A photovoltaic power station operation data processing method and system

Abstract

The application relates to a photovoltaic power station operation data processing method and system. The method comprises the following steps: collecting operation data of each inverter in a photovoltaic power station, and obtaining the time stamp of each operation data; setting an event window, aggregating the operation data from the same inverter in the event window, calculating the statistical characteristics of the aggregated data, and constructing the feature portrait of each inverter in the current event window based on the statistical characteristics; constructing a behavior mode library for each type of inverter, and updating the behavior mode library according to the actual operation data and fault feedback of each type of inverter; comparing the real-time extracted feature portrait with the early wear mode, and if the feature portrait is similar to the early wear mode, an early warning signal is sent out; the early warning signal is graded according to the comparison accuracy of the feature portrait and the early wear mode, the potential severity of the early wear mode, and the abnormal duration, and a graded warning notification is output according to the result of grading.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic data processing, in particular to a photovoltaic power station operation data processing method and system. BACKGROUND

[0002] Massive operation data processing (collection, cleaning, analysis) of large photovoltaic power stations is the key to ensuring stable and efficient power generation of the power station. The initial judgment standard of the data cleaning link is based on the performance parameters of the mainstream photovoltaic modules and inverters at the initial stage of the power station and the environmental conditions set at that time, which can effectively identify equipment failures and collection errors at that time to ensure data accuracy.

[0003] With the operation of the power station, the management party replaced the new model inverters of different manufacturers in batches to improve efficiency and follow the development of technology. These inverters are significantly different from the old model in design, material, and control method. For example, the MPPT technology is better in weak light environment, and the heat dissipation performance is better at high temperature. Even if it is within its normal operating parameter range, its data fluctuation pattern and performance curve have deviated from the initial benchmark. This directly leads to the failure of the original cleaning standard: the normal high-power data of the new model inverter may be misjudged as abnormal and excluded due to exceeding the old benchmark upper limit; and the early performance degradation of its internal key components (such as IGBT modules) will be masked by the old standard and cannot be warned of potential failure risks - such degradation may cause serious failure over time and shorten the service life of the equipment.

[0004] More complex is that the new model inverters are not a single model, but multiple models from 2-3 suppliers, and the optimal working interval, environmental response mode, and diagnostic parameter reporting format of each model are different; at the same time, the power station also deploys multiple brands of data collectors, which differ in reporting frequency (1 minute / 5 minutes), data precision, and communication protocol (Modbus RTU / TCP, etc.), leading to deviations in processing multi-source heterogeneous data by the original timestamp synchronization program, causing data "misplacement", further exacerbating the difficulty of performance comparison and fault diagnosis.

[0005] This series of problems has a chain effect on operation and management: it is difficult for operation and maintenance personnel to distinguish between normal operating conditions of the new model inverter and early component degradation signals, forcing them to increase on-site troubleshooting workload, making it impossible to implement efficiency optimization and accurate predictive maintenance of the power station, and possibly missing the best maintenance window; for the group, the uncertainty of the data of the power station makes it impossible to accurately assess the health status and power generation potential of the power station, thereby affecting cross-regional power generation prediction, power grid optimization scheduling, and power trading, which may lead to unstable power supply or economic loss. SUMMARY

[0006] The application provides a photovoltaic power station operation data processing method and system to at least solve the problems of limitations of original data cleaning standards caused by equipment updates and inability of early fault diagnosis systems to accurately identify signs of slight performance degradation of internal key components of new models of inverters in existing photovoltaic power station operation management.

[0007] In a first aspect, the application provides a photovoltaic power station operation data processing method, comprising the following steps:

[0008] Collecting operation data of each inverter in a photovoltaic power station and obtaining time stamps of each of the operation data, wherein the operation data includes power output data, working temperature data, internal circuit switch frequency data, current voltage waveform feature data and internal diagnosis information data of the inverter;

[0009] Setting an event window and aggregating the operation data from the same inverter within the event window, calculating statistical features of the aggregated data, and constructing a feature portrait of each inverter within the current event window based on the statistical features;

[0010] Constructing a behavior pattern library for each model of inverter, and updating the behavior pattern library according to actual operation data and fault feedback of each model of inverter, wherein the behavior pattern library contains normal operation modes and early wear modes under different working conditions;

[0011] Comparing the real-time extracted feature portrait with the early wear mode, and if the feature portrait is similar to the early wear mode, an early warning signal is sent out;

[0012] According to the accuracy of comparison between the feature portrait and the early wear mode, the potential severity of the early wear mode, and the duration of the anomaly, the early warning signal is graded, and a graded warning notification is output according to the result of grading.

[0013] Optionally, the setting of the event window and the aggregation of the operation data from the same inverter within the event window, the calculation of the statistical features of the aggregated data, and the construction of the feature portrait of each inverter within the current event window based on the statistical features comprise:

[0014] Running a first data processing channel and a second data processing channel in parallel; wherein the first data processing channel sets a long event window, continuously collects the operation data, and calculates the statistical features of the aggregated data to form the feature portrait of the inverter within the current long event window; the second data processing channel scans parameters sensitive to external interference in the operation data stream at a short time granularity, and when transient abnormal fluctuations are found, triggers a short time window mechanism to generate a transient abnormal portrait;

[0015] The transient abnormal image is compared with a preset interference behavior feature set to determine whether the transient abnormal fluctuation is external transient interference;

[0016] If the transient abnormal fluctuation is determined to be external transient interference, the operation data covered by the transient abnormal image in the long event window is specially processed, and statistical features of the aggregated data are calculated based on the operation data points after the special processing, and a feature image of each inverter in the current event window is constructed based on the statistical features.

[0017] Optionally, a behavior pattern library is constructed for each type of inverter, and the behavior pattern library is updated according to actual operation data and fault feedback of each type of inverter, wherein the behavior pattern library contains normal operation modes and early wear-out modes under different working conditions, including:

[0018] The existing early wear-out modes in the behavior pattern library are periodically evaluated for effectiveness, wherein the effectiveness evaluation includes:

[0019] From the operation data of the photovoltaic power station, historical early wear-out data samples corresponding to the early wear-out mode are selected;

[0020] The historical early wear-out data samples are compared with the existing early wear-out modes in the behavior pattern library to obtain a comparison result;

[0021] Based on the comparison result, it is determined whether the early wear-out mode needs to be adjusted;

[0022] If it is determined that adjustment is needed, the early wear-out mode is adjusted according to the historical early wear-out data samples and new fault feedback, and the adjustment includes updating the feature boundary of the early wear-out mode or adding a new early wear-out mode;

[0023] If it is determined that adjustment is not needed, the early wear-out mode is maintained unchanged.

[0024] Optionally, the real-time extracted feature image is compared with the early wear-out mode, and if the feature image is similar to the early wear-out mode, an early warning signal is issued, including:

[0025] According to the early wear-out mode to be compared, the weight of each feature parameter in the feature image is determined;

[0026] Using a weighted distance calculation method, the weighted distance between the real-time feature image and the early wear-out mode is calculated according to the weight;

[0027] According to the weighted distance, the similarity between the real-time feature image and the early wear-out mode is determined;

[0028] If the real-time feature image is determined to be similar to the early wear pattern, an early warning signal is sent out.

[0029] Optionally, the early warning signal is graded according to the comparison accuracy of the feature image and the early wear pattern, potential severity of the early wear pattern, and abnormal duration, comprising:

[0030] Based on the comparison accuracy, the potential severity, and the abnormal duration, a multi-dimensional grading rule is preset and a combination of warning levels of the early warning signal in the multi-dimensional grading rule is defined;

[0031] The historical fluctuation range of the comparison accuracy, the potential severity, and the abnormal duration is monitored;

[0032] When it is detected that the comparison accuracy, the potential severity, and the abnormal duration have changes beyond the historical stable fluctuation range in a short time, a grading buffer period is enabled;

[0033] In the grading buffer period, the warning level of the early warning signal is maintained, and subsequent changes of the comparison accuracy, the potential severity, and the abnormal duration are observed;

[0034] After the grading buffer period ends, the warning level is determined and adjusted according to the stable values of the comparison accuracy, the potential severity, and the abnormal duration;

[0035] A level stability factor is introduced in the multi-dimensional grading rule, and whether the warning level needs to be adjusted is determined according to the level stability factor, wherein the level stability factor is associated with the duration of the warning level, and the volatility of the comparison accuracy, the potential severity, and the abnormal duration.

[0036] Optionally, the behavior pattern library is constructed for each type of inverter, comprising:

[0037] When the historical operation data and fault feedback sample of a certain specific inverter type are insufficient, a first sample similar to the inverter type and sufficient in data is screened and obtained in the behavior pattern library;

[0038] Based on the first sample, a normal operation pattern and an early wear pattern having high correlation with the specific inverter type in key performance parameters and operation characteristics are selected as initial reference patterns of the new inverter type.

[0039] The initial reference mode is locally adjusted and refined based on the limited historical operation data and fault feedback of the new model inverter, to adapt to the unique operation characteristics of the new model inverter, and form an initial behavior mode library of the new model inverter.

[0040] After the new model inverter is put into operation, its operation data and fault feedback are continuously collected, and the initial behavior mode library is gradually improved and updated in an incremental manner until the data accumulation of the new model inverter reaches the condition for independently constructing a behavior mode library.

[0041] Optionally, the updating of the behavior mode library according to the actual operation data and fault feedback of each model of inverter includes:

[0042] The environmental parameters of the photovoltaic power station are continuously monitored, and when a significant change in the environmental parameters is detected, a working condition self-adaptive adjustment mechanism is triggered;

[0043] According to the current environmental parameters, the boundary definition of the working condition in the behavior mode library is dynamically adjusted;

[0044] Based on the adjusted boundary definition of the working condition, the mode parameters in the behavior mode library affected by the environment are corrected.

[0045] Optionally, the updating of the behavior mode library according to the actual operation data and fault feedback of each model of inverter also includes:

[0046] The operation data of the inverter in the photovoltaic power station are continuously monitored, and it is identified whether there is abnormal data in the operation data, the abnormal data including data missing, data exceeding the physical range, data mutation or data repetition;

[0047] When the abnormal data is identified, the abnormal data is classified according to the type and severity of the abnormal data;

[0048] The classification results of the data classification are data cleaned, and the data cleaning strategy includes data interpolation, data smoothing, data rejection or data weight reduction;

[0049] When the behavior mode library is updated, the operation data after the data cleaning and the fault feedback are used to update the behavior mode library.

[0050] Optionally, the comparison of the real-time extracted feature image with the early wear pattern includes:

[0051] In response to the early wear signs, multiple identification channels for different early wear patterns are started in parallel, each of which independently analyzes the real-time extracted feature image and calculates the matching degree of the real-time extracted feature image with the early wear pattern concerned by the respective identification channel;

[0052] When the matching degrees of the multiple identification channels all reach a preset threshold, a concurrent verification mechanism is triggered, which analyzes the correlation between the early wear patterns identified by the respective identification channels and outputs a correlation analysis result;

[0053] According to the correlation analysis result, the early warning priority and confidence of the early wear pattern identified by the respective identification channel are adjusted;

[0054] Based on the adjusted early warning priority and confidence, the parameters of the corresponding early wear pattern in the behavior pattern library are updated.

[0055] In a second aspect, the present application provides a photovoltaic power station operation data processing system, which comprises:

[0056] A data collection module is configured to collect operation data of each inverter in the photovoltaic power station and obtain the time stamp of each operation data, wherein the operation data includes power output data, working temperature data, internal circuit switch frequency data, current voltage waveform feature data and internal diagnosis information data of the inverter;

[0057] A feature image generation module is configured to set an event window, aggregate the operation data from the same inverter in the event window, calculate the statistical features of the aggregated data, and construct the feature image of each inverter in the current event window based on the statistical features;

[0058] A behavior pattern library construction module is configured to construct a behavior pattern library for each type of inverter and update the behavior pattern library according to the actual operation data and fault feedback of each type of inverter, wherein the behavior pattern library contains normal operation patterns and early wear patterns under different working conditions;

[0059] An early warning judgment module is configured to compare the real-time extracted feature image with the early wear patterns, and if the feature image is similar to the early wear patterns, an early warning signal is sent out;

[0060] An early warning grading notification module is configured to grade the early warning signal according to the comparison accuracy of the feature image and the early wear patterns, the potential severity of the early wear patterns, and the duration of the anomaly, and output a graded early warning notification according to the grading result.

[0061] Compared with the related art, the photovoltaic power station operation data processing method and system provided by the present application at least has the following technical effects:

[0062] By collecting multi-dimensional operation data of the inverter, obtaining accurate time stamps, setting event windows and aggregating data, calculating statistical characteristics, and then constructing the feature portrait of each inverter, the running state of the inverter is comprehensively and dynamically reflected. Subsequently, a behavior pattern library is constructed and continuously updated for each type of inverter, which contains normal operation mode and early wear mode. The performance curve and data fluctuation law specific to different types of inverters are accurately modeled, effectively overcoming the problem of limitation of original data cleaning standard caused by equipment update in the prior art. Further, by comparing the real-time feature portrait with the early wear mode, signs of slight performance degradation of the internal key components of the inverter are found in time, and an early warning signal is sent, thereby solving the technical problem that the existing early fault diagnosis system cannot accurately identify the potential risks of new types of inverters. Finally, the warning signals are classified according to the accuracy of comparison, potential severity and abnormal duration, and the classified warning notifications are output, so that the warning information is more instructive and operable, avoiding the resource waste or response lag that may be caused by the traditional "one-size-fits-all" alarm mode.

[0063] In summary, through fine data processing, personalized behavior pattern construction and intelligent warning classification mechanism, the present application greatly improves the identification ability and early warning efficiency of photovoltaic power stations for early faults of new types of inverters, effectively reduces potential operation risks and maintenance costs, and ensures stable and efficient operation of photovoltaic power stations.

[0064] The details of one or more embodiments of the present application are given in the following drawings and description, so that other features, objects and advantages of the present application are more apparent. BRIEF DESCRIPTION OF DRAWINGS

[0065] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0066] Figure 1 is a flowchart of a photovoltaic power station operation data processing method according to an exemplary embodiment.

[0067] Figure 2 is a flowchart of step S2 according to an exemplary embodiment.

[0068] Figure 3 is a flowchart of step S3 according to an exemplary embodiment.

[0069] Figure 4is a flow chart of step S4 according to an exemplary embodiment.

[0070] Figure 5 is a flow chart of step S5 according to an exemplary embodiment.

[0071] Figure 6 is a block diagram of a photovoltaic power station operation data processing system according to an exemplary embodiment. DETAILED DESCRIPTION

[0072] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is described and explained below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of the present application.

[0073] Obviously, the drawings in the following description are only some examples or embodiments of the present application, and for those of ordinary skill in the art, the present application can be applied to other similar scenarios without creative efforts based on these drawings. In addition, it can be understood that although the efforts made in this development process can be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some designs, manufacturing or production changes based on the technical content disclosed in the present application are only routine technical means and should not be understood as insufficient disclosure of the present application.

[0074] In the present application, "embodiment" means that the specific features, structures or characteristics described in combination with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase at various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment to other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0075] Unless otherwise defined, technical terms and scientific terms used in the present application shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. Unless otherwise defined, the terms "one" and "a" or "an" used in this application do not denote a singular noun, but can denote both the singular and plural. The terms "including", "containing", "having" and any variations thereof in this application are intended to cover a non-exclusive inclusion; for example, a process, method, system, product or device including a series of steps or modules (units) is not limited to the listed steps or units, but can further include steps or units not listed or can further include other steps or units inherent to these processes, methods, products or devices. The terms "connected", "connected", "coupled" and the like in this application are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The term "multiple" in this application means two or more. The term "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects. The terms "first", "second", "third" and the like in this application are only to distinguish similar objects, and do not represent a specific order for the objects.

[0076] In the related art, the replacement of the new model inverter brings a chain effect to operation and maintenance and group management: the operation and maintenance personnel are difficult to distinguish whether the data is the normal working condition performance of the new model inverter or the early attenuation signal of the component, are forced to increase the on-site troubleshooting workload, and the power plant efficiency optimization and accurate predictive maintenance cannot be implemented, which may miss the best maintenance window; for the group, the data uncertainty of the power plant makes it unable to accurately evaluate the health status and power generation potential of the power plant, thereby affecting cross-regional power generation prediction, power grid optimization scheduling and power transaction, and may cause unstable power supply or economic benefit loss.

[0077] Based on the above situation, the present application embodiment provides a photovoltaic power plant operation data processing method and system, which will be described in detail below in combination with specific embodiments and drawings.

[0078] Embodiment 1

[0079] The present application embodiment provides a photovoltaic power plant operation data processing method. Figure 1 The photovoltaic power plant operation data processing method flowchart shown is according to an exemplary embodiment. As shown in the figure, the method comprises the following steps: Figure 1

[0080] ​S1, collect operation data of each inverter in the photovoltaic power station, and obtain a timestamp of each operation data, wherein the operation data includes power output data, working temperature data, internal circuit switch frequency data, current voltage waveform characteristic data and internal diagnosis information data of the inverter;

[0081] In the embodiment, the photovoltaic power station operation data refers to real-time or historical data collected from each inverter in the photovoltaic power station, which covers various states and performance indicators of the inverter in the running process, such as power output data, working temperature data, internal circuit switch frequency data, current voltage waveform characteristic data and internal diagnosis information data. In terms of collection method, the embodiment realizes real-time monitoring and recording of the above operation parameters by integrating sensors and data collection units in the inverter, and then transmits the photovoltaic power station operation data to the central data processing platform through wired or wireless communication network. The timestamp is a time mark associated with each operation data, which is realized through a system clock synchronization mechanism, ensuring that all collected data have accurate time marks, which are used to indicate the specific time of data collection and are crucial for subsequent data aggregation and time series analysis.

[0082] S2, set an event window, aggregate operation data from the same inverter in the event window, calculate statistical features of the aggregated data, and construct a feature portrait of each inverter in the current event window based on the statistical features;

[0083] In the embodiment, the event window refers to a preset time period, and in this time period, the operation data from the same inverter will be aggregated and processed to calculate its statistical features. The feature portrait is constructed based on the statistical features of the aggregated data, which is an abstract representation of the running state and behavior pattern of the inverter in a specific event window. For example, a fixed length event window is set, for example, 10 minutes or 30 minutes. In the window, all operation data from the same inverter is aggregated and its statistical features such as mean, standard deviation, maximum, minimum, skewness and kurtosis are calculated. These statistical features serve as the basis for constructing the feature portrait of the inverter. The feature portrait is a multi-dimensional vector, each dimension of which represents a statistical feature, thereby comprehensively describing the running state of the inverter in the current event window.

[0084] S3, construct a behavior pattern library for each type of inverter, and update the behavior pattern library according to the actual operation data and fault feedback of each type of inverter, wherein the behavior pattern library contains normal operation mode and early wear mode under different working conditions;

[0085] In this embodiment, the behavior pattern library is a collection of normal operation patterns and early wear patterns of different types of inverters under different working conditions, which is the basis for early warning judgment. Early wear patterns refer to the subtle and progressive changes in performance parameters of inverters before failure, which are usually not easily identified by traditional fault diagnosis methods. For example, for each type of inverter, collect its historical operation data under different environmental temperature, light intensity, load conditions and other working conditions. Through clustering analysis or machine learning training on these historical data, the normal operation patterns of the inverter under different working conditions are identified. At the same time, combined with historical fault records and expert experience, various early wear patterns are defined and constructed, for example, the slight aging of IGBT module may be manifested as a slight fluctuation in switching frequency or a slight decline in efficiency. The update of the behavior pattern library is carried out in a regular or real-time manner, for example, when new operation data or fault feedback is obtained, the existing patterns are corrected or new patterns are added using these information.

[0086] S4, compare the real-time extracted feature image with the early wear pattern, if the feature image is similar to the early wear pattern, an early warning signal is issued;

[0087] In this embodiment, the early warning signal is a prompt information issued when the feature image of the inverter is similar to the early wear pattern, aiming to remind the operation and maintenance personnel to pay attention to the potential equipment risk. After the real-time operation data of the inverter is collected and the feature image is generated, the feature image is compared with various early wear patterns stored in the behavior pattern library. The comparison can be realized by calculating the distance or similarity between the feature image and the pattern, if the calculated similarity exceeds the preset threshold, it is considered that the inverter may be in early wear state, and the early warning signal is immediately issued.

[0088] S5, according to the comparison accuracy of the feature image and the early wear pattern, the potential severity of the early wear pattern, and the duration of the anomaly, the early warning signal is classified, and a classified warning notification is output according to the result of the classification;

[0089] In this embodiment, the classified warning notification is different levels of alarm information issued to relevant personnel after the early warning signal is classified according to the comparison accuracy, potential severity and duration of anomaly. For example, define multiple warning levels, such as "slight warning", "moderate warning" and "serious warning". The higher the comparison accuracy, the greater the potential severity of the early wear pattern, and the longer the duration of the anomaly, the higher the warning level. The classified warning notification can be output in various ways, for example, sending a message, an email to the operation and maintenance personnel or displaying different colored alarms on the monitoring interface, so that the operation and maintenance personnel can take corresponding maintenance measures according to the warning level.

[0090] The technical solutions of the above embodiments first acquire the operation data of each inverter in the photovoltaic power station continuously and obtain accurate timestamps. The photovoltaic power station operation data covers multiple dimensions of the inverter operation state, ensuring comprehensive perception of the device condition. Then, the system sets an event window, aggregates data from the same inverter, and calculates its statistical features, thereby constructing a feature portrait of each inverter in the current event window. This feature portrait is a refined description of the current operation state of the inverter and can capture subtle changes. At the same time, an action pattern library is constructed and continuously updated for each type of inverter. The action pattern library not only contains the normal operation mode of the inverter under different working conditions, but more importantly, it also contains various early wear patterns. These early wear patterns are constructed based on historical data and fault feedback and can identify subtle signs of performance degradation before failure. Through continuous updating of actual operation data and fault feedback, the action pattern library can maintain its accuracy and timeliness, thereby better adapting to the aging of the device and changes in the environment.

[0091] When the real-time extracted inverter feature portrait is generated, it is compared with the early wear patterns in the action pattern library. If the feature portrait shows high similarity to a certain early wear pattern, it means that the inverter may have entered the early wear stage, and the system will immediately issue an early warning signal. At the same time, in order to avoid false positives and provide more instructive warning information, the early warning signal is further classified according to the accuracy of the comparison, the potential severity of the early wear pattern, and the duration of the anomaly. This multi-dimensional classification mechanism can ensure the accuracy and practicality of the warning information, allowing maintenance personnel to prioritize maintenance resources and take preventive measures to avoid potential serious failures.

[0092] Through the above cooperative work, the method of the present application can effectively solve the limitations in traditional photovoltaic power station operation data processing. For example, a new model inverter may exhibit higher relative output under low light conditions, and its data points may be incorrectly judged as abnormal data and excluded due to exceeding the "normal" upper limit of the old benchmark. Conversely, if a new model inverter has a slight performance decline, manifested as a slight decrease in output power or a subtle shift in efficiency curve, its performance may still be better than that of the old model, and the decayed data may still fall within the "normal" range of the old model, or even be considered "good" performance by the old standard, thus masking the early and subtle performance decline trend and failing to identify it as a potential risk signal.

[0093] The application can more accurately capture the normal behavior and early wear characteristics of new inverter models under different working conditions by constructing an inverter feature image and a behavior pattern library, avoiding data misjudgment caused by device heterogeneity. In particular, the early wear patterns included in the behavior pattern library enable the system to identify subtle performance degradation signs that have not yet reached the traditional hard failure criteria but have predicted potential reliability risks. For example, slight aging of IGBT modules leads to increased switching loss or minor waveform distortion at a specific frequency. These changes can be discovered in time through the construction of feature images and comparison with early wear patterns. In addition, the introduction of a warning grading mechanism enables maintenance personnel to prioritize handling of anomalies with high potential severity and long duration, thereby optimizing maintenance strategies and improving maintenance efficiency.

[0094] In one possible design, Figure 2 is a flowchart of step S2 according to an example embodiment. Refer to the accompanying Figure 2 Step S2 includes:

[0095] S21, run the first data processing channel and the second data processing channel in parallel; wherein the first data processing channel sets a long event window, continuously collects running data, and calculates statistical features of aggregated data to form a feature image of the inverter within the current long event window; the second data processing channel scans parameters sensitive to external interference in the running data stream at a short time granularity, and when a transient abnormal fluctuation is found, triggers a short-time window mechanism to generate a transient abnormal image;

[0096] In this embodiment, the first data processing channel is configured to set a longer event window, such as several hours or a day, to continuously collect running data of the inverter, and calculate statistical features of aggregated data within the long event window, thereby forming a macro feature image of the inverter within the long event window. The long event window is mainly used to capture long-term and stable running trends and wear signs of the inverter.

[0097] The second data processing channel scans parameters sensitive to external interference in the running data stream at a shorter time granularity, such as seconds or minutes. These sensitive parameters include transient power fluctuations, transient distortion rates of current and voltage, or rapid changes in certain internal diagnostic signals. When the second data processing channel detects transient abnormal fluctuations in these sensitive parameters, it will immediately trigger a short-time window mechanism. The short-time window usually has a very short duration, covering only the moments before and after the abnormal fluctuation, and generates a transient abnormal image within the short-time window to describe the specific characteristics of the transient abnormal fluctuation.

[0098] S22, compare the transient abnormal image with a preset interference behavior feature set to determine whether the transient abnormal fluctuation is external transient interference;

[0099] In this embodiment, the preset interference behavior feature set pre-stores typical feature patterns of known or common external transient interference (such as power grid transient voltage drop, lightning surge, sensor transient reading error, etc.). Through comparison, the system can determine whether the detected transient abnormal fluctuation belongs to external transient interference.

[0100] S23, if it is determined that it is external transient interference, the running data covered by the transient abnormal image in the long event window is specially processed, and the statistical features of the aggregated data are calculated according to the running data points after special processing, and the feature image of each inverter in the current event window is constructed based on the statistical features;

[0101] In this embodiment, the special processing includes but is not limited to: marking the disturbed data points as invalid and excluding them from statistical calculation; using interpolation, smoothing and other methods to repair the disturbed data; or giving lower weight to the disturbed data. Through such special processing, the influence of external transient interference on the calculation of statistical features of aggregated data in the long event window is effectively eliminated or reduced. Subsequently, the statistical features of the aggregated data are recalculated according to the running data points after special processing, and more accurate statistical features are used to construct more real and reliable feature images of each inverter in the current event window.

[0102] The technical solutions of the above embodiments effectively solve the problem that the feature image is easily affected by external transient interference in the above basic scheme by introducing a first data processing channel and a second data processing channel running in parallel, and combining a transient abnormality recognition and processing mechanism. Specifically, the first data processing channel is responsible for macro, stable data aggregation and feature image construction, ensuring accurate grasp of the long-term running state of the inverter. At the same time, the second data processing channel focuses on micro, real-time anomaly detection, quickly capturing transient fluctuations in the data stream that are sensitive to external interference. When the second data processing channel identifies a transient abnormal fluctuation, it is distinguished whether it is an early sign of inverter internal wear or a transient disturbance caused by external environment by comparing with the preset interference behavior feature set. Once it is confirmed that it is external transient interference, the system performs special processing on the affected data in the long event window, thereby avoiding the interference of external noise on the evaluation of the real running state of the inverter.

[0103] Through the cooperative work of the double channels and the intelligent interference filtering mechanism, the constructed feature image can more accurately reflect the running condition and potential early wear trend of the inverter itself, improving the accuracy and robustness of the feature image of the photovoltaic power station inverter. This not only reduces unnecessary on-site inspection and maintenance costs, but also ensures timely discovery of the real early wear of the inverter, thereby more accurately performing predictive maintenance, prolonging the service life of the equipment, and improving the overall operation efficiency and safety of the photovoltaic power station.

[0104] In one example, assume that a power output data of an inverter in a certain PV power plant has a transient drop in a short time during normal operation, and the current-voltage waveform feature data also shows a short-term distortion.

[0105] At this time, the first data processing channel continues to collect data and calculate the statistical features in the long event window. Meanwhile, the second data processing channel scans the power output data, operating temperature data, internal circuit switching frequency data, current-voltage waveform feature data, and internal diagnostic information data at a second-level granularity. When the above-mentioned transient drop and distortion are detected, the second data processing channel immediately triggers the short-time window mechanism to generate a transient abnormality portrait containing these transient abnormality features.

[0106] The transient abnormality portrait is then compared with a preset interference behavior feature set. If the set contains the feature pattern of "grid transient voltage drop", and the comparison result shows a high degree of similarity, the system determines that this transient abnormal fluctuation is caused by external grid transient voltage drop, which is external transient interference.

[0107] Based on this determination, the system performs special processing on the power output data and current-voltage waveform feature data in the long event window of the first data processing channel, which overlaps with the time period of the transient abnormality portrait. For example, these disturbed data points can be marked as abnormal and excluded from the statistical calculation of the long event window, or corrected by interpolation of the data points before and after. Finally, according to the operating data after special processing, the statistical features of the aggregated data are recalculated, and a more accurate inverter feature portrait is constructed, which will no longer contain false abnormal information caused by external grid transient voltage drop, thus more truly reflecting the operating state of the inverter itself.

[0108] In one possible design, Figure 3 is a flowchart of step S3 according to an example embodiment. Refer to the accompanying Figure 3 Step S3 includes:

[0109] S31, periodically evaluate the effectiveness of the existing early wear patterns in the behavior pattern library, wherein the effectiveness evaluation includes:

[0110] S31-1, select historical early wear data samples corresponding to the early wear patterns from the operating data of the PV power plant;

[0111] S31-2, compare the historical early wear data samples with the existing early wear patterns in the behavior pattern library to obtain a comparison result;

[0112] In this embodiment, the system conducts a systematic check and verification of the early wear patterns stored in the behavior pattern library at preset time intervals (e.g., monthly, quarterly, or annually) or upon the triggering of specific events (e.g., after a large-scale failure occurs), to ensure their continued accuracy and relevance. Among them, the effectiveness evaluation aims to confirm whether the existing patterns can still accurately reflect the early wear characteristics of the current inverter; specifically, the historical database is screened for those running data generated by inverters confirmed to be in the early wear state in actual operation. These data samples should contain key characteristic parameters related to a specific early wear pattern, such as power output data, operating temperature data, internal circuit switching frequency data, current and voltage waveform feature data, and internal diagnostic information data, etc., with the corresponding time stamp, then, using pattern recognition, machine learning algorithms or statistical methods, the selected historical early wear data samples are matched with the corresponding early wear patterns in the behavior pattern library for matching degree analysis, and the comparison result is obtained. The comparison result can be expressed as a similarity score, distance metric, or classification accuracy, etc.

[0113] S32, based on the comparison result, determining whether the early wear pattern needs to be adjusted;

[0114] In this embodiment, the effectiveness of the existing early wear pattern is evaluated according to the threshold or rule set by the comparison result. For example, if the matching degree of the historical sample with the existing pattern is lower than the preset threshold, or the classification accuracy significantly decreases, it indicates that the pattern is no longer accurate and needs to be adjusted.

[0115] S33-1, if it is determined that adjustment is needed, adjusting the early wear pattern according to the historical early wear data sample and new failure feedback, the adjustment including updating the feature boundary of the early wear pattern or adding a new early wear pattern;

[0116] S33-2, if it is determined that adjustment is not needed, maintaining the early wear pattern unchanged;

[0117] In this embodiment, the adjustment includes updating the feature boundary of the early wear pattern or adding a new early wear pattern: updating the feature boundary means modifying the parameter range or threshold of the existing pattern to better adapt to the new data distribution; adding a new early wear pattern means creating and adding a new pattern to the behavior pattern library when a new type of early wear sign with unique characteristics is found that cannot be covered by the existing pattern. If it is determined that adjustment is not needed, the early wear pattern is maintained unchanged, which means that the current pattern is still effective and accurate and does not need to be modified.

[0118] The technical solutions of the above embodiments introduce a periodic effectiveness evaluation mechanism, solving the problem of early wear patterns in the behavior pattern library becoming invalid or inaccurate over time. Specifically, by periodically selecting historical early wear data samples corresponding to early wear patterns and comparing them with existing early wear patterns in the behavior pattern library, the accuracy and applicability of existing patterns are objectively measured. Subsequently, based on the comparison results, it is determined whether the pattern needs to be adjusted. When the effectiveness of the pattern decreases, the feature boundaries of the pattern are updated or new early wear patterns are added in combination with the latest historical data and fault feedback, thereby ensuring that the behavior pattern library can always accurately capture the real signs of early wear of the inverter. The dynamic adjustment and self-improvement mechanism of the technical solutions of the present application ensures that the behavior pattern library is dynamically updated and optimized as the operating state of the inverter, environmental conditions, and fault patterns evolve, enabling the early warning system to continuously provide high-quality early warning signals, adapt to changes in the operating environment and wear mechanism of the inverter, and avoid false positives and false negatives caused by outdated or inaccurate early wear patterns, significantly improving the accuracy and reliability of the early warning system of the photovoltaic power station.

[0119] In one example, assume that there is an early wear pattern of "electrolytic capacitor aging" for a certain type of inverter in the behavior pattern library, and its feature boundary is defined as a specific fluctuation range of internal circuit switching frequency data and a specific distortion rate of current-voltage waveform data. The system is set to perform effectiveness evaluation once every quarter.

[0120] In a certain evaluation, the system selects historical early wear data samples of inverters diagnosed as "electrolytic capacitor aging" in the past three months from the operation data of the photovoltaic power station. The internal circuit switching frequency data and current-voltage waveform feature data of these samples are extracted and compared with the feature boundary of the "electrolytic capacitor aging" pattern in the behavior pattern library. The comparison result shows that 20% of the historical samples, although ultimately confirmed as capacitor aging, have a slightly higher fluctuation range of switching frequency than the upper limit of the existing pattern, and a slightly lower waveform distortion rate than the lower limit of the existing pattern. Based on this comparison result, the system determines that the early wear pattern needs to be adjusted.

[0121] Specifically, the system updates the feature boundary of the "electrolytic capacitor aging" mode based on these new historical early wear data samples and recent failure feedbacks about capacitor aging, for example, appropriately relaxes the upper limit of the switching frequency fluctuation range and slightly adjusts the lower limit of the waveform distortion rate to better cover the actual early wear conditions. In addition, if new failure feedbacks are found to indicate the existence of a new form of capacitor aging that does not conform to the existing mode, the system can also choose to add a new "electrolytic capacitor slight leakage" early wear mode and define its unique feature boundary. Through this periodic evaluation and adjustment, the behavior pattern library can maintain its ability to identify early wear signs of the inverter and ensure the accuracy of the early warning.

[0122] In one possible design, Figure 4 is a flowchart of step S4 according to an example embodiment. Refer to the accompanying Figure 4 Step S4 includes:

[0123] S41, determine the weight of each feature parameter in the feature image according to the early wear mode to be compared;

[0124] In this embodiment, determining the weight of each feature parameter in the feature image means that for different early wear modes, the key indicative feature parameters will differ. For example, for early wear modes involving overheating, temperature-related feature parameters (such as operating temperature data) should be given higher weights; while for early wear modes involving electrical performance degradation, current and voltage waveform feature data or internal circuit switching frequency data are more important. The weights of each feature parameter can be determined and adjusted based on historical failure data analysis, expert experience or machine learning models to ensure that the comparison process can focus on the features that best reflect the specific wear state.

[0125] S42, using a weighted distance calculation method, calculate the weighted distance between the real-time feature image and the early wear mode according to the weights;

[0126] In this embodiment, the weighted distance calculation method can use various mathematical models, such as weighted Euclidean distance, weighted Mahalanobis distance, etc. The weighted distance calculation method takes the weight of each feature parameter into account in the distance calculation, so that the feature parameters with higher weights play a more dominant role in the distance calculation. Thus, even if some non-critical features fluctuate, as long as the key features match the early wear mode well, the weighted distance can accurately reflect the potential similarity.

[0127] S43, judge the similarity between the real-time feature image and the early wear mode according to the weighted distance;

[0128] S44, if it is judged that the real-time feature image is similar to the early wear mode, issue an early warning signal;

[0129] In this embodiment, the calculated weighted distance is compared with a preset similarity threshold. If the weighted distance is less than or equal to the similarity threshold, it is considered that the real-time feature image has sufficient similarity with the early wear pattern. The similarity threshold can be optimized according to the actual application scene requirements through historical data training and verification to balance the sensitivity and accuracy of the early warning.

[0130] The technical solutions of the above embodiments dynamically adjust the importance of the feature parameters according to the characteristics of different early wear patterns, making the comparison process more targeted and accurate. The weighted distance calculation method integrates the contributions of each feature parameter and highlights the role of key features, thereby improving the recognition accuracy of early wear signs. Subsequently, through the judgment of the weighted distance, the system can more reliably identify the operating state similar to the early wear pattern, avoiding false positives or false negatives caused by non-key feature fluctuations. Through the above technical solutions, the present application significantly improves the accuracy and reliability of the early wear warning of the photovoltaic power station inverter, not only reduces the occurrence of invalid early warnings, but also ensures the timely identification of real potential faults, providing more valuable decision-making basis for operation and maintenance personnel.

[0131] In one possible design, Figure 5 is a flowchart of step S5 according to an example embodiment. Referring to the accompanying drawings, Figure 5 Step S5 includes:

[0132] S51, based on the comparison accuracy, potential severity and abnormal duration, presetting a multi-dimensional grading rule and defining a combination of early warning levels of early warning signals within the multi-dimensional grading rule;

[0133] In this embodiment, the preset multi-dimensional grading rule means that different early warning levels are defined according to the three dimensions of comparison accuracy, potential severity and abnormal duration. For example, three levels of "low", "medium" and "high" are set, and each level corresponds to different combination intervals of the three dimension parameters. Specifically, the higher the comparison accuracy, the greater the potential severity and the longer the abnormal duration, the higher the early warning level. The specific rule can be set according to expert experience, historical fault data analysis and risk assessment, and can be adjusted according to the actual operation situation.

[0134] S52, monitoring the historical fluctuation range of the comparison accuracy, potential severity and abnormal duration;

[0135] In this embodiment, a "stable fluctuation range" is determined by calculating the mean, standard deviation or percentile of the historical data. When the real-time detected parameter value exceeds this preset historical stable fluctuation range, it indicates that there may be an abnormal situation that needs further attention.

[0136] S53, when detecting that the change of the alignment accuracy, the potential severity and the anomaly duration exceeds the range of historical stable fluctuation in a short time, enabling a hierarchical buffer period;

[0137] S54, in the hierarchical buffer period, maintaining the warning level of the early warning signal, and observing the subsequent change of the alignment accuracy, the potential severity and the anomaly duration;

[0138] In the embodiment, the purpose of the buffer period is to avoid overreaction to transient fluctuations. In the buffer period, the system will temporarily maintain the current warning level unchanged, while continuously observing the subsequent change trend of the parameters. For example, the buffer period can be set to be several minutes to several hours, and the specific length can be configured according to the characteristics of the parameters and the response requirements of the system.

[0139] S55, after the end of the hierarchical buffer period, judging and adjusting the warning level according to the stable value of the alignment accuracy, the potential severity and the anomaly duration;

[0140] In the embodiment, the stable value is obtained by averaging, median calculation or trend analysis of the data in the buffer period, so as to eliminate the influence of transient fluctuations and reflect the real change trend of the parameters.

[0141] S56, introducing a level stability factor in the multi-dimensional hierarchical rule, and judging whether the warning level needs to be adjusted according to the level stability factor, wherein the level stability factor is related to the duration of the warning level and the volatility of the alignment accuracy, the potential severity and the anomaly duration;

[0142] In the embodiment, the level stability factor is used to evaluate the stability of the current warning level. For example, if a warning level has lasted for a long time and the volatility of the related parameters is low, the level stability factor will be high, indicating that the level is relatively stable and is not easy to adjust. On the contrary, if the duration of the warning level is short or the volatility of the parameters is high, the level stability factor is low, and the system will be more cautious when adjusting the warning level. Through the level stability factor, the frequent jump of the warning level can be effectively avoided, and the robustness of the warning system is improved.

[0143] The technical solutions of the above embodiments effectively solve the problems of unstable and frequent changes in early warning levels caused by instantaneous fluctuations in parameters in traditional early warning grading methods by introducing historical fluctuation range monitoring, hierarchical buffer period mechanism, and level stability factor. Specifically, by monitoring the historical fluctuation ranges of accuracy, potential severity, and abnormal duration, the system can identify abnormal changes that exceed the normal range, thereby triggering a more detailed judgment process. When detecting short-term abnormal fluctuations, the hierarchical buffer period is enabled, so that the system does not immediately respond to transient changes, but gives a certain time to observe the persistence of parameters, avoiding misjudgment of incidental interference. After the buffer period ends, the judgment and adjustment of the early warning level are based on the stable value of the parameter, ensuring that the basis for decision-making is the real trend verified by time. Finally, the introduction of the level stability factor makes the adjustment of the early warning level not only consider the current parameter value, but also comprehensively consider the duration of the early warning level and the volatility of the parameter, thereby maintaining the timeliness of the early warning while improving the stability of the early warning level, avoiding unnecessary frequent notifications.

[0144] In one example, assume that the characteristic image of an inverter of a certain photovoltaic power station in normal operating state usually fluctuates between 0.7-0.8 in comparison accuracy with the early wear pattern, the potential severity level is 2 (a total of 5 levels), and the abnormal duration is usually less than 10 minutes. The system presets multi-dimensional grading rules and defines early warning level combinations accordingly, for example, when the comparison accuracy is lower than 0.65, the potential severity reaches level 3 or above, and the abnormal duration exceeds 30 minutes, triggering "moderate early warning".

[0145] On a certain day, the system monitors that the comparison accuracy of an inverter suddenly decreases to 0.6, the potential severity temporarily rises to level 3, and the abnormal duration is 5 minutes. Since these parameters exceed the historical stable fluctuation range in a short time, the system immediately enables a 30-minute hierarchical buffer period. During this buffer period, although the parameters fluctuate, the early warning level still remains at "no early warning". The system continues to observe and finds that the comparison accuracy stabilizes at 0.62, the potential severity stabilizes at level 2.8, and the abnormal duration stabilizes at 15 minutes.

[0146] After the hierarchical buffer period ends, the system reevaluates according to these stable values. At this time, the comparison accuracy 0.62 is still lower than the threshold of 0.65, but the potential severity level 2.8 does not reach level 3, and the abnormal duration 15 minutes does not reach 30 minutes. At the same time, the system introduces the level stability factor for judgment. Since the current early warning level (no early warning) has been maintained for a long time, and the parameter volatility is relatively small during the buffer period, the level stability factor is high, indicating that the current level is relatively stable. After comprehensive judgment, the system decides not to immediately upgrade the early warning level, but maintains the "no early warning" state, marks it as "attention state", and increases the monitoring frequency.

[0147] After several hours, the alignment accuracy of the inverter further decreases to 0.55, and the potential severity stabilizes at level 3.5, with the anomaly duration lasting more than 45 minutes. At this time, both parameters have reached or exceeded the triggering conditions of “moderate warning”. The system again enables the hierarchical buffer period, and after the buffer period ends, it determines that the warning level needs to be adjusted according to the stable value and the lower level stability factor (because the parameter continues to deteriorate, the stability decreases). Finally, the system adjusts the warning level to “moderate warning” and outputs the corresponding hierarchical warning notification.

[0148] Through the above process, the scheme of the present application avoids false alarms caused by transient fluctuations and ensures that the adjustment of the warning level is based on sustained and stable abnormal conditions.

[0149] In one possible design, in step S3, a behavior pattern library is constructed for each type of inverter, including:

[0150] S301, when the historical operation data and fault feedback sample size of a certain type of inverter are insufficient, in the behavior pattern library, the first sample similar to the inverter type and sufficient data is screened and obtained;

[0151] In this embodiment, similarity refers to high consistency in manufacturer, power level, technical architecture, key component type, or operating environment; sufficient data generally refers to a sample size that meets the preset statistical requirements, sufficient to support reliable model training. The specific screening process is based on pre-defined similarity indicators, such as distance calculation of feature vectors or expert system rules for matching.

[0152] S302, based on the first sample, selecting normal operation patterns and early wear patterns that have high correlation with the specific inverter type in terms of key performance parameters and operating characteristics as the initial reference patterns for the new type of inverter;

[0153] In this embodiment, the key performance parameters include power output efficiency, temperature response curve, current voltage harmonic content, etc.; the operating characteristics include start / stop behavior, load change response, and data fingerprints under specific fault patterns; and the high correlation is determined by statistical methods (such as correlation coefficient analysis) or machine learning algorithms (such as clustering analysis).

[0154] S303, using the limited historical operation data and fault feedback of the new type of inverter, locally adjusting and refining the initial reference patterns to adapt to the unique operating characteristics of the new type of inverter, forming an initial behavior pattern library for the new type of inverter;

[0155] In this embodiment, the local adjustment and refinement includes fine-tuning of the mode boundary, correction of the threshold value, or redistribution of the weight of specific features, to ensure that the initial mode accurately reflects the actual operation of the new model inverter. The unique operating characteristics include unique internal diagnostic information data, specific current and voltage waveform characteristic data, or power output data fluctuation patterns under specific operating conditions.

[0156] S304, after the new model inverter model is put into operation, its operation data and fault feedback are continuously collected, and the initial behavior mode library is gradually improved and updated in an incremental manner until the data accumulation of the new model inverter model reaches the condition for independently constructing the behavior mode library;

[0157] In this embodiment, the incremental manner means that the model parameters or mode definitions are dynamically updated as new data continues to flow in, for example, using online learning algorithms or periodic small batch retraining. The condition for independently constructing the behavior mode library is set to reach a predetermined data volume, the model prediction accuracy reaches a certain level, or the mode parameters converge to a stable state.

[0158] The technical solutions of the above embodiments provide a reliable initial behavior mode library for the new model inverter by using the data resources of the existing similar model inverters, effectively solving the problem that the new model equipment is difficult to perform accurate early wear-out warning when the initial data is insufficient. Specifically, when the new model inverter is put into use, its behavior mode library is not constructed from zero, but is based on the mature experience of similar models, thereby having a preliminary early warning capability in a short time. With the accumulation of the operation data of the new model inverter itself, the initial mode library will be continuously optimized and improved, and eventually evolve into a behavior mode library fully adapted to its own characteristics, ensuring the continuity and effectiveness of the early warning system.

[0159] Through the above technical solutions, even in the case of insufficient historical data of the new model inverter, an effective early wear-out warning mechanism can be quickly established, shortening the period from deployment to full monitoring of the new model equipment.

[0160] In one example, assume a photovoltaic power station introduces a batch of brand new "A-type" inverters, which are the first deployment of this model and lack historical operation data. At this time, the system will first screen the behavior pattern library for "B-type" inverters similar to the "A-type" inverters in terms of power rating, core technology platform, and manufacturer. If the "B-type" inverters have sufficient historical operation data and fault feedback, their existing normal operation patterns and early wear patterns will be used as initial reference patterns for the "A-type" inverters. Subsequently, using the limited data generated by the "A-type" inverters during the initial operation stage, these initial reference patterns are locally adjusted, for example, the characteristic boundaries of the early wear patterns are fine-tuned according to the internal circuit switching frequency data and current-voltage waveform characteristic data specific to the "A-type" inverters. Once the "A-type" inverters are formally put into operation, the system will continue to collect their power output data, operating temperature data, etc., and gradually update and improve the behavior pattern library in an incremental learning manner until the data volume and model stability reach the standard sufficient for independent construction of the behavior pattern library.

[0161] In one possible design, in step S3, updating the behavior pattern library according to the actual operation data and fault feedback of each type of inverter includes:

[0162] S311, continuously monitoring the environmental parameters of the photovoltaic power station, and triggering the working condition adaptive adjustment mechanism when a significant change in the environmental parameters is detected;

[0163] In this embodiment, the external environmental factors affecting the operating state of the inverter are collected and analyzed in real time or periodically, such as environmental temperature, humidity, solar irradiance, wind speed, etc. The environmental parameters can be obtained through various sensors deployed in the photovoltaic power station. When a significant change in the environmental parameters is detected, such as the environmental temperature continuously exceeding the preset threshold range, the solar irradiance fluctuating greatly within a short period of time, etc., the system will automatically trigger the working condition adaptive adjustment mechanism.

[0164] S312, dynamically adjusting the boundary definition of the working conditions in the behavior pattern library according to the current environmental parameters;

[0165] In this embodiment, according to the real-time or recent environmental parameter values monitored, the characteristic ranges of various working conditions (such as high-temperature working conditions, low-temperature working conditions, high-irradiance working conditions, etc.) predefined in the behavior pattern library are corrected in real time. For example, if the environmental temperature continues to rise, the lower limit of the temperature of the high-temperature working condition will be dynamically adjusted higher to more accurately reflect the "normal" high-temperature operating state under the current environment.

[0166] S313, correcting the mode parameters affected by the environment in the behavior pattern library based on the boundary definition of the adjusted working conditions;

[0167] In this embodiment, after the adjustment of the working condition boundary definition, the mode parameters in the behavior mode library that are directly related to these working conditions are updated accordingly. Among them, the directly related mode parameters generally include the power output data, working temperature data, internal circuit switching frequency data, current and voltage waveform feature data, and internal diagnostic information data of the inverter, as well as the normal fluctuation range, mean value, variance and other statistical characteristics under certain working conditions. By correcting these directly related mode parameters, it can be ensured that the behavior mode library can still accurately identify the normal operation mode and early wear mode of the inverter under different environmental conditions.

[0168] The technical solutions of the above embodiments effectively solve the adaptability problem of the behavior mode library under dynamic environment by introducing a mechanism for continuous monitoring of environmental parameters and self-adaptive adjustment of working conditions. When the environmental parameters change significantly, the system can identify and trigger the adjustment mechanism in time, avoiding the lag or inaccuracy of the behavior mode library definition caused by environmental changes. Subsequently, by dynamically adjusting the boundary definition of the working conditions, the behavior mode library can more accurately classify different operating conditions, so that the operating data of the inverter under certain environmental conditions can be more accurately classified. Finally, based on the adjusted working condition boundary, the mode parameters affected by the environment in the behavior mode library are corrected to ensure that the normal operation mode and early wear mode in the behavior mode library can reflect the real operating state under the current environment in real time, thereby improving the accuracy and reliability of early warning.

[0169] In a possible design, in step S3, updating the behavior mode library according to the actual operating data and fault feedback of each type of inverter further includes:

[0170] S321, continuously monitoring the operating data of the inverter in the photovoltaic power station, and identifying whether there is abnormal data in the operating data, the abnormal data including data missing, data exceeding the physical range, data mutation or data repetition;

[0171] In this embodiment, the system collects and analyzes the power output data, working temperature data, internal circuit switching frequency data, current and voltage waveform feature data, and internal diagnostic information data from each inverter in real time or quasi-real time. In this process, the system identifies whether there is abnormal data in the operating data. Among them, the abnormal data is the data point deviating from the normal range or mode, including: data missing, that is, failing to collect data at the expected time point; data exceeding the physical range, such as negative power output or temperature far exceeding the upper limit of the device; data mutation, which refers to a sharp and unreasonable jump in data value within a short time; and data repetition, that is, recording the same data value at different time points.

[0172] S322, when identifying that there is abnormal data, classifying the abnormal data according to the type and severity of the abnormal data;

[0173] In this embodiment, data loss is classified as "integrity anomaly" with "high severity", while slight data fluctuation is classified as "noise anomaly" with "low severity". Data classification helps to select the most appropriate data cleaning strategy later.

[0174] S323, data cleaning according to the classification result of data classification, wherein the data cleaning strategy includes data interpolation, data smoothing, data deletion or data weight reduction;

[0175] In this embodiment, the data cleaning strategy is a corrective measure taken for different types of abnormal data, including data interpolation, such as using historical data, adjacent data points or prediction models to fill in missing data; data smoothing, such as eliminating random noise or mutations in data through moving average, Gaussian filtering and other methods to make the trend more obvious; data deletion, i.e. directly deleting abnormal data points judged as serious errors or irreparable; or data weight reduction, i.e. reducing the influence weight of abnormal data points in data processing or model training, rather than completely deleting them.

[0176] S324, using the running data after data cleaning and fault feedback to update the behavior pattern library when the behavior pattern library is updated;

[0177] In this embodiment, only the data that has passed quality check and correction will be used for modeling and adjusting of the behavior pattern, thus ensuring the accuracy and effectiveness of the behavior pattern library.

[0178] The technical solutions of the above embodiments effectively solve the negative impact of abnormal data in the original running data on the update of the behavior pattern library by introducing a data cleaning mechanism. First, by continuously monitoring and identifying abnormal data, potential problems in the data source are discovered in a timely manner. Second, abnormal data is classified and severity evaluated, so that the system can select the most appropriate data cleaning strategy, avoiding information loss that may be caused by "one-size-fits-all" processing, and ensuring that the running data used to update the behavior pattern library has higher accuracy and consistency. Finally, by using the cleaned running data to update the behavior pattern library, it is ensured that the behavior pattern library can more truly and accurately reflect the normal operation mode and early wear mode of the inverter under different working conditions.

[0179] In one possible design, step S4 is followed by:

[0180] S401, when early wear signs are identified, multiple identification channels for different early wear patterns are started in parallel, wherein each identification channel independently analyzes the real-time extracted feature image and calculates the matching degree between the real-time extracted feature image and the early wear pattern concerned by each identification channel;

[0181] In this embodiment, when the system identifies the preliminary signs of early wear of the inverter, for example, when the similarity between the real-time feature portrait and any early wear pattern in the behavior pattern library reaches the preliminary threshold, it is considered to identify the early wear signs. At this time, the system no longer relies solely on a single matching result, but starts multiple identification channels in parallel. Each identification channel is designed to detect one or a class of specific early wear patterns. For example, one channel focuses on detecting power output abnormal patterns, another channel focuses on detecting internal circuit switching frequency abnormal patterns, and a channel focuses on the abnormality of current voltage waveform features. Each identification channel independently analyzes the real-time extracted feature portrait and calculates the matching degree between the feature portrait and the early wear pattern concerned by the channel. Among them, the matching degree is calculated by multiple similarity measurement methods, such as Euclidean distance, cosine similarity or Mahalanobis distance, to quantify the closeness between real-time data and known early wear patterns.

[0182] S402, when the matching degrees of multiple identification channels all reach the preset threshold, triggering a concurrent verification mechanism, wherein the concurrent verification mechanism analyzes the correlation between the early wear patterns identified by each identification channel and outputs the correlation analysis result;

[0183] In this embodiment, when the matching degrees calculated by multiple identification channels all reach the preset threshold, it indicates that multiple early wear patterns may occur simultaneously or interact with each other. The concurrent verification mechanism can deeply analyze the correlation between these early wear patterns identified by multiple channels. For example, through statistical analysis, causal inference or machine learning model to evaluate whether these patterns are independent events, cause and effect, or caused by a deeper problem, finally, the correlation analysis result output will clearly reveal the mutual relationship between different early wear patterns.

[0184] S403, according to the correlation analysis result, adjusting the early warning priority and confidence of the early wear patterns identified by each identification channel;

[0185] In this embodiment, if two early wear patterns are found to have a strong causal relationship, and one of them has a higher potential hazard, its early warning priority may be raised, and the confidence of the other pattern will be adjusted according to its degree as a precursor or accompanying phenomenon of the former. The early warning priority is used to indicate the urgency of the early wear pattern, and the confidence represents the certainty of the system to the identification result of the pattern.

[0186] S404, based on the adjusted early warning priority and confidence, updating the parameters of the corresponding early wear pattern in the behavior pattern library;

[0187] In this embodiment, the update can be dynamically adjusting the feature boundary of the early wear pattern to better adapt to the complex and variable wear performance in actual operation; it can also be updating the early warning trigger conditions, severity assessment parameters and the like related to the pattern, so that the behavior pattern library can continuously learn and evolve.

[0188] The technical solutions of the above embodiments effectively solve the limitations of a single comparison mechanism in handling complex early wear patterns by introducing parallel recognition channels and concurrent verification mechanisms. Specifically, the design of parallel recognition channels enables the system to simultaneously capture and analyze multiple potential early wear signs, avoiding the omissions that can result from a single detection mechanism. The introduction of concurrent verification mechanisms effectively reduces the false positive rate through in-depth analysis of the correlation between different early wear patterns, and more accurately identifies complex or composite early wear patterns, thereby providing more instructive early warning information for maintenance personnel. In addition, dynamically adjusting the early warning priority and confidence, and updating the behavior pattern library accordingly, enables the early warning system to have adaptive and continuous learning capabilities, better adapting to changes in the operating environment of photovoltaic power stations and the emergence of new fault patterns, further improving the efficiency and effectiveness of predictive maintenance, extending the service life of inverters, and reducing operating costs.

[0189] In one example, assume that an inverter in a certain photovoltaic power station shows a slight drop in power output data and a slight fluctuation in internal circuit switching frequency data during operation.

[0190] At this time, the system recognizes early wear signs and starts two recognition channels in parallel: the first recognition channel focuses on the "power drop type early wear pattern", and the second recognition channel focuses on the "switching frequency fluctuation type early wear pattern".

[0191] The first recognition channel analyzes the real-time feature portrait and calculates a matching degree of 85% with the "power drop type early wear pattern", reaching the preset threshold (e.g. 80%).

[0192] The second recognition channel also analyzes the real-time feature portrait and calculates a matching degree of 82% with the "switching frequency fluctuation type early wear pattern", also reaching the preset threshold.

[0193] Since the matching degrees of both recognition channels reach the preset threshold, the system triggers the concurrent verification mechanism. The concurrent verification mechanism analyzes historical data and expert knowledge base and finds that "power drop" and "switching frequency fluctuation" may jointly indicate early aging of a certain key power device (such as an IGBT module) inside the inverter in some cases. The concurrent verification mechanism outputs the correlation analysis result, indicating that there is a strong correlation between the two patterns, possibly originating from the same root problem.

[0194] According to the correlation analysis result, the system adjusts the early warning priority and confidence. For example, since the IGBT module aging is a potential serious fault, the system raises the early warning priority of the "power drop type early wear pattern" to "high" and adjusts the confidence to 90%; at the same time, the early warning priority of the "switching frequency fluctuation type early wear pattern" is adjusted to "high-medium" and the confidence is adjusted to 85%, and it is regarded as an auxiliary verification or early sign of the "power drop type early wear pattern".

[0195] Finally, based on the adjusted early warning priority and confidence, the system updates the parameters of the two early wear patterns in the behavior pattern library, for example, adjusts the feature boundary, so that it can more accurately identify similar multiple correlation fault patterns in the future. In this way, the system can issue more accurate and more instructive early warning signals to prompt the operation and maintenance personnel to check the IGBT module, thereby avoiding more serious faults.

[0196] In summary, the photovoltaic power station operation data processing method provided by the embodiment of the present application can comprehensively and dynamically reflect the operation state of the inverter by collecting multi-dimensional operation data of the inverter, obtaining accurate time stamps, setting event windows, aggregating data, calculating statistical characteristics, and then constructing the feature portrait of each inverter. Subsequently, a behavior pattern library is constructed and continuously updated for each type of inverter, which contains normal operation patterns and early wear patterns, and the performance curve and data fluctuation law specific to different types of inverters are accurately modeled, effectively overcoming the problem of the limitation of the original data cleaning standard due to equipment updates in the prior art. Further, by comparing the real-time feature portrait with the early wear pattern, signs of slight performance degradation of the internal key components of the inverter are found in time, and an early warning signal is issued, thereby solving the technical problem that the existing early fault diagnosis system cannot accurately identify the potential risks of new types of inverters. Finally, the early warning signal is classified according to the comparison accuracy, potential severity and abnormal duration, and a classified early warning notification is output, so that the early warning information is more instructive and operable, and the resource waste or response lag caused by the traditional "one-size-fits-all" alarm mode is avoided.

[0197] In summary, the embodiment of the present application greatly improves the identification ability and early warning efficiency of the photovoltaic power station for early faults of new types of inverters through fine data processing, personalized behavior pattern construction and intelligent early warning classification mechanism, effectively reduces the potential operation risk and maintenance cost, and ensures the stable and efficient operation of the photovoltaic power station.

[0198] Embodiment 2

[0199] The embodiment 2 of the present application provides a photovoltaic power station operation data processing system, Figure 6is a block diagram of a photovoltaic power station operation data processing system according to an exemplary embodiment. Referring to the accompanying drawings Figure 6 The system comprises:

[0200] a data collection module 01 for collecting operation data of each inverter in the photovoltaic power station and obtaining time stamps of each operation data, wherein the operation data comprises power output data, working temperature data, internal circuit switch frequency data, current voltage waveform feature data and internal diagnosis information data of the inverter;

[0201] a feature portrait generation module 02 for setting an event window, aggregating operation data from the same inverter within the event window, calculating statistical features of the aggregated data and constructing a feature portrait of each inverter within the current event window based on the statistical features;

[0202] a behavior pattern library construction module 03 for constructing a behavior pattern library for each type of inverter and updating the behavior pattern library according to actual operation data and fault feedback of each type of inverter, wherein the behavior pattern library contains normal operation modes and early wear-out modes under different working conditions;

[0203] a pre-warning judgment module 04 for comparing the real-time extracted feature portrait with the early wear-out mode, and issuing an early pre-warning signal if the feature portrait is similar to the early wear-out mode;

[0204] a pre-warning classification notification module 05 for classifying the early pre-warning signal according to the comparison accuracy of the feature portrait and the early wear-out mode, the potential severity of the early wear-out mode and the abnormal duration, and outputting a classified pre-warning notification according to the classified result.

[0205] In summary, the photovoltaic power station operation data processing method and system provided by the embodiment of the present application comprehensively and dynamically reflects the operation state of the inverter by collecting multi-dimensional operation data of the inverter, obtaining accurate time stamps, setting event windows, aggregating data, calculating statistical characteristics, and then constructing a feature portrait of each inverter. Subsequently, a behavior pattern library is constructed and continuously updated for each type of inverter, which includes normal operation mode and early wear mode. The unique performance curve and data fluctuation law of different types of inverters are accurately modeled, effectively overcoming the problem of limitations of original data cleaning standards caused by equipment updates in the prior art. Further, by comparing the real-time feature portrait with the early wear mode, signs of slight performance degradation of the internal key components of the inverter are found in time, and an early warning signal is issued, thereby solving the technical problem that the existing early fault diagnosis system cannot accurately identify the potential risks of new types of inverters. Finally, the warning signals are classified according to the accuracy of comparison, potential severity and abnormal duration, and the classified warning notifications are output, so that the warning information is more instructive and operable, avoiding the resource waste or response lag that may be caused by the traditional "one-size-fits-all" alarm mode.

[0206] In summary, the embodiment of the present application greatly improves the identification ability and early warning efficiency of the photovoltaic power station for early faults of new types of inverters through fine data processing, personalized behavior pattern construction and intelligent early warning classification mechanism, effectively reduces potential operation risks and maintenance costs, and ensures stable and efficient operation of the photovoltaic power station.

[0207] The technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above-described embodiments are not described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0208] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for processing photovoltaic power plant operation data, characterized in that, Includes the following steps: The operation data of each inverter in the photovoltaic power station is collected, and the timestamp of each operation data is obtained. The operation data includes the inverter's power output data, operating temperature data, internal circuit switching frequency data, current and voltage waveform characteristic data, and internal diagnostic information data. Set an event window, aggregate the operating data from the same inverter within the event window, calculate the statistical characteristics of the aggregated data, and construct a feature profile of each inverter within the current event window based on the statistical characteristics. A behavior pattern library is built for each type of inverter, and the behavior pattern library is updated according to the actual operating data and fault feedback of each type of inverter. The behavior pattern library includes normal operation mode and early wear mode under different operating conditions. The feature profile extracted in real time is compared with the early wear pattern. If the feature profile is similar to the early wear pattern, an early warning signal is issued. Based on the accuracy of the comparison between the feature profile and the early wear pattern, the potential severity of the early wear pattern, and the duration of the anomaly, the early warning signal is classified, and a classified warning notification is output based on the classification result.

2. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The process involves setting an event window, aggregating operating data from the same inverter within the event window, calculating statistical characteristics of the aggregated data, and constructing a feature profile of each inverter within the current event window based on these statistical characteristics. This includes: The first data processing channel and the second data processing channel operate in parallel. The first data processing channel sets a long event window to continuously collect the running data and calculate the statistical characteristics of the aggregated data to form a feature profile of the inverter within the current long event window. The second data processing channel scans the parameters in the running data stream that are sensitive to external interference at a short time granularity. When a transient abnormal fluctuation is detected, a short time window mechanism is triggered to generate a transient abnormal profile. The instantaneous anomaly profile is compared with a preset set of interference behavior features to determine whether the instantaneous anomaly fluctuation is an external instantaneous interference. If it is determined to be an external transient interference, the operating data covered by the transient anomaly profile within the long event window is specially processed, and the statistical characteristics of the aggregated data are calculated based on the specially processed operating data points. Based on the statistical characteristics, a feature profile of each inverter within the current event window is constructed.

3. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The method involves building a behavior pattern library for each inverter model and updating the behavior pattern library based on actual operating data and fault feedback for each inverter model. The behavior pattern library includes normal operating modes and early wear modes under different operating conditions, including: Periodically evaluate the effectiveness of existing early wear patterns in the behavioral pattern library, wherein the effectiveness evaluation includes: From the operating data of the photovoltaic power station, select historical early wear data samples corresponding to the early wear pattern; The historical early wear data samples are compared with the existing early wear patterns in the behavior pattern library to obtain the comparison results; Based on the comparison results, determine whether the early wear pattern needs to be adjusted; If it is determined that adjustment is needed, the early wear pattern is adjusted based on the historical early wear data sample and new fault feedback. The adjustment includes updating the feature boundary of the early wear pattern or adding a new early wear pattern. If it is determined that no adjustment is needed, then the early wear pattern remains unchanged.

4. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The step of comparing the real-time extracted feature profile with the early wear pattern, and issuing an early warning signal if the feature profile is similar to the early wear pattern, includes: The weights of each feature parameter in the feature profile are determined based on the early wear patterns to be compared. Using a weighted distance calculation method, the weighted distance between the real-time feature profile and the early wear pattern is calculated based on the weights. Based on the weighted distance, the similarity between the real-time feature profile and the early wear pattern is determined; If the real-time feature profile is determined to be similar to the early wear pattern, an early warning signal is issued.

5. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, According to claim 1, a method for processing photovoltaic power plant operation data is characterized in that, the step of classifying the early warning signal based on the accuracy of the comparison between the feature profile and the early wear pattern, the potential severity of the early wear pattern, and the duration of the anomaly includes: Based on the comparison accuracy, the potential severity, and the duration of the anomaly, a multidimensional grading rule is preset, and a combination of early warning levels for the early warning signals within the multidimensional grading rule is defined. Monitor the historical fluctuation range of the comparison accuracy, the potential severity, and the duration of the anomaly; When changes in the comparison accuracy, the potential severity, and the duration of the anomaly are detected to exceed the historical stable fluctuation range within a short period of time, a tiered buffer period is activated. During the tiered buffer period, the warning level of the early warning signal is maintained, and subsequent changes in the comparison accuracy, the potential severity, and the duration of the anomaly are observed. After the tiered buffer period ends, the warning level is determined and adjusted based on the stable values ​​of the comparison accuracy, the potential severity, and the duration of the anomaly. A level stability factor is introduced into the multidimensional grading rules, and the level of warning needs to be adjusted based on the level stability factor. The level stability factor is associated with the duration of the warning level, as well as the comparison accuracy, the potential severity, and the volatility of the abnormal duration.

6. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The aforementioned construction of a behavior pattern library for each inverter model includes: When the historical operating data and fault feedback sample size of a specific inverter model are insufficient, the first sample with sufficient data that is similar to the inverter model is screened from the behavior pattern library. Based on the first sample, select the normal operation mode and early wear mode that are highly correlated with the specific inverter model in terms of key performance parameters and operating characteristics, as the initial reference mode for the new inverter model. Using limited historical operating data and fault feedback from the new inverter model, the initial reference mode is locally adjusted and refined to adapt to the unique operating characteristics of the new inverter model, thus forming an initial behavior mode library for the new inverter model. After the new inverter model is put into operation, its operating data and fault feedback are continuously collected, and the initial behavior pattern library is gradually improved and updated in an incremental manner until the data accumulation of the new inverter model reaches the condition for independently building a behavior pattern library.

7. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The process of updating the behavior pattern library based on actual operating data and fault feedback for each inverter model includes: The system continuously monitors the environmental parameters of the photovoltaic power station, and triggers an adaptive adjustment mechanism when a significant change in the environmental parameters is detected. Based on the current environmental parameters, dynamically adjust the boundary definitions of the working conditions in the behavior pattern library; Based on the adjusted boundary definition of the operating conditions, the environmentally affected pattern parameters in the behavior pattern library are corrected.

8. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The process of updating the behavior pattern library based on actual operating data and fault feedback for each inverter model also includes: Continuously monitor the operating data of inverters in photovoltaic power plants and identify whether there is abnormal data in the operating data. The abnormal data includes missing data, data outside the physical range, data mutation, or data duplication. When the abnormal data is identified, the abnormal data is classified according to its type and severity. The classification results of the data classification are cleaned, wherein the data cleaning strategy includes data imputation, data smoothing, data removal, or data weight reduction; When the behavior pattern library is updated, the operational data that has been cleaned by the data and the fault feedback are used to update the behavior pattern library.

9. The photovoltaic power plant operation data processing method according to claim 1, characterized in that, The step involves comparing the real-time extracted feature profile with the early wear pattern. If the feature profile is similar to the early wear pattern, an early warning signal is issued, followed by: When early signs of wear are detected, multiple identification channels for different early wear patterns are activated in parallel. Each identification channel independently analyzes the feature profile extracted in real time and calculates the matching degree between the feature profile extracted in real time and the early wear pattern of interest of its respective identification channel. When the matching degree of multiple identification channels reaches a preset threshold, a concurrent verification mechanism is triggered, wherein the concurrent verification mechanism analyzes the correlation between the early wear patterns identified by each identification channel and outputs the correlation analysis results. Based on the correlation analysis results, adjust the warning priority and confidence level of the early wear patterns identified by each of the identification channels; Based on the adjusted warning priority and confidence level, the parameters corresponding to the early wear pattern in the behavior pattern library are updated.

10. A photovoltaic power plant operation data processing system, characterized in that, The system includes: The data collection module is used to collect the operating data of each inverter in the photovoltaic power station and obtain the timestamp of each operating data. The operating data includes the inverter's power output data, operating temperature data, internal circuit switching frequency data, current and voltage waveform characteristic data, and internal diagnostic information data. The feature profile generation module is used to set an event window, aggregate the operating data from the same inverter within the event window, calculate the statistical features of the aggregated data, and construct a feature profile of each inverter within the current event window based on the statistical features. The behavior pattern library construction module is used to build a behavior pattern library for each type of inverter and update the behavior pattern library according to the actual operating data and fault feedback of each type of inverter. The behavior pattern library includes normal operation mode and early wear mode under different operating conditions. The early warning judgment module compares the feature profile extracted in real time with the early wear pattern. If the feature profile is similar to the early wear pattern, an early warning signal is issued. The early warning classification notification module classifies the early warning signal based on the accuracy of the comparison between the feature profile and the early wear pattern, the potential severity of the early wear pattern, and the duration of the abnormality, and outputs a classified early warning notification based on the classification result.

Images