A method for detecting abnormal behavior in photovoltaic power generation systems based on smart grids

By partitioning the photovoltaic array and analyzing its dynamic characteristic parameters, the problem of high false alarm rate of environmental interference and equipment failure in photovoltaic power generation system is solved, achieving high accuracy and high sensitivity of fault detection and providing quantitative fault diagnosis support.

CN123122446APending Publication Date: 2026-11-17CHINA HYDROELECTRIC ENGINEERING CONSULTING GROUP CHENGDU RESEARCH HYDROELECTRIC INVESTIGATION DESIGN AND INSTITUTE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611230117.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing photovoltaic power generation systems have high false alarm and false alarm rates when distinguishing between environmental interference and equipment failure, which makes it impossible to effectively identify early equipment failures and poses safety hazards.

Method used

The photovoltaic array is divided into multiple logical partitions. Real-time operating data and local environmental parameters of each partition are collected to generate dynamic theoretical power reference values. Potential anomalies are identified by comparison and the spatial, temporal and rate of change trends of characteristic parameters are analyzed to distinguish between environmental shading and equipment failure, and the theoretical reference values ​​are dynamically adjusted.

Benefits of technology

It improves the accuracy and sensitivity of anomaly detection, enabling early identification of minor faults, reducing false alarms, ensuring stable system operation in complex environments, and providing quantitative fault diagnosis support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122446A_ABST
    Figure CN123122446A_ABST
Patent Text Reader

Abstract

This application relates to the field of photovoltaic power generation technology and discloses a method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid. The method includes: dividing the photovoltaic array into multiple logical partitions and synchronously collecting real-time operating data; acquiring local environmental parameters corresponding to the location of each logical partition and generating a dynamic theoretical power reference value by combining preset performance parameters of each logical partition; identifying potential anomalies by comparing real-time operating data with the dynamic theoretical power reference value and extracting feature parameters; analyzing the evolution trend of the feature parameters; determining the potential anomaly as an environmental shading event when it meets preset environmental interference characteristics, and correcting the dynamic theoretical power reference value of the corresponding logical partition using the real-time impact amplitude of the environmental interference; and determining the potential anomaly as an equipment failure event when it meets preset equipment anomaly characteristics, and outputting a fault alarm signal containing location information. This application significantly improves the detection capability for early-stage, hidden faults.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, and more specifically, to a method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid. Background Technology

[0002] Photovoltaic power generation systems are playing an increasingly important role in ensuring energy supply, and monitoring their operating status is crucial for grid stability and operation and maintenance efficiency.

[0003] In existing technologies, a common method for detecting anomalies in photovoltaic systems is to compare the real-time total power output of the entire photovoltaic array with a theoretical power reference value to determine whether an anomaly exists. This theoretical power is typically calculated based on readings from one or a few environmental sensors (such as a radiometer) within the site, combined with the nominal performance parameters of the modules. When the deviation between the actual power and the theoretical power exceeds a fixed percentage threshold, the system will trigger an alarm.

[0004] However, this method has revealed serious flaws in practical applications. Large-scale photovoltaic power plants cover vast areas, and rapid changes in local weather conditions (such as clouds, industrial emissions of steam, dust, etc.) can lead to significant differences in illumination conditions across different areas of the array. When a moving obstruction only affects some components, sensors located in unshaded areas cannot reflect this local change, resulting in an overestimation of the theoretical power, while the actual total power decreases due to the obstruction, thus generating numerous false alarms. To reduce the interference of these invalid alarms, maintenance personnel are often forced to raise the alarm threshold. While this reduces the false alarm rate, it significantly sacrifices the system's detection sensitivity, making it unable to identify small, persistent power deviations caused by early equipment failures (such as the failure of a single bypass diode or microcracks in the components). These ignored early fault signals worsen over time, potentially leading to hot spot effects, component burnout, or even fires, causing significant economic losses and safety hazards.

[0005] In other words, existing technologies suffer from high false alarm rates and high false negative rates due to the inability to effectively distinguish between environmental interference and equipment malfunctions. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application discloses an abnormal behavior detection method for photovoltaic power generation systems based on smart grids. This method aims to solve the problems of high false alarm rate and high false negative rate in abnormal detection caused by the inability to effectively distinguish between environmental interference and equipment failure in existing technologies.

[0007] This application discloses a method for detecting abnormal behavior in a photovoltaic power generation system based on a smart grid. This method is used for status monitoring of photovoltaic arrays and includes: Step S10: Divide the photovoltaic array into multiple logical partitions and collect real-time operating data of each logical partition simultaneously; Step S20: Obtain the local environmental parameters corresponding to the location of each logical partition, and combine them with the preset performance parameters of each logical partition to generate the dynamic theoretical power reference value of each logical partition. Step S30: Identify potential anomalies by comparing real-time operating data with dynamic theoretical power reference values, and extract characteristic parameters of potential anomalies; Step S40: Analyze the evolution trend of the feature parameters within a preset time window; The characteristic parameters include spatial distribution characteristic parameters, duration characteristic parameters, and rate of change characteristic parameters. When the evolution trend conforms to the preset environmental interference characteristics, the potential anomaly is determined to be an environmental occlusion event, and the dynamic theoretical power reference value of the corresponding logical partition is corrected by using the real-time impact amplitude of the environmental interference. When the evolution trend conforms to the preset equipment anomaly characteristics, the potential anomaly is determined to be an equipment failure event, and a fault alarm signal containing location information is output.

[0008] This technical solution refines the entire photovoltaic array into multiple logical partitions for independent analysis and introduces multi-dimensional characteristic parameters such as space, time, and rate to analyze the evolution trend of abnormal events. This enables the differentiation between environmental interference, which is mobile, instantaneous, and rapidly changing, and equipment faults, which are fixed, continuous, and slowly changing. It fundamentally solves the technical problem that traditional single threshold methods cannot distinguish between the two, achieving high accuracy and high sensitivity in anomaly detection.

[0009] In one implementation, the real-time operating data includes real-time power generation; step S10, which involves synchronously collecting the real-time operating data of each logical partition, specifically includes the following steps: Step S101: The real-time current and real-time voltage of each logical partition are obtained by the power monitoring unit corresponding to each logical partition at a preset acquisition frequency, and the real-time power generation is calculated. Step S102: Configure millisecond-level timestamps for real-time power generation and synchronize the timestamps of each logical partition through a network time protocol or a global positioning system time synchronization module.

[0010] This technical solution assigns a high-precision synchronization timestamp to the high-frequency collected data of each partition, ensuring that the data of all partitions are compared on a unified and accurate time base when analyzing the spatial movement trajectory and rate of change of abnormal events. This provides a high-quality data foundation for the accuracy of subsequent evolution trend analysis and avoids misjudgments caused by data time asynchrony.

[0011] In one implementation, step S20, which involves obtaining the local environment parameters corresponding to the locations of each logical partition, includes: Step S201: Acquire real-time environmental data through distributed environmental sensors deployed in the photovoltaic array; The real-time environmental data includes one or more of the following: solar irradiance, ambient temperature, wind speed, wind direction, precipitation status, and haze concentration. Step S202: Select a healthy and unshaded module string in the photovoltaic array as a virtual irradiance reference point, and calculate the effective irradiance of the corresponding area based on the actual power generation of the virtual irradiance reference point. Step S203: Combine real-time environmental data and effective irradiance to determine the local environmental parameters of each logical partition.

[0012] This technical solution combines distributed physical sensors and virtual sensors based on health component back-calculation to generate highly localized environmental parameters for each logical partition, significantly improving the accuracy of dynamic theoretical power reference values. This approach effectively overcomes the influence of uneven environmental conditions within large power plants, making initial power deviation calculations more accurate and reducing analysis errors introduced by inaccurate reference values ​​at the source.

[0013] In one implementation, step S30, the step of extracting the feature parameters of potential anomalies, includes: Step S301: Determine the number of logical partitions and geometric distribution areas covered by potential anomalies in the photovoltaic array, as spatial distribution characteristic parameters; Step S302: Record the duration from the triggering of a potential anomaly to its recovery to normal as a duration feature parameter; Step S303: Calculate the slope of the real-time running data at the moment of abnormal triggering, and use it as a characteristic parameter of the rate of change.

[0014] This technical solution concretizes the three abstract feature dimensions of space, time, and rate into quantifiable and explicit computational indicators. This provides concrete and operable numerical basis for subsequent trend analysis and logical judgment, enabling the entire differentiation algorithm to be executed stably and reliably.

[0015] In one implementation, the real-time operating data includes real-time power generation; the step of correcting the dynamic theoretical power reference value of the corresponding logical partition using the real-time impact magnitude of environmental disturbances includes: The decrease in real-time power generation relative to the dynamic theoretical power reference value is obtained as the real-time impact magnitude. Calculate the real-time attenuation coefficient based on the power decrease rate; The dynamic theoretical power reference value is weighted by the real-time attenuation coefficient to obtain the corrected dynamic theoretical power reference value, so that the dynamic theoretical power reference value is consistent with the trend of real-time operating data under environmental shading. The formula for calculating the real-time attenuation coefficient is: K1=1-(ΔP1 / 100%); where K1 is the real-time attenuation coefficient and ΔP1 is the power reduction rate.

[0016] This technical solution, upon identifying an environmental occlusion event, does not simply ignore it but dynamically adjusts the theoretical reference baseline. This allows the system to effectively monitor the health status of equipment in the occluded area even during environmental disturbances, preventing environmental occlusion from masking the possibility of a real equipment failure occurring simultaneously in that area, and enhancing the system's robustness under complex operating conditions.

[0017] In one embodiment, the step of outputting a fault alarm signal containing location information specifically includes the step of triggering a high-priority alarm signal when a potential abnormality is determined to be a device fault event and the power drop is within a preset small deviation range. Among them, the high-priority alarm signal carries at least the logical partition identifier to which the potential anomaly belongs, the preset fault type judgment result, and the corresponding physical location information.

[0018] This technical solution establishes a high-priority alarm channel specifically for minor power deviations caused by early, hidden faults. This ensures that fault signals that are most easily overlooked by traditional methods but have the greatest potential for harm receive the highest attention from maintenance personnel, enabling early warning and intervention for major faults and effectively preventing the escalation of faults.

[0019] In one implementation, step S40, which analyzes the evolution trend of the characteristic parameters within a preset time window, specifically includes the following steps: Step S401: If the spatial distribution characteristic parameters show that the influence range of the potential anomaly has shifted between different logical partitions, the duration characteristic parameter is less than the preset duration threshold, and the change rate characteristic parameter is greater than the preset rate threshold, then the evolution trend is determined to conform to the preset environmental interference characteristics. Step S402: If the spatial distribution characteristic parameters show that the potential anomaly's influence range is fixed within a specific logical partition, the duration characteristic parameters are greater than a preset duration threshold, and the change rate characteristic parameters are less than a preset rate threshold, then the evolution trend is determined to conform to the preset equipment anomaly characteristics.

[0020] This technical solution provides a clear and explicit set of logical rules that directly correlate phenomena observed in the physical world (cloud movement, equipment aging) with feature parameters extracted from data (spatial displacement, duration, rate of change). This constitutes the core decision-making logic of the entire method, making it possible to automatically classify and determine environmental disturbances and equipment failures.

[0021] In one implementation, it further includes: Step S50: Configure a standard reference module with a first spectral response characteristic and a special reference module with a second spectral response characteristic in each logical partition. Step S60: Monitor the power generation performance of the standard reference module and the special reference module, and calculate the spectral attenuation difference based on the power deviation between the two to characterize the impact of environmental aerosols on light quality.

[0022] This technical solution, by introducing the monitoring of spectral changes, deepens the analysis of environmental factors from the level of simple light intensity (irradiance) to the level of light quality (spectral composition). This enables the system to identify and compensate for spectral mismatch losses caused by aerosols such as haze and dust, further improving the accuracy of the theoretical power model and reducing misjudgments caused by changes in light quality.

[0023] In one implementation, it further includes: Step S70: Identify the logical partition with the best power performance and a spectral attenuation difference that is less than a preset difference threshold, and use it as a dynamic benchmark partition. Step S80: Generate a real-time environmental attenuation coefficient based on the average theoretical power of the dynamic reference partition, the actual power of the logical partition to which the potential anomaly belongs, and the spectral attenuation difference.

[0024] This technical solution establishes a dynamic and adaptive performance benchmark. Instead of relying on fixed theoretical values, the system searches in real time for the region with the best performance across the entire power plant as a reference. This ensures that the benchmark for performance comparison is always the most ideal actual state under the current operating conditions, greatly enhancing the system's adaptability to large-scale, uneven environmental changes.

[0025] In one implementation, real-time operating data includes real-time power generation; and also includes: Step S90: Based on the following formula, adaptively calibrate the dynamic theoretical power reference value using the environmental attenuation coefficient to obtain the environmental calibration theoretical power: P1'=P1 K2 and P1' are the theoretical power for environmental calibration, P1 is the reference value for dynamic theoretical power, and K2 is the environmental attenuation coefficient. Step S100: Based on the following formula, compare the real-time power generation with the theoretical power of environmental calibration, and calculate the percentage of equipment performance loss: ΔP2=(P1'-P2) / P1' 100%, ΔP2 is the percentage of equipment performance loss, and P2 is the real-time power generation; Step S110: Based on the following formula, quantify the percentage contribution of environmental factors according to the deviation between the dynamic theoretical power reference value and the environmental calibration theoretical power: ΔP3=(P1-P1') / P1 100%, ΔP3 is the percentage contribution of environmental factors.

[0026] This technical solution enables precise quantitative attribution of power loss. The system can not only determine whether a fault exists, but also clearly tell maintenance personnel, in percentage terms, how much of the current power loss is caused by environmental factors and how much is due to equipment performance degradation. This provides unprecedentedly refined data support for maintenance decisions, asset assessment, and performance optimization, elevating fault diagnosis to the level of quantitative analysis.

[0027] Compared with existing technologies, this application provides a method for detecting abnormal behavior in a photovoltaic power generation system based on a smart grid. By dividing the photovoltaic array into logical partitions and comprehensively analyzing the evolution trends of abnormal events in multiple dimensions such as space, time, and rate of change, it can effectively distinguish between mobile, transient environmental shading and fixed, persistent equipment failures. This fundamentally solves the false alarm problem caused by traditional methods that rely solely on power deviation thresholds, significantly improving the accuracy and reliability of alarms.

[0028] Secondly, this application significantly improves the detection capability for early-stage, hidden faults. By effectively eliminating false alarms caused by environmental interference, the system can set a more sensitive detection threshold, thereby reliably capturing minute power deviations caused by early faults such as bypass diode failure and component microcracks. This precise capture capability of minute signals enables early warning of major faults, buying valuable time for preventative maintenance and preventing fault escalation, thus ensuring the safety of power plant assets.

[0029] Finally, this application provides quantitative diagnostic and decision-making support. It not only locates faults but also, by introducing multiple reference modules and dynamic benchmarks, calibrates complex factors such as light quality and environmental degradation, quantitatively decomposing the total power loss into environmental factor contributions and equipment performance losses. This refined attribution analysis provides maintenance personnel with clear and quantifiable data support, enabling them to conduct more accurate fault diagnosis, performance evaluation, and maintenance strategy formulation, significantly improving the efficiency and quality of maintenance work. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the abnormal behavior detection method for a photovoltaic power generation system based on a smart grid in Embodiment 1 of this application; Figure 2 This is a flowchart illustrating the specific method for synchronously collecting real-time operational data from each logical partition in Embodiment 1 of this application. Figure 3 This is a flowchart illustrating the specific method for obtaining local environment parameters corresponding to the locations of each logical partition in Embodiment 1 of this application. Figure 4 This is a flowchart illustrating the specific method for extracting feature parameters of potential anomalies in Embodiment 1 of this application. Figure 5 This is a flowchart illustrating the specific method for step S40 of the embodiment of this application, which involves analyzing the evolution trend of characteristic parameters within a preset time window. Figure 6 This is a flowchart illustrating the abnormal behavior detection method for a photovoltaic power generation system based on a smart grid in Embodiment 2 of this application. Detailed Implementation

[0031] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0032] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] Example 1

[0034] This large-scale rooftop distributed photovoltaic (PV) power station, located within an industrial park, boasts a total installed capacity in the megawatt range, comprising thousands of photovoltaic modules forming a massive power generation array. The station's normal operation is crucial for the stability of the park's power grid and the reduction of energy costs for businesses.

[0035] However, a significant challenge lies in the fact that the chemical plant adjacent to the power plant periodically emits large amounts of white water vapor. These vapor plumes, carried by the wind, drift like clouds over the photovoltaic array, causing temporary, mobile shading of some photovoltaic modules and resulting in a sudden drop in power generation. Simultaneously, any photovoltaic module or its auxiliary equipment, such as the bypass diodes in the junction box, may experience performance degradation or sudden failures. These initial failures often only cause very weak but continuous power loss. Traditional monitoring systems cannot effectively distinguish between these two completely different power reduction events. They either become desensitized to false alarms due to frequent environmental interference or completely ignore early equipment failure signals indicating significant hidden dangers due to excessively high alarm thresholds. This introduces enormous uncertainty and risk to the safe and stable operation of the power plant.

[0036] In this regard, such as Figure 1 As shown, this application provides a method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid, used for monitoring the state of a photovoltaic array. The method includes the following steps: S10. Divide the photovoltaic array into multiple logical partitions and collect real-time operating data of each logical partition simultaneously; S20. Obtain the local environmental parameters corresponding to the location of each logical partition, and combine them with the preset performance parameters of each logical partition to generate the dynamic theoretical power reference value of each logical partition. S30. Identify potential anomalies by comparing real-time operating data with dynamic theoretical power reference values, and extract characteristic parameters of potential anomalies, including spatial distribution characteristic parameters, duration characteristic parameters, and rate of change characteristic parameters. S40. Analyze the evolution trend of characteristic parameters within a preset time window; Specifically, when the evolution trend conforms to the preset environmental interference characteristics, the potential anomaly is determined to be an environmental occlusion event, and the dynamic theoretical power reference value of the corresponding logical partition is corrected by using the real-time impact amplitude of the environmental interference; when the evolution trend conforms to the preset equipment anomaly characteristics, the potential anomaly is determined to be an equipment fault event, and a fault alarm signal containing location information is output.

[0037] It should be noted that the logical partitioning in this embodiment refers to the virtual division of a physically continuous large photovoltaic array into multiple independent, manageable units with clearly defined geographical locations at the data monitoring and analysis level, according to certain rules. This division does not change the physical electrical connections, but rather represents a grid-based processing method for management and analysis.

[0038] It should be noted that the dynamic theoretical power reference value in this embodiment refers to an expected power generation value that is calculated independently for each logical partition and changes in real time over time. It is no longer a single theoretical value for the entire power plant, but a highly customized performance benchmark that comprehensively considers the specific location of each partition, the real-time local environmental conditions at that location, and the inherent performance parameters of the components within that partition.

[0039] It should be noted that, in this embodiment, a potential anomaly refers to a discrepancy exceeding a preset initial deviation threshold between the real-time power generation of any logical partition and its corresponding dynamic theoretical power reference value during the initial comparison. In this case, the system only marks it as an event requiring further analysis, rather than immediately classifying it as a fault.

[0040] It should be noted that the feature parameters in this embodiment refer to a set of quantitative indicators extracted to deeply analyze the root causes of potential anomalies and describe the essential attributes of the anomaly. Spatial distribution feature parameters describe the geographical extent of the anomaly, such as whether it is fixed or mobile; duration feature parameters describe the length of the anomaly on the time axis, such as whether it is instantaneous or persistent; and rate of change feature parameters describe the drastic degree of power change when the anomaly occurs, such as whether it is abrupt or gradual.

[0041] It should be noted that the evolution trend in this embodiment refers to the comprehensive analysis of the continuous changes of the above-mentioned multiple characteristic parameters within a preset time window (e.g., several minutes or several hours). The purpose is to depict a complete dynamic profile of the abnormal event, such as whether the scope of an abnormality has migrated between different logical partitions, whether its duration is long or short, and whether its occurrence and disappearance are fast or slow.

[0042] It should be noted that the real-time impact amplitude in this embodiment refers to the real-time deviation amplitude of the real-time power generation of the logical partition relative to the dynamic theoretical power reference value of the partition during the anomaly detection process of the photovoltaic power generation system. It is used to quantify the instantaneous impact of environmental interference on the power generation capacity of the current partition and is the core calculation parameter for realizing the adaptive correction of the dynamic theoretical power reference value.

[0043] Specifically, such as Figure 2 As shown, in one embodiment, the real-time operating data includes real-time power generation; in step S10, the step of synchronously collecting the real-time operating data of each logical partition specifically includes the following steps: Step S101: The real-time current and real-time voltage of each logical partition are obtained by the power monitoring unit corresponding to each logical partition at a preset acquisition frequency, and the real-time power generation is calculated. Step S102: Configure millisecond-level timestamps for real-time power generation and synchronize the timestamps of each logical partition through a network time protocol or a global positioning system time synchronization module.

[0044] In the above steps, by assigning a high-precision synchronization timestamp to the high-frequency collected data of each partition, it is ensured that when analyzing the spatial movement trajectory and rate of change of abnormal events, the data of all partitions are compared on a unified and accurate time base. This provides a high-quality data foundation for the accuracy of subsequent evolution trend analysis and avoids misjudgment caused by data time asynchrony.

[0045] Specifically, such as Figure 3 As shown, in one embodiment, step S20, the step of obtaining the local environment parameters corresponding to the location of each logical partition, includes: Step S201: Acquire real-time environmental data through distributed environmental sensors deployed in the photovoltaic array; The real-time environmental data includes one or more of the following: solar irradiance, ambient temperature, wind speed, wind direction, precipitation status, and haze concentration. Step S202: Select a healthy and unshaded module string in the photovoltaic array as a virtual irradiance reference point, and calculate the effective irradiance of the corresponding area based on the actual power generation of the virtual irradiance reference point. Step S203: Combine real-time environmental data and effective irradiance to determine the local environmental parameters of each logical partition.

[0046] The above steps combine distributed physical sensors and virtual sensors based on health component back-calculation to generate highly localized environmental parameters for each logical partition, significantly improving the accuracy of dynamic theoretical power reference values. This approach effectively overcomes the influence of uneven environmental conditions within large power plants, making the initial power deviation calculation more accurate and reducing analysis errors introduced by inaccurate reference values ​​at the source.

[0047] Specifically, such as Figure 4 As shown, in one embodiment, step S30, the step of extracting feature parameters of potential anomalies, includes: Step S301: Determine the number of logical partitions and geometric distribution areas covered by potential anomalies in the photovoltaic array, as spatial distribution characteristic parameters; Step S302: Record the duration from the triggering of a potential anomaly to its recovery to normal as a duration feature parameter; Step S303: Calculate the slope of the real-time running data at the moment of abnormal triggering, and use it as a characteristic parameter of the rate of change.

[0048] The above steps concretize the three abstract feature dimensions of space, time, and rate into quantifiable and explicit computational indicators. This provides specific and operable numerical basis for subsequent trend analysis and logical judgment, enabling the entire differentiation algorithm to be executed stably and reliably.

[0049] Real-time operating data includes real-time power generation; the steps for correcting the dynamic theoretical power reference value of the corresponding logical partition using the real-time impact of environmental disturbances include: The decrease in real-time power generation relative to the dynamic theoretical power reference value is obtained as the real-time impact magnitude. Calculate the real-time attenuation coefficient based on the power decrease rate; The dynamic theoretical power reference value is weighted by the real-time attenuation coefficient to obtain the corrected dynamic theoretical power reference value, so that the dynamic theoretical power reference value is consistent with the trend of real-time operating data under environmental shading. The formula for calculating the real-time attenuation coefficient is: K1=1-(ΔP1 / 100%); where K1 is the real-time attenuation coefficient and ΔP1 is the power reduction rate.

[0050] In the above steps, after identifying an environmental occlusion event, the system does not simply ignore the event, but dynamically adjusts the theoretical reference baseline. This allows the system to effectively monitor the health status of equipment in the occluded area even during environmental disturbances, preventing environmental occlusion from masking the possibility of a real equipment failure occurring simultaneously in that area, and enhancing the system's robustness under complex operating conditions.

[0051] Specifically, in one embodiment, the step of outputting a fault alarm signal containing location information specifically includes the step of triggering a high-priority alarm signal when the potential abnormality is determined to be a device fault event and the power drop is within a preset small deviation range. Among them, the high-priority alarm signal carries at least the logical partition identifier to which the potential anomaly belongs, the preset fault type judgment result, and the corresponding physical location information.

[0052] In the steps described above, a high-priority alarm channel was specifically established for minor power deviations caused by early, hidden faults. This ensures that fault signals that are most easily overlooked by traditional methods but have the greatest potential for harm receive the highest attention from maintenance personnel, enabling early warning and intervention for major faults and effectively preventing the escalation of faults.

[0053] Specifically, such as Figure 5 As shown, in one embodiment, step S40, which analyzes the evolution trend of feature parameters within a preset time window, specifically includes the following steps: Step S401: If the spatial distribution characteristic parameters show that the influence range of the potential anomaly has shifted between different logical partitions, the duration characteristic parameter is less than the preset duration threshold, and the change rate characteristic parameter is greater than the preset rate threshold, then the evolution trend is determined to conform to the preset environmental interference characteristics. Step S402: If the spatial distribution characteristic parameters show that the potential anomaly's influence range is fixed within a specific logical partition, the duration characteristic parameters are greater than a preset duration threshold, and the change rate characteristic parameters are less than a preset rate threshold, then the evolution trend is determined to conform to the preset equipment anomaly characteristics.

[0054] The steps described above provide a clear and explicit set of logical rules that directly link observed phenomena from the physical world (cloud movement, equipment aging) with feature parameters extracted from the data (spatial displacement, duration, rate of change). This constitutes the core decision-making logic of the entire method, making it possible to automatically classify and determine environmental disturbances and equipment failures.

[0055] Example 2

[0056] The differences between Example 2 and Example 1 are as follows: Specifically, such as Figure 6 As shown, in one embodiment, it further includes: Step S50: Configure a standard reference module with a first spectral response characteristic and a special reference module with a second spectral response characteristic in each logical partition. Step S60: Monitor the power generation performance of the standard reference module and the special reference module, and calculate the spectral attenuation difference based on the power deviation between the two to characterize the impact of environmental aerosols on light quality.

[0057] In the steps described above, by introducing the monitoring of spectral changes, the analysis of environmental factors is deepened from the level of simple light intensity (irradiance) to the level of light quality (spectral composition). This enables the system to identify and compensate for spectral mismatch losses caused by aerosols such as haze and dust, further improving the accuracy of the theoretical power model and reducing misjudgments caused by changes in light quality.

[0058] Specifically, such as Figure 1 As shown, in one embodiment, it further includes: Step S70: Identify the logical partition with the best power performance and a spectral attenuation difference that is less than a preset difference threshold, and use it as a dynamic benchmark partition. Step S80: Generate a real-time environmental attenuation coefficient based on the average theoretical power of the dynamic reference partition, the actual power of the logical partition to which the potential anomaly belongs, and the spectral attenuation difference.

[0059] The above steps establish a dynamic and adaptive performance benchmark. The system no longer relies on fixed theoretical values, but instead searches for the best-performing region in the entire power plant in real time as a reference. This ensures that the benchmark for performance comparison is always the most ideal actual state under the current operating conditions, greatly enhancing the system's adaptability to large-scale, uneven environmental changes.

[0060] Specifically, such as Figure 1 As shown, in one embodiment, the real-time operating data includes real-time power generation; it also includes: Step S90: Based on the following formula, adaptively calibrate the dynamic theoretical power reference value using the environmental attenuation coefficient to obtain the environmental calibration theoretical power: P1'=P1 K2 and P1' are the theoretical power for environmental calibration, P1 is the reference value for dynamic theoretical power, and K2 is the environmental attenuation coefficient. Step S100: Based on the following formula, compare the real-time power generation with the theoretical power of environmental calibration, and calculate the percentage of equipment performance loss: ΔP2=(P1'-P2) / P1' 100%, ΔP2 is the percentage of equipment performance loss, and P2 is the real-time power generation; Step S110: Based on the following formula, quantify the percentage contribution of environmental factors according to the deviation between the dynamic theoretical power reference value and the environmental calibration theoretical power: ΔP3=(P1-P1') / P1 100%, ΔP3 is the percentage contribution of environmental factors.

[0061] The above steps enable precise quantitative attribution of power loss. Ultimately, the system can not only determine whether a fault exists, but also clearly tell maintenance personnel, in percentage terms, how much of the current power loss is caused by environmental factors and how much is due to equipment performance degradation. This provides unprecedentedly refined data support for maintenance decisions, asset assessment, and performance optimization, elevating fault diagnosis to the level of quantitative analysis.

[0062] It should be noted that the other steps in Example 2 are the same as those in Example 1, and will not be described in detail.

[0063] The following describes a more specific field application example of this application: First, the photovoltaic array is divided into multiple logical partitions. This step is fundamental for achieving refined monitoring. The granularity of the partitioning can be flexibly set according to the actual situation of the power plant and monitoring requirements. One feasible partitioning method is based on the physical electrical topology of the power plant. For example, a large photovoltaic power plant typically contains multiple inverters, each connected to several photovoltaic combiner boxes, and each combiner box collects the DC power from multiple photovoltaic module strings. In this case, each photovoltaic module string, each combiner box, or the entire subarray managed by each inverter can be defined as a logical partition. Partitioning by combiner box is a relatively balanced choice, providing sufficient location resolution to pinpoint problems without overloading the data processing burden due to too many partitions.

[0064] For example, for the aforementioned 1MW rooftop power station, assuming it consists of 40 combiner boxes, the entire array can be divided into 40 logical partitions, and each partition can be assigned a unique identifier, such as "partition A01" to "partition A40". At the same time, the precise physical coordinate range of each partition on the roof can be recorded in the system.

[0065] Another partitioning method is virtual grid partitioning, which ignores electrical connections and directly overlays a virtual grid onto the entire photovoltaic array's plan view. For example, a 100m by 100m array can be divided into 100 logical partitions of 10m by 10m. This method is more adaptable to power plants with irregular physical layouts. After partitioning, the system will synchronously collect real-time operational data from all logical partitions.

[0066] Next, the system needs to generate a unique dynamic theoretical power reference value for each logical partition. The core of this step lies in acknowledging and quantifying the environmental inhomogeneity within the photovoltaic power plant. Traditionally, a radiometer reading located in a corner of the power plant is used to represent the overall illumination conditions, which is clearly inaccurate. In this application, the system acquires local environmental parameters corresponding to the location of each logical partition. These parameters include at least solar irradiance and module surface temperature. After acquiring these parameters, the system calls a preset power generation performance model, which incorporates the factory calibration parameters (i.e., preset performance parameters) of the photovoltaic modules within the corresponding logical partition, such as nominal power, voltage temperature coefficient, current temperature coefficient, and fill factor.

[0067] By inputting real-time local environmental parameters into this performance model, the power value that this logical partition should theoretically output under specific environmental conditions can be calculated, i.e., the dynamic theoretical power reference value.

[0068] For example, for zone A05 with a total installed capacity of 30kW, at 10:00 AM, the system measured the solar irradiance in its area to be 850 watts per square meter, and the module surface temperature to be 45 degrees Celsius. Combining this with the performance parameters of the modules within zone A05, assuming the general temperature coefficient of the crystalline silicon modules in zone A05 is -0.38% / ℃, the standard test temperature is 25°C, and the standard irradiance is 1000 watts per square meter, then the irradiance ratio is 850 / 1000 = 0.85, the temperature difference ΔT = 45 - 25 = 20℃, the power loss is 20 × 0.38% = 7.6%, and the temperature correction factor is 1 - 0.076 = 0.924. Therefore, its dynamic theoretical power reference value is calculated to be 30 × 0.924 × 0.85 = 23.5 kilowatts. In this way, each zone has its own fair and dynamically changing evaluation standard.

[0069] Then, the system continuously compares the real-time operating data (mainly real-time power generation) of each logical partition with its dynamic theoretical power reference value. Once the actual power of a partition is lower than its theoretical reference value and exceeds an initially set, relatively sensitive deviation threshold (e.g., 5%), the system marks this event as a potential anomaly and immediately begins to extract the characteristic parameters of the anomaly.

[0070] Specifically, the system will perform the following operations to extract feature parameters: First, the spatial distribution characteristic parameters are determined. The system records the identifiers and geometric locations of all logical partitions currently exhibiting potential anomalies. If only one partition exhibits an anomaly, the spatial distribution characteristic is "fixed in a single partition." If multiple adjacent partitions exhibit anomalies simultaneously, the spatial distribution characteristic is "clustered in a certain area." The system also continuously monitors the data; if the set of partitions exhibiting anomalies changes within the next few minutes, for example, from {A05, A06} to {A06, A07}, the system updates the spatial distribution characteristic to "displacement occurs between partitions."

[0071] Second, the duration characteristic parameter is recorded. From the moment a partition is marked as potentially abnormal, the system starts a timer for that abnormality. If the partition's power generation returns to the normal range of the theoretical reference value after a period of time, the timer stops, and the recorded time is the duration characteristic parameter. If the abnormality persists, the timer will continue to run.

[0072] Third, the system calculates the rate of change characteristic parameters. It analyzes power data for several seconds before and after the moment a potential anomaly is triggered. By calculating the slope of the power curve at that point in time, the rate of power decrease is obtained.

[0073] For example, if the power drops rapidly from 20 kilowatts to 10 kilowatts in 1 second, the rate of change is very large; while if the power drops slowly from 20 kilowatts to 19.8 kilowatts in 1 hour, the rate of change is very small.

[0074] After extracting these feature parameters, the system enters the most crucial analysis and judgment stage. The system analyzes the evolution trend of these feature parameters within a preset time window.

[0075] A specific decision-making scenario is as follows: Scenario 1: Determination of Environmental Interference Events. Assume that at 10:05 AM, the system detects a sudden drop in power from the theoretical value of 23 kW to 15 kW in zones A05 and A06, with an extremely rapid rate of change. The system marks this as a potential anomaly. Within the next minute, A05 and A06 return to normal, but the adjacent zones A07 and A08 experience the same sudden power drop. One minute later, A09 and A10 are affected. The system analyzes the evolution trend of this series of events: spatial distribution parameters show a clear shift in the area of ​​impact from west to east; the duration of the impact on each zone is very short, only about one minute; and the rate of change at each anomaly is very high. This evolution trend perfectly matches the characteristics of movement, brevity, and drastic change. Based on this, the system determines that the root cause of this series of potential anomalies is an environmental obstruction event, i.e., a steam cloud from a neighboring plant drifting across the array. In this case, the system will not only not issue a fault alarm, but will also utilize this information. It dynamically and temporarily lowers the dynamic theoretical power reference value of the affected zone by 35% based on the actual power drop (e.g., from 23 kW to 15 kW, a decrease of about 35%). The significance of this is that even during the period of shading, if the zone experiences additional power drops due to equipment failure, the system can still detect new anomalies based on this revised, more realistic reference value.

[0076] Scenario 2: Determination of Equipment Failure Events. Assume the system detects that the actual power of partition B12 is consistently 2% lower than its dynamic theoretical power reference value, indicating a small but consistent deviation. This deviation remains stable for hours or even days. The system analyzes its evolution trend: spatial distribution parameters show the abnormal impact area is firmly locked within partition B12 and has never moved to other partitions; duration parameters show the anomaly has persisted for a very long time, far exceeding normal environmental fluctuation cycles; its initial rate of change may be small, or a tiny step. This evolution trend perfectly matches the characteristics of being fixed, persistent, and gradual. Based on this, the system determines the root cause of this potential anomaly to be an equipment failure event, most likely a microcrack in a component within the partition or a short circuit in a bypass diode. At this point, the system immediately generates a fault alarm signal containing precise location information (partition B12 and its physical coordinates), and sets a high priority based on the nature of the fault (e.g., a small deviation may correspond to a high-risk hotspot effect). The system then notifies maintenance personnel via a pop-up window in the monitoring center interface, SMS, or email, requiring a focused investigation of partition B12.

[0077] Using the above methods, the system can, like an experienced expert, accurately distinguish between fluctuating environmental interference and fixed equipment faults by observing the dynamic behavioral characteristics of abnormal events. This greatly reduces the false alarm rate while ensuring that no early fault signals are missed, thus solving the core technical challenges mentioned in the background section.

[0078] Furthermore, real-time operational data includes real-time power generation; the steps for synchronously collecting real-time operational data from each logical partition include: The real-time current and real-time voltage of each logical partition are acquired at a preset acquisition frequency through the power monitoring unit corresponding to each logical partition, and the real-time power generation is calculated. Configure millisecond-level timestamps for real-time power generation and synchronize the timestamps of each logical partition through a network time protocol or a global positioning system time synchronization module.

[0079] To ensure the accuracy of the analysis of abnormal evolution trends, especially the precise calculation of the spatial distribution's movement trajectory and rate of change, high-quality data acquisition is an essential prerequisite. In one specific embodiment of this application, the details of data acquisition are as follows: An intelligent power monitoring unit is installed at each logical partition (e.g., each combiner box). This unit incorporates high-precision current and voltage sensors, capable of acquiring the total current and total voltage of all component strings gathered in that partition at a preset, high acquisition frequency, e.g., once per second. After acquiring the real-time current and voltage, the microprocessor inside the monitoring unit immediately multiplies them to calculate the instantaneous real-time power generation of that logical partition. Simply acquiring the power value is insufficient; precise time synchronization is crucial. Therefore, the entire power plant monitoring system deploys a master time server. This server can acquire standard time in one of two ways. One way is by connecting to a public network time protocol server via the internet to continuously calibrate its own system clock. Another, more reliable method is to equip the server with a Global Positioning System (GPS) timing module to directly obtain nanosecond-level precise time from satellite signals. Then, all power monitoring units within the facility periodically synchronize their time with this master time server via the internal LAN, ensuring that the clock errors of all units remain within milliseconds. Once each monitoring unit calculates its real-time power generation, it immediately appends a high-precision timestamp, such as "2023-10-26 10:05:31.254". This way, when the central server aggregates data from all logical partitions, it can be certain that all data marked with the same timestamp was indeed collected at the same physical moment. This precise time synchronization is crucial for analyzing rapidly moving obstructions. For example, if an obstruction moves from partition A to partition B within 2 seconds, the system can only accurately calculate its speed and distinguish it from equipment malfunctions if the timestamps of the data in partitions A and B are sufficiently accurate.

[0080] Furthermore, the steps for obtaining the local environment parameters corresponding to the location of each logical partition include: Real-time environmental data is acquired through distributed environmental sensors deployed in the photovoltaic array; the real-time environmental data includes one or more of the following: solar irradiance, ambient temperature, wind speed, wind direction, precipitation status, and haze concentration. The healthy and unshaded component strings in the photovoltaic array are selected as virtual irradiance reference points, and the effective irradiance of the corresponding area is calculated by back-calculating the actual power generation of the virtual irradiance reference points. By combining real-time environmental data and effective irradiance, the local environmental parameters of each logical partition are determined.

[0081] To generate accurate dynamic theoretical power reference values ​​for each logical partition, obtaining precise local environmental parameters is crucial. This application provides a multi-method fusion approach to solve this problem.

[0082] First, the system utilizes distributed physical sensors deployed within the photovoltaic array. Unlike traditional power plants that install only a single weather station in a corner, this approach advocates strategically deploying multiple small environmental sensor groups within a large array, based on its area and layout. For example, in a 100m by 100m array, one sensor set can be deployed at the center and one at each of the four corners, for a total of five sets. Each sensor set can measure total solar radiation, ambient temperature, wind speed, and wind direction. For specific environments, additional sensors such as precipitation sensors (rain gauges) and haze concentration sensors (e.g., PM2.5 detectors) can be added. These physical sensors provide a macroscopic benchmark for environmental parameters in different areas of the array.

[0083] However, even deploying multiple physical sensors cannot cover the microenvironment of every logical partition. Solar irradiance, in particular, can vary dramatically in certain areas. To address this, this application introduces virtual irradiance reference points. The basic principle is that a healthy, clean, and unshaded photovoltaic module itself acts as an excellent irradiance meter. The system analyzes historical data to select several photovoltaic module strings with consistently stable performance and excellent power generation from the entire photovoltaic array, marking them as virtual irradiance reference points. These reference points are typically chosen in open locations that are not easily shaded by fixed shadows. During system operation, the actual output power, voltage, and surface temperature of these reference point module strings, measured by attached temperature sensors, are continuously monitored. Based on the performance parameters (IV curve, temperature coefficient, etc.) of the module under standard test conditions calibrated at the factory, the system can use a reverse calculation model to accurately deduce the solar irradiance value required to generate that power output, based on the current actual output power and temperature. For example, a reference point array has an actual output power of 2.5 kW at the current temperature. Using a reverse model, the effective irradiance it receives is calculated to be 890 W / m². This value more accurately reflects the lighting conditions at that location than the 850 W / m² reading from a physical irradiance meter 50 meters away.

[0084] Finally, the system fuses macroscopic data from physical sensors with microscopic data from virtual reference points. It constructs a real-time environmental parameter cloud map covering the entire photovoltaic array. For solar irradiance, the system uses spatial interpolation algorithms (such as inverse distance weighted interpolation or Kriging interpolation) based on readings from all physical and virtual sensors to calculate the most reliable effective irradiance value for the geometric center of each logical partition. For parameters with relatively stable changes, such as temperature, a partition-weighted average can be used. In this way, each logical partition obtains highly localized environmental parameters tailored to it, laying a solid foundation for the accurate calculation of subsequent theoretical power.

[0085] Furthermore, the steps for extracting feature parameters of potential anomalies include: Determine the number of logical partitions and geometric distribution areas covered by potential anomalies in the photovoltaic array, as spatial distribution characteristic parameters; Record the duration from the triggering and identification of a potential anomaly to its recovery to normal, and use this as a duration feature parameter; Calculate the slope of the real-time running data at the moment of anomaly triggering, and use it as a characteristic parameter of the rate of change.

[0086] Once a potential anomaly is identified, the system needs to extract feature parameters that describe its nature. This application defines the extraction process for these three core feature parameters in detail.

[0087] To determine the spatial distribution characteristic parameters, the system maintains a two-dimensional logical partition map that perfectly matches the physical layout of the photovoltaic array. When the power deviation of one or more partitions exceeds a threshold, the system immediately marks the corresponding partition on the map. For example, if the system detects a simultaneous decrease in power in partitions A05, A06, B05, and B06, these four partitions form a 2x2 rectangular area on the map. The system records the shape, area (covering the four logical partitions), and centroid coordinates of this affected geometric area. In the following time, if the marked area changes, for example, moving to A06, A07, B06, or B07, the system calculates the displacement vector between the centroids of the old and new areas. This information—the set of affected partitions, their geometry, area, centroid location, and their trajectory over time—together constitutes the complete spatial distribution characteristic parameters.

[0088] The process of recording duration characteristics is very intuitive. The system maintains an independent timer for each logical partition marked as potentially abnormal. When the power of partition A05 first falls below the normal range at 10:05:31, its corresponding timer starts counting from 0. If the partition's power returns to normal at 10:06:45, the timer stops, and the recorded duration is 1 minute and 14 seconds. If the partition's power fails to recover by sunset, its duration will be recorded as a very large value, or it will be marked as a persistent abnormality.

[0089] To calculate the rate of change characteristic parameters, the system retrieves a short sequence of high-frequency power data before and after the moment the anomaly is triggered.

[0090] For example, data points P(t-2), P(t-1), P(t), P(t+1), and P(t+2) are collected in seconds, where t is the trigger time. The simplest calculation method is to calculate the backward difference of the trigger time, i.e., [P(t)-P(t-1)] / Δt, where Δt is the sampling interval (1 second). To improve noise immunity, more robust calculation methods can be used, such as linear fitting of this data and taking the slope of the fitted line as the rate of change. A sudden power drop caused by cloud cover may have an absolute rate of change of several kilowatts per second. A slow performance degradation caused by component aging or dirt accumulation may have a rate of change of only a few tenths of a kilowatt per hour, almost approaching zero. By accurately quantifying these three parameters, the system provides reliable input for subsequent intelligent decision-making.

[0091] Furthermore, the steps for analyzing the evolution trend of characteristic parameters within a preset time window include: If the spatial distribution characteristic parameters show that the influence range of the potential anomaly shifts between different logical partitions, the duration characteristic parameter is less than the preset duration threshold, and the change rate characteristic parameter is greater than the preset rate threshold, then the evolution trend is determined to conform to the preset environmental interference characteristics. If the spatial distribution characteristic parameters show that the impact range of a potential anomaly is fixed within a specific logical partition, the duration characteristic parameter is greater than a preset duration threshold, and the change rate characteristic parameter is less than a preset rate threshold, then the evolution trend is determined to conform to the preset equipment anomaly characteristics.

[0092] This step is the core decision-making logic of the entire detection method, transforming the previously extracted quantitative feature parameters into a final judgment on the nature of the event. The system internally has a set of rules based on expert knowledge.

[0093] Rule 1 is used to identify environmental disturbances. The specific logic of this rule is as follows: if the spatial distribution characteristics of a potential anomaly indicate that its area of ​​influence is geographically mobile (e.g., the centroid coordinates of its affected area change continuously over time), and the duration of its stay in any single area is less than a preset duration threshold (e.g., 5 minutes), while the absolute value of its rate of change is greater than a preset rate threshold (e.g., 1 kW / s), then the system determines that the evolution trend of this anomaly conforms to the characteristics of environmental disturbances. The threshold settings here have physical significance: the 5-minute duration threshold can distinguish between fleeting clouds and long-staying cumulonimbus clouds; the 1 kW / s rate threshold can distinguish between rapidly passing shadows and naturally slow changes in sunlight intensity.

[0094] Rule 2 is used to identify equipment anomalies. The specific logic of this rule is as follows: if the spatial distribution characteristics of a potential anomaly indicate that its impact range is fixed, i.e., always confined to one or a few specific, unchanging logical partitions, and its duration characteristic parameter is greater than a preset duration threshold (e.g., 60 minutes), while the absolute value of its rate of change characteristic parameter is less than a preset rate threshold (e.g., 0.1 kW / s), then the system determines that the evolution trend of this anomaly conforms to the characteristics of an equipment anomaly. The threshold settings here are equally crucial: the 60-minute duration ensures that all short-term environmental fluctuations and power grid disturbances are excluded; the fixed spatial location directly points to the physical equipment at that location; and the extremely small rate of change is consistent with the characteristics of fault modes such as slow performance degradation or minor internal short circuits.

[0095] By combining these two core rules, the system can automatically and reliably categorize complex real-world events into the correct fault basket. It's important to note that these thresholds are not static; they can be optimized and adaptively adjusted based on the typical climate characteristics and historical data of the power plant's location during the initial system deployment phase to achieve the best classification results.

[0096] Based on the above embodiments, this application also provides some preferred technical solutions to further improve the robustness of the detection method and the accuracy of the diagnosis.

[0097] Furthermore, real-time operational data includes real-time power generation; the steps for correcting the dynamic theoretical power reference value of the corresponding logical partition using the real-time impact of environmental disturbances include: The decrease in real-time power generation relative to the dynamic theoretical power reference value is obtained as the real-time impact magnitude. Calculate the real-time attenuation coefficient based on the power decrease rate; The dynamic theoretical power reference value is weighted by the real-time attenuation coefficient to obtain the corrected dynamic theoretical power reference value, so that the dynamic theoretical power reference value is consistent with the trend of real-time operating data under environmental shading. The formula for calculating the real-time attenuation coefficient is: K1=1-(ΔP1 / 100%); where K1 is the real-time attenuation coefficient and ΔP1 is the power reduction rate.

[0098] The purpose of this step is to enable the system to intelligently adapt to this temporary adverse condition after confirming the occurrence of an environmental occlusion event, rather than simply treating the data from that period as invalid. This ensures that the system's monitoring of equipment health is not interrupted even during periods of occlusion.

[0099] The specific implementation process is as follows: Taking the aforementioned steam cloud obstruction scenario as an example, when the system determines that the power drop in partition A05 is due to environmental interference, it will immediately perform the following calculations. Assume that before being obstructed, the dynamic theoretical power reference value of partition A05 is 23.5 kilowatts, and at the moment of obstruction, its real-time power generation drops to 15.0 kilowatts.

[0100] First, the system obtains the power decrease magnitude ΔP1. This magnitude is a percentage value, calculated as: ΔP1 = (Theoretical Power - Actual Power) / Theoretical Power 100% = (23.5 - 15.0) / 23.5 100% ≈ 36.2%.

[0101] The system then calculates the real-time attenuation coefficient K1 based on this decrease. According to the formula K1=1-(ΔP1 / 100%), we can obtain K1=1-(36.2% / 100%)=1-0.362=0.638. This coefficient K1 intuitively represents that under the current shading conditions, the power generation capacity of this zone is only equivalent to 63.8% of that under unshading conditions.

[0102] Finally, the system uses this real-time attenuation coefficient K1 to correct the dynamic theoretical power reference value for this partition. The corrected reference value equals the original reference value. K1 = 23.5 kilowatts 0.638 ≈ 15.0 kW. The system will use this 15.0 kW as the new performance benchmark for partition A05 throughout the entire duration of the shading event. The significant advantage of this approach is that if, at the same time cloud shading occurs within partition A05, a component experiences a connector loosening failure, resulting in an additional 0.5 kW drop in power beyond the 15.0 kW, the system can still accurately detect this new deviation from the corrected reference value and identify it as a superimposed equipment failure event. This greatly enhances the system's robustness in monitoring complex, multi-factor-driven conditions.

[0103] Furthermore, the step of outputting a fault alarm signal containing location information includes: When a potential anomaly is determined to be a device malfunction event and the power drop is within a preset small deviation range, a high-priority alarm signal is triggered. High-priority alarm signals should carry at least the logical partition identifier of the potential anomaly, the preset fault type judgment result, and the corresponding physical location information.

[0104] This step aims to classify the risks of different types of equipment failures, ensuring that the most insidious but potentially most dangerous early-stage failures receive the highest level of attention from operations and maintenance personnel.

[0105] The specific implementation process is as follows: Taking the scenario of suspected bypass diode failure in partition B12 as an example, the system has determined through evolution trend analysis that the persistent power deviation is a device failure event.

[0106] Next, the system will assess the severity of the fault. It will calculate the power drop; assuming the theoretical power of partition B12 is 25.0 kW, while the actual power is stable at 24.5 kW, then the power drop will be (25.0 - 24.5) / 25.0. 100% = 2%. The system has a preset small deviation range, such as [0.5%, 5%]. Since 2% falls exactly within this range, the system determines that this is a typical early hidden fault characteristic. Although this fault has a negligible impact on the total power generation at present, it may hide serious safety hazards such as hot spot effects.

[0107] Therefore, the system will trigger a high-priority alarm signal. This alarm signal is not a simple "B12 partition anomaly" message, but a structured, information-rich alert. On the monitoring screen in the operations and maintenance center, it may appear as a prominent red pop-up window, the content of which includes at least: Logical partition identifier: B12.

[0108] The preset fault type assessment result is: "Suspected bypass diode breakdown or early component degradation, with hot spot risk." This assessment result can be based on a built-in fault knowledge base, which associates small deviations of varying magnitudes with possible fault modes.

[0109] The corresponding physical location information is: "Roof of Plant 3, East Array, 4th Row, Component String under Combiner Box B12". This physical location information is bound to the logical partition identifier during system initialization, allowing maintenance personnel to locate the site directly without consulting drawings.

[0110] In this way, the system frees up maintenance personnel from a large number of non-urgent general alarms, and accurately guides them to deal with serious faults that are in their infancy and require immediate intervention, thus realizing a shift from passive response to proactive prevention.

[0111] Furthermore, the method also includes: Each logical partition is configured with a standard reference module having a first spectral response characteristic and a special reference module having a second spectral response characteristic. The power generation performance of the monitoring standard reference module and the special reference module is compared, and the spectral attenuation difference is calculated based on the power deviation between the two to characterize the impact of environmental aerosols on light quality.

[0112] This step introduces a deeper method for quantifying environmental factors, aiming to address the impact of spectral changes caused by aerosols such as haze and dust on power generation performance, thereby further improving the accuracy of the theoretical power model.

[0113] In a specific implementation, in some representative logical partitions, in addition to conventional photovoltaic modules, two small reference modules are installed.

[0114] The standard reference module uses the same cell technology as the main array in the power plant, such as monocrystalline PERC silicon cells. Its spectral response characteristics (i.e., power generation efficiency for different wavelengths of light) are consistent with the main array, and it can be regarded as a miniaturized, performance-calibrated standard component.

[0115] The special reference module employs a battery technology with significantly different spectral response characteristics from the main array, such as amorphous silicon (a-Si) or cadmium telluride (CdTe). Taking amorphous silicon as an example, it is more sensitive to short-wavelength light such as blue-violet light, while the crystalline silicon cells of the main array are more sensitive to long-wavelength light such as red light and near-infrared light.

[0116] The system simultaneously and with high precision monitors the instantaneous power output of both reference modules. On a clear, cloudless day with clean air, the power output ratio of these two modules will be a relatively stable baseline value, for example, P(special) / P(standard) = 0.85. When there is a large amount of smog or dust in the air, these aerosol particles scatter sunlight, especially short-wavelength blue-violet light, leading to an increase in the proportion of long-wavelength light in the solar spectrum reaching the ground. Under this spectral change, the power generation efficiency of the amorphous silicon special reference module, which is more sensitive to blue light, will decrease more significantly than that of the crystalline silicon standard reference module, causing their power ratio to change, for example, drop to 0.80.

[0117] The system calculates the deviation of the current power ratio from the reference value, i.e., (0.85-0.80) / 0.85≈5.9%, to obtain a quantified spectral attenuation difference. The magnitude of this difference directly reflects the degree of influence of atmospheric aerosols on light quality, providing a key input parameter for subsequent more refined environmental factor calibration.

[0118] Furthermore, the method also includes: Identify logical partitions with spectral attenuation differences less than a preset difference threshold and optimal power performance, and use them as dynamic benchmark partitions; The real-time environmental attenuation coefficient is generated based on the average theoretical power of the dynamic baseline partition, the actual power of the logical partition to which the potential anomaly belongs, and the spectral attenuation difference.

[0119] The purpose of this step is to find a standard in real time throughout the entire power plant, namely the area that performs best under the current operating conditions, and use this as a benchmark to evaluate the performance of other areas more fairly, thereby generating a decay coefficient that can reflect the comprehensive environmental impact.

[0120] The specific implementation process is as follows: At any given time, the system will perform a full-field scan to determine the dynamic baseline partition.

[0121] First, the system will filter out all logical zones whose spectral attenuation difference is less than a preset difference threshold (e.g., 2%). The significance of this filtering step is to prioritize areas experiencing the cleanest air and whose spectra are closest to ideal, while excluding zones affected by severe local smog.

[0122] Then, among these spectrally selected partitions, the system further compares their performance, specifically the ratio of their actual power generation to their dynamic theoretical power reference value. The partition with the ratio closest to 100% (or the highest) is selected as the dynamic benchmark partition for the current time window. For example, partition C30 has a spectral attenuation difference of only 1% and its actual power reaches 98% of its theoretical power, making it the best performing partition among all candidates; therefore, C30 becomes the dynamic benchmark for this moment.

[0123] Once a dynamic baseline partition is selected, the system can generate a real-time environmental attenuation coefficient K2 for any other partition (such as partition B12, which has potential anomalies). This coefficient aims to quantify the sum of all factors, not related to the device itself, that cause power degradation in partition B12, including general atmospheric attenuation, local shading, and pollution. One way to generate this coefficient is to take the performance of dynamic baseline partition C30 (98%) as the best achievable performance under the current environment. Then, for partition B12, its environmental attenuation coefficient K2 can be estimated based on this best performance and corrected by incorporating local environmental information of B12 (such as spectral attenuation difference, presence of local shading, etc.) to obtain an attenuation coefficient that best represents the overall environmental conditions of B12 at present. For example, if B12 is locally unshaded and has a good spectrum, then the current environmental attenuation coefficient K2 for B12 is 0.98.

[0124] Furthermore, the real-time operational data includes real-time power generation; the method also includes: Based on the following formula, the dynamic theoretical power reference value is adaptively calibrated using the environmental attenuation coefficient to obtain the environmental calibration theoretical power: P1'=P1 K2 and P1' are the theoretical power for environmental calibration, P1 is the reference value for dynamic theoretical power, and K2 is the environmental attenuation coefficient. Based on the following formula, by comparing the real-time power generation with the theoretical power calibrated by the environment, the percentage of equipment performance loss is calculated: ΔP2=(P1'-P2) / P1' 100%, ΔP2 is the percentage of equipment performance loss, and P2 is the real-time power generation; Based on the following formula, the percentage contribution of environmental factors is quantified according to the deviation between the dynamic theoretical power reference value and the environmental calibration theoretical power: ΔP3=(P1-P1') / P1' 100%, ΔP3 is the percentage contribution of environmental factors.

[0125] This step is the final output of the entire diagnostic process. Through a series of calculations, it provides unprecedented quantitative basis for operation and maintenance decisions.

[0126] Continuing with the previous scenario, suppose the system is analyzing partition B12, and we know: The original dynamic theoretical power reference value for partition B12 is P1 = 25.0 kW (this is calculated under ideal, no-attenuation conditions).

[0127] The real-time power generation of partition B12 is P2 = 24.0 kilowatts.

[0128] The real-time environmental attenuation coefficient K2, calculated through the aforementioned steps, is 0.98 for partition B12 (representing a 2% overall environmental loss).

[0129] The first step is to calculate the environmental calibration theoretical power P1'.

[0130] P1'=P1 K2 = 25.0 kW 0.98 = 24.5 kilowatts.

[0131] The physical meaning of P1' is: if the device in partition B12 is in perfect health, then under the current overall environmental conditions, it should be able to generate 24.5 kilowatts of power.

[0132] The second step is to calculate the percentage of equipment performance loss, ΔP2.

[0133] ΔP2=(P1'-P2) / P1' 100% = (24.5 - 24.0) / 24.5 100% ≈ 2.04%.

[0134] The calculation of ΔP2 compares the power that should have been generated with the power that was actually generated. The difference between them can only be attributed to the performance issues of the equipment itself. The calculation results show that the equipment in partition B12 has a performance loss of approximately 2.04%.

[0135] The third step is to quantify the percentage contribution of environmental factors, ΔP3.

[0136] ΔP3=(P1-P1') / P1 100% = (25.0 - 24.5) / 25.0 100% = 2.0%.

[0137] The calculation of ΔP3 compares the theoretical power under ideal conditions with the power that should be generated after considering environmental factors. The difference between them is entirely due to environmental factors. The calculation results show that current environmental factors have led to a 2.0% power loss.

[0138] Ultimately, the system generates a remarkably clear diagnostic report: "The current total power deviation of partition B12 is (25.0-24.0) / 25.0=4.0%. Analysis shows that environmental factors (atmospheric decay, temperature, etc.) contribute 2.0% of the loss, while the equipment's own performance degradation contributes 2.04%." This quantitative attribution report completely resolves the pain point of unclear fault causes in traditional methods, enabling targeted maintenance work and greatly improving efficiency and accuracy.

[0139] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid, characterized in that, The abnormal behavior detection method for photovoltaic power generation systems based on smart grids is used for status monitoring of photovoltaic arrays, and includes: Step S10: Divide the photovoltaic array into multiple logical partitions and collect real-time operating data of each logical partition simultaneously; Step S20: Obtain the local environment parameters corresponding to the location of each logical partition, and combine them with the preset performance parameters of each logical partition to generate the dynamic theoretical power reference value of each logical partition. Step S30: Identify potential anomalies by comparing the real-time operating data with the dynamic theoretical power reference value, and extract the feature parameters of the potential anomalies; Step S40: Analyze the evolution trend of the feature parameters within a preset time window; The characteristic parameters include spatial distribution characteristic parameters, duration characteristic parameters, and change rate characteristic parameters. When the evolution trend conforms to the preset environmental interference characteristics, the potential anomaly is determined to be an environmental occlusion event, and the dynamic theoretical power reference value of the corresponding logical partition is corrected using the real-time impact amplitude of the environmental interference. When the evolution trend conforms to the preset equipment anomaly characteristics, the potential anomaly is determined to be an equipment failure event, and a fault alarm signal containing location information is output.

2. The smart grid based photovoltaic power system abnormal behavior detection method of claim 1, wherein, The real-time operating data includes real-time power generation; in step S10, the step of synchronously collecting the real-time operating data of each logical partition specifically includes the following steps: Step S101: The real-time current and real-time voltage of each logical partition are acquired by the power monitoring unit corresponding to each logical partition at a preset acquisition frequency, and the real-time power generation is calculated. Step S102: Configure a millisecond-level timestamp for the real-time power generation and synchronize the timestamps of each logical partition through a network time protocol or a global positioning system time synchronization module. 3.The smart grid based photovoltaic power generation system abnormal behavior detection method of claim 1, wherein, In step S20, the step of obtaining the local environment parameters corresponding to each of the logical partition locations includes: Step S201: Acquire real-time environmental data through distributed environmental sensors deployed in the photovoltaic array; The real-time environmental data includes one or more of the following: solar irradiance, ambient temperature, wind speed, wind direction, precipitation status, and haze concentration. Step S202: Select a healthy and unshaded component string in the photovoltaic array as a virtual irradiance reference point, and calculate the effective irradiance of the corresponding area based on the actual power generation of the virtual irradiance reference point. Step S203: Combine the real-time environmental data and the effective irradiance to determine the local environmental parameters of each logical partition.

4. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 1, characterized in that, In step S30, the step of extracting the feature parameters of the potential anomaly includes: Step S301: Determine the number of logical partitions and the geometric distribution area covered by the potential anomaly in the photovoltaic array, as the spatial distribution characteristic parameter; Step S302: Record the duration from the triggering of the potential anomaly to its recovery to normal as a duration feature parameter; Step S303: Calculate the slope of the real-time running data at the time of the abnormal trigger, and use it as the change rate characteristic parameter.

5. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 4, characterized in that, The real-time operating data includes real-time power generation; the step of correcting the dynamic theoretical power reference value of the corresponding logical partition using the real-time impact of environmental interference includes: The decrease in real-time power generation relative to the dynamic theoretical power reference value is obtained as the real-time impact magnitude. Calculate the real-time attenuation coefficient based on the power decrease rate; The dynamic theoretical power reference value is weighted and calculated using the real-time attenuation coefficient to obtain a corrected dynamic theoretical power reference value, so that the dynamic theoretical power reference value is consistent with the trend of real-time operating data under environmental shading. The formula for calculating the real-time attenuation coefficient is: K1=1-(ΔP1 / 100%); where K1 is the real-time attenuation coefficient and ΔP1 is the power reduction magnitude.

6. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 5, characterized in that, The step of outputting a fault alarm signal containing location information specifically includes the step of triggering a high-priority alarm signal when the potential abnormality is determined to be a device fault event and the power drop is within a preset small deviation range. The high-priority alarm signal carries at least the logical partition identifier to which the potential anomaly belongs, the preset fault type judgment result, and the corresponding physical location information.

7. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 1, characterized in that, Step S40, which analyzes the evolution trend of the feature parameters within a preset time window, specifically includes the following steps: Step S401: If the spatial distribution characteristic parameter shows that the influence range of the potential anomaly has shifted between different logical partitions, the duration characteristic parameter is less than a preset duration threshold, and the change rate characteristic parameter is greater than a preset rate threshold, then the evolution trend is determined to conform to the preset environmental interference characteristics. Step S402: If the spatial distribution characteristic parameter shows that the influence range of the potential anomaly is fixed in a specific logical partition, the duration characteristic parameter is greater than a preset duration threshold, and the change rate characteristic parameter is less than a preset rate threshold, then the evolution trend is determined to conform to the preset equipment anomaly characteristics.

8. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 1, characterized in that, Also includes: Step S50: Configure a standard reference module with a first spectral response characteristic and a special reference module with a second spectral response characteristic in each of the logical partitions. Step S60: Monitor the power generation performance of the standard reference module and the special reference module, and calculate the spectral attenuation difference based on the power deviation between the two to characterize the impact of environmental aerosols on light quality.

9. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 8, characterized in that, Also includes: Step S70: Identify the logical partition with the spectral attenuation difference less than a preset difference threshold and the best power performance, and use it as a dynamic benchmark partition; Step S80: Generate a real-time environmental attenuation coefficient based on the average theoretical power of the dynamic reference partition, the actual power of the logical partition to which the potential anomaly belongs, and the spectral attenuation difference.

10. The method for detecting abnormal behavior of a photovoltaic power generation system based on a smart grid according to claim 9, characterized in that, The real-time operating data includes real-time power generation; and also includes: Step S90: Based on the following formula, adaptively calibrate the dynamic theoretical power reference value using the environmental attenuation coefficient to obtain the environmental calibration theoretical power: P1'=P1 K2, P1' is the environmental calibration theoretical power, P1 is the dynamic theoretical power reference value, and K2 is the environmental attenuation coefficient; Step S100: Based on the following formula, compare the real-time power generation with the environmental calibration theoretical power to calculate the percentage of equipment performance loss: ΔP2=(P1'-P2) / P1' 100%, ΔP2 is the percentage of equipment performance loss, and P2 is the real-time power generation; Step S110: Based on the following formula, quantify the percentage contribution of environmental factors according to the deviation between the dynamic theoretical power reference value and the environmental calibration theoretical power: ΔP3=(P1-P1') / P1 100%, ΔP3 is the percentage contribution of environmental factors.