Photovoltaic power generation component fault automatic alarm method and system

Abstract

The present invention discloses a method and system for automatically alarming for photovoltaic power generation component faults. The method comprises: performing PCA cluster analysis on operating sample data from a photovoltaic power generation system to extract first and second principal component features; extracting characteristic values ​​for each sample, counting the total number of samples of the collected operating sample data and the number of extracted sample features, and inputting the combined data into a first fault prediction model to obtain a first fault prediction result; inputting the first and second principal component features into a convolutional neural network to extract spatial information, and into a GRU network to extract time series information, then calculating the total similarity and inputting it into a pre-established knowledge graph model to obtain a second fault prediction result; and synthesizing the two results to generate a graded fault alarm. By establishing a precise analysis model algorithm, the accuracy and efficiency of fault analysis are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a method and system for automatically alarming photovoltaic power generation component failures. Background Art

[0002] Photovoltaic power generation technology, a current research priority within renewable energy, has played a significant role in promoting social progress and large-scale energy conservation and emission reduction. In the field of data monitoring and real-time alarming for photovoltaic power generation systems, existing technologies mostly analyze various data points based on thresholds or incorporating standards within the field, generating alarms when problems are detected. For example, CN116111951A (published on May 12, 2023) discloses a retrieval unit that uses a data detection unit to obtain operating parameters of various components within the photovoltaic power generation unit to monitor the photovoltaic power generation system in real time and promptly identify the cause of any system failures. A first analysis unit initially determines the operating status of the photovoltaic power generation unit based on the voltage and frequency at the power consumption end. A second analysis unit further determines the cause of the photovoltaic power generation unit failure based on the average temperature of each photovoltaic panel, the battery charge level, and the battery consumption rate. An alarm unit then issues an alarm based on the fault cause output by the second analysis unit. However, there are some defects in this type of method, such as not fully considering the correlation between fault data, not establishing a model algorithm for precise analysis, not incorporating the spatiotemporal data in the photovoltaic power generation system into the fault analysis data, etc. These will make the data fault monitoring efficiency for the photovoltaic power generation system relatively low and cannot guarantee the accuracy of data analysis and fault monitoring. Summary of the Invention

[0003] To address the aforementioned shortcomings of the prior art, the present invention aims to provide a method and system for automatically alerting photovoltaic module faults. This method fully considers data correlation, establishes a precise analysis model algorithm, and incorporates temporal and spatial data from the photovoltaic system into fault analysis, thereby significantly improving the accuracy and efficiency of fault analysis.

[0004] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a method for automatically alarming a photovoltaic power generation component failure, the method comprising:

[0006] Step 1: Collect various operating sample data in the photovoltaic power generation system;

[0007] Step 2: Perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature; wherein the first principal component feature and the second principal component feature represent two different types of data after PCA cluster analysis;

[0008] Step 3: Based on the MapReduce parallel computing mode, extract the feature value of each sample in the first principal component feature and the second principal component feature in the Map phase, and count the total number of samples of the running sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce phase;

[0009] Step 4: superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result; the first fault prediction model is an SVM fault classification model;

[0010] Step 5: Input the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h; input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p;

[0011] Step 6: Calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is:

[0012]

[0013] Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA;

[0014] Step 7: Calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is:

[0015]

[0016] Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB;

[0017] Step 8: Calculate the total similarity sim(h,p). The formula is:

[0018]

[0019] in, 、 Indicates the weight value;

[0020] Step 9: Input the total similarity sim(h,p) into a pre-established knowledge graph model to obtain a second fault prediction result;

[0021] Step 10: Combining the first fault prediction result and the second fault prediction result, a fault classification alarm of the photovoltaic power generation component is performed.

[0022] Furthermore, the method further includes: Step 1, collecting various operation sample data in the photovoltaic power generation system, specifically including:

[0023] The monitoring device collects real-time on-site information of the photovoltaic power generation system, including the voltage, current, power, battery capacity, ambient temperature and surface temperature of the photovoltaic equipment during operation of the photovoltaic array.

[0024] Furthermore, the step 2 further includes:

[0025] Perform K-means cluster analysis on the collected running sample data to extract the first principal component features and the second principal component features; K-means cluster analysis uses the square error criterion, specifically:

[0026] ,

[0027] Among them, E is the square error of all sample data in the data set, q represents a given sample, Represents cluster D i The average value of , k is the number of cluster categories, i=1,2,...,k.

[0028] Furthermore, it also includes: when the PCA cluster analysis or the K-means cluster analysis performs information extraction processing, specifically taking the temperature-related data in the photovoltaic power generation system as the extracted first principal component features, and the remaining features as the extracted second principal component features.

[0029] Furthermore, the method further includes: in step 3, based on the MapReduce parallel operation mode, extracting each sample feature value in the first principal component feature and the second principal component feature in the Map phase, and counting the total number of samples of the running sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce phase, specifically including:

[0030] The data processing of the MapReduce parallel computing model is abstracted into a Map function and a Reduce function; the sample feature values ​​of the first principal component feature and the second principal component feature are read through the Map function, and the read data is transmitted to the Reduce function for integration and statistics.

[0031] Furthermore, the method further includes: superimposing and combining the characteristic value of each sample, the total number of samples, and the number of sample characteristics, specifically performing data splicing, including:

[0032] Arrange the characteristic values ​​of each sample in order, insert the total number of samples in front of the arranged data, and insert the sample characteristic number in the back of the arranged data to complete data splicing.

[0033] Furthermore, it also includes: building an SVM fault classification model, specifically including:

[0034] Extract some of the superimposed combined data as training data for the SVM model.

[0035] Design loss function to calculate the output value p of SVM model j Compared with the actual fault category value q in the training set j When the error is less than the set threshold, the training ends and the fault classification model SVM is obtained;

[0036] Among them, the loss function is:

[0037]

[0038] Where J is the number of training sets, j=1,2,...,J,w j It is an adjustable parameter.

[0039] Furthermore, it also includes:

[0040] The total similarity is used as the input of the knowledge graph analysis model, and the fault classification label is used as the output of the knowledge graph analysis model. The knowledge graph analysis model is trained to obtain a fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label.

[0041] Furthermore, the method further includes: Step 10, combining the first fault prediction result and the second fault prediction result to perform a fault classification alarm for the photovoltaic power generation component, specifically including:

[0042] comparing the first fault prediction result with a first threshold, and comparing the second fault prediction result with a second threshold;

[0043] If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is greater than or equal to a second threshold, a level one alarm is issued;

[0044] If the first fault prediction result is less than the first threshold value, and the second fault prediction result is less than the second threshold value, then there is no fault and no automatic alarm is issued;

[0045] If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is less than a second threshold, a level 2 alarm is issued;

[0046] If the first fault prediction result is less than a first threshold value and the second fault prediction result is greater than or equal to a second threshold value, a level 3 alarm is issued.

[0047] In a second aspect, the present invention further provides a photovoltaic power generation component fault automatic alarm system, the system comprising:

[0048] The acquisition module is used to collect various operating sample data in the photovoltaic power generation system;

[0049] A clustering module is used to perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature; wherein the first principal component feature and the second principal component feature represent two different types of data after the PCA cluster analysis;

[0050] A MapReduce unit module, configured to extract, in a Map phase, the first principal component feature and each sample feature value of the second principal component feature based on a MapReduce parallel operation mode, and to count the total number of samples of the collected running sample data and the number of extracted sample features in a Reduce phase;

[0051] A first prediction module is configured to superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result; the first fault prediction model is an SVM fault classification model;

[0052] a conversion module, configured to input the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h; input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p;

[0053] The first calculation module is used to calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is:

[0054]

[0055] Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA;

[0056] The second calculation module is used to calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is:

[0057]

[0058] Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB;

[0059] The third calculation module is used to calculate the total similarity sim(h,p), the formula is:

[0060]

[0061] in, 、 Indicates the weight value;

[0062] A second prediction module is used to input the total similarity sim(h, p) into a pre-established knowledge graph model to obtain a second fault prediction result;

[0063] The hierarchical alarm module is used to comprehensively analyze the first fault prediction result and the second fault prediction result to perform a hierarchical fault alarm for the photovoltaic power generation component.

[0064] Beneficial effects:

[0065] 1. The present invention collects various types of operating sample data in a photovoltaic power generation system through step 1; step 2, performs PCA cluster analysis on the collected operating sample data to extract the first principal component feature and the second principal component feature; step 3, based on the MapReduce parallel operation mode, extracts each sample feature value in the first principal component feature and the second principal component feature in the Map stage, and counts the total number of samples of the operating sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce stage; step 4, superimposes and combines each sample feature value, the total number of samples and the number of sample features, and inputs the superimposed data into a first fault prediction model to obtain a first fault prediction result; the first fault prediction model is an SVM fault classification model; step 5, inputs the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and converts the spatial information into a spatial feature vector h; inputs the first principal component feature and the second principal component feature into a GRU network to extract time series information , and convert the time series information into a time series vector p; step 6, calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA, step 7, calculate the similarity L(p, setB) between the time series vector p and the vector set setB, step 8, combine the similarity L(h, setA) and the similarity L(p, setB) to calculate the total similarity sim(h, p), step 9, input the total similarity sim(h, p) into a pre-established knowledge graph model to obtain a second fault prediction result; step 10, comprehensively integrate the first fault prediction result and the second fault prediction result to perform a fault classification alarm for the photovoltaic power generation component. The present invention obtains a first fault prediction result by implementing PCA-SVM data clustering and fault analysis algorithms in the MapReduce parallel computing mode; calculates similarities respectively by extracting the spatial feature vectors and temporal feature vectors of the first principal component feature and the second principal component feature in combination with the corresponding vector sets, and inputs the total similarity after weighted summation of the two similarities into the knowledge graph model to obtain a second fault prediction result; finally, the two fault prediction results are comprehensively analyzed to realize fault classification alarm of photovoltaic power generation components; the method fully considers the correlation between data, establishes a precise analysis model algorithm, and incorporates the spatiotemporal data in the photovoltaic power generation system into the fault analysis considerations, thereby greatly improving the accuracy and efficiency of fault analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 The figure is a flow chart of an automatic alarm method for photovoltaic power generation component failure. DETAILED DESCRIPTION

[0067] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only the portions relevant to the invention are shown in the accompanying drawings.

[0068] like Figure 1 As shown, this embodiment provides a method for automatically alarming a photovoltaic power generation component failure, the method comprising:

[0069] Step 1: Collect various operating sample data in the photovoltaic power generation system;

[0070] Specifically, the monitoring device collects real-time on-site information of the photovoltaic power generation system, including the voltage, current, power, battery capacity, ambient temperature and surface temperature of the photovoltaic equipment during operation of the photovoltaic array.

[0071] The detection device specifically includes sensors and converters connected to the photovoltaic power generation system, a signal conditioning unit, a data acquisition unit interface, a communication interface, a database, and a host computer. Specifically, the temperature sensor uses a highly precise platinum resistance temperature sensor. The signal conditioning unit connects to the data acquisition unit interface via an RS485 bus, which in turn connects to the database / host computer via a communication interface. The data acquisition unit interface stores collected data such as the voltage, current, power, battery capacity, ambient temperature, and surface temperature of the photovoltaic array during operation in the database.

[0072] Step 2: Perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature; wherein the first principal component feature and the second principal component feature represent two different types of data after PCA cluster analysis;

[0073] Specifically, K-means cluster analysis can be performed on the collected running sample data to extract the first principal component feature and the second principal component feature; the K-means cluster analysis adopts the square error criterion, specifically:

[0074] ,

[0075] Among them, E is the square error of all sample data in the data set, q represents a given sample, Represents cluster D i The average value of , k is the number of cluster categories, i=1,2,...,k.

[0076] Specifically, PCA cluster analysis and K-means cluster analysis achieve the same function, which is to perform cluster analysis on massive amounts of basic operating data.

[0077] Step 3: Based on the MapReduce parallel computing mode, extract the feature value of each sample in the first principal component feature and the second principal component feature in the Map phase, and count the total number of samples of the running sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce phase;

[0078] Specifically, the PCA-SVM clustering classification algorithm or the K-means-SVM clustering classification algorithm is used to implement fault type identification and fault analysis, and the parallel computing is implemented on the MapReduce framework to improve the computing speed.

[0079] Step 4: superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result; the first fault prediction model is an SVM fault classification model;

[0080] Data superposition and combination specifically involves data splicing, including:

[0081] Arrange the characteristic values ​​of each sample in order, insert the total number of samples in front of the arranged data, and insert the sample characteristic number in the back of the arranged data to complete data splicing, and input the spliced ​​data set into SVM for fault prediction.

[0082] By means of data splicing, the characteristic data are effectively combined and the correlation between the data is mined, thereby improving the accuracy of fault prediction.

[0083] Step 5: Input the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h; input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p;

[0084] Specifically, the convolutional neural network includes several convolutional layers and pooling layers; by utilizing the sensitivity of the convolutional neural network to spatial features and the sensitivity of the GRU network to temporal features, the temporal data characteristics and spatial data characteristics related to photovoltaic power generation system failures are mined.

[0085] Step 6: Calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is:

[0086]

[0087] Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA;

[0088] Step 7: Calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is:

[0089]

[0090] Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB;

[0091] Specifically, vector set setA and vector set setB are both feature sets that are pre-counted and stored in a database; through the above steps 6 and 7, the present invention innovatively calculates the similarity between the relevant spatial feature vectors and time series vectors in the photovoltaic power generation system and their corresponding vector sets, and then performs a weighted summation on the two similarities to achieve the effect of fully utilizing the spatiotemporal feature information for fault prediction, thereby greatly improving the fault prediction accuracy for the photovoltaic power generation system.

[0092] Step 8: Calculate the total similarity sim(h,p). The formula is:

[0093]

[0094] in, 、 Indicates the weight value;

[0095] Step 9: Input the total similarity sim(h,p) into a pre-established knowledge graph model to obtain a second fault prediction result;

[0096] Specifically, accurate fault monitoring is achieved by pre-building a fault-type knowledge graph analysis model that reflects the mapping relationship between total similarity and fault classification labels, and using the calculated total similarity sim(h,p) as the specific input feature of the knowledge graph model to output the corresponding fault classification label.

[0097] Step 10: Combining the first fault prediction result and the second fault prediction result, a fault classification alarm of the photovoltaic power generation component is performed.

[0098] In an optional embodiment, it also includes: when the PCA cluster analysis or the K-means cluster analysis performs information extraction processing, specifically using the temperature-related data in the photovoltaic power generation system as the extracted first principal component features, and the remaining features as the extracted second principal component features.

[0099] Temperature is the primary monitoring data for faults in photovoltaic power generation systems. To a certain extent, temperature plays a crucial role in fault prediction. Therefore, to enhance the importance of temperature, this paper specifically extracts temperature-related data as the first principal component feature, and the remaining features as the second principal component features.

[0100] In an optional embodiment, the method further includes: in step 3, based on the MapReduce parallel operation mode, extracting each sample feature value of the first principal component feature and the second principal component feature in the Map phase, and counting the total number of samples of the running sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce phase, specifically including:

[0101] The data processing of the MapReduce parallel computing model is abstracted into a Map function and a Reduce function; the sample feature values ​​of the first principal component feature and the second principal component feature are read through the Map function, and the read data is transmitted to the Reduce function for integration and statistics.

[0102] Specifically, MapReduce offers advantages such as automatic distributed parallel computing and high fault tolerance. Furthermore, corresponding weights can be assigned to the sample feature values ​​of the first and second principal component features. Because the first principal component feature is temperature-related data, a higher weight can be assigned to the sample feature values ​​of the first principal component feature, enabling timely and accurate fault prediction.

[0103] In an optional embodiment, the method further includes: constructing an SVM fault classification model, specifically including:

[0104] Extract some of the superimposed combined data as training data for the SVM model.

[0105] Design loss function to calculate the output value p of SVM model j Compared with the actual fault category value q in the training set j When the error is less than the set threshold, the training ends and the fault classification model SVM is obtained;

[0106] Among them, the loss function is:

[0107]

[0108] Where J is the number of training sets, j=1,2,...,J,w j It is an adjustable parameter.

[0109] In an optional embodiment, the method further includes:

[0110] The total similarity is used as the input of the knowledge graph analysis model, and the fault classification label is used as the output of the knowledge graph analysis model. The knowledge graph analysis model is trained to obtain a fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label.

[0111] Specifically, based on knowledge graph technology, by constructing a fault-type knowledge graph analysis model that reflects the mapping relationship between total similarity and fault classification labels, it is proposed for the first time to apply the knowledge graph model to fault prediction of photovoltaic power generation systems, and good prediction results have been achieved.

[0112] In an optional embodiment, the method further includes: Step 10, combining the first fault prediction result and the second fault prediction result to perform a fault classification alarm for the photovoltaic power generation component, specifically including:

[0113] comparing the first fault prediction result with a first threshold, and comparing the second fault prediction result with a second threshold;

[0114] If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is greater than or equal to a second threshold, a level one alarm is issued;

[0115] If the first fault prediction result is less than the first threshold value, and the second fault prediction result is less than the second threshold value, then there is no fault and no automatic alarm is issued;

[0116] If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is less than a second threshold, a level 2 alarm is issued;

[0117] If the first fault prediction result is less than a first threshold value and the second fault prediction result is greater than or equal to a second threshold value, a level 3 alarm is issued.

[0118] Specifically, by using a level-based alarm mechanism, it is possible to refine the handling measures for various types of faults in photovoltaic power generation units, thereby improving the efficiency and timeliness of fault handling.

[0119] Based on the same inventive concept, this embodiment provides an automatic fault alarm system for photovoltaic power generation components, the system comprising:

[0120] The acquisition module is used to collect various operating sample data in the photovoltaic power generation system;

[0121] A clustering module is used to perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature; wherein the first principal component feature and the second principal component feature represent two different types of data after the PCA cluster analysis;

[0122] A MapReduce unit module, configured to extract, in a Map phase, the first principal component feature and each sample feature value of the second principal component feature based on a MapReduce parallel operation mode, and to count the total number of samples of the collected running sample data and the number of extracted sample features in a Reduce phase;

[0123] A first prediction module is configured to superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result; the first fault prediction model is an SVM fault classification model;

[0124] a conversion module, configured to input the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h; input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p;

[0125] The first calculation module is used to calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is:

[0126]

[0127] Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA;

[0128] The second calculation module is used to calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is:

[0129]

[0130] Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB;

[0131] The third calculation module is used to calculate the total similarity sim(h,p), the formula is:

[0132]

[0133] in, 、 Indicates the weight value;

[0134] A second prediction module is used to input the total similarity sim(h, p) into a pre-established knowledge graph model to obtain a second fault prediction result;

[0135] The hierarchical alarm module is used to comprehensively analyze the first fault prediction result and the second fault prediction result to perform a hierarchical fault alarm for the photovoltaic power generation component.

Claims

1. A photovoltaic power generation component fault automatic alarm method, characterized in that: The method includes: Step 1: Collect various operating sample data in the photovoltaic power generation system; Step 2: Perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature, wherein the first principal component feature and the second principal component feature represent two different types of data after PCA cluster analysis; Step 3: Based on the MapReduce parallel computing mode, extract the feature value of each sample in the first principal component feature and the second principal component feature in the Map phase, and count the total number of samples of the running sample data collected in step 1 and the number of sample features extracted in step 3 in the Reduce phase; Step 4: superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result, where the first fault prediction model is an SVM fault classification model; Step 5: Input the first principal component feature and the second principal component feature in step 2 into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h; input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p; Step 6: Calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is: ， Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA; Step 7: Calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is: ， Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB; Step 8: Calculate the total similarity sim(h,p). The formula is: ， in, 、 Indicates the weight value; Step 9: Input the total similarity sim(h, p) into the pre-established knowledge graph model to obtain a second fault prediction result. A fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label is pre-built, and the calculated total similarity sim(h, p) is used as a specific input feature of the knowledge graph model to output the corresponding fault classification label. Step 10: Based on the first fault prediction result and the second fault prediction result, a fault classification alarm is issued for the photovoltaic power generation component; It also includes: taking the total similarity as the input of the knowledge graph analysis model, taking the fault classification label as the output of the knowledge graph analysis model, and training the knowledge graph analysis model to obtain a fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label.

2. The method according to claim 1, characterized in that The step 1 is to collect various operating sample data in the photovoltaic power generation system, specifically including: The monitoring device collects real-time on-site information of the photovoltaic power generation system, including the voltage, current, power, battery capacity, ambient temperature and surface temperature of the photovoltaic equipment during operation of the photovoltaic array.

3. The method according to claim 1, characterized in that The step 2 further comprises: Perform K-means cluster analysis on the collected running sample data to extract the first principal component features and the second principal component features. The K-means cluster analysis adopts the square error criterion, specifically: ， Among them, E is the square error of all sample data in the data set, q represents a given sample, Represents cluster D i The average value of , k is the number of cluster categories, i=1,2,...,k.

4. The method according to claim 3, characterized in that Also includes: When performing information extraction processing, the PCA cluster analysis or the K-means cluster analysis specifically uses the temperature-related data in the photovoltaic power generation system as the extracted first principal component features, and the remaining features as the extracted second principal component features.

5. The method according to claim 1, wherein Also includes: The step 3, based on the MapReduce parallel operation mode, extracts each sample feature value in the first principal component feature and the second principal component feature in the Map phase, and counts the total number of samples of the running sample data collected in the step 1 and the number of sample features extracted in the step 3 in the Reduce phase, specifically including: The data processing of the MapReduce parallel computing model is abstracted into a Map function and a Reduce function. The sample feature values ​​of the first principal component feature and the second principal component feature are read through the Map function, and the read data is transmitted to the Reduce function for integration and statistics.

6. The method according to claim 1, wherein The superimposing and combining the characteristic value of each sample, the total number of samples and the number of sample characteristics, specifically performing data splicing, includes: Arrange the characteristic values ​​of each sample in order, insert the total number of samples in front of the arranged data, and insert the sample characteristic number in the back of the arranged data to complete data splicing.

7. The method according to claim 6, characterized in that Construct an SVM fault classification model, including: Extract part of the superimposed and combined data as training data for the SVM model; Design loss function to calculate the output value p of SVM model j Compared with the actual fault category value q in the training set j When the error is less than the set threshold, the training ends and the fault classification model SVM is obtained; Among them, the loss function is: ， Where J is the number of training sets, j=1,2,...,J,w j It is an adjustable parameter.

8. The method according to claim 1, characterized in that The step 10, combining the first fault prediction result and the second fault prediction result to perform a fault classification alarm for the photovoltaic power generation component, specifically includes: comparing the first fault prediction result with a first threshold, and comparing the second fault prediction result with a second threshold; If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is greater than or equal to a second threshold, a level one alarm is issued; If the first fault prediction result is less than the first threshold value, and the second fault prediction result is less than the second threshold value, then there is no fault and no automatic alarm is issued; If the first fault prediction result is greater than or equal to a first threshold, and the second fault prediction result is less than a second threshold, a level 2 alarm is issued; If the first fault prediction result is less than a first threshold value and the second fault prediction result is greater than or equal to a second threshold value, a level 3 alarm is issued.

9. An automatic alarm system for photovoltaic power generation component failure, characterized in that: The system includes: The acquisition module is used to collect various operating sample data in the photovoltaic power generation system; A clustering module is used to perform PCA cluster analysis on the collected running sample data to extract the first principal component feature and the second principal component feature, wherein the first principal component feature and the second principal component feature represent two different types of data after the PCA cluster analysis; A MapReduce unit module, configured to extract, in a Map phase, the first principal component feature and each sample feature value of the second principal component feature based on a MapReduce parallel operation mode, and to count the total number of samples of the collected running sample data and the number of extracted sample features in a Reduce phase; A first prediction module is configured to superimpose and combine the characteristic value of each sample, the total number of samples, and the number of sample characteristics, and input the superimposed data into a first fault prediction model to obtain a first fault prediction result, wherein the first fault prediction model is an SVM fault classification model; a conversion module, configured to input the first principal component feature and the second principal component feature into a convolutional neural network to extract spatial information, and convert the spatial information into a spatial feature vector h, input the first principal component feature and the second principal component feature into a GRU network to extract time series information, and convert the time series information into a time series vector p; The first calculation module is used to calculate the similarity L(h, setA) between the spatial feature vector h and the vector set setA. The formula is: ， Among them, the vector set setA represents the spatial feature set under normal operation of the photovoltaic power generation system, and a is a vector in the vector set setA; The second calculation module is used to calculate the similarity L(p, setB) between the time series vector p and the vector set setB. The formula is: ， Among them, the vector set setB represents the time feature set under normal operation of the photovoltaic power generation system, and b is a vector in the vector set setB; The third calculation module is used to calculate the total similarity sim(h,p), the formula is: ， in, 、 Indicates the weight value; The second prediction module is used to input the total similarity sim(h, p) into a pre-established knowledge graph model to obtain a second fault prediction result, by pre-building a fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label, and using the calculated total similarity sim(h, p) as a specific input feature of the knowledge graph model to output the corresponding fault classification label; A hierarchical alarm module for synthesizing the first fault prediction result and the second fault prediction result to perform a hierarchical fault alarm for the photovoltaic power generation component; It also includes: taking the total similarity as the input of the knowledge graph analysis model, taking the fault classification label as the output of the knowledge graph analysis model, and training the knowledge graph analysis model to obtain a fault-type knowledge graph analysis model that reflects the mapping relationship between the total similarity and the fault classification label.

Images