A photovoltaic power station fault analysis and early warning method and system based on big data

Abstract

The present invention relates to the field of photovoltaic power station management technology, and in particular to a photovoltaic power station fault analysis and early warning method and system based on big data. The method comprises the following steps: obtaining photovoltaic power station operation data and corresponding photovoltaic power station early warning data; extracting equipment operation characteristics and environmental operation characteristics according to the photovoltaic power station operation data, and obtaining equipment operation characteristic data and environmental operation characteristic data respectively; constructing a recursive tree according to the equipment operation characteristic data and the environmental operation characteristic data, and obtaining equipment operation decision tree data and environmental operation decision tree data respectively; constructing a decision forest according to the equipment operation decision tree data, the environmental operation decision tree data and the corresponding photovoltaic power station early warning data, and obtaining a photovoltaic analysis model of the photovoltaic power station, so as to perform photovoltaic power station fault early warning operations. The present invention improves the operation efficiency and safety of photovoltaic power stations, reduces the occurrence rate of failures and maintenance costs, and promotes the stable development of clean energy.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station management, and in particular to a photovoltaic power station fault analysis and early warning method and system based on big data. Background Art

[0002] With the continuous increase in global energy demand and the improvement of environmental protection awareness, the development and utilization of clean energy has become an important direction for the development of today's society. As an important part of renewable energy, photovoltaic power stations have been widely used due to their clean, efficient and low-carbon characteristics. However, the operation and maintenance of photovoltaic power stations face many challenges, such as equipment failures, environmental changes, etc. These factors will affect the power generation efficiency and safety of photovoltaic power stations. Therefore, it is of great practical significance to establish an efficient photovoltaic power station fault analysis and early warning system.

[0003] Traditional photovoltaic power plant monitoring and fault detection methods mainly rely on manual inspection and conventional monitoring systems, which have many shortcomings. First, manual inspection is inefficient and difficult to detect potential faults in a timely manner; second, conventional monitoring systems can usually only detect obvious fault signals, but it is difficult to effectively identify and warn of early fault signs or complex multi-factor faults. In addition, photovoltaic power plants are usually distributed over a wide area, with a large number of equipment and complex and changeable environmental conditions, which further increases the difficulty of fault detection and maintenance. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a photovoltaic power station fault analysis and early warning method and system based on big data to solve at least one of the above technical problems.

[0005] The present application provides a photovoltaic power station fault analysis and early warning method based on big data, which is applied to a big data platform. The method includes:

[0006] S1. Obtaining photovoltaic power station operation data and corresponding photovoltaic power station early warning data;

[0007] S2. Extract equipment operation characteristics and environmental operation characteristics according to the photovoltaic power station operation data to obtain equipment operation characteristic data and environmental operation characteristic data respectively;

[0008] S3, constructing a recursive tree according to the equipment operation characteristic data and the environment operation characteristic data, and obtaining equipment operation decision tree data and environment operation decision tree data respectively;

[0009] S4. Construct a decision forest based on the equipment operation decision tree data, the environment operation decision tree data and the corresponding photovoltaic power station early warning data to obtain a photovoltaic analysis model of the photovoltaic power station to perform photovoltaic power station fault early warning operations.

[0010] In the present invention, by acquiring and analyzing the operation data and early warning data of the photovoltaic power station, the method can monitor the operation status and environmental conditions of the equipment in real time. Once an abnormality or potential fault risk is detected, the system can quickly issue an early warning to help operators take timely maintenance and adjustment measures. The extraction of equipment operation characteristics and environmental operation characteristics covers multiple dimensions of the system, such as voltage, current, temperature, humidity, wind speed, etc. These characteristics provide comprehensive data support for fault analysis. By constructing recursive trees and decision forest models, the method can effectively process high-dimensional and complex operation data. The recursive tree can reveal the complex relationship between features, while the decision forest improves the generalization ability and prediction accuracy of the model by integrating multiple decision trees. The method is based on a big data platform and realizes real-time processing and analysis of large-scale data. The system can process massive data from multiple photovoltaic power stations and has good scalability.

[0011] Optionally, S1 includes:

[0012] Obtain original photovoltaic power station operation data through the Internet of Things protocol and cloud interface;

[0013] Clean the original photovoltaic power station operation data and convert its format to obtain the photovoltaic power station operation data;

[0014] Obtain photovoltaic power station early warning data corresponding to photovoltaic power station operation data.

[0015] In the present invention, the operation data of the photovoltaic power station can be obtained in real time through the Internet of Things protocol and the cloud interface, ensuring the timeliness of the data and enabling the system to continuously monitor the equipment status and environmental conditions. A high-quality data foundation helps to improve the accuracy and reliability of the model, reduce false positives and omissions caused by data problems, and improve the effectiveness of fault analysis and early warning. Format conversion converts data from different sensors or communication devices and equipment into a unified format so that the data can be seamlessly integrated and analyzed on a big data platform. Combined with early warning data, abnormal patterns in operating data can be better understood, which helps to identify and predict potential faults. Early warning data also provides labeled data for model training and verification, improving the accuracy of the prediction model.

[0016] Optionally, S2 includes:

[0017] Extracting power station operation data and environmental operation data based on the photovoltaic power station operation data to obtain power station operation data and environmental operation data;

[0018] Extracting features from power plant operation data to obtain preliminary equipment operation feature data, and extracting features from environmental operation data to obtain preliminary environmental operation feature data;

[0019] Perform multi-scale feature conversion on preliminary equipment operation characteristic data to obtain equipment operation characteristic data;

[0020] The extreme value features are extracted from the preliminary environmental operation characteristic data to obtain the environmental operation characteristic data.

[0021] In the present invention, by dividing the photovoltaic power station operation data into power station operation data and environmental operation data, the operation status of the equipment and the changes in environmental conditions can be analyzed respectively, which helps to more accurately understand the impact of various types of data on the performance of the photovoltaic power station. Feature extraction of the power station operation data can extract key information reflecting the health status of the equipment, such as voltage, current, temperature, etc. Accurate feature extraction can better reflect the actual operating status of the equipment, help to timely discover abnormal conditions or potential faults, and improve the sensitivity of fault detection. Features such as temperature, humidity, wind speed, etc. are extracted from the environmental operation data to obtain preliminary environmental operation feature data. Analyzing these features can help understand the impact of the external environment on the operation of the photovoltaic power station. Multi-scale feature conversion is performed on the preliminary equipment operation feature data, and multi-level information in the data is captured by analyzing the feature changes on different time scales. Extreme value features are extracted from the preliminary environmental operation feature data to identify extreme events or abnormal conditions in the environmental data.

[0022] Optionally, the multi-scale feature transformation includes:

[0023] Perform initial feature decomposition based on preliminary equipment operation feature data to obtain decomposed feature data;

[0024] Perform multi-scale time series segmentation on the decomposed feature data to obtain multi-scale time series segmentation data;

[0025] Perform multi-scale feature extraction on multi-scale time series segmentation data to obtain multi-scale feature data;

[0026] Perform frequency domain feature analysis on the multi-scale feature data to obtain frequency domain feature data;

[0027] Perform feature combination according to the multi-scale feature data to obtain combined feature data;

[0028] Multi-scale feature aggregation is performed based on the frequency domain feature data and the combined feature data to obtain the equipment operation feature data.

[0029] In the present invention, the preliminary equipment operation feature data is decomposed to extract features of different scales and dimensions. This helps to identify complex structures and potential patterns in the data. The decomposed feature data is segmented according to different time scales so that short-term and long-term change features can be analyzed separately. Features are extracted at each time scale, including average value, fluctuation amplitude, etc., and multi-scale feature data is extracted to enrich the feature set. The multi-scale feature data is converted to the frequency domain, its frequency components are analyzed, and periodic features and main frequency components are identified. Multi-scale and frequency domain features are combined to generate more complex combined feature data to capture the interaction between different features. Multi-scale feature aggregation provides a complete feature perspective that covers all aspects of equipment operation. Comprehensive analysis can improve the accuracy and comprehensiveness of fault detection, enabling the system to better cope with complex operating environments.

[0030] Optionally, the extreme value feature extraction includes:

[0031] Conduct extreme frequency analysis on preliminary environmental operation characteristic data to obtain extreme frequency data;

[0032] Constructing extreme value event sequence for extreme value frequency data and preliminary environmental operation characteristic data to obtain extreme value event sequence data;

[0033] Calculate the extreme value impact factor according to the extreme value event sequence data to obtain the extreme value impact factor data;

[0034] Perform extreme value clustering processing based on extreme value impact factor data and extreme value event sequence data to obtain extreme value category data;

[0035] Extreme value trend processing is performed based on extreme value category data and preliminary environmental operation characteristic data to obtain environmental operation characteristic data.

[0036] In the present invention, the frequency and time distribution of extreme events are identified by analyzing the extreme value frequency in the preliminary environmental operation characteristic data. Based on the extreme value frequency data and the preliminary environmental operation characteristic data, an extreme value event sequence is constructed to record detailed information such as the time, duration, and amplitude of the extreme value. Using the extreme value event sequence data, the impact factor of each extreme value event is calculated to quantify its impact on the operation of the photovoltaic power station. Extreme value clustering processing helps to identify and classify different types of extreme environmental events, so that differentiated management and response strategies can be adopted for different categories of extreme situations. Extreme value trend processing helps to identify the long-term development trends and possible future changes of extreme events. By analyzing historical data and trends, the system can predict extreme environmental conditions that may occur in the future, providing a basis for advance preparation and risk prevention.

[0037] Optionally, S3 includes:

[0038] Initialize the recursive tree according to the equipment operation characteristic data and the environment operation characteristic data to obtain equipment operation decision tree initialization data and environment operation decision tree initialization data;

[0039] Perform node splitting standard selection on the equipment operation characteristic data and the environment operation characteristic data to obtain equipment operation splitting standard data and environment operation splitting standard data respectively;

[0040] Perform node splitting according to the equipment operation decision tree initialization data, equipment operation splitting standard data, equipment operation characteristic data and corresponding photovoltaic power station early warning data to obtain preliminary equipment operation decision tree data, and perform node splitting according to the environment operation decision tree initialization data, environment operation characteristic data, environment operation splitting standard data and corresponding photovoltaic power station early warning data to obtain preliminary environment operation decision tree data;

[0041] Performing tree structure optimization on preliminary equipment operation decision tree data and preliminary equipment operation decision tree data to obtain optimized equipment operation decision tree data and optimized equipment operation decision tree data respectively;

[0042] The optimized equipment operation decision tree data and the optimized equipment operation decision tree data are post-pruned to obtain equipment operation decision tree data and environment operation decision tree data, respectively.

[0043] The recursive tree initialization in the present invention provides a basic classification and regression framework to ensure that the model has a reasonable structure from the beginning. Through proper initialization, the training time can be reduced and the initial performance of the model can be improved. Select appropriate node splitting criteria (such as information gain, Gini index, etc.) for the equipment and environmental operation feature data to ensure that the splitting process can maximize purity or minimize error. Node splitting is the core of the formation of the decision tree model, which can effectively separate different categories or regression values ​​in the data set. Through multiple splits, the model can gradually refine the classification and regression of the data and capture more details and patterns. The structure of the initially generated decision tree is optimized, including adjusting the depth and branch structure of the tree to improve the performance of the tree. The optimized decision tree is post-pruned to remove unnecessary branches and nodes, thereby simplifying the model.

[0044] Optionally, the recursive tree initialization includes:

[0045] Sorting the importance of the features according to the equipment operation feature data and the environment operation feature data, and obtaining equipment operation feature ranking data and environment operation feature ranking data respectively;

[0046] Performing feature interaction effect processing on the equipment operation feature data and the environment operation feature data to obtain equipment operation feature interaction effect data and environment operation feature interaction effect data respectively;

[0047] Performing initial node screening on the equipment operation feature sorting data and the equipment operation feature interaction effect data to obtain the equipment operation feature initial node data, and performing initial node screening on the environment operation feature sorting data and the environment operation feature interaction effect data to obtain the environment operation feature initial node data;

[0048] Splitting candidate sets are generated for the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics, and obtaining equipment operation characteristic classification candidate set data and environment operation characteristic classification candidate set data respectively;

[0049] An initial tree depth is set according to the equipment operation feature classification candidate set data and the environment operation feature classification candidate set data to obtain equipment operation feature initial tree depth data and environment operation feature initial tree depth data;

[0050] Initializing the recursive tree according to the initial node data of the equipment operation characteristics, the classification candidate set data of the equipment operation characteristics, and the initial tree depth data of the equipment operation characteristics to obtain the initialization data of the equipment operation decision tree;

[0051] Recursive tree initialization is performed according to the initial node data of the environment operation characteristics, the classification candidate set data of the environment operation characteristics and the initial tree depth data of the environment operation characteristics to obtain the initial data of the environment operation decision tree.

[0052] The feature importance ranking in the present invention helps to identify key features, so that the model can focus more on the features that contribute most to the classification or regression task, and improve the efficiency and accuracy of the model. Feature interaction effect processing can identify and utilize the complex relationships between different features, help the model better understand the deep patterns and associations in the data, and improve the prediction performance. Initial node screening ensures that the recursive tree starts to split from the most important features and key interaction effects, lays a good foundation for the model and improves the initial performance of the model. Split candidate set generation provides a variety of split path options, allowing the model to find the optimal splitting solution, maximize information gain or minimize error, and improve the decision-making ability of the model. Reasonable tree depth setting can prevent model overfitting, maintain the simplicity and generalization ability of the model, and ensure that the model has sufficient complexity to capture patterns in the data. Recursive tree initialization constructs a decision tree with a good initial structure by integrating key features and the best splitting solution, providing a solid foundation for subsequent model training and optimization.

[0053] Optionally, the device operation feature classification candidate set data includes first device operation feature classification candidate set data and second device operation feature classification candidate set data, the environment operation feature classification candidate set data includes first environment operation feature classification candidate set data and second environment operation feature classification candidate set data, and the split candidate set generation includes:

[0054] Clustering the equipment operation feature initial node data and the environment operation feature initial node data to obtain first equipment operation feature classification candidate set data and first environment operation feature classification candidate set data;

[0055] A chi-square test is performed on the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics to obtain the second equipment operation characteristic classification candidate set data and the second environment operation characteristic classification candidate set data.

[0056] Clustering and chi-square test in the present invention provide two different feature analysis perspectives. Clustering analyzes from the perspective of data distribution and similarity, while chi-square test analyzes from the perspective of the correlation between features and target variables. Combining these two methods, effective splitting points can be identified more comprehensively. By using clustering, natural categories in the data can be identified, reducing the risk of overfitting the model to specific training data. The chi-square test ensures that the model focuses on the features most relevant to the target variable, thereby improving the generalization ability of the model. The split candidate sets generated by multiple methods provide a wealth of choices, providing more possibilities for the model to find the best splitting point, and the diversified splitting point selection strategies help improve the classification and regression performance of the model.

[0057] Optionally, S4 includes:

[0058] Perform multiple decision tree integration based on equipment operation decision tree data and environment operation decision tree data to obtain decision forest structure data;

[0059] Calculate the decision tree weights on the decision forest structure data to obtain the decision tree weight data;

[0060] Diversity optimization is performed based on decision tree weight data and decision forest structure data to obtain decision forest diversity data;

[0061] The voting mechanism is constructed according to the decision forest diversity data to obtain the decision forest voting mechanism data;

[0062] Based on the decision forest diversity data, decision forest voting mechanism data and the corresponding photovoltaic power station early warning data, the model is integrated to obtain the photovoltaic analysis model of the photovoltaic power station for photovoltaic power station fault early warning operations.

[0063] The integration of multiple decision trees in the present invention can average the prediction errors of individual trees and significantly improve the accuracy and stability of the model. By assigning weights, decision trees with better performance can be given priority to improve the overall prediction accuracy. Diversity optimization can increase the robustness of the model and combat the influence of noise and abnormal data. The voting mechanism integrates the opinions of multiple models to reduce the impact of a single erroneous prediction on the final result. Through weighted voting, the voting results can be adjusted according to the accuracy and weight of each tree to improve the overall prediction accuracy. Model fusion combines the advantages of multiple decision trees and early warning data to provide more comprehensive fault analysis and prediction for photovoltaic power plants.

[0064] Optionally, the present application further provides a photovoltaic power station fault analysis and early warning system based on big data, which is used to execute the photovoltaic power station fault analysis and early warning method based on big data as described above, and the photovoltaic power station fault analysis and early warning system based on big data includes:

[0065] Photovoltaic power station data acquisition module, used to obtain photovoltaic power station operation data and corresponding photovoltaic power station early warning data;

[0066] The photovoltaic power station data feature extraction module is used to extract equipment operation features and environmental operation features according to the photovoltaic power station operation data, and obtain equipment operation feature data and environmental operation feature data respectively;

[0067] A photovoltaic power station recursive tree construction module is used to construct a recursive tree according to the equipment operation characteristic data and the environment operation characteristic data, and obtain equipment operation decision tree data and environment operation decision tree data respectively;

[0068] The photovoltaic analysis model construction module of the photovoltaic power station is used to construct a decision forest based on the equipment operation decision tree data, the environmental operation decision tree data and the corresponding photovoltaic power station early warning data to obtain the photovoltaic analysis model of the photovoltaic power station for photovoltaic power station fault early warning operations.

[0069] The objects of the present invention are:

[0070] 1. Through the Internet of Things protocol and cloud interface, the operation data and early warning data of the photovoltaic power station can be obtained in real time, ensuring the system's continuous perception of the latest status and avoiding information lag.

[0071] 2. By extracting equipment operation characteristics and environmental operation characteristics data, the operation status of the photovoltaic power station can be analyzed from multiple dimensions and scales, and the potential complex relationships and influencing factors in the system can be captured. In the recursive tree initialization process, the key initial nodes are screened out by combining feature importance sorting and feature interaction effect analysis, which can more accurately capture key features and the complex relationships between them, and improve the model's explanatory power and prediction accuracy.

[0072] 3. By integrating multiple devices and environments to run decision trees, a decision forest with high diversity is constructed. Diversity optimization further ensures the robustness of the model and can effectively deal with different types of abnormal situations and data noise. The construction of the voting mechanism combines the weight distribution of the decision tree, so that the model can make more accurate predictions while considering multiple trees comprehensively. This mechanism improves the overall decision quality of the model and reduces the impact of single model bias.

[0073] 4. By combining the decision tree data of equipment and environment and early warning data, the photovoltaic analysis model formed can accurately predict and warn of faults. The model integrates multiple data sources and prediction results, providing a more comprehensive analysis perspective, which can timely detect potential faults and issue early warning signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0075] Figure 1 A flowchart showing a photovoltaic power station fault analysis and early warning method based on big data according to an embodiment of the present invention is provided;

[0076] Figure 2 A flowchart showing a method for collecting data from a photovoltaic power station according to an embodiment of the present invention is provided;

[0077] Figure 3 A flowchart showing a method for extracting features from photovoltaic power station data according to an embodiment of the present invention is shown;

[0078] Figure 4 A flowchart showing a method for constructing a recursive tree of a photovoltaic power station according to an embodiment of the present invention is shown;

[0079] Figure 5 A flowchart showing a method for constructing a photovoltaic analysis model for a photovoltaic power station according to an embodiment of the present invention is shown;

[0080] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0081] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0082] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0083] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0084] See also Figures 1 to 5 The present application provides a photovoltaic power station fault analysis and early warning method based on big data, which is applied to a big data platform. The method includes:

[0085] S1. Obtaining photovoltaic power station operation data and corresponding photovoltaic power station early warning data;

[0086] Specifically, install sensors on photovoltaic modules, inverters and other equipment in photovoltaic power stations or use communication devices to connect with the equipment to monitor the operating status and environmental parameters of the equipment in real time, such as current, voltage, temperature, etc. Deploy meteorological stations or environmental monitoring equipment to obtain environmental data such as solar radiation, wind speed, humidity, etc.

[0087] Use wired communication (such as optical fiber) or wireless communication (such as 4G / 5G, Wi-Fi) technology to transmit real-time data to the data center or cloud big data platform. Use distributed databases (such as Hadoop HDFS) to store the collected massive data to ensure the reliability and scalability of data storage.

[0088] Clean the collected data to remove noise and outliers. Use smoothing filtering and other methods to handle mutation points and missing values ​​in the data. Standardize and normalize the data to ensure the consistency of data from different sources.

[0089] Based on historical fault data and expert experience, establish an early warning model library that includes the characteristics and early warning thresholds of various faults. Collect and store early warning data, including historical warning records and current warning threshold information.

[0090] S2. Extract equipment operation characteristics and environmental operation characteristics according to the photovoltaic power station operation data to obtain equipment operation characteristic data and environmental operation characteristic data respectively;

[0091] Specifically, the acquired raw data is classified into equipment operation data and environmental data. Data mining technology is used to perform time series analysis on the data to extract valid data within a specific time period.

[0092] Extract the voltage, current, power, inverter efficiency and other characteristics of photovoltaic modules. Calculate the statistical characteristics of the equipment such as the average value, standard deviation, maximum value, minimum value, etc.

[0093] Extract environmental characteristics such as solar radiation intensity, temperature, humidity, wind speed, etc. Use time series analysis to extract the changing trends of environmental characteristics.

[0094] The extracted eigenvalues ​​are combined into eigenvectors to form equipment operation feature data and environment operation feature data. The feature data is further processed by dimensionality reduction, such as using principal component analysis (PCA) to reduce redundant features.

[0095] S3, constructing a recursive tree according to the equipment operation characteristic data and the environment operation characteristic data, and obtaining equipment operation decision tree data and environment operation decision tree data respectively;

[0096] Specifically, the extracted feature data set is divided into a training set and a test set, usually in a ratio of 8:2 or 7:3. Select a recursive tree algorithm, such as a decision tree, random forest, or gradient boosting tree. For complex multidimensional feature data, use random forest or gradient boosting tree to improve the robustness of the model.

[0097] The recursive tree model is trained using the equipment operation feature data to learn the relationship between each feature and the equipment operation status, and to generate equipment operation decision tree data.

[0098] The recursive tree model is trained using environmental operation feature data to analyze the impact of environmental features on equipment operation and generate environmental operation decision tree data.

[0099] Use the test set to evaluate the predictive accuracy of the recursive tree model using metrics such as precision, recall, and F1 value. Adjust model parameters such as tree depth, splitting criteria, etc. to optimize model performance.

[0100] S4. Construct a decision forest based on the equipment operation decision tree data, the environment operation decision tree data and the corresponding photovoltaic power station early warning data to obtain a photovoltaic analysis model of the photovoltaic power station to perform photovoltaic power station fault early warning operations.

[0101] Specifically, the decision forest model is constructed by combining the equipment operation decision tree and the environment operation decision tree. The model is validated using out-of-bag data to evaluate the generalization ability of the model.

[0102] The photovoltaic analysis model of the photovoltaic power station is trained using historical warning data and the decision forest model. The model input includes real-time equipment operation characteristic data and environmental operation characteristic data, and outputs fault warning information.

[0103] According to the prediction accuracy of the model for different types of faults, the weights of different decision trees are adjusted to improve the accuracy of the overall model. Cross-validation technology is used to further optimize model parameters.

[0104] Real-time application of photovoltaic analysis models to monitor and analyze the operating status of photovoltaic power stations. Provide real-time fault warning information and maintenance suggestions to help operation and maintenance personnel take timely measures to avoid equipment failures and production losses.

[0105] In the present invention, by acquiring and analyzing the operation data and early warning data of the photovoltaic power station, the method can monitor the operation status and environmental conditions of the equipment in real time. Once an abnormality or potential fault risk is detected, the system can quickly issue an early warning to help operators take timely maintenance and adjustment measures. The extraction of equipment operation characteristics and environmental operation characteristics covers multiple dimensions of the system, such as voltage, current, temperature, humidity, wind speed, etc. These characteristics provide comprehensive data support for fault analysis. By constructing recursive trees and decision forest models, the method can effectively process high-dimensional and complex operation data. The recursive tree can reveal the complex relationship between features, while the decision forest improves the generalization ability and prediction accuracy of the model by integrating multiple decision trees. The method is based on a big data platform and realizes real-time processing and analysis of large-scale data. The system can process massive data from multiple photovoltaic power stations and has good scalability.

[0106] Optionally, S1 includes:

[0107] S11, obtaining the original photovoltaic power station operation data through the Internet of Things protocol and the cloud interface;

[0108] Specifically, install sensors at key locations of the photovoltaic power station (such as photovoltaic modules and inverters) or use communication devices to connect with equipment to monitor real-time operating parameters such as current, voltage, temperature, irradiance, etc. Configure IoT communication modules (such as LoRa, NB-IoT, Zigbee, etc.) to ensure the remote transmission capability of data.

[0109] Select IoT protocols based on actual needs, such as MQTT (Message Queuing Telemetry Transport) for low-latency, lightweight data transmission, and CoAP (Constrained Application Protocol) for communication between resource-constrained devices. By configuring the collection cycle and data sending frequency, the collected raw data is sent to the cloud through the IoT gateway.

[0110] Design a cloud data receiving API interface to support real-time data reception and storage. The interface should support the parsing and standardization of different types of data. Use encryption technologies such as HTTPS and TLS to ensure the security and integrity of data during transmission.

[0111] S12, cleaning and formatting the original photovoltaic power station operation data to obtain the photovoltaic power station operation data;

[0112] Specifically, use statistical methods (such as the 3σ rule) or machine learning models (such as Isolation Forest) to identify and remove abnormal data points. Check the uniqueness of data records and delete duplicate records. Use interpolation, mean filling, or nearest neighbor filling methods to handle missing values ​​in the data.

[0113] Parse raw data from formats such as JSON and XML into structured data, such as relational databases or time series databases. Perform unit conversion and format standardization (such as unified timestamp format) on data to ensure data consistency.

[0114] Select a suitable database management system (such as MySQL, Hadoop HDFS) to store the cleaned PV power station operation data according to the data volume and real-time requirements. Establish indexes and optimize query performance to support subsequent data analysis and model training.

[0115] S13, obtaining photovoltaic power station early warning data corresponding to the photovoltaic power station operation data.

[0116] Specifically, historical operation data is analyzed to identify typical failure modes and establish a failure feature library. Based on historical failure data and statistical analysis, early warning thresholds for various types of failures are set.

[0117] Use supervised learning algorithms (such as decision trees, SVMs, neural networks) to train early warning models, with real-time operation data as model input and early warning signals as output. Alternatively, build a rule engine (such as Drools) based on expert experience and fault feature libraries for real-time monitoring and early warning. Alternatively, generate an input interface window to obtain corresponding PV power plant early warning data.

[0118] Combine real-time PV power station operation data with early warning models to generate early warning data and identify potential failure risks. Record early warning information, including warning time, warning type, related equipment, etc., for subsequent analysis and maintenance decisions.

[0119] In the present invention, the operation data of the photovoltaic power station can be obtained in real time through the Internet of Things protocol and the cloud interface, ensuring the timeliness of the data and enabling the system to continuously monitor the equipment status and environmental conditions. A high-quality data foundation helps to improve the accuracy and reliability of the model, reduce false positives and omissions caused by data problems, and improve the effectiveness of fault analysis and early warning. Format conversion converts data from different sensors or communication devices and equipment into a unified format so that the data can be seamlessly integrated and analyzed on a big data platform. Combined with early warning data, abnormal patterns in operating data can be better understood, which helps to identify and predict potential faults. Early warning data also provides labeled data for model training and verification, improving the accuracy of the prediction model.

[0120] Optionally, S2 includes:

[0121] S21, extracting power station operation data and environmental operation data according to the photovoltaic power station operation data, to obtain power station operation data and environmental operation data;

[0122] Specifically, the operating data of photovoltaic modules, inverters, transformers and other equipment are extracted from the cleaned data, such as current, voltage, power output, etc. Data related to the power station environment, such as solar radiation intensity, temperature, humidity, wind speed, etc., are extracted.

[0123] S22, extracting features from the power station operation data to obtain preliminary equipment operation feature data, and extracting features from the environmental operation data to obtain preliminary environmental operation feature data;

[0124] Specifically, calculate the statistical characteristics of the equipment, such as the average value, maximum value, minimum value, standard deviation, etc. Extract the trend characteristics of the equipment during operation, such as the slope change of the power curve.

[0125] Calculate the average, extreme value, and fluctuation range of environmental data. Extract seasonal variation characteristics, such as daily variation trends of temperature and radiation.

[0126] S23, performing multi-scale feature conversion on the preliminary equipment operation characteristic data to obtain equipment operation characteristic data;

[0127] Specifically, wavelet transform is performed on the preliminary equipment operation characteristic data to extract multi-scale equipment operation characteristics. Fourier transform is applied to extract frequency domain features to identify periodic and non-periodic components in equipment operation.

[0128] S24. Extract extreme value features from the preliminary environmental operation characteristic data to obtain environmental operation characteristic data.

[0129] Specifically, extreme value analysis is performed on the preliminary environmental operation characteristic data to detect abnormal conditions above or below a predetermined threshold, and cluster analysis is performed on the detected extreme value characteristics to identify common environmental abnormal patterns.

[0130] In the present invention, by dividing the photovoltaic power station operation data into power station operation data and environmental operation data, the operation status of the equipment and the changes in environmental conditions can be analyzed respectively, which helps to more accurately understand the impact of various types of data on the performance of the photovoltaic power station. Feature extraction of the power station operation data can extract key information reflecting the health status of the equipment, such as voltage, current, temperature, etc. Accurate feature extraction can better reflect the actual operating status of the equipment, help to timely discover abnormal conditions or potential faults, and improve the sensitivity of fault detection. Features such as temperature, humidity, wind speed, etc. are extracted from the environmental operation data to obtain preliminary environmental operation feature data. Analyzing these features can help understand the impact of the external environment on the operation of the photovoltaic power station. Multi-scale feature conversion is performed on the preliminary equipment operation feature data, and multi-level information in the data is captured by analyzing the feature changes on different time scales. Extreme value features are extracted from the preliminary environmental operation feature data to identify extreme events or abnormal conditions in the environmental data.

[0131] Optionally, the multi-scale feature transformation includes:

[0132] Perform initial feature decomposition based on preliminary equipment operation feature data to obtain decomposed feature data;

[0133] Specifically, the preliminary equipment operation characteristic data are decomposed using eigendecomposition methods, such as principal component analysis (PCA), empirical mode decomposition (EMD), or independent component analysis (ICA).

[0134] PCA: Perform a linear transformation on the feature data and project the original data into a new coordinate system. The principal components represent the direction with the largest variance in the data. The first few principal components are extracted as the decomposed feature data.

[0135] EMD: Decomposes nonlinear and non-stationary signals into several intrinsic mode functions (IMFs) that represent the characteristics of the data at different time scales.

[0136] ICA: Separates statistically independent signal sources and identifies independent characteristic components in device operation.

[0137] Perform multi-scale time series segmentation on the decomposed feature data to obtain multi-scale time series segmentation data;

[0138] Specifically, a sliding window or recursive segmentation technique is used to perform multi-scale time series segmentation on the decomposed feature data.

[0139] Set multiple time windows of different sizes, slide the feature data, and extract the data fragments in each window. According to the data change pattern, recursively split the data into time fragments of multiple scales to ensure that the dynamic changes of data are captured at different time scales.

[0140] Perform multi-scale feature extraction on multi-scale time series segmentation data to obtain multi-scale feature data;

[0141] Specifically, wavelet transform and multi-scale entropy analysis are applied. Wavelet transform is performed on each time series segment to extract signal features of different frequency bands, such as low-frequency trend features and high-frequency emergencies. The data complexity and uncertainty at different time scales are calculated to quantify features at multiple scales.

[0142] Perform frequency domain feature analysis on the multi-scale feature data to obtain frequency domain feature data;

[0143] Specifically, Fast Fourier Transform (FFT) and Power Spectrum Analysis are used. FFT: Converts multi-scale feature data in the time domain to the frequency domain and calculates the amplitude and phase of different frequency components. Power Spectrum Analysis: Calculates the power spectrum density, analyzes the energy distribution of data at different frequencies, and identifies periodic features and sudden events.

[0144] Perform feature combination according to the multi-scale feature data to obtain combined feature data;

[0145] Specifically, linear combination and nonlinear combination methods are adopted.

[0146] Linear combination: Perform linear weighted combination on multi-scale feature data to integrate multiple features into a feature vector.

[0147] Nonlinear combination: Use nonlinear methods (such as multi-layer perceptron or support vector machine) to perform complex nonlinear combinations of multi-scale features to generate more expressive combined features.

[0148] Multi-scale feature aggregation is performed based on the frequency domain feature data and the combined feature data to obtain the equipment operation feature data.

[0149] Specifically, an integrated learning method such as random forest or gradient boosting is used to aggregate the frequency domain feature data and the combined feature data to generate aggregate feature data. Cluster analysis is performed on the aggregate feature data to identify feature patterns under different operating states, and finally aggregate them into device operation feature data.

[0150] In the present invention, the preliminary equipment operation feature data is decomposed to extract features of different scales and dimensions. This helps to identify complex structures and potential patterns in the data. The decomposed feature data is segmented according to different time scales so that short-term and long-term change features can be analyzed separately. Features are extracted at each time scale, including average value, fluctuation amplitude, etc., and multi-scale feature data is extracted to enrich the feature set. The multi-scale feature data is converted to the frequency domain, its frequency components are analyzed, and periodic features and main frequency components are identified. Multi-scale and frequency domain features are combined to generate more complex combined feature data to capture the interaction between different features. Multi-scale feature aggregation provides a complete feature perspective that covers all aspects of equipment operation. Comprehensive analysis can improve the accuracy and comprehensiveness of fault detection, enabling the system to better cope with complex operating environments.

[0151] Optionally, the extreme value feature extraction includes:

[0152] Conduct extreme frequency analysis on preliminary environmental operation characteristic data to obtain extreme frequency data;

[0153] Specifically, calculate the Z value of each data point, define a Z value threshold (such as ±3), and identify extreme data points that exceed the threshold. Calculate the moving average of the data, and determine a deviation threshold to detect data points that deviate significantly from the moving average as extreme values. Count the number of extreme values ​​detected over a period of time, calculate the frequency of occurrence of extreme values, and generate extreme value frequency data.

[0154] Specifically, the moving average of the data is calculated, and a deviation threshold is determined to detect data points that deviate significantly from the moving average as extreme values. The number of extreme values ​​detected within a period of time is counted, and the frequency of occurrence of extreme values ​​is calculated to generate extreme value frequency data.

[0155] Specifically, more importantly, a first extreme value frequency analysis is performed on the preliminary environmental operation characteristic data to obtain first extreme value frequency data; a second extreme value frequency analysis is performed on the preliminary environmental operation characteristic data to obtain second extreme value frequency data, wherein the first extreme value frequency analysis and the second extreme value frequency analysis are different extreme value frequency analysis methods; clustering processing is performed based on the first extreme value frequency data and the second extreme value frequency data to obtain extreme value frequency clustering data; and weighted fusion is performed on the first extreme value frequency data and the second extreme value frequency data based on the extreme value frequency clustering data to obtain extreme value frequency data.

[0156] Constructing extreme value event sequence for extreme value frequency data and preliminary environmental operation characteristic data to obtain extreme value event sequence data;

[0157] Specifically, the extreme value frequency data is combined with the preliminary environmental operation characteristic data to identify the occurrence time, duration and related characteristics of each extreme value. The timestamp of the extreme value occurrence is matched with the preliminary environmental operation characteristic data to form an event record. Based on the matching results, an extreme value event sequence is constructed, which contains information such as the time, characteristic value and frequency of the event.

[0158] Calculate the extreme value impact factor according to the extreme value event sequence data to obtain the extreme value impact factor data;

[0159] Specifically, an impact factor model for extreme value events is established to analyze the contribution of each environmental feature to the occurrence of extreme value events. Based on the calculation results, the impact factor data of each extreme value event is generated to describe the degree of influence of each environmental feature on the extreme value.

[0160] Specifically, the correlation coefficient between each environmental feature and the occurrence of extreme value events is calculated to determine the main influencing factors. Based on the calculation results, the influencing factor data of each extreme value event is generated to describe the degree of influence of each environmental feature on the extreme value.

[0161] Perform extreme value clustering processing based on extreme value impact factor data and extreme value event sequence data to obtain extreme value category data;

[0162] Specifically, extreme value events are grouped according to the influencing factor data, and similar extreme value events are clustered into the same category using distance measurement methods such as Euclidean distance. Each clustering result is labeled to form extreme value category data to describe the characteristics of different types of extreme value events.

[0163] Specifically, a clustering tree is constructed based on the similarity of extreme value events, and extreme value events are gradually merged or split to form different clustering levels. Each clustering result is labeled to form extreme value category data to describe the characteristics of different types of extreme value events.

[0164] Specifically, more importantly, a probability graph is constructed based on extreme value impact factor data and extreme value event sequence data to obtain extreme value time probability graph data;

[0165] Perform dynamic probability evaluation based on extreme value event probability graph data to obtain extreme value event dynamic probability data;

[0166] According to the extreme value event dynamic probability data, cluster calculation is performed on the extreme value influencing factor data and the extreme value event sequence data to obtain the extreme value event clustering feature data;

[0167] According to the cluster center data in the extreme event cluster feature data, the extreme value influencing factor data and the extreme value event sequence data are classified to obtain the extreme value category data. First, the time point, event intensity, occurrence frequency, duration and other features of all extreme value events are extracted from the extreme value influencing factor data and the extreme value event sequence data. Each extreme value event is regarded as a node, and the feature vector of the node is composed of its influencing factor and time series features. For each pair of extreme value events, an edge is generated based on its time interval, event intensity difference, influencing factor similarity and other features. The weight of the edge can be calculated by the similarity or correlation function of the above features. Based on the generated nodes and edges, a time-related probability graph structure is constructed, in which the edge weight represents the correlation strength between extreme value events.

[0168] Based on historical data, an initial probability of occurrence is assigned to each extreme event. The initial probability can be estimated based on factors such as the frequency of occurrence and the scope of impact of the event. Using recursive relationships, the probability of the event at the current moment is updated based on the occurrence of the extreme event at the previous moment. The update process is based on the historical pattern of events, time intervals, and the correlation between events (represented by the edge weights of the graph). The system gradually updates the probability of events through time steps, taking into account the temporal correlation of events and evaluating their probability of occurrence in the future. Based on the probability of the extreme event at the previous moment, plus the influence of adjacent nodes in the graph, the probability of the current event is dynamically adjusted. The probabilities of all events are gradually updated according to the time series to form a time-following probability evaluation sequence. The probability data of each extreme event at different time points are summarized to obtain dynamic probability evaluation results.

[0169] The dynamic probability, influencing factors and time series data of extreme events are combined into a multidimensional feature vector, with each event corresponding to an independent vector. The similarity between events is calculated based on the distance metric between feature vectors. The distance metric can be combined with the difference in dynamic probability, the difference in influencing factors and the similarity of time series patterns for comprehensive calculation. Several cluster centers are initially assigned, and the positions of the centers are determined by statistical analysis of the feature vectors of all events to determine the preliminary cluster centers. Based on the similarity of the feature vectors, each event is classified into the closest cluster center, and then the eigenvalues ​​of each cluster center are recalculated based on the classification results. This process is iterated until the clustering results converge. A comprehensive vector containing dynamic probability, influencing factors and time series features is generated for each extreme event. All events are compared pairwise, their similarities are calculated, and a similarity matrix is ​​generated. Events are classified into the closest center, and the features of each center are updated until the clustering results are stable.

[0170] Extreme value trend processing is performed based on extreme value category data and preliminary environmental operation characteristic data to obtain environmental operation characteristic data.

[0171] Specifically, we conduct time series analysis on extreme events of each category to identify the occurrence patterns and trends of extreme events. Based on the trend analysis results of extreme events, we update the environmental operation characteristic data and provide prediction information on future environmental changes.

[0172] Specifically, we use the historical data of extreme events to establish a prediction model (such as the ARIMA model) to predict the trend of extreme events in the future. Based on the trend analysis results of extreme events, we update the environmental operation characteristic data and provide prediction information on future environmental changes.

[0173] In the present invention, the frequency and time distribution of extreme events are identified by analyzing the extreme value frequency in the preliminary environmental operation characteristic data. Based on the extreme value frequency data and the preliminary environmental operation characteristic data, an extreme value event sequence is constructed to record detailed information such as the time, duration, and amplitude of the extreme value. Using the extreme value event sequence data, the impact factor of each extreme value event is calculated to quantify its impact on the operation of the photovoltaic power station. Extreme value clustering processing helps to identify and classify different types of extreme environmental events, so that differentiated management and response strategies can be adopted for different categories of extreme situations. Extreme value trend processing helps to identify the long-term development trends and possible future changes of extreme events. By analyzing historical data and trends, the system can predict extreme environmental conditions that may occur in the future, providing a basis for advance preparation and risk prevention.

[0174] Optionally, S3 includes:

[0175] S31, performing recursive tree initialization according to the device operation characteristic data and the environment operation characteristic data, to obtain device operation decision tree initialization data and environment operation decision tree initialization data;

[0176] Specifically, a binary tree or a multi-branch tree structure is used as a basic form of the decision tree. An initial split feature is selected from the equipment operation feature data and the environment operation feature data to establish a root node.

[0177] Using the equipment operation feature data, select the feature that can best distinguish between fault and normal operation status as the root node feature, and initialize the equipment operation decision tree. Using the environment operation feature data, select the feature that can best distinguish the equipment operation status under different environmental conditions as the root node feature, and initialize the environment operation decision tree.

[0178] S32, performing node splitting standard selection on the device operation characteristic data and the environment operation characteristic data, and obtaining device operation splitting standard data and environment operation splitting standard data respectively;

[0179] Specifically, the degree to which each feature reduces the uncertainty of classification is calculated, and the feature with the highest information gain is selected for node splitting.

[0180] Specifically, the purity gain of the feature is calculated, and the feature with the smallest Gini index is selected for node splitting.

[0181] Specifically, the association between the features and the target variable is evaluated, and the feature with the largest chi-square statistic is selected.

[0182] S33, performing node splitting according to the equipment operation decision tree initialization data, equipment operation splitting standard data, equipment operation characteristic data and corresponding photovoltaic power station early warning data to obtain preliminary equipment operation decision tree data, and performing node splitting according to the environment operation decision tree initialization data, environment operation characteristic data, environment operation splitting standard data and corresponding photovoltaic power station early warning data to obtain preliminary environment operation decision tree data;

[0183] Specifically, the root node is split according to the equipment operation decision tree initialization data, equipment operation splitting standard data, equipment operation characteristic data and warning data. The above process is repeated for each child node until the stop condition is met (such as homogeneity of data within the node or too little data in the node).

[0184] According to the initialization data of the environment operation decision tree, the environment operation characteristic data, the environment operation splitting standard data and the early warning data, the root node is split. Similarly, each child node is recursively split.

[0185] S34, performing tree structure optimization on the preliminary equipment operation decision tree data and the preliminary equipment operation decision tree data to obtain optimized equipment operation decision tree data and optimized equipment operation decision tree data respectively;

[0186] Specifically, overfitted decision trees are pruned to remove nodes that do not contribute much to prediction. The maximum tree depth is set to avoid the increase in complexity caused by too deep a tree structure.

[0187] S35, performing post-pruning processing on the optimized equipment operation decision tree data and the optimized equipment operation decision tree data to obtain equipment operation decision tree data and environment operation decision tree data respectively.

[0188] Specifically, the dataset is divided into multiple subsets, cross-validation is performed, the contribution of each node is evaluated, and a decision is made whether to prune. The post-pruning error of each node is calculated, and if the error decreases after pruning, the node is pruned.

[0189] The recursive tree initialization in the present invention provides a basic classification and regression framework to ensure that the model has a reasonable structure from the beginning. Through proper initialization, the training time can be reduced and the initial performance of the model can be improved. Select appropriate node splitting criteria (such as information gain, Gini index, etc.) for the equipment and environmental operation feature data to ensure that the splitting process can maximize purity or minimize error. Node splitting is the core of the formation of the decision tree model, which can effectively separate different categories or regression values ​​in the data set. Through multiple splits, the model can gradually refine the classification and regression of the data and capture more details and patterns. The structure of the initially generated decision tree is optimized, including adjusting the depth and branch structure of the tree to improve the performance of the tree. The optimized decision tree is post-pruned to remove unnecessary branches and nodes, thereby simplifying the model.

[0190] Optionally, the recursive tree initialization includes:

[0191] Sorting the importance of the features according to the equipment operation feature data and the environment operation feature data, and obtaining equipment operation feature ranking data and environment operation feature ranking data respectively;

[0192] Specifically, a random forest model is trained to rank the features by calculating the importance score of each feature in the model.

[0193] Specifically, the degree to which each feature improves the classification effect of the target variable is calculated, and the features are ranked based on information gain.

[0194] Specifically, the chi-square statistic is used to assess the correlation of each feature with the target variable and to rank the features.

[0195] Performing feature interaction effect processing on the equipment operation feature data and the environment operation feature data to obtain equipment operation feature interaction effect data and environment operation feature interaction effect data respectively;

[0196] Specifically, high-order interaction features of features are generated to analyze nonlinear relationships between features. Polynomial combinations of features are generated, including square terms, cubic terms, and higher-order terms. For example, given two features, quadratic interaction features are generated. , , or cubic interaction feature , , , .

[0197] Specifically, a partial dependency plot is drawn to show the effect of a specific feature on the target variable while other features are fixed. Select the feature of interest, and the other features you want to keep fixed. For each value of a specific feature, calculate the prediction for all data points while keeping other features fixed, and take the average. The vertical axis represents the predicted output of the model, and the horizontal axis represents the value of the feature. By drawing the PDP, you can observe how the feature affects the prediction of the model.

[0198] Specifically, calculate the contribution of each feature and its interaction to the prediction result and evaluate the interaction effect. For each feature and its interaction feature, calculate the marginal contribution in all feature combinations. Summarize the marginal contributions of all combinations to get the Shapley value of each feature. Get the contribution of each feature and its interaction feature, and quantitatively show the impact of the feature on the model prediction.

[0199] Performing initial node screening on the equipment operation feature sorting data and the equipment operation feature interaction effect data to obtain the equipment operation feature initial node data, and performing initial node screening on the environment operation feature sorting data and the environment operation feature interaction effect data to obtain the environment operation feature initial node data;

[0200] Specifically, the feature with the largest contribution is selected from the equipment operation feature ranking data and the interaction effect data as the initial node. The feature with the largest contribution is selected from the environment operation feature ranking data and the interaction effect data as the initial node.

[0201] Splitting candidate sets are generated for the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics, and obtaining equipment operation characteristic classification candidate set data and environment operation characteristic classification candidate set data respectively;

[0202] Specifically, based on the initial node data, possible splitting conditions are generated. For continuous features (such as current and voltage), discretization methods are determined, including equal frequency binning and equal width binning. Equal frequency binning is to divide the eigenvalues ​​into several intervals evenly according to the frequency, so that the number of samples in each interval is roughly the same. Equal width binning is to divide the eigenvalue range into several intervals with equal width.

[0203] For each continuous feature, calculate all possible split points. For example, for a sequence of feature values, you can take the average of every two adjacent values ​​as a potential split point.

[0204] For discrete features (such as equipment type), generate all possible category combinations. For example, if the feature has three categories (A, B, C), possible split conditions include {A}, {B, C}, and so on. Calculate the Gini index of each split point, and select the split point with the lowest Gini index as the candidate split point. Calculate the information gain of the split point, and select the split point with the largest information gain. Obtain all valid split points for each feature to form the candidate set data for equipment operation feature classification.

[0205] The generation of classification candidate sets of environmental operating characteristics adopts processing similar to that of equipment operating characteristics, and possible splitting conditions are generated for the category combinations of environmental characteristics.

[0206] An initial tree depth is set according to the equipment operation feature classification candidate set data and the environment operation feature classification candidate set data to obtain equipment operation feature initial tree depth data and environment operation feature initial tree depth data;

[0207] Specifically, the initial depth of the equipment operation decision tree is set according to the complexity of the equipment operation feature classification candidate set data and the target variable. The initial depth of the environment operation decision tree is set according to the complexity of the environment operation feature classification candidate set data and the target variable.

[0208] Initializing the recursive tree according to the initial node data of the equipment operation characteristics, the classification candidate set data of the equipment operation characteristics, and the initial tree depth data of the equipment operation characteristics to obtain the initialization data of the equipment operation decision tree;

[0209] Specifically, the initial structure of the equipment operation decision tree is constructed using the equipment operation feature initial node data, classification candidate set data, and initial tree depth data.

[0210] Recursive tree initialization is performed according to the initial node data of the environment operation characteristics, the classification candidate set data of the environment operation characteristics and the initial tree depth data of the environment operation characteristics to obtain the initial data of the environment operation decision tree.

[0211] Specifically, the initial structure of the environment operation decision tree is constructed using the environment operation feature initial node data, classification candidate set data and initial tree depth data.

[0212] The feature importance ranking in the present invention helps to identify key features, so that the model can focus more on the features that contribute most to the classification or regression task, and improve the efficiency and accuracy of the model. Feature interaction effect processing can identify and utilize the complex relationships between different features, help the model better understand the deep patterns and associations in the data, and improve the prediction performance. Initial node screening ensures that the recursive tree starts to split from the most important features and key interaction effects, lays a good foundation for the model and improves the initial performance of the model. Split candidate set generation provides a variety of split path options, allowing the model to find the optimal splitting solution, maximize information gain or minimize error, and improve the decision-making ability of the model. Reasonable tree depth setting can prevent model overfitting, maintain the simplicity and generalization ability of the model, and ensure that the model has sufficient complexity to capture patterns in the data. Recursive tree initialization constructs a decision tree with a good initial structure by integrating key features and the best splitting solution, providing a solid foundation for subsequent model training and optimization.

[0213] Optionally, the device operation feature classification candidate set data includes first device operation feature classification candidate set data and second device operation feature classification candidate set data, the environment operation feature classification candidate set data includes first environment operation feature classification candidate set data and second environment operation feature classification candidate set data, and the split candidate set generation includes:

[0214] Clustering the equipment operation feature initial node data and the environment operation feature initial node data to obtain first equipment operation feature classification candidate set data and first environment operation feature classification candidate set data;

[0215] Specifically, the initial node data of the device operation characteristics and the initial node data of the environment operation characteristics are standardized to reduce the impact of the difference in feature dimensions. K initial cluster centers are randomly selected, and the selection of the K value can be determined according to the elbow rule or the silhouette coefficient. The distance from each data point to the cluster center is calculated, the data point is assigned to the nearest cluster center, and then the position of the cluster center is updated. When the cluster center no longer changes or the preset maximum number of iterations is reached, the iteration is stopped. The first device operation feature classification candidate set data generated is the device operation feature category represented by each cluster center. The first environment operation feature classification candidate set data generated is the environment operation feature category represented by each cluster center.

[0216] A chi-square test is performed on the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics to obtain the second equipment operation characteristic classification candidate set data and the second environment operation characteristic classification candidate set data.

[0217] Specifically, the initial node data of the equipment operation characteristics and the initial node data of the environmental operation characteristics are discretized, and the continuous features are converted into categorical features. The chi-square statistic of each feature and the target variable (such as fault status) is calculated. According to the chi-square statistic and the degree of freedom, the corresponding p-value is found, and compared with the significance level (such as 0.05) to determine whether there is a significant difference between the feature and the target variable. The obtained second equipment operation characteristic classification candidate set data includes the significant equipment operation characteristics screened out by the chi-square test. The obtained second environmental operation characteristic classification candidate set data includes the significant environmental operation characteristics screened out by the chi-square test.

[0218] Clustering and chi-square test in the present invention provide two different feature analysis perspectives. Clustering analyzes from the perspective of data distribution and similarity, while chi-square test analyzes from the perspective of the correlation between features and target variables. Combining these two methods, effective splitting points can be identified more comprehensively. By using clustering, natural categories in the data can be identified, reducing the risk of overfitting the model to specific training data. The chi-square test ensures that the model focuses on the features most relevant to the target variable, thereby improving the generalization ability of the model. The split candidate sets generated by multiple methods provide a wealth of choices, providing more possibilities for the model to find the best splitting point, and the diversified splitting point selection strategies help improve the classification and regression performance of the model.

[0219] Optionally, S4 includes:

[0220] S41, performing multi-decision tree integration according to the equipment operation decision tree data and the environment operation decision tree data to obtain decision forest structure data;

[0221] Specifically, by introducing randomness, the data and features of the device operation decision tree and the environment operation decision tree are randomly sampled to build multiple independent decision trees. Multiple decision trees are built in sequence, and each tree attempts to correct the errors of the previous tree. The structure and parameter information of multiple decision trees are included to form a forest structure.

[0222] S42, performing decision tree weight calculation on the decision forest structure data to obtain decision tree weight data;

[0223] Specifically, a performance evaluation metric (such as accuracy, F1 value, AUC, etc.) is used to assign a weight to each decision tree. The performance of each tree in the decision forest is evaluated and its performance on the validation set is calculated. A weight is assigned to each tree based on its performance, and the better the performance, the higher the weight of the tree. The decision tree weight data contains the weight information of each decision tree, which is used for subsequent voting and fusion.

[0224] S43, performing diversity optimization according to the decision tree weight data and the decision forest structure data to obtain decision forest diversity data;

[0225] Specifically, we calculate the diversity metrics between decision trees, such as the Q statistic (used to measure the correlation of errors between trees) and the correlation coefficient (used to measure the correlation of tree predictions). We use the greedy algorithm to find a balance between weight and diversity, select decision trees with both high accuracy and high diversity, and optimize the diversity of the decision forest. The optimized decision forest structure has higher diversity and robustness.

[0226] S44, constructing a voting mechanism according to the decision forest diversity data to obtain decision forest voting mechanism data;

[0227] Specifically, according to the weight of the decision tree, each tree votes on the prediction result, and the vote of the tree with a higher weight has a greater influence. The prediction probability of each tree is weighted averaged according to the weight, and the category with the highest probability is selected as the final prediction. The decision forest voting mechanism data contains the voting mechanism rules of the decision tree, which guides the integrated decision process of the model.

[0228] S45. Model fusion is performed according to the decision forest diversity data, the decision forest voting mechanism data and the corresponding photovoltaic power station early warning data to obtain a photovoltaic analysis model of the photovoltaic power station to perform photovoltaic power station fault early warning operations.

[0229] Specifically, the optimized decision forest model is combined with the early warning data to construct a photovoltaic analysis model for photovoltaic power stations. The voting mechanism of the decision forest is used to predict the operating status of the photovoltaic power station based on the voting results. The early warning data is integrated to comprehensively consider the operating history and environmental conditions of the equipment to improve the accuracy of fault prediction. The photovoltaic analysis model for photovoltaic power stations can analyze the operating data of photovoltaic power stations in real time and provide accurate fault warning information.

[0230] The integration of multiple decision trees in the present invention can average the prediction errors of individual trees and significantly improve the accuracy and stability of the model. By assigning weights, decision trees with better performance can be given priority to improve the overall prediction accuracy. Diversity optimization can increase the robustness of the model and combat the influence of noise and abnormal data. The voting mechanism integrates the opinions of multiple models to reduce the impact of a single erroneous prediction on the final result. Through weighted voting, the voting results can be adjusted according to the accuracy and weight of each tree to improve the overall prediction accuracy. Model fusion combines the advantages of multiple decision trees and early warning data to provide more comprehensive fault analysis and prediction for photovoltaic power plants.

[0231] Optionally, the present application further provides a photovoltaic power station fault analysis and early warning system based on big data, which is used to execute the photovoltaic power station fault analysis and early warning method based on big data as described above, and the photovoltaic power station fault analysis and early warning system based on big data includes:

[0232] Photovoltaic power station data acquisition module, used to obtain photovoltaic power station operation data and corresponding photovoltaic power station early warning data;

[0233] The photovoltaic power station data feature extraction module is used to extract equipment operation features and environmental operation features according to the photovoltaic power station operation data, and obtain equipment operation feature data and environmental operation feature data respectively;

[0234] A photovoltaic power station recursive tree construction module is used to construct a recursive tree according to the equipment operation characteristic data and the environment operation characteristic data, and obtain equipment operation decision tree data and environment operation decision tree data respectively;

[0235] The photovoltaic analysis model construction module of the photovoltaic power station is used to construct a decision forest based on the equipment operation decision tree data, the environmental operation decision tree data and the corresponding photovoltaic power station early warning data to obtain the photovoltaic analysis model of the photovoltaic power station for photovoltaic power station fault early warning operations.

[0236] The objects of the present invention are:

[0237] 1. Through the Internet of Things protocol and cloud interface, the operation data and early warning data of the photovoltaic power station can be obtained in real time, ensuring the system's continuous perception of the latest status and avoiding information lag.

[0238] 2. By extracting equipment operation characteristics and environmental operation characteristics data, the operation status of the photovoltaic power station can be analyzed from multiple dimensions and scales, and the potential complex relationships and influencing factors in the system can be captured. In the recursive tree initialization process, the key initial nodes are screened out by combining feature importance sorting and feature interaction effect analysis, which can more accurately capture key features and the complex relationships between them, and improve the model's explanatory power and prediction accuracy.

[0239] 3. By integrating multiple devices and environments to run decision trees, a decision forest with high diversity is constructed. Diversity optimization further ensures the robustness of the model and can effectively deal with different types of abnormal situations and data noise. The construction of the voting mechanism combines the weight distribution of the decision tree, so that the model can make more accurate predictions while considering multiple trees comprehensively. This mechanism improves the overall decision quality of the model and reduces the impact of single model bias.

[0240] 4. By combining the decision tree data of equipment and environment and early warning data, the photovoltaic analysis model formed can accurately predict and warn of faults. The model integrates multiple data sources and prediction results, providing a more comprehensive analysis perspective, which can timely detect potential faults and issue early warning signals.

[0241] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0242] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A photovoltaic power station fault analysis and early warning method based on big data, characterized in that: Applied to a big data platform, the method includes: S1. Obtaining photovoltaic power station operation data and corresponding photovoltaic power station early warning data; S2. Extract equipment operation characteristics and environmental operation characteristics according to the photovoltaic power station operation data to obtain equipment operation characteristic data and environmental operation characteristic data respectively; S3, constructing a recursive tree according to the equipment operation characteristic data and the environment operation characteristic data, and obtaining equipment operation decision tree data and environment operation decision tree data respectively; S4, constructing a decision forest based on the equipment operation decision tree data, the environment operation decision tree data and the corresponding photovoltaic power station early warning data to obtain a photovoltaic analysis model of the photovoltaic power station to perform photovoltaic power station fault early warning operations; S3 includes: Initialize the recursive tree according to the equipment operation characteristic data and the environment operation characteristic data to obtain equipment operation decision tree initialization data and environment operation decision tree initialization data; Perform node splitting standard selection on the equipment operation characteristic data and the environment operation characteristic data to obtain equipment operation splitting standard data and environment operation splitting standard data respectively; Perform node splitting according to the equipment operation decision tree initialization data, equipment operation splitting standard data, equipment operation characteristic data and corresponding photovoltaic power station early warning data to obtain preliminary equipment operation decision tree data, and perform node splitting according to the environment operation decision tree initialization data, environment operation characteristic data, environment operation splitting standard data and corresponding photovoltaic power station early warning data to obtain preliminary environment operation decision tree data; Performing tree structure optimization on preliminary equipment operation decision tree data and preliminary equipment operation decision tree data to obtain optimized equipment operation decision tree data and optimized equipment operation decision tree data respectively; Post-pruning the optimized equipment operation decision tree data and the optimized equipment operation decision tree data to obtain equipment operation decision tree data and environment operation decision tree data respectively; Recursive tree initialization includes: Sorting the importance of the features according to the equipment operation feature data and the environment operation feature data, and obtaining equipment operation feature ranking data and environment operation feature ranking data respectively; Performing feature interaction effect processing on the equipment operation feature data and the environment operation feature data to obtain equipment operation feature interaction effect data and environment operation feature interaction effect data respectively; Performing initial node screening on the equipment operation feature sorting data and the equipment operation feature interaction effect data to obtain the equipment operation feature initial node data, and performing initial node screening on the environment operation feature sorting data and the environment operation feature interaction effect data to obtain the environment operation feature initial node data; Splitting candidate sets are generated for the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics, and obtaining equipment operation characteristic classification candidate set data and environment operation characteristic classification candidate set data respectively; An initial tree depth is set according to the equipment operation feature classification candidate set data and the environment operation feature classification candidate set data to obtain equipment operation feature initial tree depth data and environment operation feature initial tree depth data; Initializing the recursive tree according to the initial node data of the equipment operation characteristics, the classification candidate set data of the equipment operation characteristics, and the initial tree depth data of the equipment operation characteristics to obtain the initialization data of the equipment operation decision tree; Recursive tree initialization is performed according to the initial node data of the environment operation characteristics, the classification candidate set data of the environment operation characteristics and the initial tree depth data of the environment operation characteristics to obtain the initial data of the environment operation decision tree.

2. The method according to claim 1, characterized in that S1 includes: Obtain original photovoltaic power station operation data through the Internet of Things protocol and cloud interface; Clean the original photovoltaic power station operation data and convert its format to obtain the photovoltaic power station operation data; Obtain photovoltaic power station early warning data corresponding to photovoltaic power station operation data.

3. The method according to claim 1, characterized in that S2 include: Extracting power station operation data and environmental operation data based on the photovoltaic power station operation data to obtain power station operation data and environmental operation data; Extracting features from power plant operation data to obtain preliminary equipment operation feature data, and extracting features from environmental operation data to obtain preliminary environmental operation feature data; Perform multi-scale feature conversion on preliminary equipment operation characteristic data to obtain equipment operation characteristic data; The extreme value features are extracted from the preliminary environmental operation characteristic data to obtain the environmental operation characteristic data.

4. The method according to claim 3, characterized in that Multi-scale feature transformation includes: Perform initial feature decomposition based on preliminary equipment operation feature data to obtain decomposed feature data; Perform multi-scale time series segmentation on the decomposed feature data to obtain multi-scale time series segmentation data; Perform multi-scale feature extraction on multi-scale time series segmentation data to obtain multi-scale feature data; Perform frequency domain feature analysis on the multi-scale feature data to obtain frequency domain feature data; Perform feature combination according to the multi-scale feature data to obtain combined feature data; Multi-scale feature aggregation is performed based on the frequency domain feature data and the combined feature data to obtain the equipment operation feature data.

5. The method according to claim 3, characterized in that: Extreme value feature extraction includes: Conduct extreme frequency analysis on preliminary environmental operation characteristic data to obtain extreme frequency data; Constructing extreme value event sequence for extreme value frequency data and preliminary environmental operation characteristic data to obtain extreme value event sequence data; Calculate the extreme value impact factor according to the extreme value event sequence data to obtain the extreme value impact factor data; Perform extreme value clustering processing based on extreme value impact factor data and extreme value event sequence data to obtain extreme value category data; Extreme value trend processing is performed based on extreme value category data and preliminary environmental operation characteristic data to obtain environmental operation characteristic data.

6. The method according to claim 1, characterized in that The device operation feature classification candidate set data includes first device operation feature classification candidate set data and second device operation feature classification candidate set data, the environment operation feature classification candidate set data includes first environment operation feature classification candidate set data and second environment operation feature classification candidate set data, and the split candidate set generation includes: Clustering the equipment operation feature initial node data and the environment operation feature initial node data to obtain first equipment operation feature classification candidate set data and first environment operation feature classification candidate set data; A chi-square test is performed on the initial node data of the equipment operation characteristics and the initial node data of the environment operation characteristics to obtain the second equipment operation characteristic classification candidate set data and the second environment operation characteristic classification candidate set data.

7. The method according to claim 1, characterized in that S4 includes: Perform multiple decision tree integration based on equipment operation decision tree data and environment operation decision tree data to obtain decision forest structure data; Calculate the decision tree weights on the decision forest structure data to obtain the decision tree weight data; Diversity optimization is performed based on decision tree weight data and decision forest structure data to obtain decision forest diversity data; The voting mechanism is constructed according to the decision forest diversity data to obtain the decision forest voting mechanism data; Based on the decision forest diversity data, decision forest voting mechanism data and the corresponding photovoltaic power station early warning data, the model is integrated to obtain the photovoltaic analysis model of the photovoltaic power station for photovoltaic power station fault early warning operations.

8. A photovoltaic power station fault analysis and early warning system based on big data, characterized in that: Used to execute the photovoltaic power station fault analysis and early warning method based on big data as claimed in claim 1, the photovoltaic power station fault analysis and early warning system based on big data includes: Photovoltaic power station data acquisition module, used to obtain photovoltaic power station operation data and corresponding photovoltaic power station early warning data; The photovoltaic power station data feature extraction module is used to extract equipment operation features and environmental operation features according to the photovoltaic power station operation data, and obtain equipment operation feature data and environmental operation feature data respectively; The photovoltaic power station recursive tree construction module is used to initialize the recursive tree according to the equipment operation characteristic data and the environment operation characteristic data to obtain the equipment operation decision tree initialization data and the environment operation decision tree initialization data; select the node splitting standard for the equipment operation characteristic data and the environment operation characteristic data to obtain the equipment operation splitting standard data and the environment operation splitting standard data respectively; perform node splitting according to the equipment operation decision tree initialization data, the equipment operation splitting standard data, the equipment operation characteristic data and the corresponding photovoltaic power station early warning data to obtain preliminary equipment operation decision tree data, and perform node splitting according to the environment operation decision tree initialization data, the environment operation characteristic data, the environment operation splitting standard data and the corresponding photovoltaic power station early warning data to obtain preliminary environment operation decision tree data; perform tree structure optimization on the preliminary equipment operation decision tree data and the preliminary equipment operation decision tree data to obtain optimized equipment operation decision tree data and optimized equipment operation decision tree data respectively; perform post-pruning processing on the optimized equipment operation decision tree data and the optimized equipment operation decision tree data to obtain equipment operation decision tree data and environment operation decision tree data respectively; the recursive tree initialization includes: performing feature importance sorting according to the equipment operation characteristic data and the environment operation characteristic data to obtain the equipment operation feature sorting data and the environment operation feature sorting data respectively. Operation feature sorting data; performing feature interaction effect processing on the equipment operation feature data and the environment operation feature data, respectively obtaining equipment operation feature interaction effect data and environment operation feature interaction effect data; performing initial node screening on the equipment operation feature sorting data and the equipment operation feature interaction effect data, obtaining equipment operation feature initial node data, and performing initial node screening on the environment operation feature sorting data and the environment operation feature interaction effect data, obtaining environment operation feature initial node data; performing split candidate set generation on the equipment operation feature initial node data and the environment operation feature initial node data, respectively obtaining equipment operation feature classification candidate set data and environment operation feature classification candidate set data; performing initial tree depth setting according to the equipment operation feature classification candidate set data and the environment operation feature classification candidate set data, obtaining equipment operation feature initial tree depth data and environment operation feature initial tree depth data; performing recursive tree initialization according to the equipment operation feature initial node data, the equipment operation feature classification candidate set data and the equipment operation feature initial tree depth data, obtaining equipment operation decision tree initialization data; performing recursive tree initialization according to the environment operation feature initial node data, the environment operation feature classification candidate set data and the environment operation feature initial tree depth data, obtaining environment operation decision tree initialization data; The photovoltaic analysis model construction module of the photovoltaic power station is used to construct a decision forest based on the equipment operation decision tree data, the environmental operation decision tree data and the corresponding photovoltaic power station early warning data to obtain the photovoltaic analysis model of the photovoltaic power station for photovoltaic power station fault early warning operations.

Images