Photovoltaic power station anomaly detection method and system based on multi-source heterogeneous data

Abstract

The invention relates to a photovoltaic power station anomaly detection method and system based on multi-source heterogeneous data, and belongs to the field of photovoltaic system monitoring. The multi-source heterogeneous data comprises photovoltaic module infrared images (unstructured data) under normal power generation, inverter abnormity and photovoltaic module abnormity, meteorological sensor data and inverter data (structured data). When the model is trained, a fuzzy time matching algorithm is adopted to align the meteorological data and the inverter data, the high-dimensional feature vector of the infrared image is extracted, and the high-dimensional feature vector is spliced and fused with the aligned meteorological data and the inverter data for training the anomaly detection model. And when the photovoltaic power station monitoring system gives an alarm, collecting multi-source heterogeneous data in parallel, identifying fusion features by using the trained model, and outputting a result. According to the scheme, through feature layer fusion and fuzzy time matching, the method adapts to a high-altitude complex environment, the operation and maintenance efficiency and the fault diagnosis accuracy are remarkably improved, the operation and maintenance cost is reduced, and the method has universality and expandability.

Description

Technical Field

[0001] The present invention belongs to the field of photovoltaic system monitoring, and in particular relates to a photovoltaic power station anomaly detection method and system based on multi-source heterogeneous data. Background Art

[0002] With the intensification of the global energy crisis and the acceleration of climate change, the importance of energy transition has become increasingly prominent. Photovoltaic power generation, as a clean, renewable energy source, has experienced rapid growth worldwide. Leveraging its abundant solar resources, western my country has seen a continuous expansion in the scale of photovoltaic power station construction, gradually moving towards higher-altitude, larger-scale projects. The commissioning of high-altitude photovoltaic power stations, such as the Xingchuan Photovoltaic Power Station in Ganzi, Sichuan, not only provides clean energy for the region but also promotes regional economic development.

[0003] However, the unique environmental conditions at high altitudes, such as low oxygen levels, large temperature swings, and unpredictable weather, pose significant challenges to the stable operation of photovoltaic systems. These environmental factors can easily lead to failures in key equipment such as photovoltaic modules and inverters, significantly increasing the complexity of operation and maintenance. Achieving efficient operation and maintenance, as well as fault diagnosis, for photovoltaic systems in high-altitude and complex terrain environments has become a major technical challenge urgently needed to be addressed by the photovoltaic industry.

[0004] Furthermore, as PV power plants connect to the grid and operate, the number and types of sensors continue to increase, generating exponentially more data. This data, including meteorological data, inverter data, and image data, encompasses both structured and unstructured forms, exhibiting distinct multi-source heterogeneity. The key to improving the efficiency and quality of high-altitude PV power plant operation and maintenance is to integrate and mine this massive, heterogeneous data in real time, and apply it to fault diagnosis scenarios involving power plant data anomalies.

[0005] In summary, there are significant deficiencies in the operation and maintenance and fault diagnosis of photovoltaic power stations in high-altitude environments. There is an urgent need for a technical solution that can effectively cope with complex environmental conditions and realize multi-source heterogeneous data fusion and mining to improve the operation and maintenance efficiency and reliability of photovoltaic power stations. Summary of the Invention

[0006] To solve the above problems, the present invention proposes a photovoltaic power station anomaly detection method and system based on multi-source heterogeneous data, which adopts the following technical solutions:

[0007] In a first aspect, the present invention proposes a photovoltaic power station anomaly detection method based on multi-source heterogeneous data, comprising the following steps:

[0008] Acquiring multi-source heterogeneous data for training a photovoltaic power station anomaly detection model, the multi-source heterogeneous data comprising unstructured infrared images of photovoltaic modules acquired under normal power generation conditions, abnormal inverter conditions, and abnormal photovoltaic module conditions, as well as structured meteorological sensor data and inverter data;

[0009] During model training, a fuzzy time matching algorithm is first used to time-align meteorological sensor data with inverter data, extracting high-dimensional feature vectors from infrared images. This high-dimensional feature vector is then combined with the corresponding time-aligned meteorological and inverter data to form a fusion. This combined data is then used to train the photovoltaic power station anomaly detection model.

[0010] When an alarm occurs in the photovoltaic power station monitoring system, multi-source heterogeneous data are collected in parallel, and the trained photovoltaic power station anomaly detection model is used to perform row recognition on the features after fusion of multi-source heterogeneous data, and the recognition results are output.

[0011] Furthermore, the meteorological data includes ambient temperature, irradiance and wind speed, and the inverter data includes working status, current and voltage amplitudes, temperature in the warehouse, and inverter efficiency.

[0012] Furthermore, before training the photovoltaic power station anomaly detection model, it also includes structured data preprocessing and data enhancement operations at the data level.

[0013] Furthermore, the preprocessing of the structured data includes:

[0014] Perform missing value filling, noise filtering and data standardization on meteorological data and inverter data;

[0015] Through timestamp matching, the meteorological data and inverter data are time-aligned to form a unified feature sequence.

[0016] Furthermore, the data enhancement operation at the data level uses oversampling, undersampling or mixed sampling methods to balance the category distribution of the data.

[0017] Furthermore, the photovoltaic power station anomaly detection model adopts a support vector machine, a random forest or a deep learning model.

[0018] Furthermore, a CNN network is used to extract high-dimensional feature vectors of infrared images, and the CNN network is trained together with a photovoltaic power station anomaly detection model.

[0019] Furthermore, the collection of infrared images is achieved through drone inspections, and the photovoltaic components corresponding to each inverter are covered by multiple infrared images.

[0020] In a second aspect, the present invention proposes a photovoltaic power station anomaly detection system based on multi-source heterogeneous data, which is used to implement the above-mentioned photovoltaic power station anomaly detection method.

[0021] The beneficial effects of the present invention are:

[0022] This method uses a deep convolutional neural network (CNN) to extract infrared image features, fuses them with meteorological and inverter data at the feature layer, and combines them with a classification algorithm to achieve high-precision anomaly diagnosis for photovoltaic power plants. During classification algorithm training, the SMOTE method addresses the imbalance of dataset categories and improves model performance. This solution is adaptable to complex high-altitude environments and achieves real-time alignment of multi-source data through fuzzy time matching, significantly improving operation and maintenance efficiency and fault diagnosis accuracy while reducing operation and maintenance costs. It possesses strong versatility and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of the photovoltaic power station anomaly detection method based on multi-source heterogeneous data proposed in the present invention. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in the various embodiments of the present invention can be combined accordingly without conflicting with each other.

[0025] like Figure 1 As shown, the photovoltaic power station anomaly detection method based on multi-source heterogeneous data proposed in the present invention, when an alarm occurs in the photovoltaic power station monitoring system, multi-source heterogeneous data is collected in parallel, and the data includes unstructured data (such as infrared images) and structured data (such as meteorological sensor data and inverter data); a deep convolutional neural network (CNN) is used to extract features of the unstructured image data, and the image information is converted into a high-dimensional feature vector; the structured data (including meteorological data and inverter data) is preprocessed and cleaned, and then converted into a unified feature sequence through time matching and data integration; the high-dimensional feature vector extracted from the image data is fused with the feature sequence of the structured data at the feature level; a classification algorithm is used to identify the fused features, and the identification type is output, including but not limited to normal power generation, inverter abnormality, and photovoltaic component abnormality, so as to achieve rapid diagnosis of photovoltaic power station faults.

[0026] 1. Data Collection

[0027] The dataset in this example was collected from the Xingchuan PV Power Plant in Ganzi, Sichuan. The data collection area covers 16 inverters from a single subarray, along with corresponding weather station data and infrared imagery from the subarray. To ensure data diversity and integrity, multiple drone infrared image acquisitions were conducted on the same subarray on the same day, focusing on areas with significant defects. The data types and characteristics are described in detail below.

[0028] During operation, a photovoltaic power station generates meteorological data, inverter data, and infrared image data. The first two are structured data presented in tabular form, while the latter is unstructured image data. During data collection, the data collected for a single inverter includes current and voltage amplitudes, inverter efficiency, chamber temperature, and inverter operating status. This data is collected at a rate of 1 minute per scan. Meteorological data includes ambient temperature, irradiance, and wind speed. The data is collected at a rate of 15 minutes per scan. Infrared data is collected on-site, and each photovoltaic module contained in an inverter can be captured by three infrared images. In this embodiment, due to the limited battery life of the drone and other inspection tasks, a total of eight flights were conducted over two days, capturing infrared images of the same subarray at different time periods. Because irradiance varies at different times, the power generation conditions (brightness and darkness) of the photovoltaic panels also vary. By combining current and voltage data with meteorological data, a dataset of the same inverter over the same time period can be obtained.

[0029] During data collection at the Ganzi Power Plant, the big data monitoring platform uses a VPN for internal login, requiring manual download requests for inverter data. Weather station data is logged in via Linux, requiring downloading of relevant meteorological data from the system's webpage. For infrared image data, after receiving abnormal data, a drone is controlled to fly to the PV array area for data collection. Each inverter is associated with three infrared images. Data fusion at the power plant is performed on the same computer where data is collected. This data fusion helps power plant operations and maintenance personnel determine the cause of the abnormal data.

[0030] 2. Dataset Construction

[0031] This process involves the fusion of unstructured data (such as infrared images) and structured data (such as meteorological sensor data and inverter data). Traditional data layer fusion is to fuse the various collected data at the original data level, such as calculating variance, weighted average, etc., and then extract features and classify the fused data to obtain the final result. The disadvantage of data layer fusion is that it is heavily dependent on the acquisition of data by the sensor. If the sensor fails to obtain data or obtains inaccurate values due to a malfunction, the fusion result will be greatly deviated. Feature layer fusion is higher than the data layer fusion level. On the basis of independently completing the feature extraction of each data source, it integrates the obtained features into a single feature vector or a set input model for training and prediction. The present invention adopts a feature layer fusion method to extract features from infrared images and convert them into structured data. Through data splicing and weighted fusion, it forms a comprehensive data set that integrates multiple data features together with meteorological data and inverter data.

[0032] In the construction of the preliminary data set of this embodiment, the inverter data with a more detailed time scale is first used as the main body of the data set, and the fuzzy time matching algorithm is used to extract the most recent meteorological data; secondly, each infrared image is subjected to the CNN algorithm to extract a high-dimensional feature vector, which is structured data and together with the other two structured data constitute a CSV file. By manually constructing an artificial data set containing three results: "normal power generation", "inverter abnormality", and "photovoltaic component abnormality", it can be used as a preliminary data set for classifier training. Since the preliminary data set may have category imbalance, training the model with it will cause the learning model to be biased towards the "normal power generation" category, resulting in poor performance. Therefore, before training the photovoltaic power station anomaly detection model, the structured meteorological sensor data and inverter data are preprocessed and data enhancement operations are performed at the data level.

[0033] The preprocessing of structured data includes:

[0034] (1) Perform missing value filling, noise filtering and data standardization on meteorological data and inverter data.

[0035] Here, due to sensor failures or data transmission interruptions, meteorological and inverter data may contain missing values. To address this issue, appropriate filling methods are used. For example, for time series data, linear interpolation or the mean of the preceding and following data can be used for filling; for non-time series data, the global mean or median can be used to ensure data integrity. Secondly, sensor data may be affected by environmental interference or equipment errors during collection and transmission, resulting in noise. To eliminate the impact of noise on model training, filtering algorithms (such as moving average filtering or wavelet transform) are used to smooth the data, preserving valid information while removing abnormal fluctuations. Furthermore, meteorological and inverter data have significant differences in dimensions and numerical ranges. For example, the units of temperature, irradiance, current, and voltage are different. Directly using these data may lead to unstable model training. Therefore, data standardization is performed to scale the values of different features to the same range. Common methods include min-max scaling and z-score standardization to ensure balanced contributions of each feature to the model.

[0036] (2) Through timestamp matching, the meteorological data and inverter data are time-aligned to form a unified feature sequence.

[0037] Data enhancement operations are performed at the data level to solve the problem of imbalanced data set categories and improve the generalization ability of the model. The data enhancement operation at the data level uses oversampling, undersampling or mixed sampling methods to balance the category distribution of the data, wherein oversampling balances the category distribution by expanding the minority class samples, thereby achieving better classification results. Undersampling compresses the number of majority class samples, but when the sample size gap is large, the undersampling method may inadvertently discard a large amount of valuable information. Mixed sampling combines the ideas of oversampling and undersampling, first expanding the minority class samples, and then further cleaning the data. In order to maximize the retention of the original data of each category, this embodiment uses the SMOTE algorithm for oversampling data enhancement. Unlike simply copying minority class samples, it generates new synthetic samples by performing random interpolation between minority class samples and their adjacent minority class samples to expand the number of minority class samples and balance the data set structure.

[0038] By using a fuzzy time matching algorithm to time-align meteorological data with inverter data, and combining it with real-time collected infrared image data, real-time monitoring and rapid diagnosis of the operating status of the photovoltaic power station are achieved, significantly improving operation and maintenance efficiency. The present invention achieves efficient fusion of multi-source heterogeneous data, extracts infrared image features through a deep convolutional neural network (CNN), and combines meteorological data with inverter data to achieve feature layer fusion of multi-source heterogeneous data, solving the problem that traditional methods are difficult to handle the fusion of unstructured data and structured data, and significantly improving data utilization efficiency. The SMOTE algorithm is used to oversample minority class samples, which solves the problem of class imbalance in the data set, avoids the phenomenon of the model being biased towards majority class samples, and further improves the performance and generalization ability of the classification model.

[0039] 3. Screening classification models

[0040] The classification algorithms used include support vector machine (SVM), random forest and K-nearest neighbor (KNN) algorithms. The optimal classification model was selected by comparing the performance of different algorithms.

[0041] The classifiers were evaluated based on three criteria: precision, recall, and F1-score. The analysis showed that all classifiers achieved higher F1 scores in the "normal power generation" category than in the other two categories. This is consistent with the high proportion of real-world datasets in this category, as data augmentation still has some impact on classification. Compared to support vector machines and SVMs, the random forest method performed better. In this example, a random forest was used as the PV power plant anomaly detection model.

[0042] The present invention uses a variety of classification algorithms such as support vector machines, random forests and K-nearest neighbors to process the fused data. By comparing and optimizing the optimal model, it can accurately identify various fault types such as "normal power generation", "inverter abnormality" and "photovoltaic component abnormality", and realize the fault category judgment when abnormalities occur in the power station monitoring data. In view of the special environment of photovoltaic power stations in high-altitude areas (such as low oxygen, large temperature differences, and changeable weather), through multi-source data fusion and classification algorithms, it can effectively respond to fault diagnosis needs in complex environments and improve the robustness and adaptability of the system. Through drone inspections and automated data processing, the dependence on manual inspections is reduced, the difficulty and cost of operation and maintenance of photovoltaic power stations in high-altitude areas are reduced, and the timeliness and accuracy of fault detection are improved. This solution is not only suitable for photovoltaic power stations in high-altitude areas, but can also be extended to the operation and maintenance of photovoltaic systems in other complex environments. It has strong versatility and scalability.

[0043] This embodiment also provides a photovoltaic power station anomaly detection system based on multi-source heterogeneous data, including:

[0044] Data acquisition module, used to collect unstructured photovoltaic module infrared images, as well as structured meteorological sensor data and inverter data;

[0045] Feature extraction module, used to extract high-dimensional feature vectors of infrared images;

[0046] A data preprocessing module is used to time-align meteorological sensor data with inverter data using a fuzzy time matching algorithm;

[0047] Feature fusion module, used to combine high-dimensional feature vectors with meteorological data and inverter data aligned with the corresponding time;

[0048] The photovoltaic power station anomaly detection model is used to perform row recognition on the fused features and output the recognition results.

[0049] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment, and the implementation methods of the remaining modules will not be repeated here. The system embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Ordinary technicians in this field can understand and implement it without paying any creative work.

[0050] Embodiments of the system of the present invention can be applied to any device with data processing capabilities, such as a computer or other device. System embodiments can be implemented through software, hardware, or a combination of software and hardware. For example, a software implementation, as a logical device, is implemented by a processor of any device with data processing capabilities, reading corresponding computer program instructions from non-volatile memory into internal memory and executing them.

[0051] It should also be noted that the photovoltaic power plant anomaly detection method based on multi-source heterogeneous data in the above-mentioned embodiment can essentially be implemented via a computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the method provided in the above-mentioned embodiment. The storage medium stores a computer program. When the computer program is executed by a processor, it can implement the photovoltaic power plant anomaly detection method based on multi-source heterogeneous data in the above-mentioned embodiment.

[0052] Specifically, in the computer-readable storage medium of the above two embodiments, the stored computer program is executed by a processor to perform the steps of the above-mentioned photovoltaic power station anomaly detection method based on multi-source heterogeneous data.

[0053] It is understood that the storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage medium may be any medium capable of storing program code, such as a USB flash drive, a mobile hard drive, a magnetic disk, or an optical disk.

[0054] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0055] It should also be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division. In actual implementation, there may be other division methods, for example, multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0056] The embodiment described above is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Persons skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent substitution or equivalent transformation falls within the scope of protection of the present invention.

Claims

1. A photovoltaic power station anomaly detection method based on multi-source heterogeneous data, characterized in that: The following steps are involved: Acquiring multi-source heterogeneous data for training a photovoltaic power station anomaly detection model, the multi-source heterogeneous data comprising unstructured infrared images of photovoltaic modules acquired under normal power generation conditions, abnormal inverter conditions, and abnormal photovoltaic module conditions, as well as structured meteorological sensor data and inverter data; During model training, a fuzzy time matching algorithm is first used to time-align meteorological sensor data with inverter data, extracting high-dimensional feature vectors from infrared images. This high-dimensional feature vector is then combined with the corresponding time-aligned meteorological and inverter data to form a fusion. This combined data is then used to train the photovoltaic power station anomaly detection model. When an alarm occurs in the photovoltaic power station monitoring system, multi-source heterogeneous data are collected in parallel, and the trained photovoltaic power station anomaly detection model is used to perform row recognition on the features after fusion of multi-source heterogeneous data, and the recognition results are output.

2. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 1 is characterized in that: The meteorological data includes ambient temperature, irradiance and wind speed, and the inverter data includes working status, current and voltage amplitudes, temperature in the warehouse, and inverter efficiency.

3. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 1 is characterized in that: Before training the photovoltaic power station anomaly detection model, it also includes structured data preprocessing and data enhancement operations at the data level.

4. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 3 is characterized in that: The preprocessing of structured data includes: Perform missing value filling, noise filtering and data standardization on meteorological data and inverter data; Through timestamp matching, the meteorological data and inverter data are time-aligned to form a unified feature sequence.

5. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 3 is characterized in that: The data enhancement operation at the data level uses oversampling, undersampling or mixed sampling methods to balance the category distribution of the data.

6. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 1 is characterized in that: The photovoltaic power station anomaly detection model adopts a support vector machine, a random forest or a deep learning model; the photovoltaic power station anomaly detection model is trained together with a CNN network for extracting high-dimensional feature vectors of infrared images.

7. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 1 is characterized in that: The method of using the trained photovoltaic power station anomaly detection model to perform row recognition on the features after fusion of multi-source heterogeneous data includes: Extract high-dimensional feature vectors from unstructured image data; The structured meteorological data and inverter data are pre-processed and converted into a unified feature sequence through time matching; Fuse the high-dimensional feature vectors extracted from unstructured image data with the feature sequences of structured data at the feature level; The trained photovoltaic power station anomaly detection model is used to identify the fused features and output recognition results, which are normal power generation, inverter anomaly, and photovoltaic component anomaly.

8. The photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to claim 6 is characterized in that: The collection of infrared images is achieved through drone inspections, and the photovoltaic components corresponding to each inverter are covered by multiple infrared images.

9. A photovoltaic power station anomaly detection system based on multi-source heterogeneous data, characterized in that: include: Data acquisition module, used to collect unstructured photovoltaic module infrared images, as well as structured meteorological sensor data and inverter data; Feature extraction module, used to extract high-dimensional feature vectors of infrared images; A data preprocessing module is used to time-align meteorological sensor data with inverter data using a fuzzy time matching algorithm; Feature fusion module, used to combine high-dimensional feature vectors with meteorological data and inverter data aligned with the corresponding time; The photovoltaic power station anomaly detection model is used to perform row recognition on the fused features and output the recognition results.

10. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the photovoltaic power station anomaly detection method based on multi-source heterogeneous data according to any one of claims 1 to 8 is implemented.

Images