Photovoltaic power station photovoltaic module operation parameter anomaly detection method based on edge end

Abstract

The invention discloses a photovoltaic power station photovoltaic module operation parameter anomaly detection method based on an edge end. The method comprises the following specific steps: S1, data acquisition; s2, data preprocessing; s3, feature extraction: extracting features beneficial to anomaly detection from a large amount of original data; the characteristics comprise indexes related to component output power and environmental parameters; s4, performing unsupervised anomaly detection; s5, quantitative sensing distillation; light-weight edge end deployment: aiming at a large-scale photovoltaic system, carrying out light-weight processing on the model through a quantitative perception knowledge mechanism so as to enable the model to adapt to a resource-constrained environment of an edge end; through multi-dimensional feature fusion and in combination with time relevance, the accuracy of anomaly detection can be remarkably improved when the operation parameters of the photovoltaic power station assembly are processed; according to the method, a minimum and maximum strategy is designed to amplify the distinguishability of the associated anomaly and the normal condition, so that an advanced result is obtained on an unsupervised time sequence anomaly detection benchmark.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power stations, and in particular relates to an edge-end-based method for detecting abnormal operation parameters of photovoltaic components in a photovoltaic power station. Background Art

[0002] Traditional photovoltaic power station anomaly detection is mainly based on electrical parameters, image detection, infrared detection of water vapor transmission rate, and weather condition fault perception. It is monitored through conventional electrical parameters, image acquisition equipment conducts a large number of inspections and image detection, infrared detection modules detect water vapor transmission rate, and fault perception is performed based on sunny day data.

[0003] These traditional anomaly detection methods often rely on a large number of hardware devices, specific weather conditions, or use traditional machine learning methods (such as support vector machines, recurrent neural networks, and graph neural networks) on electrical parameters to analyze the abnormal conditions of photovoltaic components. However, these methods have obvious limitations in terms of deployment costs, external environmental changes, edge deployment, and time information of parameters. Although anomaly detection technology based on deep learning has developed rapidly in recent years, there are few studies on real-time detection in large-scale photovoltaic power station environments. In particular, there are still many problems with existing technologies when facing the complex environment of large-scale photovoltaic power stations:

[0004] Extremely high cost: Both image detection and water vapor transmission rate detection rely on the deployment of a large number of hardware devices, which makes the cost of abnormal detection of large-scale photovoltaic module operating parameters extremely high.

[0005] Difficulties in edge deployment: The resource environment at the edge of the PV system is limited, the performance of the equipment itself is not high, and large-scale models are difficult to deploy.

[0006] Strict requirements for the environment: When image detection and weather condition failures are detected, it is necessary to ensure that the solar irradiance, air visibility, etc. during the detection period meet the detection requirements.

[0007] Low detection accuracy: When monitoring anomalies through conventional current and voltage methods, it is difficult to fully capture the various fault characteristics of photovoltaic modules, especially when facing complex environments (such as wind, sand, shading, etc.).

[0008] Anomalies are difficult to distinguish: Some existing methods focus on learning point representations through carefully designed recurrent networks and self-supervising through reconstruction or autoregressive tasks. However, anomalies are rare, and point-by-point representations are less informative for complex temporal patterns and may be dominated by normal time points, making anomalies difficult to distinguish.

[0009] Low real-time performance: Existing methods based on electrical parameter detection all compare the electrical parameters within a fixed unit time with their estimated values. The learned information is still limited to a single time point and has low real-time performance.

[0010] Inefficiency: In recent years, some large Transformer-based models have achieved general results in time series anomaly detection methods, but they are inefficient when processing long time series and have obvious deficiencies in extracting long-range time series dependent features. Summary of the invention

[0011] The purpose of the present invention is to provide an edge-based photovoltaic power station photovoltaic module operating parameter anomaly detection method, by processing the multi-dimensional operating parameters of photovoltaic, designing a minimax strategy to amplify the distinguishability of normal and abnormal, so as to achieve efficient detection of photovoltaic module operating parameter anomalies.

[0012] To achieve the above purpose, the present invention provides the following technical solution: a method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal, the specific steps of which include:

[0013] S1: Data collection:

[0014] Collect real-time operating data from various components of the PV power plant. The data is uploaded to the central control platform in real time through sensors or monitoring systems to form a data set for subsequent analysis.

[0015] S2: Data preprocessing:

[0016] In the case of missing, noisy or inconsistent data during the collection process, the data needs to be cleaned and preprocessed. The missing parts of the data are filled by interpolation and mean filling, and the noise in the data is removed by filtering algorithm to ensure the stability and accuracy of the data. Finally, to ensure that each feature is compared at the same scale, the data is normalized.

[0017] S3: Feature extraction:

[0018] Extract features that are useful for anomaly detection from large amounts of raw data; features include metrics related to component output power, as well as environmental parameters;

[0019] S4: Unsupervised Anomaly Detection:

[0020] The time series is trained for unsupervised anomaly detection through the Light-Anomaly model; the model uses the minimax strategy to amplify the distinguishability of associated anomalies and normal situations, and the Gated-SSM Block module can capture the long-range dependency of information;

[0021] S5: Quantization-aware distillation:

[0022] OneBit quantize the previously trained unsupervised anomaly detection model;

[0023] S6: Edge deployment:

[0024] The lightweight model after quantization-aware distillation is converted into the PyTorchMobile format suitable for edge devices before deployment and integrated with other components of the PV system to ensure that the model can detect anomalies on real-time data on edge devices;

[0025] S7: Parameter anomaly detection:

[0026] After the lightweight model with quantitative perception is deployed at the edge of the photovoltaic system, anomaly detection is performed on the real-time data. Once an abnormality is detected, the edge device immediately takes response measures and sends an abnormal notification to alert the operation and maintenance personnel.

[0027] Preferably, in S1, the data includes electrical characteristic operating parameters and environmental information of the photovoltaic module, the electrical characteristic operating parameters include voltage, current and power; the environmental information includes temperature, wind speed and radiation.

[0028] Preferably, in S3, the indicators related to the component output power include but are not limited to the ratio of output power to expected power, power fluctuation amplitude and power stability.

[0029] Preferably, in S3, the environmental parameters include but are not limited to the effect of temperature on the efficiency of photovoltaic modules and output power fluctuations caused by changes in radiation intensity.

[0030] Preferably, in S6, other components include but are not limited to a power generation monitoring system and a control system.

[0031] Preferably, in S5, the weight parameters of the model are quantized with 1 bit to reduce the memory usage and computing power consumption of the model; then, through quantization-aware knowledge distillation, the knowledge of the basic model is transferred to the 1-bit quantized model, so that it can be applied to the resource-constrained environment at the edge of the photovoltaic system, and ensure the balance between the model's spatiotemporal efficiency and detection performance.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The method of the present invention has the following advantages: Lighter edge deployment: For large-scale photovoltaic systems, the model is lightweighted through a quantized perception knowledge mechanism to adapt it to the resource-constrained environment of the edge.

[0034] The detection accuracy is significantly improved: through multi-dimensional feature fusion and combined with time correlation, the present invention can significantly improve the accuracy of anomaly detection when processing the operating parameters of photovoltaic power station components.

[0035] Efficient anomaly distinction: Since anomalies are very rare and their associations are mainly concentrated at adjacent time points, they are difficult to identify. This paper designs a minimax strategy to amplify the distinguishability of associated anomalies from normal situations, achieving advanced results on the unsupervised time series anomaly detection benchmark.

[0036] High computational efficiency: Compared with traditional recurrent neural network (RNN) and convolutional neural network (CNN) methods, the Light-Anomaly designed in the present invention has linear time complexity, reduces the occupation of computing resources, and can maintain a low computing cost.

[0037] Scalability and adaptability: As the scale of photovoltaic power plants increases, the dimensions of monitored parameters and environmental data will also increase. By integrating multiple data sources, the system can more easily handle larger-scale power plants and has stronger adaptability to power plants of different sizes and different environmental conditions. The fusion model can be effectively deployed in a variety of environments, improving the scalability of the system.

[0038] Save operation and maintenance costs: Through more accurate anomaly detection and fault diagnosis, the system can reduce ineffective maintenance caused by false alarms or missed alarms; at the same time, potential faults can be discovered in advance for preventive maintenance, reducing maintenance costs and extending equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] See also Figure 1 The present invention provides a technical solution: a method for detecting abnormal operation parameters of photovoltaic modules in photovoltaic power stations based on the edge. The method draws on the idea of ​​Anomaly-Transformer, mainly processes multi-dimensional operation parameters of photovoltaics, and designs a minimax strategy to amplify the distinguishability between normal and abnormal, so as to achieve efficient detection of abnormal operation parameters of photovoltaic modules.

[0042] When deploying the model, in order to achieve high real-time performance, low latency, and as few computing resources as possible, and considering the resource-constrained environment at the edge of the photovoltaic system, the present invention uses quantitative perception and knowledge distillation to further compress and lightweight the model, so that it can be applied to the resource-constrained environment at the edge of the photovoltaic system, and ensure the balance between the model's spatiotemporal efficiency and detection performance. The specific steps are as follows:

[0043] S1: Data collection:

[0044] Collecting real-time operating data from various components of the PV power station will involve the electrical characteristics operating parameters of the PV modules, as well as environmental information. The electrical characteristics operating parameters include voltage, current and power; environmental information includes temperature, wind speed and radiation. These data will be uploaded to the central control platform in real time through sensors or monitoring systems to form a data set for subsequent analysis.

[0045] S2: Data preprocessing:

[0046] Since the data may be missing, noisy or inconsistent during the collection process, the data is cleaned and preprocessed, the missing parts of the data are filled by interpolation and mean filling, and the noise in the data is removed by filtering algorithm to ensure the stability and accuracy of the data. Finally, to ensure that each feature is compared at the same scale, the data is normalized.

[0047] S3: Feature extraction:

[0048] Extract features that are helpful for anomaly detection from a large amount of raw data. These features include indicators related to the output power of the component, such as the ratio of output power to expected power, power fluctuation amplitude, and power stability. At the same time, environmental parameters may also have an impact on the operating status of the component, such as the impact of temperature on the efficiency of photovoltaic components and the output power fluctuations that may be caused by changes in radiation intensity, which are all important features.

[0049] S4: Unsupervised Anomaly Detection:

[0050] The time series is trained for unsupervised anomaly detection through the Light-Anomaly model. The model designs a minimax strategy to amplify the distinguishability of associated anomalies and normal situations, and through the Gated-SSM Block module, it can capture the long-range dependency of information and achieve efficiency far higher than the Transformer model.

[0051] S5: Quantization-aware distillation:

[0052] The previously trained unsupervised anomaly detection model is quantized by OneBit. This method uses 1 bit to quantize the model's weight parameters, greatly reducing the model's memory usage and computing power consumption. Then, through quantization-aware knowledge distillation, the knowledge of the basic model (i.e., the teacher model) is transferred to the 1-bit quantized model (i.e., the student model), so that it can be applied to the resource-constrained environment at the edge of the photovoltaic system and ensure the balance between the model's spatiotemporal efficiency and detection performance.

[0053] S6: Edge deployment:

[0054] The lightweight model after quantization-aware distillation is converted into the PyTorchMobile format suitable for edge devices before deployment, and integrated with other components of the PV system (such as power generation monitoring system, control system, etc.) to ensure that the model can detect anomalies in real-time data on edge devices.

[0055] S7: Parameter anomaly detection:

[0056] After the lightweight model with quantitative perception is deployed at the edge of the photovoltaic system, anomaly detection is performed on the real-time data. Once an abnormality is detected, the edge device immediately takes response measures and sends an abnormal notification to remind the operation and maintenance personnel, so as to timely discover potential faults and perform preventive maintenance, reduce maintenance costs, and extend the service life of the equipment.

[0057] Although the embodiments of the present invention have been shown and described, as detailed above, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal, characterized in that: The specific steps include: S1: Data collection: Collect real-time operating data from various components of the PV power plant. The data is uploaded to the central control platform in real time through sensors or monitoring systems to form a data set for subsequent analysis. S2: Data preprocessing: In the case of missing, noisy or inconsistent data during the collection process, the data needs to be cleaned and preprocessed. The missing parts of the data are filled by interpolation and mean filling, and the noise in the data is removed by filtering algorithm to ensure the stability and accuracy of the data. Finally, to ensure that each feature is compared at the same scale, the data is normalized. S3: Feature extraction: Extract features that are useful for anomaly detection from large amounts of raw data; features include metrics related to component output power, as well as environmental parameters; S4: Unsupervised Anomaly Detection: The time series is trained for unsupervised anomaly detection through the Light-Anomaly model; the model uses the minimax strategy to amplify the distinguishability of associated anomalies and normal situations, and the Gated-SSM Block module can capture the long-range dependency of information; S5: Quantization-aware distillation: OneBit quantize the previously trained unsupervised anomaly detection model; S6: Edge deployment: The lightweight model after quantization-aware distillation is converted into PyTorch Mobile format suitable for edge devices before deployment and integrated with other components of the PV system to ensure that the model can detect anomalies on real-time data on edge devices; S7: Parameter anomaly detection: After the lightweight model with quantitative perception is deployed at the edge of the photovoltaic system, anomaly detection is performed on the real-time data. Once an abnormality is detected, the edge device immediately takes response measures and sends an abnormal notification to alert the operation and maintenance personnel.

2. The method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal according to claim 1 is characterized in that: In S1, the data includes electrical characteristic operating parameters of the photovoltaic module and environmental information. The electrical characteristic operating parameters include voltage, current and power; the environmental information includes temperature, wind speed and radiation.

3. The method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal according to claim 1 is characterized in that: In S3, the indicators related to the component output power include but are not limited to the ratio of output power to expected power, power fluctuation amplitude and power stability.

4. The method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal according to claim 1, characterized in that: In S3, the environmental parameters include but are not limited to the impact of temperature on the efficiency of photovoltaic modules and output power fluctuations caused by changes in radiation intensity.

5. The method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal according to claim 1, characterized in that: In S6, other components include but are not limited to a power generation monitoring system and a control system.

6. The method for detecting abnormal operation parameters of photovoltaic modules in a photovoltaic power station based on an edge terminal according to claim 1, characterized in that: In S5, the weight parameters of the model are quantized with 1 bit to reduce the memory usage and computing power consumption of the model; then, through quantization-aware knowledge distillation, the knowledge of the basic model is transferred to the 1-bit quantized model, so that it can be applied to the resource-constrained environment at the edge of the photovoltaic system and ensure the balance between the model's spatiotemporal efficiency and detection performance.

Images