Intelligent photovoltaic fuse operation state monitoring method based on deep learning

Abstract

The invention discloses an intelligent photovoltaic fuse operation state monitoring method based on deep learning. The method comprises the following steps: obtaining operation data and recording a corresponding timestamp; executing time alignment and association processing, and generating a photovoltaic fuse operating state data sequence; state features are extracted and arranged to form a photovoltaic fuse state feature sequence; inputting the state characteristic sequence of the photovoltaic fuse into the improved sequential relation regression network, and modeling the operation state; in the modeling process, according to the long-term degradation mechanism of the running state of the photovoltaic fuse, state evolution direction constraint is applied to the running state level; the continuous degradation position characterization quantity is output, and the operation state grade of the photovoltaic fuse is determined; in the modeling process, cross-time consistency judgment is executed, and the current running state level is updated; and outputting the monitoring result of the running state of the photovoltaic fuse. According to the invention, an improved sequential relation regression deep learning method is adopted, and stable and continuous monitoring of the operation state of the photovoltaic fuse is realized.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation system operation monitoring technology, and in particular to a method for monitoring the operation status of intelligent photovoltaic fuses based on deep learning. Background Technology

[0002] With the continuous expansion of photovoltaic (PV) power generation system installations, PV arrays place higher demands on the safety and stability of the power system during long-term operation. As a key electrical component for current protection in PV power generation systems, the operating status of PV fuses directly affects the safe operation of module branches, combiner units, and downstream power equipment. Existing methods for monitoring the operating status of PV fuses largely rely on threshold judgments of single or limited operating parameters such as current and temperature, or on periodic inspections based on manual experience. These methods struggle to effectively characterize the gradual degradation of fuses caused by thermal stress, electrical aging, and other factors during long-term operation.

[0003] While some existing technologies incorporate data analysis or machine learning methods to identify the status of photovoltaic fuses, most treat operating status as a discrete and reversible classification problem. This ignores the objective fact that fuse operating status exhibits ordered and irreversible degradation characteristics over time, making them susceptible to misjudgments due to short-term operating condition fluctuations or noise interference. Furthermore, existing technologies typically lack mechanisms to constrain the consistency of operating status across time, making it difficult to achieve stable, continuous, and equipment degradation-compliant determinations of fuse operating status, and thus failing to meet the practical needs of long-term online monitoring of photovoltaic power generation systems. Summary of the Invention

[0004] One objective of this invention is to propose a deep learning-based intelligent photovoltaic fuse operation status monitoring method. This invention employs an improved order relation regression deep learning method to achieve stable and continuous monitoring of the photovoltaic fuse operation status.

[0005] A method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to an embodiment of the present invention includes the following steps:

[0006] Output the photovoltaic fuse operation data after cross-time consistency determination, obtain the target photovoltaic fuse operation data and record the corresponding timestamp;

[0007] Perform time alignment and correlation processing on the operating data to generate a sequence of photovoltaic fuse operating status data.

[0008] Based on the photovoltaic fuse operating status data sequence, the status features are extracted and arranged in chronological order to form a photovoltaic fuse status feature sequence;

[0009] The state feature sequence of photovoltaic fuses is input into an improved order relation regression network to model the operating state of photovoltaic fuses.

[0010] In the modeling process of the improved ordinal regression network, the state evolution direction constraint is applied to the operating state level based on the long-term degradation mechanism of the photovoltaic fuse operating state.

[0011] The improved ordinal regression network outputs a continuous degradation location representation quantity, and the operating status level of the photovoltaic fuse is determined based on the correspondence between the continuous degradation location representation quantity and the preset operating status interval.

[0012] In the process of modeling the operating status of photovoltaic fuses, based on the determination results of the operating status level under the continuous time index, cross-time consistency determination is performed on the operating status level and the current operating status level of the photovoltaic fuse is updated.

[0013] Status level, as the result of monitoring the operating status of photovoltaic fuses.

[0014] Optionally, the acquisition of the runtime data includes:

[0015] Acquire the target photovoltaic fuse in the photovoltaic power generation system and determine the target photovoltaic fuse as the object of operation data collection;

[0016] Within the preset monitoring time window, current data of the target photovoltaic fuse is collected according to the preset sampling cycle, and a corresponding timestamp is recorded for each current data collection to form current data with timestamps.

[0017] Within a preset monitoring time window, temperature data of the target photovoltaic fuse is collected according to the same preset sampling period as that for current data collection, and a timestamp is recorded for each temperature data collection to form time-stamped temperature data.

[0018] Within the preset monitoring time window, the operating condition data of the target photovoltaic fuse at the corresponding acquisition time is acquired, and the corresponding timestamp is recorded for the operating condition data to form operating condition data with timestamps.

[0019] The current data, temperature data, and operating condition data are combined with their corresponding timestamps to form the operating data of the target photovoltaic fuse within a preset monitoring time window.

[0020] Optionally, the generation of the photovoltaic fuse operating status data sequence includes:

[0021] Read the timestamps corresponding to the current data, temperature data, and operating condition data in the operation data;

[0022] Based on the timestamp, the start time of the unified time index is determined as the earliest time in the timestamp, and the end time of the unified time index is determined as the latest time in the timestamp. A unified time index covering the start time to the end time is generated with a preset sampling period as the time step.

[0023] Time alignment is performed on current data, temperature data, and operating condition data according to a unified time index.

[0024] The current data, temperature data, and operating condition data that have completed time alignment processing under the same time index are associated and arranged in the order of the unified time index to generate a sequence of photovoltaic fuse operating status data arranged in chronological order.

[0025] Optionally, the generation of the photovoltaic fuse state characteristic sequence includes:

[0026] Obtain the photovoltaic fuse operation status data sequence, and read the photovoltaic fuse operation status data records in the photovoltaic fuse operation status data sequence in chronological order;

[0027] For each photovoltaic fuse operation status data record, the corresponding current data, temperature data, and operating condition data are read as input data for status feature extraction.

[0028] Based on current data, the characteristics of current change are calculated according to the relationship between current data under adjacent time indices;

[0029] Based on temperature data, temperature change characteristics are calculated according to the relationship between temperature data under adjacent time indices.

[0030] Based on the operating condition data, the operating condition data is converted into operating condition features according to the preset operating condition coding rules;

[0031] The current change characteristics, temperature change characteristics, and operating condition characteristics obtained under the same time index are combined to generate the state characteristics under the corresponding time index.

[0032] The state features generated under each time index are arranged in chronological order to form a photovoltaic fuse state feature sequence.

[0033] Optionally, modeling the operating status of the photovoltaic fuse includes:

[0034] Obtain the state feature sequence of the photovoltaic fuse, and keep the time order of each state feature in the photovoltaic fuse state feature sequence unchanged. Use the photovoltaic fuse state feature sequence as the input of the improved order relation regression network.

[0035] Determine the operating status level of the photovoltaic fuse and clarify the order of operation status levels;

[0036] According to the time sequence of the photovoltaic fuse state feature sequence, the state features under each time index are sequentially input into the improved order relation regression network, and network mapping processing is performed on each state feature to generate a continuous state representation that corresponds one-to-one with each time index.

[0037] Based on the sequential relationship between the operational state levels, the execution order relationship constraint processing of continuous state representation is performed.

[0038] The continuous state representation after being processed by the order relation constraint is used as the output of the modeling result of the photovoltaic fuse operation state by the improved order relation regression network.

[0039] Optionally, applying state evolution direction constraints to the operating state level includes:

[0040] In the process of modeling the improved order relation regression network, the operating state level of the photovoltaic fuse under the continuous time index is obtained, and the preset evolution direction of the operating state level is determined according to the sequential relationship between the operating state levels.

[0041] For the running status levels participating in modeling under two adjacent time indices, the running status level corresponding to the previous time index is determined as the current running status level, and the running status level corresponding to the next time index is determined as the candidate running status level.

[0042] Based on the sequential relationship between the operational status levels, determine the positional relationship of the candidate operational status level relative to the current operational status level in the preset evolution direction;

[0043] When a candidate running status level is the same as the current running status level, or when a candidate running status level is located in a preset evolution direction, the candidate running status level is confirmed as the running status level corresponding to that time index during the modeling process.

[0044] When a candidate operating state level is in the opposite position to the current operating state level, the operating state level corresponding to that time index is limited to the current operating state level during the modeling process, thereby completing the state evolution direction constraint imposed on the operating state level.

[0045] Optionally, determining the operating status level of the photovoltaic fuse includes:

[0046] Obtain the characterization values ​​of consecutive degradation locations and arrange them in chronological order;

[0047] Define a preset operating state range, and each preset operating state range corresponds to a unique operating state level;

[0048] For each time index, the continuous degradation position representation quantity is determined according to the interval boundary of the preset running state interval.

[0049] The operating state level corresponding to the preset operating state interval where the continuous degradation position characterization quantity is located is determined as the operating state level under this time index;

[0050] Output the operating status level determined under each time index, which will be used as the operating status level of the photovoltaic fuse.

[0051] Optionally, the update of the current operating status level of the photovoltaic fuse includes:

[0052] Obtain the running status level determined under continuous time index, and arrange the running status levels according to the order of the time index to form a running status level sequence;

[0053] Determine the number of consecutive time indices used for cross-time consistency determination, and construct a consecutive time index window in the running state level sequence based on the number of consecutive time indices;

[0054] For a continuous time index window, determine whether all the running status levels contained in the continuous time index window are the same to obtain the cross-time consistency determination result;

[0055] When the cross-time consistency determination result is consistent, the running status level corresponding to the last time index in the continuous time index window is confirmed as the current running status level;

[0056] When the cross-time consistency determination result is inconsistent, the current running status level remains unchanged, and cross-time consistency determination is performed on the running status level in subsequent continuous time index windows.

[0057] Optionally, the generation of the photovoltaic fuse operation status monitoring results includes:

[0058] Obtain the photovoltaic fuse operating status level after cross-time consistency determination, and determine the photovoltaic fuse operating status level as the valid operating status level under the current time index;

[0059] The effective operating status level is associated with the corresponding time index to generate photovoltaic fuse operating status monitoring results;

[0060] Under continuous time indexing, the monitoring results of the operating status of multiple photovoltaic fuses are arranged in the order of the time index to form a sequence of photovoltaic fuse operating status monitoring results;

[0061] Output the photovoltaic fuse operation status monitoring results or photovoltaic fuse operation status monitoring result sequence as the photovoltaic fuse operation status monitoring results.

[0062] The beneficial effects of this invention are:

[0063] This invention addresses the problem of unstable, continuous, and degradation-compliant operation status determination of photovoltaic fuses in photovoltaic power generation systems over long-term operation. It proposes a deep learning-based intelligent photovoltaic fuse operation status monitoring method. By performing time-series modeling and constraint processing of multi-source operating data, reliable perception and accurate determination of the fuse's operating status are achieved. Compared to traditional monitoring methods relying on fixed thresholds or human experience, this invention comprehensively utilizes current data, temperature data, and operating condition data to form a time-series of operating status data. This comprehensively reflects the dynamic changes of photovoltaic fuses under real-world operating conditions, providing a more complete data foundation for operation status analysis.

[0064] Furthermore, this invention introduces an improved order relation regression network to model the operating state of photovoltaic fuses as an evolutionary process with sequential and irreversible characteristics. This ensures that the operating state determination results maintain a consistent evolutionary direction over time, avoiding repeated state jumps caused by short-term fluctuations or abnormal operating conditions. By continuously characterizing the degree of fuse degradation through continuous degradation location representations and combining this with preset operating state intervals to determine the operating state level, the operating state determination is transformed from traditional discrete classification to interval mapping determination based on a continuous degradation process. This improves the continuity and engineering interpretability of the operating state determination.

[0065] Furthermore, this invention introduces a cross-time consistency determination mechanism during the operational status determination process. The current operational status level is only updated when it remains consistent across multiple consecutive time indices. This effectively suppresses the impact of noise interference and instantaneous operating condition changes on the monitoring results, improving the stability and reliability of the operational status monitoring results. Through the above technical solution, this invention enables long-term online and stable monitoring of the operational status of photovoltaic fuses, providing reliable technical support for intelligent operation and maintenance, risk warning, and safety management of photovoltaic power generation systems, and possesses significant engineering application value. Attached Figure Description

[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0067] Figure 1 This is an overall flowchart of a deep learning-based intelligent photovoltaic fuse operation status monitoring method proposed in this invention.

[0068] Figure 2 This is a schematic diagram of the improved order relation regression network used to model the operating status of a photovoltaic fuse in a deep learning-based intelligent photovoltaic fuse operating status monitoring method proposed in this invention.

[0069] Figure 3 This is a schematic diagram illustrating the mapping relationship between the continuous degradation position representation quantity and the preset operating state interval in the intelligent photovoltaic fuse operating status monitoring method based on deep learning proposed in this invention. Detailed Implementation

[0070] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0071] refer to Figures 1-3 A deep learning-based method for monitoring the operating status of intelligent photovoltaic fuses includes the following steps:

[0072] The system acquires the operating data of the target photovoltaic fuse in the photovoltaic power generation system within a preset monitoring time window, and records the corresponding timestamps for the operating data. The operating data includes current data, temperature data, and operating condition data.

[0073] The operating data is time-aligned, and the current data, temperature data, and operating condition data are correlated based on a unified time index to generate a sequence of photovoltaic fuse operating status data arranged in chronological order.

[0074] Based on the photovoltaic fuse operating status data sequence, state features that characterize the changes in the operating status of photovoltaic fuses are extracted, and the state features are arranged in chronological order to form a photovoltaic fuse state feature sequence.

[0075] The photovoltaic fuse state feature sequence is input into an improved order relation regression network to model the photovoltaic fuse operating state. The order relation regression network uses the operating state levels with a sequential order as modeling constraints.

[0076] In the modeling process of the improved order relation regression network, based on the long-term degradation mechanism of the photovoltaic fuse operating state, the state evolution direction constraint is applied to the operating state level, so that the photovoltaic fuse operating state level is only allowed to maintain the current operating state level or evolve towards a preset higher risk operating state level.

[0077] An improved ordinal regression network outputs a continuous degradation location characterization quantity to characterize the degree of degradation of a photovoltaic fuse. Based on the correspondence between the continuous degradation location characterization quantity and the preset operating state interval, the operating state level of the photovoltaic fuse is determined.

[0078] In the process of modeling the operating status of photovoltaic fuses, based on the determination results of the operating status level under continuous time index, cross-time consistency determination is performed on the operating status level. The current operating status level of the photovoltaic fuse is updated only when the operating status level is consistent under multiple consecutive time indices.

[0079] Output the photovoltaic fuse operation status level after cross-time consistency determination, as the photovoltaic fuse operation status monitoring result.

[0080] In this embodiment, the acquisition of the running data includes:

[0081] Acquire the target photovoltaic fuse in the photovoltaic power generation system and determine the target photovoltaic fuse as the object of operation data collection;

[0082] Within the preset monitoring time window, current data of the target photovoltaic fuse is collected according to the preset sampling cycle, and a corresponding timestamp is recorded for each current data collection to form current data with timestamps.

[0083] Within a preset monitoring time window, temperature data of the target photovoltaic fuse is collected according to the same preset sampling period as that for current data collection, and a timestamp is recorded for each temperature data collection to form time-stamped temperature data.

[0084] Within the preset monitoring time window, the operating condition data of the target photovoltaic fuse at the corresponding acquisition time is acquired, and the corresponding timestamp is recorded for the operating condition data to form operating condition data with timestamps.

[0085] The current data, temperature data, and operating condition data are combined with their corresponding timestamps to form the operating data of the target photovoltaic fuse within a preset monitoring time window.

[0086] In this embodiment, the generation of the photovoltaic fuse operating status data sequence includes:

[0087] Read the timestamps corresponding to the current data, temperature data, and operating condition data in the operation data;

[0088] Based on the timestamp, the start time of the unified time index is determined as the earliest time in the timestamp, and the end time of the unified time index is determined as the latest time in the timestamp. A unified time index covering the start time to the end time is generated with a preset sampling period as the time step.

[0089] According to the unified time index, the current data is time aligned so that the current data corresponding to each time index is consistent with the time position of that time index.

[0090] According to a unified time index, time alignment processing is performed on the temperature data to ensure that the temperature data corresponding to each time index is consistent with the time position of that time index.

[0091] Based on a unified time index, time alignment processing is performed on the operating condition data to ensure that the operating condition data corresponding to each time index is consistent with the time position of that time index.

[0092] Based on a unified time index, current data, temperature data, and operating condition data that have completed time alignment processing under the same time index are associated and arranged in the order of the unified time index to generate a sequence of photovoltaic fuse operating status data arranged in chronological order.

[0093] In this embodiment, the generation of the photovoltaic fuse state characteristic sequence includes:

[0094] Obtain the photovoltaic fuse operation status data sequence, and read the photovoltaic fuse operation status data records in the photovoltaic fuse operation status data sequence in chronological order;

[0095] For each photovoltaic fuse operation status data record, the corresponding current data, temperature data, and operating condition data are read as input data for status feature extraction.

[0096] Based on current data, according to the relationship between current data under adjacent time indices, a current change characteristic is calculated to characterize the current change of the photovoltaic fuse. The current change characteristic includes the absolute value of the difference between current data under adjacent time indices and the ratio of the difference to a preset sampling period.

[0097] Based on temperature data, according to the relationship between temperature data under adjacent time indices, a temperature change feature is calculated to characterize the temperature change of the photovoltaic fuse. The temperature change feature includes the absolute value of the difference between temperature data under adjacent time indices and the ratio of the difference to a preset sampling period.

[0098] Based on the operating condition data, the operating condition data is converted into operating condition features according to the preset operating condition coding rules;

[0099] The preset operating condition coding rule is based on the discrete categories of the operating conditions of photovoltaic fuses in the photovoltaic power generation system. The corresponding operating condition data is mapped in a regular way, and different operating conditions are converted into numerical operating condition features that can be used for state feature extraction and modeling according to a predetermined coding method.

[0100] The current change characteristics, temperature change characteristics, and operating condition characteristics obtained under the same time index are combined to generate the state characteristics under the corresponding time index.

[0101] The state features generated under each time index are arranged in chronological order to form a photovoltaic fuse state feature sequence.

[0102] In this embodiment, modeling the operating status of the photovoltaic fuse includes:

[0103] Obtain the state feature sequence of the photovoltaic fuse, and keep the time order of each state feature in the photovoltaic fuse state feature sequence unchanged. Use the photovoltaic fuse state feature sequence as the input of the improved order relation regression network.

[0104] The improved sequence regression network takes the photovoltaic fuse state feature sequence as input and introduces the sequential relationship between the operating state levels in the modeling process. It constrains the direction of change of the operating state under adjacent time indices, so that the operating state is only allowed to remain or evolve towards a higher risk direction, thereby making the modeling process conform to the long-term degradation mechanism of the photovoltaic fuse operating state.

[0105] The operating status levels of photovoltaic fuses are determined, and the sequential relationship between the operating status levels is clarified. The sequential relationship is used as a modeling constraint for the improved order relationship regression network.

[0106] According to the time sequence of the photovoltaic fuse state feature sequence, the state features under each time index are sequentially input into the improved order relation regression network, and network mapping processing is performed on each state feature to generate a continuous state representation that corresponds one-to-one with each time index.

[0107] Based on the sequential relationship between the operational state levels, the sequential relationship constraint processing is performed on the continuous state representation to ensure that the continuous state representation under adjacent time indices satisfies the sequential relationship of the operational state levels in the direction of change.

[0108] The order relation constraint processing specifically includes continuously checking and restricting the direction of change of continuous state representation in the time dimension during the modeling process, so that the evolution of the running state always conforms to the preset order of the running state levels, thereby avoiding reverse order jumps in the running state.

[0109] The continuous state representation after being processed by the order relation constraint is used as the output of the modeling result of the photovoltaic fuse operation state by the improved order relation regression network.

[0110] In this embodiment, applying state evolution direction constraints to the operating state level includes:

[0111] In the modeling process of the improved order relation regression network, the operating state level of the photovoltaic fuse under the continuous time index is obtained, and the preset evolution direction of the operating state level is determined according to the sequential relationship between the operating state levels. The preset evolution direction points to a higher risk operating state level.

[0112] For the running status levels participating in modeling under two adjacent time indices, the running status level corresponding to the previous time index is determined as the current running status level, and the running status level corresponding to the next time index is determined as the candidate running status level.

[0113] Based on the sequential relationship between the operational status levels, determine the positional relationship of the candidate operational status level relative to the current operational status level in the preset evolution direction;

[0114] When a candidate running status level is the same as the current running status level, or when a candidate running status level is located in a preset evolution direction, the candidate running status level is confirmed as the running status level corresponding to that time index during the modeling process.

[0115] When a candidate operating state level is in the opposite position to the current operating state level, the operating state level corresponding to that time index is limited to the current operating state level during the modeling process, thereby completing the state evolution direction constraint imposed on the operating state level.

[0116] In this embodiment, determining the operating status level of the photovoltaic fuse includes:

[0117] Obtain continuous degradation location characterization quantities to characterize the degree of degradation of photovoltaic fuses, and arrange the continuous degradation location characterization quantities in chronological order;

[0118] The continuous degradation position representation is a continuous numerical representation output by the improved order relation regression network during the operation state modeling process. It is used to characterize the degradation position of the photovoltaic fuse in the long-term operation state evolution path under the premise of satisfying the order of operation state levels and irreversible degradation constraints, and serves as a continuous basis for determining the operation state level.

[0119] A preset operating state interval is determined, which consists of multiple intervals arranged in a sequential order, and each preset operating state interval corresponds to a unique operating state level.

[0120] For each time index, the continuous degradation position representation quantity is determined according to the interval boundary of the preset running state interval.

[0121] The operating state level corresponding to the preset operating state interval where the continuous degradation position characterization quantity is located is determined as the operating state level under this time index;

[0122] Output the operating status level determined under each time index, which will be used as the operating status level of the photovoltaic fuse.

[0123] In this embodiment, updating the current operating status level of the photovoltaic fuse includes:

[0124] Obtain the running status level determined under continuous time index, and arrange the running status levels according to the order of the time index to form a running status level sequence;

[0125] The number of continuous time indices used for cross-time consistency determination is determined. The number of continuous time indices corresponds to the preset monitoring time window and preset sampling period of the photovoltaic fuse. Based on the number of continuous time indices, a continuous time index window is constructed in the operating status level sequence.

[0126] For a continuous time index window, determine whether all the running status levels contained in the continuous time index window are the same to obtain the cross-time consistency determination result;

[0127] When the cross-time consistency determination result is consistent, the running status level corresponding to the last time index in the continuous time index window is confirmed as the current running status level;

[0128] When the cross-time consistency determination result is inconsistent, the current running status level remains unchanged, and cross-time consistency determination is performed on the running status level in subsequent continuous time index windows.

[0129] In this embodiment, the generation of the photovoltaic fuse operation status monitoring results includes:

[0130] Obtain the photovoltaic fuse operating status level after cross-time consistency determination, and determine the photovoltaic fuse operating status level as the valid operating status level under the current time index;

[0131] The effective operating status level is associated with the corresponding time index to generate photovoltaic fuse operating status monitoring results;

[0132] Under continuous time indexing, the monitoring results of the operating status of multiple photovoltaic fuses are arranged in the order of the time index to form a sequence of photovoltaic fuse operating status monitoring results;

[0133] Output the photovoltaic fuse operation status monitoring results or photovoltaic fuse operation status monitoring result sequence as the photovoltaic fuse operation status monitoring results.

[0134] Example 1:

[0135] To verify the feasibility of this invention in practice, it was applied to a scenario involving the monitoring of the DC-side branch operation status of a centralized photovoltaic power plant. This photovoltaic power plant has a long operating time, a large photovoltaic array, and numerous DC-side branches. Fuses are distributed at key locations in each branch, playing a crucial protective role during long-term grid-connected operation. Because photovoltaic power generation systems are significantly affected by factors such as sunlight conditions, ambient temperature, and load variations, fuses continuously endure current surges and thermal stresses during long-term operation. Their operating status often exhibits slow, phased, and irreversible degradation characteristics. Traditional methods based on threshold judgment or manual experience analysis are insufficient for continuous and stable monitoring of this degradation process, easily leading to misjudgments of the operating status due to short-term fluctuations, and failing to meet the actual requirements for the long-term safe operation of photovoltaic power generation systems.

[0136] In this application scenario, the photovoltaic power generation system already possesses the capability to perform basic data collection on the branch circuit operation status, continuously acquiring current, temperature, and operating condition information corresponding to each fuse. After deploying the method of this invention into the existing operation monitoring system, operating data of the target photovoltaic fuse is continuously collected within a preset monitoring time window, and the collected operating data is uniformly recorded with a time identifier. Addressing the data differences arising from different sensor sources and different sampling times, a unified time index is constructed through time alignment processing, enabling accurate correlation between current data, temperature data, and operating condition data within the same time dimension, thereby forming an operating status data sequence that reflects the actual operation process of the fuse.

[0137] During the operational status analysis, the system extracts state features based on the generated operational status data sequence to characterize changes in the fuse's operational status. These extracted features not only reflect changes in the operational status at a single moment but also demonstrate the evolutionary trend of the operational status over time, enabling the features to comprehensively reflect the overall state changes of the fuse at different operational stages. Subsequently, the chronologically ordered sequence of state features is input into an improved order regression network to model the photovoltaic fuse's operational status. This network introduces sequential constraints between operational status levels during the modeling process, ensuring that the model, when learning the patterns of state changes, always follows the objective law of the fuse's operational status gradually evolving from low risk to high risk, thereby avoiding reverse changes in operational status due to short-term anomalies or noise interference.

[0138] In the modeling process, the improved ordinal regression network outputs a continuous degradation location representation quantity to characterize the degradation degree of the photovoltaic fuse. This continuous degradation location representation quantity changes continuously with the time index and is used to describe the position of the fuse in the evolution path of its operating state. Through this continuous degradation location representation quantity, the originally discrete and difficult-to-compare operating states are transformed into a degradation location description in continuous space, making the operating states under different time indices comparable and continuous. Based on this, the continuous degradation location representation quantity is mapped to pre-defined operating state intervals to obtain the operating state level determination results under each time index, thereby achieving hierarchical identification of the fuse's operating state.

[0139] Considering the challenges of varying sunlight, load adjustments, and environmental disturbances in the actual operation of photovoltaic power generation systems, this invention further introduces a cross-time consistency determination mechanism during the operation status determination process. By confirming the consistency of the operation status levels under a continuous time index, the current operation status level of the photovoltaic fuse is only updated when the operation status level remains consistent over a continuous period. This mechanism effectively suppresses the impact of short-term fluctuations on the operation status determination results, making the final output operation status monitoring results more stable and able to truly reflect the long-term trend of the fuse's operation status.

[0140] During the continuous operation of this photovoltaic power plant, by applying the method of this invention online over a long period, operators can continuously acquire the overall situation of the photovoltaic fuse's operating status evolution over time, and analyze and judge the operating risks of the fuse based on stable operating status monitoring results. Compared with previous methods that relied on manual inspection or simple rule analysis, this invention can achieve continuous perception, orderly modeling, and stable output of the photovoltaic fuse's operating status without increasing human intervention. It effectively solves the problem of reliably monitoring the long-term degradation state of fuses in existing technologies, providing strong technical support for the safe operation and intelligent maintenance of photovoltaic power generation systems.

[0141] Table 1. Comparative Experimental Data of Photovoltaic Fuse Operation Status Monitoring Methods

[0142] Performance indicators Traditional threshold determination method Traditional machine learning methods Method of the present invention Operational status misjudgment rate (%) 14.6 8.9 2.7 Number of status fluctuations (times / day) 11.3 6.4 4.2 Duration of continuous state (in hours) 4.1 7.6 14.9 Consistency score of degradation trend (0-100) 62 78 93 Operation and maintenance intervention frequency (times / month) 9.2 5.7 2.1

[0143] As shown in Table 1, the difference in accuracy among the three methods can be directly observed from the comparison of the false judgment rates of the operating status. The false judgment rate of the traditional threshold method is 14.6%, indicating that in actual operation, a false judgment occurs approximately once every seven status judgments. This is mainly due to the high sensitivity of the threshold method to instantaneous fluctuations in current and temperature. The traditional machine learning method reduces the false judgment rate to 8.9%, which is an improvement over the threshold method, but there is still a significant margin for error. The method of this invention further reduces the false judgment rate to 2.7%, indicating that by introducing order relation regression modeling and state evolution direction constraints, the operating status judgment is transformed from "instantaneous judgment" to evolutionary trend judgment, weakening the impact of short-term anomalies on the judgment results.

[0144] The difference is even more apparent in the metric of state fluctuation frequency. Traditional thresholding methods result in 11.3 state fluctuations per day, meaning the running state switches frequently throughout the day, changing almost every two hours, severely impacting readability. Traditional machine learning methods reduce the frequency to 6.4 times per day, but significant state jitter still exists. The method of this invention exhibits only 4.2 state fluctuations per day, directly reflecting the effectiveness of the continuous degradation position representation and cross-temporal consistency determination mechanism in suppressing frequent state jumps.

[0145] The data on the duration of continuous state maintenance further validates the improved state stability from a reverse perspective. Under traditional thresholding methods, the duration of continuous state maintenance is only 4.1 hours, indicating that the operating state is difficult to maintain consistency over a long period. Traditional machine learning methods increase this duration to 7.6 hours, but this is still insufficient to support long-term operational judgments. The continuous state maintenance duration of the method in this invention reaches 14.9 hours, demonstrating that the method can stably characterize the operating stage of the fuse on a longer timescale, better reflecting the continuous characteristics of the actual degradation process.

[0146] In terms of consistency scoring of degradation trends, the method of this invention also demonstrates a significant advantage. The traditional threshold method scores 62, indicating a low degree of matching between its judgment results and the actual degradation trend; the traditional machine learning method improves to 78, but still suffers from the problem of trend recurrence. The method of this invention achieves a score of 93, an improvement of 31 points compared to the threshold method and 15 points compared to the traditional machine learning method. This indicates that through order relation constraints and irreversible evolution modeling, the trend of changes in the operating state over time is more consistent and can more realistically reflect the long-term degradation behavior of the fuse.

[0147] From the perspective of the operation and maintenance intervention frequency, a key engineering application metric, the traditional threshold method requires 9.2 interventions per month, the traditional machine learning method requires 5.7 interventions per month, while the method of this invention requires only 2.1 interventions per month. This difference indicates that because the operational status results output by this invention are more stable and have a lower error rate, operators do not need to perform frequent manual reviews and interventions. Compared to the traditional threshold method, the number of operation and maintenance interventions is reduced by approximately 77%, and compared to the traditional machine learning method, it is reduced by approximately 63%, directly demonstrating the application value of this invention in actual operation and maintenance.

[0148] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring the operating status of intelligent photovoltaic fuses based on deep learning, characterized in that, Includes the following steps: Output the photovoltaic fuse operation data after cross-time consistency determination, obtain the target photovoltaic fuse operation data and record the corresponding timestamp; Perform time alignment and correlation processing on the operating data to generate a sequence of photovoltaic fuse operating status data. Based on the photovoltaic fuse operating status data sequence, the status features are extracted and arranged in chronological order to form a photovoltaic fuse status feature sequence; The state feature sequence of photovoltaic fuses is input into an improved order relation regression network to model the operating state of photovoltaic fuses. In the modeling process of the improved ordinal regression network, the state evolution direction constraint is applied to the operating state level based on the long-term degradation mechanism of the photovoltaic fuse operating state. The improved ordinal regression network outputs a continuous degradation location representation quantity, and the operating status level of the photovoltaic fuse is determined based on the correspondence between the continuous degradation location representation quantity and the preset operating status interval. In the process of modeling the operating status of photovoltaic fuses, based on the determination results of the operating status level under the continuous time index, cross-time consistency determination is performed on the operating status level and the current operating status level of the photovoltaic fuse is updated. Status level, as the result of monitoring the operating status of photovoltaic fuses.

2. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The acquisition of the operational data includes: Acquire the target photovoltaic fuse in the photovoltaic power generation system and determine the target photovoltaic fuse as the object of operation data collection; Within the preset monitoring time window, current data of the target photovoltaic fuse is collected according to the preset sampling cycle, and a corresponding timestamp is recorded for each current data collection to form current data with timestamps. Within a preset monitoring time window, temperature data of the target photovoltaic fuse is collected according to the same preset sampling period as that for current data collection, and a timestamp is recorded for each temperature data collection to form time-stamped temperature data. Within the preset monitoring time window, the operating condition data of the target photovoltaic fuse at the corresponding acquisition time is acquired, and the corresponding timestamp is recorded for the operating condition data to form operating condition data with timestamps. The current data, temperature data, and operating condition data are combined with their corresponding timestamps to form the operating data of the target photovoltaic fuse within a preset monitoring time window.

3. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The generation of the photovoltaic fuse operating status data sequence includes: Read the timestamps corresponding to the current data, temperature data, and operating condition data in the operation data; Based on the timestamp, the start time of the unified time index is determined as the earliest time in the timestamp, and the end time of the unified time index is determined as the latest time in the timestamp. A unified time index covering the start time to the end time is generated with a preset sampling period as the time step. Time alignment is performed on current data, temperature data, and operating condition data according to a unified time index. The current data, temperature data, and operating condition data that have completed time alignment processing under the same time index are associated and arranged in the order of the unified time index to generate a sequence of photovoltaic fuse operating status data arranged in chronological order.

4. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The generation of the photovoltaic fuse state characteristic sequence includes: Obtain the photovoltaic fuse operation status data sequence, and read the photovoltaic fuse operation status data records in the photovoltaic fuse operation status data sequence in chronological order; For each photovoltaic fuse operation status data record, the corresponding current data, temperature data, and operating condition data are read as input data for status feature extraction. Based on current data, the characteristics of current change are calculated according to the relationship between current data under adjacent time indices; Based on temperature data, temperature change characteristics are calculated according to the relationship between temperature data under adjacent time indices. Based on the operating condition data, the operating condition data is converted into operating condition features according to the preset operating condition coding rules; The current change characteristics, temperature change characteristics, and operating condition characteristics obtained under the same time index are combined to generate the state characteristics under the corresponding time index. The state features generated under each time index are arranged in chronological order to form a photovoltaic fuse state feature sequence.

5. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The modeling of the operating status of photovoltaic fuses includes: Obtain the state feature sequence of the photovoltaic fuse, and keep the time order of each state feature in the photovoltaic fuse state feature sequence unchanged. Use the photovoltaic fuse state feature sequence as the input of the improved order relation regression network. Determine the operating status level of the photovoltaic fuse and clarify the order of operation status levels; According to the time sequence of the photovoltaic fuse state feature sequence, the state features under each time index are sequentially input into the improved order relation regression network, and network mapping processing is performed on each state feature to generate a continuous state representation that corresponds one-to-one with each time index. Based on the sequential relationship between the operational state levels, the execution order relationship constraint processing of continuous state representation is performed. The continuous state representation after being processed by the order relation constraint is used as the output of the modeling result of the photovoltaic fuse operation state by the improved order relation regression network.

6. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The constraint on the state evolution direction of the operating state level includes: In the process of modeling the improved order relation regression network, the operating state level of the photovoltaic fuse under the continuous time index is obtained, and the preset evolution direction of the operating state level is determined according to the sequential relationship between the operating state levels. For the running status levels participating in modeling under two adjacent time indices, the running status level corresponding to the previous time index is determined as the current running status level, and the running status level corresponding to the next time index is determined as the candidate running status level. Based on the sequential relationship between the operational status levels, determine the positional relationship of the candidate operational status level relative to the current operational status level in the preset evolution direction; When a candidate running status level is the same as the current running status level, or when a candidate running status level is located in a preset evolution direction, the candidate running status level is confirmed as the running status level corresponding to that time index during the modeling process. When a candidate operating state level is in the opposite position to the current operating state level, the operating state level corresponding to that time index is limited to the current operating state level during the modeling process, thereby completing the state evolution direction constraint imposed on the operating state level.

7. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The determination of the operating status level of the photovoltaic fuse includes: Obtain the characterization values ​​of consecutive degradation locations and arrange them in chronological order; Define a preset operating state range, and each preset operating state range corresponds to a unique operating state level; For each time index, the continuous degradation position representation quantity is determined according to the interval boundary of the preset running state interval. The operating state level corresponding to the preset operating state interval where the continuous degradation position characterization quantity is located is determined as the operating state level under this time index; Output the operating status level determined under each time index, which will be used as the operating status level of the photovoltaic fuse.

8. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The updated current operating status level of the photovoltaic fuse includes: Obtain the running status level determined under continuous time index, and arrange the running status levels according to the order of the time index to form a running status level sequence; Determine the number of consecutive time indices used for cross-time consistency determination, and construct a consecutive time index window in the running state level sequence based on the number of consecutive time indices; For a continuous time index window, determine whether all the running status levels contained in the continuous time index window are the same to obtain the cross-time consistency determination result; When the cross-time consistency determination result is consistent, the running status level corresponding to the last time index in the continuous time index window is confirmed as the current running status level; When the cross-time consistency determination result is inconsistent, the current running status level remains unchanged, and cross-time consistency determination is performed on the running status level in subsequent continuous time index windows.

9. The method for monitoring the operating status of an intelligent photovoltaic fuse based on deep learning according to claim 1, characterized in that, The generation of the photovoltaic fuse operation status monitoring results includes: Obtain the photovoltaic fuse operating status level after cross-time consistency determination, and determine the photovoltaic fuse operating status level as the valid operating status level under the current time index; The effective operating status level is associated with the corresponding time index to generate photovoltaic fuse operating status monitoring results; Under continuous time indexing, the monitoring results of the operating status of multiple photovoltaic fuses are arranged in the order of the time index to form a sequence of photovoltaic fuse operating status monitoring results; Output the photovoltaic fuse operation status monitoring results or the photovoltaic fuse operation status monitoring result sequence as the photovoltaic fuse operation status monitoring results.

Images