A method for judging heliostat failure by comparing historical heliostat data

Abstract

The application discloses a kind of to judge heliostat fault method using historical heliostat data contrast, comprising the following steps: S1, camera installation and configuration;S2, image acquisition;S3, image pre-processing;S4, image segmentation;S5, shape recognition;S6, feature extraction;S7, historical data contrast;S8, anomaly detection;S9, machine learning detects abnormal heliostat;S10, real-time monitoring and feedback.The application introduces historical data contrast and machine learning technology, significantly improves the accuracy and efficiency of heliostat fault detection, and the specific advantages include: high accuracy, combined with image processing and historical data contrast effectively identify a variety of abnormal types, reduce the misjudgment rate;High degree of automation, through automatic exposure algorithm, region growing algorithm and machine learning model realize full-process automatic detection;Wide adaptability, considering the influence of weather, season and other factors, suitable for complex and changeable operating environment.

Description

Technical Field

[0001] This invention relates to the field of concentrated solar power generation technology, and in particular to a method for determining heliostat malfunctions by comparing historical heliostat data. Background Technology

[0002] In solar thermal power generation systems, heliostats are critical components, and their normal operation directly affects the efficiency and stability of the entire system. To ensure efficient operation of heliostats, timely detection of their operating status and identification of potential faults is crucial. Traditional heliostat fault detection methods mainly rely on manual inspection or sensor-based data analysis. These methods suffer from low efficiency, limited coverage, and insufficient adaptability to complex operating conditions. In recent years, with the development of image analysis technology, using cameras to acquire and analyze images of the heliostat's reflective surface has become an effective detection method. However, existing image analysis methods typically rely on single-shot image data for judgment, lacking comparative analysis of historical data. This limits the accuracy of the detection results, especially under complex weather conditions or when multiple anomalies overlap, easily leading to misjudgments or missed detections.

[0003] Furthermore, existing technologies face numerous challenges in image acquisition and processing. For instance, the choice of camera installation location directly affects image clarity and coverage; improper installation may result in partial obstruction of heliostats or a decrease in image quality. Simultaneously, the bright spots on the heliostat's reflective surface can interfere with image recognition, affecting the accuracy of feature extraction. In the image preprocessing stage, achieving both noise removal and contrast enhancement is often difficult, easily leading to loss of detail or artifacts. In image segmentation and shape recognition, due to the discontinuity of the heliostat's reflective surface edge information and interference from complex backgrounds, traditional algorithms struggle to accurately extract the heliostat's geometric features. These problems further limit the reliability and practicality of image analysis-based fault detection methods.

[0004] On the other hand, existing technologies typically focus only on the static characteristics of heliostats, neglecting the dynamic patterns of their behavior. For example, the operating angle of a heliostat may not exceed the normal range in the short term, but abnormal trends may emerge over time, and such changes are difficult to capture through image analysis at a single time point. Furthermore, for some complex anomalies, such as curvature anomalies or coating peeling, morphological analysis alone may not fully reflect the essence of the problem. Therefore, how to combine time-series data with advanced technologies such as machine learning to achieve accurate modeling and anomaly detection of heliostat dynamic behavior has become a pressing technical challenge.

[0005] In summary, existing technologies for heliostat fault detection suffer from insufficient image acquisition and processing accuracy, limited feature extraction capabilities, and a lack of dynamic behavior analysis. To improve the accuracy and reliability of detection, there is an urgent need for a heliostat fault detection method that can comprehensively consider historical data comparison, optimize image processing workflows, and incorporate machine learning techniques. Summary of the Invention

[0006] This invention provides a method for judging heliostat malfunctions by comparing historical heliostat data, in order to solve the problems of insufficient accuracy in heliostat malfunction detection and susceptibility to environmental interference mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for determining heliostat malfunctions by comparing historical heliostat data includes the following steps: S1: Camera Installation and Configuration. The camera installation location must meet the field-of-view coverage requirements. Specifically, the camera is fixed to the top of the tower, ensuring that both the inner and outer ring heliostats are within the field of view. The coverage range is optimized by simulating the camera's field of view and installation angle. The camera bracket is made of rigid materials and designed with an adjustable structure to adapt to wind and vibration environments. After installation, a stability test is conducted to ensure that the image is not affected by bracket shift or blurring, thus maintaining recognition accuracy. Furthermore, the camera is equipped with a cooling system and filters. Protective devices are regularly inspected, cleaned, and replaced to prevent damage from strong light or high temperatures.

[0008] S2: Image Acquisition. Dynamic adjustment of exposure parameters is the core of image acquisition. Specifically, this involves developing an automatic exposure algorithm based on real-time shooting results, adjusting the aperture and exposure time to ensure the heliostat's reflective surface is clearly visible. In particular, when light spots interfere with image recognition, the system records the heliostat number causing the interference and marks it as an abnormal heliostat; if the light spots cause interference for a prolonged period, recognition is delayed until the interference is eliminated.

[0009] S3: Image Preprocessing. Denoising is achieved using Gaussian filtering or median filtering algorithms, selected based on the noise type to smooth noisy pixels while preserving the heliostat's contour information. Furthermore, an adaptive contrast enhancement algorithm is applied, dynamically adjusting the image based on the brightness and contrast characteristics of local areas to highlight the heliostat's reflecting surface and improve its distinction from the background. Brightness adjustment is performed on each heliostat's region of interest (ROI) to avoid over-adjustment that could lead to detail loss or artifacts.

[0010] S4: Image Segmentation. Edge detection uses the Canny operator, filtering noise and preserving edge pixels of the heliostat's reflective surface by setting high and low thresholds. Furthermore, reflective surface features are summarized to assist the region growing algorithm, including features such as area, shape, and location distribution. The region growing algorithm uses detected edges as seed points, merging adjacent regions to form a complete heliostat reflective surface region, and performing area and shape recognition during the merging process.

[0011] S5: Shape Recognition. The four edges are fitted using a Hough transform, and the fitting results are screened to ensure they conform to historical patterns. Furthermore, sub-pixel-level coordinate calculations are performed on the edge pixels of each edge to redetermine edge positions and improve shape recognition accuracy.

[0012] S6: Feature Extraction. Based on the fitted four edges, calculate the geometric features of the heliostat's reflecting surface, including area, perimeter, symmetry, and edge angles. Specifically, the area is calculated based on the coordinates of the intersection of the four edges, the perimeter is calculated based on the sum of the edge lengths, the symmetry is calculated based on the difference in diagonal lengths, and the edge angles are calculated based on the included angle between adjacent edges.

[0013] S7: Historical Data Comparison. Establish a historical feature database, storing normal operating status characteristics and various fault status characteristics under different time periods and weather conditions according to the heliostat number. Furthermore, use statistical analysis methods to compare current heliostat characteristics with historical data, and reasonably set thresholds to identify abnormal heliostats. In particular, consider the impact of weather and seasonal changes, and label data for different operating conditions to improve the accuracy of judgment.

[0014] S8: Anomaly Detection. Based on feature comparison results, identify conditions such as dust accumulation, coating peeling or damage, abnormal curvature, and abnormal tracking angle. Dust accumulation is manifested as a hazy substance on the reflective surface, recorded as needing cleaning; coating peeling or damage is manifested as localized color abnormalities or cracks, recorded as coating peeling or damage; abnormal curvature is manifested as irregular arrangement of sub-mirrors or differences from adjacent heliostats, recorded as abnormal curvature; abnormal tracking angle is manifested as a large angular difference between individual heliostats and surrounding heliostats over a prolonged period, recorded as abnormal tracking angle.

[0015] S9: Machine Learning for Detecting Abnormal Heliostats. A large amount of historical image data is collected and labeled. Preprocessing operations such as denoising and contrast enhancement are performed on the images to create time-series data. Further, a model suitable for processing time-series features (such as a Long Short-Term Memory (LSTM) network) is selected to capture the patterns of heliostat angle changes over time. Model performance is evaluated and optimized through small-batch training and validation sets. The trained model is then deployed to analyze newly acquired images in real time to determine whether the heliostats are in an abnormal state.

[0016] S10: Real-time monitoring and feedback. The system periodically acquires images of all heliostats in the field and uses morphological and machine learning methods for detection. When an abnormal heliostat is detected, the data is categorized and saved to the alarm history database, and the alarm information is displayed on the program interface. Furthermore, operators can view real-time images through the monitoring system and annotate and process abnormal heliostats.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention significantly improves the accuracy and efficiency of heliostat fault detection by introducing historical data comparison and machine learning technology. Specific advantages include: high precision, effectively identifying various anomaly types by combining image processing and historical data comparison, reducing the false positive rate; high automation, achieving fully automated detection through automatic exposure algorithms, region growing algorithms, and machine learning models; wide adaptability, considering the influence of weather, seasons, and other factors, suitable for complex and ever-changing operating environments; and good real-time performance, supporting real-time monitoring and feedback, facilitating timely fault detection and handling.

[0018] 2. The technical solution of this invention constructs a complete fault detection system by performing multi-level processing of heliostat reflection surface images, combining historical data comparison and machine learning models. This system not only meets the real-time monitoring needs of large-scale heliostat fields, but also has strong scalability and can be adapted to heliostat fields of different sizes and types.

[0019] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a method for determining heliostat malfunctions by comparing historical heliostat data, as proposed in this invention. Detailed Implementation

[0021] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0022] This invention relates to a method for determining heliostat malfunctions by comparing historical heliostat data. Detailed embodiments will be described in conjunction with the accompanying drawings. Figure 1 A detailed description is provided. This embodiment uses a heliostat in a solar thermal power generation system as an application scenario, and achieves accurate monitoring and anomaly detection of the heliostat's operating status through image acquisition, processing, analysis, and machine learning technologies.

[0023] Camera installation is a fundamental step in the entire system. During implementation, it's crucial to ensure the camera covers the entire reflective surface area of ​​the heliostats. The camera is installed at the top of the tower, with its height and angle optimized through simulation calculations to ensure both the inner and outer ring heliostats are within the field of view. The rigid support is constructed of high-strength steel and designed with an adjustable structure for easy adjustment of the camera's angle and position. During installation, the support is secured with bolts, and the camera angle is calibrated using a level to ensure its shooting direction is parallel to the ground and without deviation. A cooling system and filters are installed in front of the camera lens as protective measures. The cooling system uses liquid cooling circulation to lower the equipment temperature, while the filters effectively block direct sunlight. After installation, stability tests are conducted, including simulating wind impact and vibration environments, to ensure the image is not shifted or blurred due to external interference.

[0024] The core of the image acquisition phase lies in the dynamic adjustment of exposure parameters. The system employs an automatic exposure algorithm module to control the aperture and exposure time in real time. When the image is overexposed due to excessive light on the heliostat's reflective surface, the algorithm automatically reduces the aperture or shortens the exposure time; conversely, it increases the exposure time to improve brightness. For light spot interference, the system records the heliostat number causing the interference and marks it as an abnormal heliostat. If the light spot interference persists for an extended period, identification is delayed until the interference is eliminated. This process ensures that the acquired image clarity meets the requirements of subsequent processing.

[0025] The image preprocessing stage includes three parts: denoising, contrast enhancement, and brightness adjustment. A Gaussian filtering module handles Gaussian noise, while a median filtering module smooths salt-and-pepper noise. The two filtering algorithms are selected based on the noise type, preserving the heliostat's contour information while removing noisy pixels. An adaptive contrast enhancement module dynamically adjusts the image based on the brightness and contrast characteristics of local areas, highlighting the heliostat's reflective surface and improving its distinction from the background. The brightness adjustment module operates on each heliostat's region of interest (ROI) to avoid over-adjustment that could lead to detail loss or artifacts. These operations work together to ensure the image quality meets the requirements for segmentation and recognition.

[0026] In the image segmentation stage, the Canny operator edge detection module is used to detect the edge information of the heliostat's reflecting surface. The high and low threshold settings were optimized through multiple experiments. The low threshold ensures that more pixels belonging to the reflecting surface edges are included in the detection range, while the high threshold filters out false edges caused by noise or other interference factors. The region growing algorithm module uses the detected edges as seed points to merge adjacent regions into a complete heliostat reflecting surface region, performing area and shape recognition during the merging process. The Hough transform fitting module is used to fit the four edges, screening whether the fitting results conform to historical patterns. The sub-pixel calculation module performs sub-pixel level coordinate calculations for the edge pixels of each edge, redefining the edge positions to improve shape recognition accuracy.

[0027] The feature extraction stage is completed by the geometric feature calculation module, which calculates the geometric features of the heliostat's reflecting surface based on the coordinates of the intersection points of the four sides.

[0028] For example, suppose the intersection points of the four edges are (x1, y1), (x2, y2), (x3, y3), and (x4, y4), then:

[0029] Abnormal heliostats can be identified by comparing and analyzing the characteristics of the heliostat's reflecting surface area, perimeter, and edge angle with data in a historical feature database.

[0030] The historical data comparison phase is completed by the statistical analysis module. The system establishes a historical feature database to store the normal working status characteristics and various fault status characteristics under different time periods and weather conditions. When comparing the current heliostat characteristics with historical data, reasonable thresholds are set to identify abnormal heliostats. For example, dust accumulation is manifested as a hazy substance on the reflecting surface, recorded as needing cleaning; coating peeling or damage is manifested as local color abnormalities or cracks, recorded as coating peeling or damage; curvature abnormality is manifested as irregular arrangement of sub-mirrors or differences from neighboring heliostats, recorded as curvature abnormality; tracking angle abnormality is manifested as a large angle difference between individual heliostats and surrounding heliostats for a long time, recorded as tracking angle abnormality.

[0031] The machine learning-based anomaly detection stage for heliostats is implemented through an LSTM model training module. The system collects and annotates a large amount of historical image data to create time-series data. A Long Short-Term Memory (LSTM) network, suitable for processing time-series features, is selected to capture the patterns of heliostat angle changes over time. During training, the system divides the data into mini-batches for training. In each training step, the loss function is calculated based on the current batch of data, and the optimizer is used to update the model parameters. A validation set evaluates the model performance; training is stopped early or regularization measures are implemented when the validation set performance no longer improves or begins to decline. The real-time analysis module deploys the trained model to analyze newly acquired images in real time to determine whether the heliostat is in an abnormal state.

[0032] The real-time monitoring and feedback phase is implemented through the monitoring system interface. The system periodically acquires images of all heliostats and uses morphological and machine learning methods for detection. When an abnormal heliostat is detected, the data is categorized and saved to the alarm history database, and the alarm information is displayed on the program interface. Operators can view real-time images through the monitoring system and annotate and handle abnormal heliostats. The abnormal heliostat annotation module marks the abnormal location in the image and provides detailed information, facilitating operators to quickly locate and handle problems.

[0033] In this embodiment, the specific operation process of the system is as follows: First, the camera acquires full-field heliostat images at a preset frequency, and the automatic exposure algorithm module adjusts the exposure parameters in real time to ensure image clarity. The acquired images are processed by a Gaussian filtering module or a median filtering module to remove noise, and then the image quality is improved by an adaptive contrast enhancement module and a brightness adjustment module. The Canny operator edge detection module detects the edges of the heliostat's reflecting surface, and the region growing algorithm module merges adjacent regions to form a complete reflecting surface. The Hough transform fitting module fits four edge lines, and the sub-pixel calculation module improves edge accuracy. The geometric feature calculation module extracts features and compares them with data in the historical feature database, and the statistical analysis module identifies abnormal heliostats. The LSTM model training module captures the pattern of the heliostat's operating angle changing over time, and the real-time analysis module judges the heliostat's status. Finally, alarm information is displayed on the monitoring system interface, and the abnormal heliostat annotation module provides detailed annotation information.

[0034] Through the above implementation methods, the system achieves comprehensive monitoring and accurate judgment of the heliostat's operating status, significantly improving the accuracy of fault detection and reducing false alarms and missed alarms. The system can adapt to different weather and seasonal changes, covers various operating conditions, and the introduction of a time dimension enhances the understanding of the heliostat's behavioral patterns, enabling automated and intelligent real-time monitoring and reducing the cost of manual intervention.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A method for determining heliostat malfunctions by comparing historical heliostat data, characterized in that, Includes the following steps: S1. Camera installation and configuration: The camera is installed at the top of the tower to ensure that the heliostats of both the inner and outer rings are within the field of view; S2, Image Acquisition: Images are acquired through the aperture adjustment module and the automatic exposure algorithm module; S3. Image preprocessing: Use a Gaussian filter module or a median filter module to preprocess the image; S4. Image segmentation: The Canny operator edge detection module is used to segment the image and extract the heliostat reflective surface region. S5. Shape Recognition: Hough transform is used to fit the four sides, and the fitting results are screened to see if they conform to historical patterns. S6. Feature Extraction: Based on the four fitted edges, calculate the geometric features of the heliostat's reflecting surface, including area, perimeter, symmetry, and edge angles. S7. Historical data comparison: Establish a historical feature database to store the normal working status characteristics and various fault status characteristics under different time periods and weather conditions according to the heliostat number. S8. Anomaly Detection: Based on the feature comparison results, identify dust accumulation, coating peeling or damage, abnormal curvature, and abnormal tracking angle. S9. Machine Learning Anomaly Detection Heliostat: Collects and annotates a large amount of historical image data, performs denoising and contrast enhancement preprocessing on the images, and produces time series data; Select a model suitable for processing time series features to capture the pattern of heliostat operating angle changes over time; evaluate and optimize model performance through small-batch training and validation sets; deploy the trained model to perform real-time analysis on newly acquired images to determine whether the heliostat is in an abnormal state. S10. Real-time monitoring and feedback: The system periodically acquires full-field heliostat images and uses morphological and machine learning methods for detection; The real-time monitoring and feedback phase is achieved through the monitoring system interface. The system periodically collects images of the entire field heliostat and uses morphological and machine learning methods for detection. When an abnormal heliostat is detected, the data is classified and saved to the alarm history database and the alarm information is displayed on the program interface. Operators can view real-time images and mark and process abnormal heliostats through the monitoring system. The abnormal heliostat annotation module marks abnormal locations in the image and provides detailed information, making it easier for operators to quickly locate and handle problems.

2. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 1, characterized in that, The camera is fixed to the top of the tower. The bracket is made of rigid material and designed as an adjustable structure. The bracket is equipped with a cooling system and a filter.

3. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 2, characterized in that, After the camera is installed, a stability test should be conducted, and protective devices should be checked, cleaned, and replaced regularly to prevent damage to the equipment.

4. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 3, characterized in that, During image acquisition, an automatic exposure algorithm module is used to dynamically adjust the aperture size and exposure time to ensure that the heliostat's reflective surface is clearly visible.

5. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 4, characterized in that, When light spot interference occurs during image recognition, the heliostat number causing the interference is recorded and marked as an abnormal heliostat. If the light spot interference persists for a long time, the recognition is delayed until the interference is eliminated.

6. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 5, characterized in that, In the image preprocessing stage, a Gaussian filter module or a median filter module is used to remove noise, and an adaptive contrast enhancement module and a brightness adjustment module are used to improve image quality.

7. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 6, characterized in that, In the image segmentation stage, reflective surface features are summarized to assist the region growing algorithm in making judgments, including area, shape, and location distribution features.

8. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 7, characterized in that, The region growing algorithm uses detected edges as seed points to merge adjacent regions to form a complete heliostat reflector surface region, and performs area and shape recognition during the merging process.

9. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 8, characterized in that, In the shape recognition stage, the Hough transform fitting module is used to fit the four edges, and the edge positions are redetermined through the sub-pixel calculation module to improve the shape recognition accuracy.

10. The method for determining heliostat malfunctions by comparing historical heliostat data according to claim 9, characterized in that, In the historical data comparison phase, a historical feature database is established, and a statistical analysis module is used to compare the current heliostat features with historical data to identify abnormal heliostats.

Images