation of abnormal states is crucial during the operation of solar cells and is the basis for ensuring the stable operation of the system and achieving high-efficiency power generation.

[0003] During the operation of solar panels, real-time monitoring and abnormal state evaluation are the keys to ensuring their efficient and stable operation. Traditional abnormal monitoring methods mainly rely on clustering analysis based on Euclidean distance, which clusters the multi-dimensional operation monitoring data of the panels to identify normal and abnormal states. However, these methods usually do not fully consider the temporal characteristics in the data and are easily affected by short-term anomalies caused by external environmental factors (such as weather changes, cloud cover, etc.), resulting in misidentification of abnormal states. In particular, when the operation data of solar panels shows fluctuations caused by short-term weather changes, traditional methods may misjudge these short-term anomalies as equipment failures, thus affecting subsequent fault monitoring and maintenance decisions. At the same time, existing methods also lack sensitivity to long-term trends (such as equipment aging or efficiency decline) and are prone to ignoring the gradually deteriorating trend of the panels.

[0004] Therefore, how to ensure the accuracy of abnormal pattern recognition in the operation monitoring data of solar cells has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method for evaluating abnormal states during the operation of solar cells to solve the problem of how to ensure the accuracy of abnormal pattern recognition in the operation monitoring data of solar cells.

[0006] Embodiments of the present invention provide a method for evaluating abnormal states during the operation of solar cells. A method for evaluating abnormal states during the operation of solar cells includes the following steps:

[0007] Collect all the monitoring data during the operation of the solar panels in the power station, respectively obtain the multi-dimensional operation monitoring time series data of each solar panel, divide the multi-dimensional operation monitoring time series data of all solar panels into windows, and obtain the multi-dimensional operation monitoring time series data windows of all solar panels;

[0008] For any solar panel, obtain the multi-dimensional operation monitoring time series data window of the solar panel, and determine the importance degree of each dimension according to the change information of each dimension data in the multi-dimensional operation monitoring time series data window of the solar panel; Determine the short-term abnormal degree of the window according to the importance degree of each dimension in the multi-dimensional operation monitoring time series data window of the solar panel; Perform a threshold judgment according to the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel, obtain the short-term abnormal window and determine the trend analysis factor according to the frequency change of the short-term abnormal window;

[0009] Determine the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window according to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel;

[0010] Determine the time series optimization distance metric between the multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional abnormal monitoring time series data window, and complete the clustering process based on the time series optimization distance metric, and perform an abnormal state evaluation according to the result of the clustering process.

[0011] Preferably, the determining the importance degree of each dimension according to the change information of each dimension data in the multi-dimensional operation monitoring time series data window of the solar panel includes:

[0012] According to the distance between the monitoring data points in each dimension in the multi-dimensional operation monitoring time series data window of the solar panel, determine the connected path distance sequence of each monitoring data point in each monitoring dimension in the multi-dimensional monitoring time series data window; Take the absolute value of the difference between the corresponding positions of the connected path distance sequences of each multi-dimensional monitoring data point corresponding to each timestamp in the window in each dimension and the connected path distance sequences of other timestamps in this dimension as the first difference, and use the reciprocal of the path number in the path distance sequence as the weight to perform a weighted sum of the first difference and divide by the number of paths in the path distance sequence to obtain the first path distance difference mean value; Perform a normalization process on the mean value of the first path distance difference means between the connected path distance sequences of each multi-dimensional monitoring data point corresponding to each timestamp in the window in each dimension and the connected path distance sequences of all other timestamps in this dimension, obtain the normalization result, and use the normalization result as the importance degree of each multi-dimensional monitoring data point corresponding to each timestamp in the window in each dimension.

[0013] Preferably, determining the connected path distance sequence of each monitoring data point in each monitoring dimension in the multi-dimensional monitoring time series data window according to the distances between the monitoring data points in each dimension in the multi-dimensional operation monitoring time series data window of the solar panel includes:

[0014] For any monitoring data point in any dimension in the window, taking this monitoring data point as the starting point, traversing the path according to the minimum Euclidean distance in the window, and taking the corresponding path of the traversal result as the connected path; taking the Euclidean distance between the two data points corresponding to each sub-path in the connected path as the path distance of the sub-path, and obtaining the connected path distance sequence according to the path distances of each sub-path in the connected path.

[0015] Preferably, determining the short-term anomaly degree of the window according to the importance degree of each dimension in the multi-dimensional operation monitoring time series data window of the solar panel includes:

[0016] Obtaining the importance degree of each multi-dimensional monitoring data point corresponding to each timestamp in the window in each dimension, and taking the average value of the importance degrees of all multi-dimensional monitoring data points in any dimension in any window as the window dimension importance degree of this dimension; taking the average value of the window dimension importance degrees of any dimension in all windows as the overall dimension importance degree of this dimension; dividing the overall dimension importance degree of any dimension by the sum of the overall importance degrees of all dimensions as the fusion importance degree of this dimension; taking the average value of the monitoring data of any dimension in the window as the first average value of the window;

[0017] For any dimension in the window, starting from the second data point, taking the result of multiplying the importance degree of the data point by the difference in the monitoring values between this data point and the previous data point as the first difference value; taking the weighted average calculation result of all the first difference values in any dimension in the window with the importance degree as the weight as the second difference value of this dimension; taking the square of the result of subtracting the monitoring value corresponding to any timestamp in any dimension in the window from the first average value of the window as the first mean difference; taking the square root of the weighted average of all the first mean differences of all timestamps in any dimension in the window with the importance degree corresponding to each timestamp as the first fluctuation degree of this dimension;

[0018] Taking the square root of the weighted average of the squares of the first fluctuation degrees of all dimensions in the window according to the fusion importance degree as the second fluctuation degree of the window; taking the square root of the weighted average of the second difference values of all dimensions in the window according to the fusion importance degree as the third difference value of the window;

[0019] Obtain the second fluctuation degree of the window and the third difference value of the window, and use the result of dividing the second fluctuation degree of the window by the third difference value of the window and performing normalization processing as the short-term anomaly degree of the window.

[0020] Preferably, performing threshold judgment according to the short-term anomaly degree of the window of the multi-dimensional operation monitoring time series data of the solar panel, obtaining short-term anomaly windows, and determining a trend analysis factor according to the frequency change of the short-term anomaly windows, including:

[0021] Obtain a short-term anomaly threshold, judge the short-term anomaly degree of the window according to the short-term anomaly threshold, and obtain all short-term anomaly windows; obtain a second window, and obtain the short-term anomaly frequency change factor in the second window according to the short-term anomaly change information of the first window in the second window; obtain the least squares fitting result of the multi-dimensional monitoring time series data in the second window according to the least squares fitting result of the multi-dimensional monitoring time series data in the second window, and use the result of calculating the weighted mean of the average slopes in the fitting results corresponding to each dimension in the monitoring time series data in the second window through the fusion importance degree of the dimension as the window trend feature factor in the second window; obtain the trend item decomposition result of all multi-dimensional monitoring time series data in the second window through trend item decomposition according to all multi-dimensional monitoring time series data of the solar panel, and use the result of calculating the weighted mean of the average slopes of the trend item decomposition results of each dimension in the second window through the fusion importance degree of the dimension as the overall window trend feature factor of the trend item; obtain the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window, and use the calculation result of the absolute value of the subtraction between the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window as the first trend feature factor;

[0022] Obtain the short-term anomaly frequency change factor in the second window and the first trend feature factor of the second window, use the product of the short-term anomaly frequency change factor and the first trend feature factor as the second trend feature factor, and use the calculation result of subtracting the constant 1 from the second trend feature factor as the trend analysis factor of the second window.

[0023] Preferably, determining a distance metric evaluation factor for the window of the multi-dimensional operation monitoring time series data of the solar panel according to the trend analysis factor and the short-term anomaly degree of the window of the multi-dimensional operation monitoring time series data of the solar panel, including:

[0024] Obtain the trend analysis factor of the second window and the short-term anomaly degree of the first window. Take the trend analysis factor of the second window as the trend analysis factor of all the first windows in the second window; take the calculation result of subtracting the constant 1 from the trend analysis factor of the first window as the first trend feature item of the first window; take the calculation result of multiplying the short-term anomaly degree of the first window by the first trend feature item of the first window as the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window.

[0025] Preferably, determining the time series optimization distance metric between multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional anomaly monitoring time series data window, and completing the clustering process based on the distance metric, includes:

[0026] Obtain two multi-dimensional monitoring data that need to be distance-measured and the first windows corresponding to the two multi-dimensional monitoring data. Take the mean of the distance metric evaluation factors of the first windows corresponding to the two multi-dimensional monitoring data as the time series difference weight, and multiply the time series difference weight by the DTW distance between the first windows corresponding to the two multi-dimensional monitoring data as the time series difference metric; take the result of subtracting the constant 1 from the time series difference weight as the Euclidean distance difference weight, and multiply the Euclidean distance difference weight by the Euclidean distance between the two multi-dimensional monitoring data as the Euclidean distance difference metric;

[0027] Obtain the time series difference metric and the Euclidean distance difference metric of the two multi-dimensional monitoring data and the two multi-dimensional operation monitoring time series data windows corresponding to the two multi-dimensional monitoring data. Add the time series difference metric and the Euclidean distance difference metric as the time series optimization distance metric; and perform clustering to obtain the clustering result.

[0028] Preferably, the abnormal state evaluation according to the result of the clustering process includes:

[0029] Obtain the result of the clustering process. Take the average Euclidean distance between the cluster center point of any cluster in the clustering result and the cluster center points of all other clusters as the first Euclidean distance of the cluster; take the average Euclidean distance between the data points within the cluster of any cluster in the clustering result and the cluster center point of the cluster as the second Euclidean distance of the cluster;

[0030] Obtain the first Euclidean distance and the second Euclidean distance of the cluster. Take the result of multiplying the first Euclidean distance and the second Euclidean distance and performing normalization processing as the abnormal degree of the cluster;

[0031] Obtain the threshold of the anomaly degree of the cluster class. Compare the anomaly degree of the cluster class with the threshold of the anomaly degree of the cluster class. If the anomaly degree of the cluster class is greater than or equal to the threshold of the anomaly degree of the cluster class, determine that the cluster class is an abnormal state cluster class; if the anomaly degree of the cluster class is less than the threshold of the anomaly degree of the cluster class, determine that the cluster class is a normal state cluster class, so as to complete the evaluation of the abnormal state during the operation of the solar cell.

[0032] Preferably, for the obtaining of the second window, according to the short-term anomaly change information of the first window in the second window, obtain the short-term anomaly frequency change factor in the second window; according to the short-term anomaly frequency change factor in the second window, obtain the window trend feature factor in the second window through the least squares fitting result, including:

[0033] The second window contains at least three first windows. Take the number of short-term abnormal first windows in the second window as the short-term anomaly frequency change evaluation factor of the second window. According to the short-term anomaly frequency change evaluation factors of all second windows in the multi-dimensional operation monitoring time series data of the solar panel, obtain the short-term anomaly frequency factor sequence. Perform least squares fitting on the short-term anomaly frequency factor sequence to obtain the fitting result of the short-term anomaly frequency change evaluation factor. Take the normalized calculation result of the slope corresponding to the second window in the fitting result of the short-term anomaly frequency factor as the short-term anomaly frequency change factor in the second window.

[0034] Preferably, for the window division of all the multi-dimensional operation monitoring time series data of the solar panels, to obtain all the multi-dimensional operation monitoring time series data windows of the solar panels, including:

[0035] Respectively obtain the preset window length of the multi-dimensional operation monitoring time series data of the solar panel; perform a traversal with a continuous step size of 1 on the multi-dimensional operation monitoring time series data of the solar panel according to the preset window length. Take the subsequence contained in each movement of the window with the preset window length as the multi-dimensional operation monitoring time series data window corresponding to the center point of the window.

[0036] The beneficial effects of the embodiments of the present invention compared with the prior art are:

[0037] The present invention collects all the monitoring data during the operation of the solar panels in the power station, and respectively obtains the multi-dimensional operation monitoring time-series data of each solar panel. Window partitioning is performed on the multi-dimensional operation monitoring time-series data of all solar panels to obtain the multi-dimensional operation monitoring time-series data windows of all solar panels. For any solar panel, the multi-dimensional operation monitoring time-series data window of the solar panel is obtained, and the importance degree of the dimension is determined according to the change information of each dimension data within the multi-dimensional operation monitoring time-series data window of the solar panel. The short-term abnormal degree of the window is determined according to the importance degree of each dimension within the multi-dimensional operation monitoring time-series data window of the solar panel. Threshold judgment is performed according to the short-term abnormal degree of the multi-dimensional operation monitoring time-series data window of the solar panel, and the short-term abnormal window is obtained and the trend analysis factor is determined according to the frequency change of the short-term abnormal window. According to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time-series data window of the solar panel, the distance metric evaluation factor of the multi-dimensional operation monitoring time-series data window is determined. According to the distance metric evaluation factor of the multi-dimensional abnormal monitoring time-series data window, the time-series optimization distance metric between the multi-dimensional operation monitoring time-series data windows is determined, and the clustering process is completed based on the time-series optimization distance metric, and the abnormal state evaluation is performed according to the result of the clustering process. Among them, by introducing the time-series difference metric of the multi-dimensional monitoring time-series data in the clustering process, the interference of the fluctuation caused by the short-term abnormality to the clustering is eliminated, thereby reducing the misjudgment and improving the accuracy of the abnormal state recognition. In the clustering process for abnormal state evaluation, the Euclidean distance and the time-series difference metric are dynamically evaluated through the distance metric evaluation factor, which can effectively handle the environmental fluctuations caused by seasonal changes or weather factors and enhance the adaptability to these environmental factors. By dynamically adjusting the weight of the distance metric evaluation factor, the short-term environmental fluctuations are avoided being misjudged as abnormal states, thereby improving the adaptability of the abnormal evaluation method to environmental changes. In the above, by introducing the trend analysis factor, the distance metric evaluation factor is adaptively adjusted according to different operation data and abnormal patterns, so as to ensure that the abnormal detection method can dynamically adjust the detection strategy, distinguish the short-term weather fluctuations from the long-term decline trend, and improve the accuracy of the abnormal state evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and the descriptions thereof are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0039] Figure 1 It is a flowchart of an abnormal state evaluation method for the operation of solar cells provided in Embodiment 1 of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] Embodiments of the present disclosure will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals designate like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present disclosure, and should not be construed as a limitation to the present disclosure.

[0041] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above accompanying drawings are used to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0042] To illustrate the technical solution of the present invention, it will be described below through specific embodiments.

[0043] The specific scenario targeted by the present invention is:

[0044] See Figure 1 , which is a method flowchart of an abnormal state evaluation method for a solar cell during operation provided in the first embodiment of the present invention. As Figure 1 shown, an abnormal state evaluation method for a solar cell during operation may include:

[0045] Step S101, collect all the monitoring data during the operation of the solar panels in the power station, respectively obtain the multi-dimensional operation monitoring time series data of each solar panel, and perform window partitioning on the multi-dimensional operation monitoring time series data of all solar panels to obtain the multi-dimensional operation monitoring time series data windows of all solar panels.

[0046] In a solar power station, a variety of monitoring sensors are installed on each solar panel to collect the operation monitoring data of the solar panel. Among them, there are temperature sensors, voltage sensors, current sensors, power sensors, irradiance sensors, etc. Therefore, in the embodiment of the present invention, all the multi-dimensional monitoring time series data of each solar panel in the power station are collected. It should be noted that the multi-dimensional operation monitoring time series data of the solar panel are all known historical monitoring data, and are collected once per minute. There is no limitation here and it can be set according to the real-time scenario.

[0047] After obtaining the multi-dimensional operation monitoring time-series data of each solar panel, it is necessary to divide the window of the multi-dimensional operation monitoring time-series data of the solar panel according to the preset time-series window length. Preferably, in the embodiment of the present invention, the window length is set The entire multi-dimensional operation monitoring time-series data of the solar panel is traversed with a step size of 1 using a window of

[0048] Step S102, for any solar panel, obtain the multi-dimensional operation monitoring time-series data window of the solar panel, and determine the importance degree of the dimension according to the change information of each dimension data in the window; determine the short-term abnormal degree of the window according to the importance degree of each dimension in the window; perform threshold judgment according to the short-term abnormal degree of the window to obtain a short-term abnormal window; determine a trend analysis factor according to the frequency change of the short-term abnormal window.

[0049] After obtaining the multi-dimensional operation monitoring time-series data of each solar panel and the multi-dimensional operation monitoring time-series data window, the operation mode of the multi-dimensional monitoring data points of the solar panel can be recognized and divided by using a clustering algorithm according to the difference between the multi-dimensional monitoring data points of the solar panel. When the traditional clustering algorithm is used to recognize and divide the operation mode of the solar panel operation monitoring data, it is realized according to the Euclidean distance between the multi-dimensional operation monitoring data points of the solar panel. However, when there is a sudden change in light or a change in environmental wind speed in the solar panel, the multi-dimensional monitoring data points may mutate, resulting in the short-term abnormality being recognized as an abnormal mode in the process of recognizing the operation mode of the solar panel based on clustering. Therefore, when measuring the distance between the multi-dimensional operation monitoring data points of the solar panel during the clustering process, it is necessary to comprehensively measure the Euclidean distance between the multi-dimensional operation monitoring data points of the solar panel by combining the time-series change difference between the multi-dimensional operation monitoring time-series data windows corresponding to the multi-dimensional operation monitoring time-series data points of the solar panel, so as to achieve the purpose of optimizing the Euclidean distance in the traditional clustering process.

[0050] In the process of optimizing the temporal variation difference of the Euclidean distance between multi-dimensional monitoring data points of a solar panel, since different degrees of temporal variation difference are required for optimization under different operating states, it is necessary to evaluate the distance metric weight, that is, the distance metric evaluation factor, between the Euclidean distance between multi-dimensional monitoring data points and the temporal difference between the multi-dimensional operating monitoring temporal data window corresponding to the multi-dimensional monitoring data points, which is used for optimizing the distance metric between the multi-dimensional operating monitoring data of the solar panel during the clustering process. During the evaluation process of the above-mentioned distance metric evaluation factor, since the monitoring data of the solar panel is multi-dimensional data, and each timestamp in the multi-dimensional operating monitoring temporal data corresponds to the data of multiple monitoring dimensions of the panel at the sampling time, it is necessary to evaluate the importance degree of each monitoring data point in each multi-dimensional operating monitoring temporal data window during the analysis process, so as to perform multi-dimensional information fusion through the importance degree of dimensions, thereby ensuring the accuracy of the distance metric evaluation factor during the distance optimization process. After obtaining the importance degree of the data points in each dimension in the temporal window, since the performance degradation caused by long-term use of the solar panel leads to abnormal changes in multi-dimensional monitoring data points similar to short-term anomalies, which will present as a continuous increase in the occurrence frequency of short-term anomaly windows, after obtaining the multi-dimensional operating monitoring temporal data of each solar panel, further determine the short-term anomaly degree of the window according to the importance degree of each dimension in the window, and perform threshold judgment according to the short-term anomaly degree to obtain the short-term anomaly temporal data window in all the multi-dimensional operating monitoring temporal data windows of the solar panel, and evaluate the short-term anomaly frequency of the short-term anomaly temporal data window through the second window, so as to obtain the trend analysis factor of the short-term anomaly window.

[0051] The specific optimization process is as follows: determine the importance degree of dimensions according to the change information of each dimension data in the window; determine the short-term anomaly degree of the window according to the importance degree of each dimension in the window; perform threshold judgment according to the short-term anomaly degree of the window to obtain the short-term anomaly window; determine the trend analysis factor according to the frequency change of the short-term anomaly window.

[0052] Among them, determining the importance degree of dimensions according to the change information of each dimension data in the window includes:

[0053] Determine the connected path distance sequence of each monitoring data point in each monitoring dimension in the multi-dimensional monitoring time series data window according to the distances between the monitored data points in the window in each dimension; take the absolute value of the difference between the corresponding positions of the connected path distance sequences of the multi-dimensional monitored data points corresponding to each timestamp in the window in each dimension and the connected path distance sequences of other timestamps in that dimension as the first difference, and perform a weighted sum of the first difference by taking the reciprocal of the path number in the path distance sequence and dividing by the number of paths in the path distance sequence to obtain the first path distance difference mean; perform normalization processing on the mean of the first path distance difference means between the connected path distance sequences of the multi-dimensional monitored data points corresponding to each timestamp in the window in each dimension and the connected path distance sequences of all other timestamps in that dimension, and obtain the normalization result, and use the normalization result as the importance degree of the multi-dimensional monitored data points corresponding to each timestamp in the window in each dimension.

[0054] In one embodiment, taking the multi-dimensional operation monitoring time series data window of the th solar panel as an example, denote the th window as , the th path distance in the connected path distance sequence of the data point corresponding to the th dimension and the th timestamp in the th window is , then the calculation expression for the importance degree of the data point corresponding to the th window, the th dimension, and the th timestamp of the solar panel is:

[0055] ;

[0056] where represents the importance degree of the data point corresponding to the th window, the th dimension, and the th timestamp of the solar panel, represents the normalization function, represents the number of data points in the window, represents the path number in the path distance sequence, represents the th path distance in the path distance sequence corresponding to the th dimension and the th timestamp in the th window, represents the th window, the The first difference between the th path distances in the path distance sequence corresponding to a time stamp.

[0057] Wherein, the method for determining the connected path distance sequence of each monitoring data point in each monitoring dimension in the multi-dimensional monitoring time series data window according to the distances between the monitoring data points in the window in each dimension includes:

[0058] For any monitoring data point in any dimension in the window, taking this monitoring data point as the starting point, traversing the path according to the minimum Euclidean distance in the window, and taking the path corresponding to the traversal result as the connected path; taking the Euclidean distance between the two data points corresponding to each sub-path in the connected path as the path distance of the sub-path, and obtaining the connected path distance sequence according to the path distances of each sub-path in the connected path.

[0059] After obtaining the importance degree of each multi-dimensional monitoring data point in each dimension corresponding to each time stamp in the window, it is necessary to determine the short-term anomaly degree of the window according to the importance degrees of each dimension in the window, so as to judge the short-term anomaly window through the short-term anomaly degree of the window, and thus evaluate the trend analysis factor through the frequency change presented by the short-term anomaly window.

[0060] Wherein, the method for determining the short-term anomaly degree of the window according to the importance degrees of each dimension in the window includes:

[0061] Obtaining the importance degree of each multi-dimensional monitoring data point in each dimension corresponding to each time stamp in the window, and taking the average value of the importance degrees of all multi-dimensional monitoring data points in any dimension in any window as the window dimension importance degree of this dimension;

[0062] In an embodiment, the calculation expression of the window dimension importance degree of the th dimension in the th window is:

[0063] ;

[0064] Wherein, represents the window dimension importance degree of the th dimension in the th window, represents the number of time stamps in the window, that is, the window length, represents the importance degree of the th dimension in the multi-dimensional data point corresponding to the th time stamp in the th time series data window.

[0065] Take the average of the window dimension importance degrees of any dimension in all windows as the overall dimension importance degree of this dimension;

[0066] In one embodiment, the calculation expression of the overall dimension importance degree of the

[0067] -th dimension is:

[0068] Among them, represents the overall dimension importance degree of the -th dimension, represents the window dimension importance degree of the -th window in the -th dimension, represents the total number of windows in the multi-dimensional operation monitoring time series data of all solar panels.

[0069] Take the overall dimension importance degree of any dimension divided by the sum of the overall dimension importance degrees of all dimensions as the fusion importance degree of this dimension;

[0070] In one embodiment, the calculation expression of the fusion importance degree of the

[0071] -th dimension is:

[0072] Among them, represents the fusion importance degree of the -th dimension, represents the overall dimension importance degree of the -th dimension, represents the overall dimension importance degree of the -th dimension, represents the total number of dimensions.

[0073] Take the average of the monitoring data of any dimension in the said window as the first window average;

[0074] In one embodiment, assume that the monitoring data corresponding to the -th time stamp in the -th dimension in the -th window is , then the -th window in the -th dimension, the calculation expression of the first window average is:

[0075] ;

[0076] Among them, represents the -th window in the The first mean value of the window in the dimension represents the monitoring data corresponding to the -th timestamp in the -th dimension of the -th window, and represents the number of timestamps in the said window.

[0077] For any dimension in the said window, starting from the second data point, multiply the importance level of the data point by the difference in the monitoring value between this data point and the previous data point, and the result is used as the first difference value;

[0078] In one implementation, let the monitoring data corresponding to the -th timestamp in the -th dimension of the -th window be , then the calculation expression for the first difference value of the data point corresponding to the -th timestamp in the -th dimension of the -th time series data window is:

[0079] ;

[0080] where represents the first difference value of the data point corresponding to the -th timestamp in the -th dimension of the -th time series data window, represents the monitoring data corresponding to the -th timestamp in the -th dimension of the -th window, represents the monitoring data corresponding to the -th timestamp in the -th dimension of the -th window, represents the importance level of the -th dimension among the multi-dimensional data points corresponding to the -th timestamp in the -th time series data window, represents the absolute value calculation symbol.

[0081] It should be noted that the first difference value of the data point corresponding to the -th timestamp in the -th dimension of the -th time series data window is only obtained starting from the -th data point in the said time series data window.

[0082] Taking the importance as the weight, the weighted mean calculation result of all the first difference values of any dimension in the window is used as the second difference value of this dimension;

[0083] In an embodiment, the calculation expression of the second difference value of the th window for the th dimension is:

[0084] ;

[0085] Wherein, represents the second difference value of the th window for the th dimension, represents the importance of the th multi-dimensional data point corresponding to the th timestamp in the th time series data window for the th dimension, represents the first difference value of the th dimension at the data point corresponding to the th timestamp in the th time series data window, represents the absolute value calculation symbol.

[0086] Taking the square of the result of subtracting the monitored value corresponding to any timestamp of any dimension in the window from the first mean of the window as the first mean difference; taking the square root of the weighted mean of all the first mean differences of any dimension in the window through the importance corresponding to each timestamp as the first fluctuation degree of this dimension;

[0087] In an embodiment, assuming that the monitored data corresponding to the th window for the th dimension and the th timestamp is , then the calculation expression of the first fluctuation degree of the th window for the th dimension is:

[0088] ;

[0089] Wherein, represents the first fluctuation degree of the th window for the th dimension, represents the importance of the th multi-dimensional data point corresponding to the th timestamp in the th time series data window for the Indicates the monitoring data corresponding to the timestamp in the th dimension within the th window, the first window mean value of the th window in the th dimension,

[0090] Calculates the square root of the weighted mean of the squares of the first fluctuation degrees of all dimensions in the window according to the fusion importance degree as the second fluctuation degree of the window;

[0091] In one embodiment, the calculation expression for the second fluctuation degree of the th window is:

[0092] ;

[0093] wherein, represents the second fluctuation degree of the th window, represents the fusion importance degree of the th dimension, represents the square of the first fluctuation degree of the th dimension in the th window, represents the total number of dimensions.

[0094] Calculates the square root of the weighted mean of the second difference values of all dimensions in the window according to the fusion importance degree as the third difference value of the window;

[0095] In one embodiment, the calculation expression for the third difference value of the th window is:

[0096] ;

[0097] wherein, represents the third difference value of the th window, represents the fusion importance degree of the th dimension, represents the second difference value of the th dimension in the th window, represents the total number of dimensions.

[0098] Obtain the second fluctuation degree of the window and the third difference value of the window, and use the result of dividing the second fluctuation degree of the window by the third difference value of the window and performing normalization processing as the short-term anomaly degree of the window.

[0099] In one embodiment, the calculation expression for the short-term anomaly degree of the

[0100] th window is:

[0101] wherein, represents the short-term anomaly degree of the th window, represents the second fluctuation degree of the th window, represents the third difference value of the th window. represents the normalization function.

[0102] It should be noted that by using the fluctuation degree of the monitoring values in the time-series data window and the change information of the monitoring data in the window to evaluate the short-term anomaly degree of the time-series data window, the higher the second fluctuation degree , the higher the short-term anomaly degree. Based on the short-term anomaly degree of the window, the larger the third difference value of the window, the lower the short-term anomaly degree of the window.

[0103] After obtaining the short-term anomaly degree of the multi-dimensional operation time-series data window of the solar panel, perform short-term anomaly window judgment according to the short-term anomaly degree of the time-series data window through a short-term anomaly threshold, so as to evaluate the trend analysis factor through the change in the number of short-term windows and the overall trend of the multi-dimensional operation monitoring time-series data of the solar panel.

[0104] Among them, the threshold judgment according to the short-term anomaly degree of the window, obtaining the short-term anomaly window and determining the trend analysis factor according to the frequency change of the short-term anomaly window includes:

[0105] Obtain the short-term anomaly threshold, determine the short-term anomaly degree of the window according to the short-term anomaly threshold, and obtain all short-term anomaly windows; obtain the second window, and obtain the short-term anomaly frequency change factor in the second window according to the short-term anomaly change information of the first window in the second window; obtain the least squares fitting result of the multi-dimensional monitoring time series data in the second window through the least squares fitting result of the multi-dimensional monitoring time series data in the second window, and use the result of weighted mean calculation of the average slope in the fitting results corresponding to each dimension in the monitoring time series data in the second window by the fusion importance degree of the dimension as the window trend feature factor in the second window; obtain the trend item decomposition result of all multi-dimensional monitoring time series data in the second window through trend item decomposition according to all multi-dimensional monitoring time series data of the solar panel, and use the result of weighted mean calculation of the average slope of the trend item decomposition results of each dimension in the second window by the fusion importance degree of the dimension as the overall window trend feature factor of the trend item; obtain the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window, and use the calculation result of the absolute value of the subtraction between the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window as the first trend feature factor;

[0106] In one embodiment, set the window short-term anomaly degree threshold for short-term anomaly window judgment. Preferably, set the window short-term anomaly degree threshold to , and mark the window with a short-term anomaly degree higher than the short-term anomaly degree threshold as a short-term anomaly window, so as to complete the short-term anomaly window judgment.

[0107] Obtain the short-term anomaly frequency change factor in the second window and the first trend feature factor of the second window, use the product of the short-term anomaly frequency change factor and the first trend feature factor as the second trend feature factor, and use the calculation result of subtracting the constant 1 from the second trend feature factor as the trend analysis factor of the second window.

[0108] Among them, the step of obtaining the second window and obtaining the short-term anomaly frequency change factor in the second window according to the short-term anomaly change information of the first window in the second window includes:

[0109] The second window contains at least three first windows. The number of short-term abnormal first windows in the second window is used as the short-term abnormal frequency change evaluation factor of the second window. According to the short-term abnormal frequency change evaluation factors of all second windows in the multi-dimensional operation monitoring time series data of the solar panel, a short-term abnormal frequency factor sequence is obtained. The short-term abnormal frequency factor sequence is subjected to least squares fitting to obtain the fitting result of the short-term abnormal frequency change evaluation factor. The normalized calculation result of the slope corresponding to the second window in the fitting result of the short-term abnormal frequency factor is used as the short-term abnormal frequency change factor in the second window.

[0110] In one embodiment, let the short-term abnormal frequency change factor in the th second time series data window be , and the first trend feature factor in the th second time series data window be . Then the calculation expression of the trend analysis factor of the th second time series data window is:

[0111] ;

[0112] where represents the trend analysis factor of the th second time series data window, represents the short-term abnormal frequency change factor in the th second time series data window, represents the first trend feature factor in the th second time series data window, and represents the normalization function, and 1 represents the constant 1.

[0113] Step S103: Determine the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window according to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel.

[0114] After obtaining the trend analysis factor of the multi-dimensional operation monitoring time series data window of the solar panel, the long-term trend change information appearing in the window can be optimally evaluated during the evaluation process of the distance metric evaluation factor for distance optimization metric through the trend analysis factor, so as to avoid the elimination of the long-term attenuation trend occurring in the operation process of the solar panel in the distance metric based on time series optimization, thereby ensuring accurate abnormal state evaluation of the multi-dimensional operation monitoring data of the solar panel.

[0115] As described above, the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window of the solar panel is determined according to the trend analysis factor and the short-term anomaly degree of the multi-dimensional operation monitoring time series data window, including:

[0116] Obtain the trend analysis factor of the second window and the short-term anomaly degree of the first window, and use the trend analysis factor of the second window as the trend analysis factor of all the first windows in the second window; use the calculation result of subtracting the constant 1 from the trend analysis factor of the first window as the first trend feature term of the first window; use the calculation result of multiplying the short-term anomaly degree of the first window by the first trend feature term of the first window as the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window.

[0117] In one embodiment, the calculation expression of the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window of the solar panel is:

[0118] ;

[0119] wherein, represents the distance metric evaluation factor of the th multi-dimensional operation monitoring time series data window of the solar panel, represents the short-term anomaly degree of the th window, represents the trend analysis factor of the th time series data window.

[0120] It should be noted that when using the short-term anomaly degree and the trend analysis factor of the multi-dimensional operation monitoring time series data window of the solar panel to evaluate the distance metric evaluation factor, the higher the short-term anomaly degree, the more likely it is that there are short-term anomaly situations in this window, such as short-term anomalies such as sudden changes in monitoring data caused by weather changes. Then the higher the distance metric evaluation factor. The higher the trend analysis factor of the time series data window, the higher the trend change situation in the current window, and the smaller the distance metric evaluation factor.

[0121] Step S104: Determine the distance metric between the multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window, and complete the clustering process based on the distance metric. Evaluate the abnormal state according to the result of the clustering process.

[0122] After obtaining the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window, use the distance metric evaluation factor of the window to optimize the distance metric in the clustering process of the multi-dimensional monitoring data of the solar panel, so as to obtain an accurate clustering result.

[0123] Among them, determining the temporal optimization distance metric between multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional anomaly monitoring time series data window, and completing the clustering process based on the distance metric includes:

[0124] Obtain two multi-dimensional monitoring data that need to perform distance metric and the first window corresponding to the two multi-dimensional monitoring data, take the mean of the distance metric evaluation factors of the first windows corresponding to the two multi-dimensional monitoring data as the temporal difference weight, and multiply the temporal difference weight by the DTW distance between the first windows corresponding to the two multi-dimensional monitoring data as the temporal difference metric; subtract the constant 1 from the temporal difference weight as the Euclidean distance difference weight, and multiply the Euclidean distance difference weight by the Euclidean distance between the two multi-dimensional monitoring data as the Euclidean distance difference metric;

[0125] Obtain the temporal difference metric and the Euclidean distance difference metric of the two multi-dimensional monitoring data and the two multi-dimensional monitoring time series data windows corresponding to the two multi-dimensional monitoring data, and add the temporal difference metric and the Euclidean distance difference metric as the temporal optimization distance metric;

[0126] In one embodiment, assume that the multi-dimensional operation monitoring data of the th solar panel is , and the multi-dimensional operation monitoring data of the th solar panel is , then the calculation expression of the temporal optimization distance metric between the two multi-dimensional operation monitoring data points is:

[0127] ;

[0128] Among them, represents the temporal optimization distance metric between the multi-dimensional operation monitoring data of the th solar panel and the multi-dimensional operation monitoring data of the th solar panel, represents the mean of the distance metric evaluation factors between the two multi-dimensional operation monitoring data points, that is, the temporal difference weight, represents the Euclidean distance between two multi-dimensional monitoring data points, represents the distance between the multi-dimensional operation monitoring time series data windows corresponding to two multi-dimensional monitoring data points, and 1 represents the constant 1.

[0129] After obtaining the temporal optimization distance metric between the two multi-dimensional operation monitoring data points, perform clustering according to the temporal optimization distance metric between the two multi-dimensional monitoring data to obtain the clustering result.

[0130] It should be noted that the value of clustering is determined by the elbow method. The method of determining the value of clustering by the elbow method is a well-known technology, so it will not be elaborated here.

[0131] After obtaining the clustering results of the solar panel operation monitoring data, the abnormal state assessment can be carried out according to the results of the clustering process.

[0132] Among them, the abnormal state assessment according to the results of the clustering process includes:

[0133] Obtain the results of the clustering process, and take the average Euclidean distance between the cluster center point of any cluster in the clustering results and the cluster center points of all other clusters as the first Euclidean distance of the cluster; take the average Euclidean distance between the data points within the cluster of any cluster in the clustering results and the cluster center point of the cluster as the second Euclidean distance of the cluster;

[0134] Obtain the first Euclidean distance and the second Euclidean distance of the cluster, and take the result of multiplying the first Euclidean distance and the second Euclidean distance and performing normalization processing as the abnormal degree of the cluster;

[0135] In an embodiment, the calculation expression of the abnormal degree of the th cluster is:

[0136] ;

[0137] Among them, represents the abnormal degree of the th cluster, represents the first Euclidean distance of the th cluster, represents the second Euclidean distance of the th cluster, represents the normalization function.

[0138] It should be noted that when using the first Euclidean distance and the second Euclidean distance of the cluster to evaluate the abnormal degree of the cluster, the larger the first Euclidean distance , it indicates that the distance between the th cluster and other clusters is farther, and the higher the abnormal degree. The larger the second Euclidean distance , it indicates that the data points in the th cluster are more dispersed, and the higher the abnormal degree.

[0139] Obtain the threshold of the anomaly degree of the cluster class. Compare the anomaly degree of the cluster class with the threshold of the anomaly degree of the cluster class. If the anomaly degree of the cluster class is greater than or equal to the threshold of the anomaly degree of the cluster class, determine that the cluster class is an abnormal state cluster class; if the anomaly degree of the cluster class is less than the threshold of the anomaly degree of the cluster class, determine that the cluster class is a normal state cluster class, so as to complete the evaluation of the abnormal state in the operation process of the solar cell.

[0140] In one embodiment, set the threshold of the anomaly degree of the cluster class to evaluate the abnormal state of the cluster class in the clustering result. Preferably, set the threshold of the anomaly degree of the cluster class to , and mark the cluster class with an anomaly degree greater than the threshold of the anomaly degree of the cluster class as an abnormal state in the operation process of the solar cell, thus completing the evaluation of the abnormal state in the operation process of the solar cell.

[0141] In summary, the embodiments of the present invention collect all the monitoring data during the operation of the solar panels in the power station, respectively obtain the multi-dimensional operation monitoring time series data of each solar panel, divide the multi-dimensional operation monitoring time series data of all solar panels into windows to obtain the multi-dimensional operation monitoring time series data windows of all solar panels; for any solar panel, obtain the multi-dimensional operation monitoring time series data window of the solar panel, and determine the importance degree of the dimension according to the change information of each dimension data within the multi-dimensional operation monitoring time series data window of the solar panel; determine the short-term abnormal degree of the window according to the importance degree of each dimension within the multi-dimensional operation monitoring time series data window of the solar panel; perform threshold judgment according to the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel, obtain the short-term abnormal window and determine the trend analysis factor according to the frequency change of the short-term abnormal window; determine the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window according to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel; determine the time series optimization distance metric between the multi-dimensional abnormal monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional abnormal monitoring time series data window, and complete the clustering process based on the time series optimization distance metric, and perform abnormal state evaluation according to the result of the clustering process. Among them, by introducing the time series difference metric of multi-dimensional monitoring time series data in the clustering process, the interference of fluctuations caused by short-term abnormalities to clustering is eliminated, thereby reducing misjudgments and improving the accuracy of abnormal state recognition. In the clustering process for abnormal state evaluation, the Euclidean distance and the time series difference metric are dynamically weighted by the distance metric evaluation factor, which can effectively handle environmental fluctuations caused by seasonal changes or weather factors and enhance the adaptability to these environmental factors. By dynamically adjusting the weight of the distance metric evaluation factor, short-term environmental fluctuations are avoided from being misjudged as abnormal states, thereby improving the adaptability of the abnormal evaluation method to environmental changes. In the above, by introducing the trend analysis factor, the distance metric evaluation factor is adaptively adjusted according to different operation data and abnormal patterns, so as to ensure that the abnormal detection method can dynamically adjust the detection strategy, distinguish short-term weather fluctuations from long-term decline trends, and improve the accuracy of abnormal state evaluation.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for evaluating abnormal states during the operation of a solar cell, characterized in that, The abnormal state evaluation method for the operation of a solar cell includes: Collect all the monitoring data during the operation of the solar panels in the power station, respectively obtain the multi-dimensional operation monitoring time series data of each solar panel, and perform window partitioning on the multi-dimensional operation monitoring time series data of all solar panels to obtain the multi-dimensional operation monitoring time series data windows of all solar panels; For any solar panel, obtain the multi-dimensional operation monitoring time series data window of the solar panel, and determine the importance degree of each dimension according to the change information of the data in each dimension within the multi-dimensional operation monitoring time series data window of the solar panel; determine the short-term abnormal degree of the window according to the importance degree of each dimension within the multi-dimensional operation monitoring time series data window of the solar panel; perform threshold judgment according to the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel, obtain the short-term abnormal window, and determine the trend analysis factor according to the frequency change of the short-term abnormal window; Determine the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window according to the trend analysis factor and the short-term abnormal degree of the multi-dimensional operation monitoring time series data window of the solar panel; Determine the time series optimized distance metric between the multi-dimensional operation monitoring time series data windows according to the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window, and complete the clustering process based on the time series optimized distance metric, and perform abnormal state evaluation according to the result of the clustering process.

2. The abnormal state evaluation method for the operation of a solar cell according to claim 1, wherein The determining the importance degree of each dimension according to the change information of the data in each dimension within the multi-dimensional operation monitoring time series data window of the solar panel includes: According to the distance between the monitoring data points in each dimension within the multi-dimensional operation monitoring time series data window of the solar panel, determine the connected path distance sequence of each monitoring data point in each monitoring dimension within the multi-dimensional monitoring time series data window; take the absolute value of the difference between the corresponding positions of the connected path distance sequences of the multi-dimensional monitoring data points corresponding to each timestamp in each dimension within the window and the connected path distance sequences in the same dimension at other timestamps as the first difference, multiply the first difference by the reciprocal of the path number in the path distance sequence as the weight, sum them up and divide by the number of paths in the path distance sequence to obtain the first path distance difference mean; perform normalization processing on the mean of the first path distance difference means between the connected path distance sequences of the multi-dimensional monitoring data points corresponding to each timestamp in each dimension within the window and the connected path distance sequences in the same dimension at all other timestamps, obtain the normalization processing result, and take the normalization processing result as the importance degree of each monitoring data point corresponding to each timestamp in each dimension within the window.

3. The abnormal state evaluation method for the operation of a solar cell according to claim 2, wherein The determining the connected path distance sequence of each monitoring data point in each monitoring dimension within the multi-dimensional monitoring time series data window according to the distance between the monitoring data points in each dimension within the multi-dimensional operation monitoring time series data window of the solar panel includes: For any monitored data point in any dimension within the window, take this monitored data point as the starting point, and perform path traversal in the window according to the minimum Euclidean distance. Take the path corresponding to the traversal result as the connected path; take the Euclidean distance between the two data points corresponding to each sub-path in the connected path as the path distance of the sub-path, and obtain the connected path distance sequence according to the path distances of each sub-path in the connected path.

4. A method for evaluating abnormal states during the operation of a solar cell according to claim 1, wherein The determination of the short-term anomaly degree of the window according to the importance of each dimension within the multi-dimensional operation monitoring time series data window of the solar panel includes: Obtain the importance of each multi-dimensional monitored data point corresponding to each timestamp in the window in each dimension. Take the average value of the importance of all multi-dimensional monitored data points in any dimension within any window as the window dimension importance of this dimension; take the average value of the window dimension importance of any dimension in all windows as the overall dimension importance of this dimension; divide the overall dimension importance of any dimension by the sum of the overall importance of all dimensions as the fusion importance of this dimension; take the average value of the monitored data in any dimension within the window as the first average value of the window; For any dimension within the window, starting from the second data point, take the result of multiplying the importance of the data point by the difference in the monitored values between this data point and the previous data point as the first difference value; according to the importance as the weight, take the result of weighted average calculation of all the first difference values in any dimension within the window as the second difference value of this dimension; take the square of the result of subtracting the monitored value corresponding to any timestamp in any dimension within the window from the first average value of the window as the first average difference; take the result of the square root of the weighted average of all the first average differences of all timestamps in any dimension within the window by the importance corresponding to each timestamp as the first fluctuation degree of this dimension; Take the result of the square root of the weighted average of the squares of the first fluctuation degrees of all dimensions within the window according to the fusion importance as the second fluctuation degree of the window; take the result of the square root of the weighted average of the second difference values of all dimensions within the window according to the fusion importance as the third difference value of the window; Obtain the second fluctuation degree and the third difference value of the window, and take the result of dividing the second fluctuation degree of the window by the third difference value of the window and performing normalization processing as the short-term anomaly degree of the window.

5. A method for evaluating abnormal states during the operation of a solar cell according to claim 1, characterized in that, The threshold judgment according to the short-term anomaly degree of the multi-dimensional operation monitoring time series data window of the solar panel, obtaining the short-term anomaly window and determining the trend analysis factor according to the frequency change of the short-term anomaly window includes: Obtain the short-term anomaly threshold, determine the short-term anomaly degree of the window according to the short-term anomaly threshold, and obtain all short-term anomaly windows; obtain the second window, and obtain the short-term anomaly frequency change factor in the second window according to the short-term anomaly change information of the first window in the second window; obtain the least squares fitting result of the multi-dimensional monitoring time series data in the second window through the least squares fitting result of the multi-dimensional monitoring time series data in the second window, and use the weighted mean calculation result of the average slope in the fitting results corresponding to each dimension in the monitoring time series data in the second window by the fusion importance degree of the dimension as the window trend feature factor in the second window; obtain the trend item decomposition result of all multi-dimensional monitoring time series data in the second window through trend item decomposition according to all multi-dimensional monitoring time series data of the solar panel, and use the weighted mean calculation result of the average slope of the trend item decomposition results of each dimension in the second window by the fusion importance degree of the dimension as the overall window trend feature factor of the trend item; obtain the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window, and use the absolute value calculation result of the subtraction between the window trend feature factor in the second window and the overall window trend feature factor of the trend item in the second window as the first trend feature factor; Obtain the short-term anomaly frequency change factor in the second window and the first trend feature factor of the second window, use the product of the short-term anomaly frequency change factor and the first trend feature factor as the second trend feature factor, and use the calculation result of subtracting the constant 1 from the second trend feature factor as the trend analysis factor of the second window.

6. The abnormal state evaluation method for the operation of a solar cell according to claim 5, wherein Determine the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window according to the trend analysis factor and the short-term anomaly degree of the multi-dimensional operation monitoring time series data window of the solar panel, including: Obtain the trend analysis factor of the second window and the short-term anomaly degree of the first window, and use the trend analysis factor of the second window as the trend analysis factor of all first windows in the second window; use the calculation result of subtracting the constant 1 from the trend analysis factor of the first window as the first trend feature term of the first window; use the calculation result of multiplying the short-term anomaly degree of the first window by the first trend feature term of the first window as the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window.

7. A method for evaluating abnormal states during the operation of a solar cell according to claim 1, characterized in that, Based on the distance metric evaluation factor of the multi-dimensional operation monitoring time series data window, determine the time series optimization distance metric between multi-dimensional operation monitoring time series data windows, and complete the clustering process based on the distance metric, including: Obtain two multi-dimensional monitoring data that need to be distance-measured and the first window corresponding to the two multi-dimensional monitoring data. Take the mean of the distance-measurement evaluation factors of the first window corresponding to the two multi-dimensional monitoring data as the time-series difference weight, and multiply the time-series difference weight by the DTW distance between the first windows corresponding to the two multi-dimensional monitoring data as the time-series difference measurement; Subtract the constant 1 from the time-series difference weight to obtain the Euclidean distance difference weight, and multiply the Euclidean distance difference weight by the Euclidean distance between the two multi-dimensional monitoring data as the Euclidean distance difference measurement; Obtain the time-series difference measurement and the Euclidean distance difference measurement of the two multi-dimensional monitoring data and the two multi-dimensional monitoring time-series data windows corresponding to the two multi-dimensional monitoring data. Add the time-series difference measurement and the Euclidean distance difference measurement as the time-series optimized distance measurement; And perform K-means clustering based on the time-series optimized distance measurement between the two multi-dimensional monitoring data to obtain the clustering result.

8. A method for evaluating abnormal states during the operation of a solar cell according to claim 1, characterized in that, The abnormal state evaluation according to the result of the clustering process includes: Obtain the result of the clustering process. Take the average Euclidean distance between the cluster center point of any cluster in the clustering result and the center points of all other clusters as the first Euclidean distance of the cluster; Take the average Euclidean distance between the data points within the cluster of any cluster in the clustering result and the cluster center point of the cluster as the second Euclidean distance of the cluster; Obtain the first Euclidean distance and the second Euclidean distance of the cluster. Take the result of multiplying the first Euclidean distance and the second Euclidean distance and performing normalization processing as the abnormal degree of the cluster; Obtain the cluster abnormal degree threshold. Compare the abnormal degree of the cluster with the cluster abnormal degree threshold. If the abnormal degree of the cluster is greater than or equal to the cluster abnormal degree threshold, determine that the cluster is an abnormal state cluster; If the abnormal degree of the cluster is less than the cluster abnormal degree threshold, determine that the cluster is a normal state cluster, thus completing the abnormal state evaluation during the operation of the solar cell.

9. A method for evaluating abnormal states during the operation of a solar cell according to claim 5, characterized in that, The obtaining of the second window and obtaining the short-term abnormal frequency change factor in the second window according to the short-term abnormal change information of the first window in the second window includes: The second window contains at least three first windows. Take the number of short-term abnormal first windows in the second window as the short-term abnormal frequency change evaluation factor of the second window. Obtain the short-term abnormal frequency factor sequence according to the short-term abnormal frequency change evaluation factors of all second windows in the multi-dimensional operation monitoring time-series data of the solar cell panel. Perform least squares fitting on the short-term abnormal frequency factor sequence to obtain the fitting result of the short-term abnormal frequency change evaluation factor. Take the normalized calculation result of the slope corresponding to the second window in the fitting result of the short-term abnormal frequency factor as the short-term abnormal frequency change factor in the second window.

10. A method for evaluating abnormal states during the operation of a solar cell according to claim 5, characterized in that, The window division of all multi-dimensional operation monitoring time-series data of the solar cell panel to obtain all multi-dimensional operation monitoring time-series data windows of the solar cell panel includes: Obtain the multi-dimensional operation monitoring time series data of the solar panel and the preset window length respectively; perform a traversal with a constant step size of 1 on the multi-dimensional operation monitoring time series data of the solar panel according to the preset window length, and use the subsequence included in each movement of the window with the preset window length as the multi-dimensional operation monitoring time series data window corresponding to the center point of the window.

Images