Inverter Aging Monitoring Data Processing Method and System Based on Big Data
By constructing a healthy reference state and utilizing the Oliker-Prussner algorithm and Laguerre cell migration technology, the problem of distinguishing between operating condition changes and aging changes in inverter aging monitoring was solved, achieving accurate aging identification under complex operating conditions.
Patent Information
- Application Number
- CN202611237062.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing inverter aging monitoring methods are unable to accurately distinguish between normal operating condition changes and irreversible aging changes under different operating conditions, leading to misjudgments and inaccurate interpretations of aging status.
A healthy reference state is constructed by employing Oliker-Prussner semi-discrete optimal transport and Laguerre cell migration, combined with the return residual under the same operating condition and the source quality transport eligibility classification. Through node weight adjustment and Laguerre cell update, the healthy reference attribution is corrected and irreversible aging is identified.
It improves the accuracy and stability of inverter aging monitoring data processing, and can accurately identify irreversible aging changes under complex operating conditions, reducing the risk of misjudgment.
Smart Images

Figure CN123066448A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of inverter condition monitoring technology, and in particular to a data processing method and system for inverter aging monitoring based on big data. Background Technology
[0002] With the large-scale application of photovoltaic power generation, energy storage systems, and power electronic equipment, inverters are subject to the combined effects of load changes, ambient temperature fluctuations, switching frequency variations, and device performance degradation during long-term operation, continuously generating a large amount of operational monitoring data. Current inverter aging monitoring typically involves collecting operating parameters and health performance parameters, comparing historical health data with current monitoring data, and determining the inverter's performance degradation based on characteristic offsets, threshold changes, or state classification results. As monitoring cycles lengthen and operating conditions increase, historical health data and continuous monitoring data are characterized by large data volumes, complex state distributions, and frequent changes in operating conditions, necessitating the establishment of stable health reference relationships under different operating conditions.
[0003] Existing aging monitoring methods typically use fixed reference values, uniform state distances, or direct comparisons of health response differences to determine abnormal states when processing monitoring data under different operating conditions. This approach fails to adequately consider the sensitivity of different health condition locations to changes in health response. When a target inverter experiences health response fluctuations due to changes in normal operating conditions, it is easy to confuse recoverable changes caused by operating conditions with irreversible changes caused by device aging, leading to misjudgments of the aging state. Furthermore, when performing health reference matching based on state distance or data distribution, the reference attribution of the target monitoring state may change with data distribution and matching boundaries. Continuous health response offsets that have already occurred may be reassigned to another health condition, causing actual aging changes to be interpreted as differences in normal operating conditions.
[0004] Therefore, how to provide a data processing method and system for inverter aging monitoring based on big data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose a data processing method and system for inverter aging monitoring based on big data. This invention constructs a health reference state and introduces Oliker-Prussner semi-discrete optimal transmission, combined with Laguerre cell migration, same-condition return residuals, and source quality transport eligibility classification, to distinguish between normal operating condition changes and irreversible aging changes. Furthermore, it utilizes adjustable quality to correct the health reference attribution, reducing the interference of operating condition fluctuations and incorrect attribution on aging judgment, thereby improving the accuracy, stability, and reliability of inverter aging monitoring data processing.
[0006] The inverter aging monitoring data processing method based on big data according to an embodiment of the present invention includes the following steps: Acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status; A health reference node is constructed based on the inverter monitoring status, and a mapping relationship between changes in operating conditions and changes in health response is established to generate a health reference status. The transport cost between the working condition and the health reference node is corrected by using the health reference state. The semi-discrete optimal transmission is performed by Oliker-Prussner. Laguerre cells are constructed by node weights to obtain the initial health reference home state. The Laguerre cell migration is tracked based on the initial health reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is subtracted to obtain the same operating condition return residual state. Based on the return residual status under the same working conditions, the source quality transportation qualification is divided to obtain the quality ownership status, which includes fixed ownership quality and adjustable quality. By utilizing the quality attribution state, the fixed attribution quality is separated from the quality balance, and the node weights and Laguerre cells are updated with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The health reference state is reconstructed by modifying the health reference attribution state, and irreversible aging is identified by combining target monitoring data and quality attribution state, and the inverter aging monitoring results are output.
[0007] Optionally, obtaining the inverter monitoring status specifically includes: Acquire health operation data and target monitoring data, perform time alignment and monitoring window division on the corresponding monitoring quantities, and organize the data within the same monitoring window according to the collection order to form a status alignment sequence; A health baseline is determined for the health operation data based on the state alignment sequence. The relative state changes of the health operation data and the target monitoring data are characterized according to the health baseline to form a state characterization sequence. The operating condition changes and health response changes are extracted from the state characterization sequence, and a correspondence is established between the operating condition changes and health response changes within the same monitoring window. The state is then organized according to the continuous monitoring time sequence to obtain the inverter monitoring state.
[0008] Optionally, obtaining the health reference status specifically includes: Extract the operating condition status and health response status corresponding to the health operation data from the inverter monitoring status, divide the health reference area according to the distribution location of the operating condition status, determine the representative operating condition location in each health reference area and associate it with the corresponding health response status to form a set of health reference nodes. Call the set of health reference nodes, extract the health operation status with continuous operation relationship in the health reference area corresponding to each health reference node, determine the operating condition change and health response change of the health operation status relative to the corresponding health reference node, establish the correspondence between the operating condition change and the health response change belonging to the same operation change process, and perform local fitting on the correspondence to form a health response mapping relationship. Associat the set of health reference nodes with the mapping relationship of health response, organize the corresponding relationship according to the working condition position of the health reference nodes, and generate the health reference status.
[0009] Optionally, obtaining the initial health reference attribution status specifically includes: The mapping relationship between the health reference nodes and health response in the health reference status is invoked to determine the direction and degree of change of the operating condition status of the target monitoring data relative to each health reference node, and the association between the health reference node and the corresponding operating condition change is retained to form the node operating condition offset status. The node operating condition offset status is processed by using the health response mapping relationship corresponding to each health reference node, and the sensitivity of different operating condition change directions to normal health response changes under healthy operating conditions is determined. The node-level scale recalibration is performed on the operating condition change direction corresponding to each health reference node according to the sensitivity, and the transportation cost of correcting the target operating condition status to the health response of the corresponding health reference node is calculated using the recalibrated operating condition offset. The health response-corrected transportation cost is used as the transportation cost of the Oliker-Prussner method. Node weights are configured for each health reference node. The target operating condition is assigned to the health reference node with the minimum transportation result after the health response-corrected transportation cost is adjusted according to the corresponding node weights. Laguerre cells are constructed, and the node-level scale recalibration relationship determined by the health operation data remains unchanged during the node weight adjustment process, thus forming the initial Laguerre state. The source quality carried by each Laguerre cell in the initial Laguerre state is statistically analyzed. The source quality is compared with the discrete target quality of the corresponding healthy reference node. The node weight is adjusted node by node according to the quality deviation. After each adjustment of the node weight, the unchanging healthy response is called to correct the transportation cost and redetermine the Laguerre cell boundary and source quality. The node weight adjustment is repeated until the quality balance condition is reached, and the balanced Laguerre state is obtained. Read the balanced Laguerre state, assign the operating conditions in the target monitoring data to the health reference node according to the corresponding Laguerre unit, organize the attribution relationship between the operating conditions, Laguerre units and health reference nodes along the continuous acquisition time sequence of the target monitoring data, and obtain the initial health reference attribution state.
[0010] Optionally, obtaining the residual state under the same working condition specifically includes: Read the initial health reference home status, track the Laguerre unit migration corresponding to the operating condition status according to the continuous acquisition time sequence of the target monitoring data, identify the return process of the target inverter after leaving and re-entering the same Laguerre unit, associate the home position before leaving and the home position after returning to form the same unit return status. For the same unit return status, compare the working condition status of the home position before leaving and the home position after returning, determine the working condition differences that still exist after re-entering the same Laguerre unit, call the health response mapping relationship of the corresponding health reference node to determine the normal health response change corresponding to the working condition difference, and correspond it with the actual health response change of the two home positions to form the return response status. Subtract the corresponding normal health response changes from the actual health response changes in the return response state to obtain the return residuals that cannot be explained by the difference in return operating conditions. Collect the return residuals formed during the continuous return process according to the corresponding Laguerre unit and health reference node, and organize the change direction and continuity relationship of the return residuals along the acquisition time sequence of the target monitoring data to obtain the return residual state under the same operating conditions.
[0011] Optionally, obtaining the quality assignment status, which includes both fixed and adjustable quality, specifically includes: Read the status of the return residual under the same working condition, collect the return residual generated by continuous return according to the same Laguerre cell and healthy reference node, compare the change direction, residual amplitude and recovery trend of the continuous return residual, and associate the return residual that still maintains the same direction of offset after repeated return and has not recovered to the healthy operating fluctuation range with the corresponding source quality to determine the continuous offset source quality. Determine the Laguerre cell and healthy reference node corresponding to the persistent offset source quality, revoke the retransport eligibility of the persistent offset source quality with other healthy reference nodes, maintain the affiliation relationship between the persistent offset source quality and the current healthy reference node and determine it as a fixed affiliation quality, determine the source quality that does not meet the persistent offset conditions as an adjustable quality, and form a source quality transport eligibility status. Based on the health reference node and the Laguerre unit's source quality transportation eligibility status, a fixed association is established between the fixed assigned quality and the corresponding return residual status under the same working condition. The adjustable quality is then associated with the current Laguerre unit's assignment, thus obtaining the quality assignment status.
[0012] Optionally, obtaining the corrected health reference attribution status specifically includes: Read the quality attribution status, calculate the fixed attribution quality and adjustable quality according to the healthy reference nodes, separate the fixed attribution quality from the Oliker-Prussner quality balance object, and deduct the quality share already occupied by the fixed attribution quality from the discrete target quality of the corresponding healthy reference node, determine the remaining target quality of each healthy reference node borne by the adjustable quality, and at the same time correct the remaining target quality of each healthy reference node according to the total amount of all adjustable quality, forming an adjustable quality balance state; Compare the adjustable quality of the current Laguerre cell carried by each healthy reference node with the corresponding remaining target quality to determine the adjustable quality difference. Adjust the Oliker-Prussner node weights according to the adjustable quality difference. When the adjustable quality exceeds the corresponding remaining target quality, adjust the node weights in the direction of shrinking the carrying range of the corresponding Laguerre cell. When the adjustable quality is lower than the corresponding remaining target quality, adjust the node weights in the direction of expanding the carrying range of the corresponding Laguerre cell. After each node weight adjustment, keep the fixed assigned quality and the corresponding healthy reference assignment unchanged. Only for the adjustable quality, call the health response to correct the transportation cost and re-divide the Laguerre cells. Repeatedly execute adjustable quality statistics, quality difference comparison, node weight adjustment, and Laguerre cell update until the difference between the adjustable quality of each healthy reference node and the corresponding remaining target quality falls within the allowable deviation range. Retain the node weights and Laguerre cell partitioning results after quality balancing is completed. Merge the health reference attribution of the fixed attribution quality with the health reference attribution of the rebalanced adjustable quality. Organize the correspondence according to the continuous acquisition time sequence of the target monitoring data to form the corrected health reference attribution status.
[0013] Optionally, obtaining the inverter aging monitoring results specifically includes: Read the corrected health reference attribution status, call the corresponding health response mapping relationship according to the updated health reference node attribution, reconstruct the normal health response corresponding to the target operating condition status, and form the reconstructed health reference status. By comparing the actual health response of the target monitoring data with the normal health response of the reconstructed health reference state, and combining the quality attribution state to screen for health response deviations, the irreversible aging state is obtained. The irreversible aging state is organized according to the acquisition time sequence of the target monitoring data, the health reference attribution, duration range and degree of change corresponding to irreversible aging are determined, and the inverter aging monitoring results are output.
[0014] According to an embodiment of the present invention, a big data-based inverter aging monitoring data processing system includes: The data status characterization module is used to acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status. The health reference construction module is used to build health reference nodes for inverter monitoring status, establish a mapping relationship between changes in operating conditions and changes in health response, and generate health reference status. The transportation cost optimization module is used to correct the transportation cost between the working condition and the health reference node using the health reference state. It uses Oliker-Prussner to perform semi-discrete optimal transmission, constructs Laguerre cells through node weights, and obtains the initial health reference home state. The return residual extraction module is used to track the migration of the Laguerre cell in the initial healthy reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is deducted to obtain the return residual state under the same operating condition. The transportation qualification classification module is used to classify the source quality transportation qualification based on the return residual status under the same working conditions, and obtain the quality classification status including fixed quality and adjustable quality. The quality balance correction module is used to separate the fixed-attribute quality from the quality balance using the quality attribution state, and update the node weights and Laguerre cells with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The aging status identification module is used to reconstruct the health reference status by using the corrected health reference attribution status, identify irreversible aging by combining target monitoring data and quality attribution status, and output the inverter aging monitoring results.
[0015] The beneficial effects of this invention are: This invention utilizes healthy operating data to establish a mapping relationship between health reference nodes and changes in operating conditions and health responses. It employs the Oliker-Prussner algorithm to solve the semi-discrete optimal transmission problem between the target monitoring data's operating condition distribution and a finite number of health reference nodes. Laguerre cells are constructed using node weights to assign health references to the monitoring status of the target inverter under different operating conditions from a data distribution perspective. Compared to methods using fixed thresholds, uniform distances, or single health benchmarks for aging assessment, the Oliker-Prussner algorithm can adapt to different operating conditions through Laguerre cells and node weights. The data distribution changes under different operating conditions, making it suitable for processing large-scale, multi-condition inverter operation data generated through long-term continuous monitoring. Addressing the issue that conventional transportation costs are insufficient to reflect the differences in sensitivity to changes in health response under different health conditions, this invention further utilizes the health response mapping relationship corresponding to each health reference node to perform node-level scale recalibration of the direction of operating condition changes. This enables transportation costs to reflect the local normal response characteristics at different health condition locations, while simultaneously eliminating the reverse influence of actual health response offsets in the target monitoring data on transportation geometry. This reduces the possibility of true aging offsets being incorrectly interpreted as differences in health conditions, thereby improving the accuracy of health reference attribution.
[0016] This invention addresses the problem in the conventional Oliker-Prussner algorithm where source quality can be continuously redistributed as node weights change and Laguerre cell boundaries move. It tracks the migration of Laguerre cells during the continuous operation of the target inverter and, when the target inverter returns to the previous Laguerre cell, uses a health response mapping relationship to subtract the normal health response changes caused by differences in remaining operating conditions before and after the return, forming a return residual state under the same operating conditions. By comparing the direction of change, residual amplitude, and recovery trend of the continuous return residuals, it can further distinguish between normal health response changes that can recover with operating conditions and health response offsets that are continuously retained under the same health reference conditions. This avoids misjudgments caused by judging aging solely based on instantaneous parameter deviations, allowing irreversible performance changes of the inverter to be continuously verified under complex operating condition switching conditions.
[0017] This invention modifies the transport eligibility of source quality in semi-discrete optimal transmission based on the return residual under the same operating condition. It determines the source quality corresponding to the continuous return offset as the fixed assigned quality and revokes its eligibility for retransmission across healthy reference nodes. The remaining source quality is retained as adjustable quality. Furthermore, the fixed assigned quality is removed from the Oliker-Prussner quality balancing object, and the node weights and Laguerre cells are updated only based on the quality difference between the adjustable quality and the remaining target quality. As a result, the continuous aging offset that has been verified by the return under the same operating condition will not be redistributed to other healthy reference nodes due to subsequent node weight adjustments and Laguerre cell repartitioning. At the same time, the adjustable quality corresponding to the normal operating condition changes can still maintain the dynamic reassignment capability.
[0018] After completing the quality rebalancing, the normal health response corresponding to the target inverter is reconstructed and compared with the actual health response. This can further identify irreversible aging states that cannot be eliminated by changes in normal operating conditions and reassignment of health references. This improves the ability to separate normal operating condition responses from irreversible aging responses in inverter aging monitoring data processing, and enhances the accuracy, stability, and reliability of aging identification under complex operating conditions and long-term continuous monitoring conditions. Attached Figure Description
[0019] 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: Figure 1 This is a flowchart of the inverter aging monitoring data processing method based on big data proposed in this invention; Figure 2 This is a comparison curve of the actual health response and the reconfigured normal health response of the target inverter under the same operating condition return-back condition, based on the big data-based inverter aging monitoring data processing method proposed in this invention. Detailed Implementation
[0020] 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.
[0021] refer to Figures 1-2 The data processing method for inverter aging monitoring based on big data includes the following steps: Acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status; A health reference node is constructed based on the inverter monitoring status, and a mapping relationship between changes in operating conditions and changes in health response is established to generate a health reference status. The transport cost between the working condition and the health reference node is corrected by using the health reference state. The semi-discrete optimal transmission is performed by Oliker-Prussner. Laguerre cells are constructed by using node weights to obtain the initial health reference home state. The Laguerre cell migration is tracked based on the initial health reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is subtracted to obtain the same operating condition return residual state. Based on the return residual status under the same working conditions, the source quality transportation qualification is divided to obtain the quality ownership status, which includes fixed ownership quality and adjustable quality. By utilizing the quality attribution state, the fixed attribution quality is separated from the quality balance, and the node weights and Laguerre cells are updated with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The health reference state is reconstructed by modifying the health reference attribution state, and irreversible aging is identified by combining target monitoring data and quality attribution state, and the inverter aging monitoring results are output.
[0022] In this embodiment, obtaining the inverter monitoring status specifically includes: Acquire health operation data and target monitoring data, perform time alignment and monitoring window division on the corresponding monitoring quantities, organize the data within the same monitoring window according to the collection order to form a state alignment sequence. The health operation data consists of large-scale historical operation data formed by different operating cycles and different operating condition intervals. The target monitoring data is the operation monitoring data stream formed by continuous collection according to the monitoring time sequence during the continuous operation of the target inverter. The time alignment establishes the time position relationship between the health operation data and the target monitoring data according to the collection time of the corresponding monitoring quantity. The monitoring window divides the data range according to the continuous monitoring time sequence and maintains the original collection order of the monitoring quantities within the window. The state alignment sequence is a data sequence formed by organizing the health operation data and the target monitoring data under a unified time position and monitoring window. A health benchmark is determined for the health operation data based on the state alignment sequence. The relative state changes of the health operation data and the target monitoring data are characterized according to the health benchmark to form a state characterization sequence. The health benchmark is determined from the health operation data in the state alignment sequence and is used to represent the reference state of the corresponding monitoring quantity under the health operation conditions. When determining the health benchmark, the health operation data are collected according to the same monitoring quantity, and the reference level and normal change range of the corresponding monitoring quantity are determined by the state range that continuously appears in the health operation stage. The relative state change is the direction and degree of change of the current state of the monitoring quantity relative to the corresponding health benchmark. The health operation data and the target monitoring data are characterized using the same health benchmark so that the state changes in the target monitoring data retain the offset relationship relative to the health operation state. The state characterization sequence is a state sequence formed by reorganizing the data after the health benchmark characterization according to the monitoring window and acquisition order in the state alignment sequence. Operating condition changes and health response changes are extracted from the state representation sequence. A correspondence is established between operating condition changes and health response changes within the same monitoring window. The state is organized according to the continuous monitoring sequence to obtain the inverter monitoring state. The operating condition change is the state change formed in the state representation sequence due to the change in the target inverter's operating conditions. It is used to represent the operating condition position corresponding to the subsequent health reference node. The health response change is the health performance response change generated by the inverter under the corresponding operating conditions. It is used to represent the response position required for the subsequent health response mapping relationship. The correspondence maintains the mutual association between operating condition changes and health response changes within the same monitoring window, so that the health response change can be traced back to the operating condition change that generated the corresponding response. The inverter monitoring state is a state object formed by organizing the correspondence between operating condition changes and health response changes according to the continuous monitoring sequence.
[0023] Health operation data can be selected from historical data accumulated during 180 consecutive days of healthy operation of the target type inverter. The sampling period of the original monitoring data is set to 1 second, and no less than 1,000,000 valid monitoring records are generated for a single inverter. The target monitoring data is continuously collected according to the 1-second sampling period. During time alignment, all monitoring quantities are unified to the 1-second sampling position. Data that is missing for no more than 2 seconds is supplemented by adjacent sampling positions. Monitoring segments that are missing for more than 5 seconds are not included in the state representation. The monitoring window length is set to 300 seconds, and the moving step size of adjacent monitoring windows is set to 60 seconds, so that adjacent monitoring windows maintain 240 seconds of overlapping data, which is used to maintain the temporal correlation of the continuous operating state changes of the target inverter.
[0024] In this embodiment, obtaining the health reference status specifically includes: The operating condition status and health response status corresponding to the health operation data are extracted from the inverter monitoring status. Health reference areas are divided according to the distribution of operating condition status. Representative operating condition locations are determined in each health reference area and associated with the corresponding health response status to form a set of health reference nodes. The health reference area is a data area with continuous operating condition coverage, divided according to the relative positional relationship between operating condition statuses in the health operation data. The representative operating condition location is determined from the location in the corresponding health reference area where the health operation data coverage is stable and can characterize the overall operating conditions of the area. The health reference node uses the representative operating condition location as the operating condition benchmark and retains the corresponding health response status, so that each health reference node represents the operating condition location and normal health response under a certain health operation condition. The set of health reference nodes is composed of the health reference nodes corresponding to each health reference area and covers the distribution range of operating condition status of the health operation data. The system calls upon a set of health reference nodes and extracts healthy operating states with continuous operating relationships within the health reference regions corresponding to each health reference node. It then determines the changes in operating conditions and health responses relative to the corresponding health reference nodes, establishing a correspondence between operating condition changes and health response changes belonging to the same operating change process. Local fitting is then performed on this correspondence to form a health response mapping relationship. The continuous operating relationship is the correlation between the preceding and following states of health operating data formed according to the acquisition sequence. Local fitting is performed independently within each health reference region. By adjusting the mapping relationship between operating condition changes and health response changes, the deviation between predicted and actual health response changes is reduced, and excessive amplification of the mapping caused by insufficient local data is limited. This health response mapping relationship is used to characterize the normal health response changes corresponding to operating condition changes near the health reference nodes and is determined solely using healthy operating data. The set of associated health reference nodes is mapped to the health response. The corresponding relationship is organized according to the operating condition position of the health reference nodes to generate a health reference state. The health reference state is used to characterize the health operating condition position and normal health response change of each health reference node. The corresponding normal health response is determined according to the operating condition change of the target inverter relative to the health reference nodes. The aging offset in the target monitoring data does not participate in the construction of the health reference state.
[0025] After the health operation data is characterized, it is divided into 24 health reference regions according to the distribution of operating conditions. Each health reference region retains no less than 3,000 health operation status records. When there are fewer than 3,000 valid records in a health reference region, the corresponding region is merged with the nearest neighboring health reference region. One health reference node is determined for each health reference region, and no less than 500 continuous health operation statuses near the health reference node are used to establish a local health response mapping relationship. During the local fitting process, 80% of the health operation statuses are used for mapping relationship establishment, and 20% of the health operation statuses are used for deviation verification. When the average deviation between the predicted health response change and the actual health response change exceeds 5%, the sampling range of the local health operation status is expanded again.
[0026] In this embodiment, obtaining the initial health reference attribution status specifically includes: The mapping relationship between health reference nodes and health response in the health reference state is invoked to determine the direction and degree of change of the operating condition status of the target monitoring data relative to each health reference node, and the association between the health reference node and the corresponding operating condition change is retained to form a node operating condition offset state. The node operating condition offset state is established for each health reference node according to the same target operating condition status, so that the same target operating condition status has an independent operating condition offset for different health reference nodes. The operating condition change direction is used to indicate the direction of change of the target operating condition status from the corresponding healthy operating condition position, and the operating condition change degree is used to indicate the state offset magnitude generated along the corresponding change direction. The node operating condition offset status is processed using the health response mapping relationship corresponding to each health reference node. The sensitivity of different operating condition change directions to normal health response changes under healthy operating conditions is determined. The node-level scale recalibration is performed on the operating condition change direction corresponding to each health reference node according to the sensitivity. The recalibrated operating condition offset is used to calculate the health response correction transportation cost from the target operating condition state to the corresponding health reference node. The node-level scale recalibration is as follows: under the local health response relationship corresponding to each health reference node, the relative scale of different operating condition change directions participating in the transportation cost calculation is changed. The operating condition direction with more sensitive normal health response changes increases the corresponding transportation change amplitude, and the operating condition direction with weaker normal health response changes decreases the corresponding transportation change amplitude, so that different health reference nodes form different local transportation scales according to their own health response characteristics. The health response correction transportation cost is only corrected using the health response mapping relationship determined by the healthy operating data. The health response offset actually generated in the target monitoring data does not participate in the node-level scale recalibration and transportation cost calculation. The health response-corrected transportation cost is used as the transportation cost in the Oliker-Prussner method. Node weights are configured for each health reference node. The target operating condition is assigned to the health reference node with the minimum transportation result after the health response-corrected transportation cost is adjusted by the corresponding node weights. Laguerre cells are constructed, and the node-level scale recalibration relationship determined by the health operation data remains unchanged during the node weight adjustment process, forming an initial Laguerre state. The Oliker-Prussner method is used to perform semi-discrete quality allocation of the operating condition distribution formed by the target monitoring data to a finite number of health reference nodes. The node weights are iterative variables that adjust the transportation carrying capacity of the health reference nodes. Changes in node weights are used to move the Laguerre cell boundaries between adjacent health reference nodes. A Laguerre cell is an operating condition region belonging to the same health reference node under the joint constraints of the health response-corrected transportation cost and node weights. The node weights only change the belonging boundaries between adjacent Laguerre cells during each round of adjustment, ensuring that the Laguerre cells are always updated according to the local transportation scale determined by the health operation data. The source quality carried by each Laguerre cell in the initial Laguerre state is statistically analyzed. The source quality is compared with the discrete target quality of the corresponding healthy reference node. The node weights are adjusted node by node according to the quality deviation. After each adjustment, the unchanging health response is called to correct the transportation cost and redetermine the Laguerre cell boundary and source quality. The node weight adjustment is repeated until the quality balance condition is reached, resulting in a balanced Laguerre state. The source quality represents the target operating condition share currently belonging to the corresponding healthy reference node. The discrete target quality represents the reference share that the corresponding healthy reference node should bear under the support of healthy operating data. The quality deviation is the degree of insufficiency or excess of the source quality relative to the discrete target quality. The quality balance condition is that the quality deviation of each Laguerre cell is within the allowable deviation range determined based on the operating condition coverage fluctuation of the healthy operating data, and continuous node weight adjustments no longer cause changes in the Laguerre cell affiliation that exceed the allowable range. When the quality balance condition is met, the node weight adjustment is stopped and the current Laguerre cell boundary and corresponding node weights are retained. The balanced Laguerre state is read, and the operating conditions in the target monitoring data are assigned to the health reference nodes according to the corresponding Laguerre units. The attribution relationship between the operating conditions, Laguerre units, and health reference nodes is organized along the continuous acquisition time sequence of the target monitoring data to obtain the initial health reference attribution state. The initial health reference attribution state retains the Laguerre unit attribution and health reference node attribution determined by the target operating conditions under the health response correction transport geometry, so that the maintenance, migration, and return of the health reference attribution during the continuous operation of the target inverter can be tracked according to the monitoring time sequence.
[0027] This invention performs local scale recalibration of the operating condition change direction of different health reference nodes through a health response mapping relationship, and embeds the corrected transport cost into the Oliker-Prussner semi-discrete optimal transmission and Laguerre cell partitioning. This allows the health reference assignment to adapt to the differences in normal health response sensitivity under different operating conditions, while avoiding the actual aging offset of the target inverter from changing the transport geometry in the opposite direction. This reduces the erroneous assignment caused by confusion between operating condition changes and aging changes, and improves the accuracy and stability of inverter health reference matching under complex operating conditions.
[0028] In this embodiment, obtaining the residual state under the same working condition specifically includes: The initial health reference home status is read, and the migration of Laguerre units corresponding to the operating conditions is tracked according to the continuous acquisition time sequence of the target monitoring data. The return process of the target inverter re-entering the same Laguerre unit after leaving is identified. The home position before leaving and the home position after returning are associated to form the same unit return status. The Laguerre unit migration is determined according to the Laguerre unit home status change of the target operating conditions in adjacent monitoring windows. The same unit return status keeps the home position before leaving and the home position after returning correspond to the same Laguerre unit and the same health reference node, and records the Laguerre unit migration relationship experienced between the two homes. For the same unit return status, compare the operating condition status of the home position before departure and the home position after return, determine the operating condition difference that still exists after re-entering the same Laguerre unit, call the health response mapping relationship of the corresponding health reference node to determine the normal health response change corresponding to the operating condition difference, and correspond it with the actual health response change of the two home positions to form the return response status. The operating condition difference is the local operating condition offset that the target inverter retains after returning to the same health reference home area, and the actual health response change is the health response difference generated by the target inverter between the home position before departure and the home position after return. The corresponding normal health response change is subtracted from the actual health response change in the return response state to obtain the return residual that cannot be explained by the difference in return operating conditions. The return residual formed during the continuous return process is collected according to the corresponding Laguerre unit and health reference node. The change direction and continuity relationship of the return residual are organized along the acquisition time sequence of the target monitoring data to obtain the same operating condition return residual state. The same operating condition return residual state represents the health response offset that the target inverter retains after re-entering the same Laguerre unit, under the condition of deducting the normal operating condition response change, and maintains the correlation between the return residual and the corresponding Laguerre unit, health reference node and target monitoring data.
[0029] This invention tracks the Laguerre unit feedback relationship, deducts normal health response changes caused by operating condition differences under the same health reference attribution, and extracts continuous feedback residuals that cannot be explained by operating condition changes. This suppresses the interference of load fluctuations and operating condition switching on aging judgment, enhances the ability to identify irreversible health response offsets, and improves the accuracy and reliability of inverter aging monitoring under complex operating conditions.
[0030] In this embodiment, obtaining the quality assignment state, which includes both fixed and adjustable quality, specifically includes: Read the status of the return residual under the same operating condition, collect the return residuals generated by continuous return according to the same Laguerre cell and healthy reference node, compare the change direction, residual amplitude and recovery trend of the continuous return residuals, associate the return residuals that still maintain the same direction after repeated return and have not recovered to the healthy operation fluctuation range with the corresponding source quality, and determine the continuous offset source quality. The continuous offset source quality is: the source quality corresponding to the return residual under the same operating condition that maintains the same direction during the continuous return process and has not recovered to the healthy operation fluctuation range. The healthy operation fluctuation range of the return residual is determined according to the normal residual change generated by the healthy operation data under the same operating condition return conditions. When the return residual enters the healthy operation fluctuation range with repeated return or a continuous reverse recovery occurs, the corresponding source quality is uncertain as the continuous offset source quality. The Laguerre cell and healthy reference node corresponding to the persistent offset source quality are determined. The retransport eligibility of the persistent offset source quality with other healthy reference nodes is revoked. The affiliation relationship between the persistent offset source quality and the current healthy reference node is maintained and determined as a fixed-affiliation quality. Source qualities that do not meet the persistent offset conditions are determined as adjustable qualities, forming a source quality transport eligibility state. The source quality transport eligibility indicates whether the source quality is allowed to re-participate in the transport competition of other healthy reference nodes when the Laguerre cell affiliation changes. The fixed-affiliation quality only retains the valid transport affiliation with the current healthy reference node and will not reselect other healthy reference nodes due to changes in the Laguerre cell boundary. The adjustable quality retains the transport competition relationship with adjacent healthy reference nodes and continues to determine the Laguerre cell affiliation according to the health response and adjusted transport cost. Based on the health reference node and the Laguerre unit organization source quality transportation eligibility status, a fixed association is established between the fixed assigned quality and the corresponding return residual status under the same working condition. The adjustable quality is then associated with the current Laguerre unit assignment to obtain the quality assignment status. The quality assignment status is used to record the health reference assignment, transportation eligibility, and corresponding return residual of different source qualities.
[0031] This invention classifies source quality transport eligibility based on the continuous direction and recovery trend of the return residual under the same operating condition. It fixes the continuously offset source quality that cannot be explained by changes in normal operating conditions to the current healthy reference attribution, while retaining the retransport capability of the remaining source quality. This avoids the real aging offset being rematched to other healthy operating conditions, and improves the accuracy of irreversible aging characteristic retention and healthy reference attribution determination.
[0032] In this embodiment, obtaining the corrected health reference attribution status specifically includes: Read the quality attribution status, and calculate the fixed attribution quality and adjustable quality according to the healthy reference nodes. Remove the fixed attribution quality from the Oliker-Prussner quality balance object, and deduct the quality share already occupied by the fixed attribution quality from the discrete target quality of the corresponding healthy reference node. Determine the remaining target quality of each healthy reference node to be borne by the adjustable quality. At the same time, correct the remaining target quality of each healthy reference node according to the total amount of all adjustable quality to form an adjustable quality balance state. The remaining target quality is the quality share that still needs to be borne by the adjustable quality after deducting the part already occupied by the fixed attribution quality from the original discrete target quality of the healthy reference node. The adjustable quality balance state maintains the correspondence between the total amount of all adjustable quality and the total amount of adjustable target quality of each healthy reference node, so that the quality conservation relationship of the remaining source quality is not destroyed after the fixed attribution quality is withdrawn and re-transported. The adjustable quality carried by each healthy reference node's current Laguerre cell is compared with the corresponding remaining target quality to determine the adjustable quality difference. The Oliker-Prussner node weights are then adjusted according to this difference. When the adjustable quality exceeds the corresponding remaining target quality, the node weights are adjusted in the direction of reducing the carrying range of the corresponding Laguerre cell. When the adjustable quality is lower than the corresponding remaining target quality, the node weights are adjusted in the direction of expanding the carrying range of the corresponding Laguerre cell. After each node weight adjustment, the fixed assigned quality and corresponding healthy reference assignment remain unchanged. The health response is invoked only for the adjustable quality to correct the transportation cost and re-divide the Laguerre cells. The adjustable quality difference reflects the deviation between the source quality that the current Laguerre cell can re-transport and the corresponding remaining target quality. The node weights are updated only based on the adjustable quality difference. This changes the conventional node weight update that balances all source qualities to balancing only the adjustable quality that is still eligible for transportation. The process of adjusting quality statistics, quality difference comparison, node weight adjustment, and Laguerre cell update is repeated until the difference between the adjustable quality of each healthy reference node and the corresponding remaining target quality falls within the allowable deviation range. The node weights and Laguerre cell partitioning results after quality balancing are retained. The health reference attribution of the fixed assigned quality is merged with the health reference attribution of the adjusted quality after rebalancing. The correspondence is organized according to the continuous acquisition time sequence of the target monitoring data to form a corrected health reference attribution state. The allowable deviation range is determined based on the normal source quality distribution fluctuations generated by the health operation data near the corresponding healthy reference node. The corrected health reference attribution state retains the fixed correspondence between the fixed assigned quality and the original healthy reference node that does not migrate with changes in node weight, as well as the Laguerre cell attribution of the adjustable quality re-determined by the health response correction transportation cost and the updated node weights. This ensures that the health response offset confirmed by the same operating condition return cannot be redistributed to other healthy reference nodes through subsequent Oliker-Prussner quality balancing.
[0033] This invention separates the fixed attribution quality confirmed by the same operating condition return from the quality balance object, and only uses the adjustable quality difference update node weight and Laguerre cell with retransmission qualification, so that the confirmed continuous health response offset is no longer redistributed to other healthy reference nodes in subsequent quality balance, while retaining the dynamic adjustment capability of the source quality corresponding to normal operating condition changes, thereby improving the accuracy and stability of inverter aging state maintenance and healthy reference attribution correction.
[0034] In this embodiment, obtaining the inverter aging monitoring results specifically includes: The corrected health reference attribution status is read, and the corresponding health response mapping relationship is invoked according to the updated health reference node attribution. The normal health response corresponding to the target operating condition is reconstructed to form a reconstructed health reference status. The reconstructed health reference status maintains the correspondence between the operating condition, health reference node, Laguerre unit and normal health response according to the continuous acquisition sequence of the target monitoring data. The normal health response is determined according to the operating condition change of the target operating condition relative to the current health reference node and the corresponding health response mapping relationship, and is combined with the health response benchmark corresponding to the health reference node. This allows the health reference attribution after the fixed attribution quality stripping and adjustable quality rebalancing to be converted back into the health response that the target inverter should have under the corresponding operating condition. The actual health response of the target monitoring data is compared with the normal health response of the reconstructed health reference state. The health response offset is screened in combination with the quality attribution state to obtain the irreversible aging state. The health response offset is the remaining difference between the actual health response and the corresponding normal health response. When the health response offset corresponding to the fixed attribution quality is in the same direction as the return residual under the same working condition and continues to exist, it is determined to be irreversible aging. The change of the quality corresponding to the adjustable quality and the change of the normal health response range as the Laguerre unit is re-divided and restored is not included in the irreversible aging state. Irreversible aging states are organized along the acquisition time sequence of target monitoring data. The health reference attribution, duration range, and degree of change corresponding to irreversible aging are determined, and the inverter aging monitoring results are output. The inverter aging monitoring results record the monitoring location, duration range, degree of change, and health reference attribution corresponding to the irreversible aging of the target inverter, and maintain the correlation between the irreversible aging state and the corresponding target monitoring data. This is used to separate normal operating condition response changes and irreversible health response changes from the target monitoring data continuously collected under different operating conditions.
[0035] The actual health response of the target inverter is compared with the reconstructed normal health response in 300-second monitoring windows. When the health response offset corresponding to the fixed home quality maintains the same direction of change as the return residual under the same operating condition for three consecutive monitoring windows and the duration reaches 900 seconds, the corresponding state is determined to be an irreversible aging state. For the health response offset corresponding to the adjustable quality, it is not included in the irreversible aging state when it recovers to the normal health response fluctuation range determined by the health operation data for two consecutive monitoring windows after the Laguerre cell re-division is completed. The inverter aging monitoring results are updated according to a 60-second output cycle, and the monitoring timing position, duration, degree of health response offset, and corresponding health reference node of the first occurrence of irreversible aging are recorded.
[0036] The big data-based inverter aging monitoring data processing system includes: The data status characterization module is used to acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status. The health reference construction module is used to build health reference nodes for inverter monitoring status, establish a mapping relationship between changes in operating conditions and changes in health response, and generate health reference status. The transportation cost optimization module is used to correct the transportation cost between the working condition and the health reference node using the health reference state. It uses Oliker-Prussner to perform semi-discrete optimal transmission, constructs Laguerre cells through node weights, and obtains the initial health reference home state. The return residual extraction module is used to track the migration of the Laguerre cell in the initial healthy reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is deducted to obtain the return residual state under the same operating condition. The transportation qualification classification module is used to classify the source quality transportation qualification based on the return residual status under the same working conditions, and obtain the quality classification status including fixed quality and adjustable quality. The quality balance correction module is used to separate the fixed-attribute quality from the quality balance using the quality attribution state, and update the node weights and Laguerre cells with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The aging status identification module is used to reconstruct the health reference status by using the corrected health reference attribution status, identify irreversible aging by combining target monitoring data and quality attribution status, and output the inverter aging monitoring results.
[0037] Example 1: To verify the feasibility of this invention in practice, it was applied to a long-term operation monitoring scenario of an inverter of the same model in a photovoltaic power generation system. The invention processed the operation monitoring data generated under the combined effects of load changes, DC input fluctuations, changes in heat dissipation conditions, and device performance degradation. The test data included historical healthy operation data from a healthy inverter and target monitoring data generated from continuous operation of the target inverter. The healthy operation data contained approximately 1.56 million valid records, and the target monitoring data contained approximately 720,000 valid records. Monitored parameters included DC-side voltage, DC-side current, AC output current, active power, power factor, radiator temperature, DC bus ripple, conversion efficiency, and output current distortion. The sampling period was set to 1 second, and continuous monitoring data was organized according to a 300-second monitoring window and a 60-second moving step.
[0038] After characterizing the health operation data, 24 health reference regions were divided according to the distribution of operating conditions, and a health reference node and corresponding health response mapping relationship were determined in each region. Since different operating condition regions have different sensitivities to changes in health response, for example, a 5% change in output power in some high-load regions corresponds to a normal radiator temperature rise change of approximately 3.8% to 5.2%, while in low-load regions the temperature rise change is approximately 1.1% to 2.0%. Therefore, this invention does not use a uniform operating condition distance for health reference matching. Instead, it utilizes the health response mapping relationship of each health reference node to perform local scale recalibration of the direction of operating condition changes, and uses the obtained health response-corrected transport cost for Oliker-Prussner semi-discrete optimal transmission. Each health reference node constructs a corresponding Laguerre cell, with a mass balance allowable deviation set to 3%. The node weights stabilized after 63 to 118 adjustments. Compared with uniform state distance matching, the number of unnecessary switching of health reference nodes within the test interval was reduced from 37 to 14.
[0039] During continuous operation of the target inverter, the migration of the Laguerre cell is further tracked. When the target inverter leaves and returns to the same Laguerre cell, the actual health response before and after the return is compared, and the normal health response change caused by the remaining operating condition difference is deducted. In one set of test data, the conversion efficiency was 97.42% when the target inverter was first in the corresponding healthy reference area, and 96.91% when it returned to the same Laguerre cell. The remaining operating condition difference can explain about 0.13 percentage points of change. After deducting, there is still about 0.38 percentage points of return residual. The return residuals for three consecutive times were 0.38 percentage points, 0.41 percentage points, and 0.44 percentage points, respectively, all maintaining the same direction of change, indicating that the corresponding health response deviation cannot be recovered with normal operating conditions.
[0040] For the source masses corresponding to the aforementioned persistent offsets, this invention revokes their eligibility for retransmission to other healthy reference nodes and identifies them as fixed-attribute masses, while the remaining source masses are retained as adjustable masses. In the test data, approximately 7.8% of the source masses were identified as fixed-attribute masses, and approximately 92.2% of the source masses remained in an adjustable state. When re-performing Oliker-Prussner mass balancing, the fixed-attribute masses are removed from the mass balancing object, and only the mass difference between the adjustable masses and the remaining target masses is used to update the node weights and Laguerre cells. The allowable deviation for adjustable mass balancing is set to 2%, and stability is achieved after 38 to 76 node weight adjustments, thereby preventing the already confirmed persistent health response offsets from being reassigned to other healthy operating conditions.
[0041] To further verify the processing effect of the present invention, the same target monitoring data were processed using a fixed health threshold comparison method, a conventional Oliker-Prussner method with uniform transportation costs and all source masses participating in the balance, and the method of the present invention. The fixed health threshold comparison method determines a uniform normal range based on the health operation data. The conventional Oliker-Prussner method uses a uniform operating condition distance for Laguerre cell partitioning, does not perform node-level health response scale recalibration, does not change the source mass transportation eligibility based on the return residual under the same operating condition, and maintains the participation of all source masses in quality balance during node weight adjustment. Specific comparison results are shown in Table 1. Table 1 Comparison of Aging Identification and Health Reference Attribution Performance of Different Methods
[0042] As shown in Table 1, under the same target monitoring data conditions, the aging state identification accuracy of this invention reaches 95.1%, which is 13.5 percentage points and 7.2 percentage points higher than the fixed health threshold method and the conventional Oliker-Prussner method, respectively. The false positive rate for normal operating conditions is reduced to 3.2%, and the false negative rate for irreversible aging is reduced to 3.6%, indicating that the health response correction transportation cost can reduce the interference of different operating condition response differences on aging judgment. The number of unnecessary switching of health reference is reduced from 46 times in the fixed health threshold method and 31 times in the conventional Oliker-Prussner method to 14 times, and the stability rate of health reference assignment is improved to 97.0%. Meanwhile, the retention rate of persistent migration under the same operating conditions reached 96.4%, which is 15.1 percentage points higher than the conventional Oliker-Prussner method. This indicates that by restricting the retransport eligibility of persistent migration source quality and separating fixed-attribution quality from quality balance, the situation where persistent aging migration is reassigned to other healthy reference nodes during the Laguerre cell repartitioning process can be reduced. The normal operating condition recovery recognition rate reached 97.3%, and the average confirmation delay of irreversible aging was shortened to 3.4 monitoring windows. This further demonstrates that the present invention improves the accuracy, stability, and timeliness of irreversible aging state recognition while maintaining the dynamic matching capability under normal operating conditions.
[0043] Depend on Figure 2It can be seen that the reconstructed normal health response remains generally stable within the continuous monitoring window and is always near the normal fluctuation range of healthy operation. However, the actual health response of the target inverter is basically consistent with the reconstructed normal health response in the early stage. With continuous operation, it gradually shows a continuous downward trend and gradually deviates from the normal fluctuation range of healthy operation. This indicates that the target inverter has formed a continuous health response offset under similar operating conditions that cannot be fully explained by changes in normal operating conditions. This change characteristic corresponds to the judgment logic of the return residual under the same operating conditions. It can be explained that after the present invention compensates for the normal operating condition response through health reference reconstruction, it can further retain irreversible health response changes, providing a data basis for continuous offset source quality identification, fixed attribution quality determination, and inverter irreversible aging state judgment.
[0044] 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 data processing method for inverter aging monitoring based on big data, characterized in that, Includes the following steps: Acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status; A health reference node is constructed based on the inverter monitoring status, and a mapping relationship between changes in operating conditions and changes in health response is established to generate a health reference status. The transport cost between the working condition and the health reference node is corrected by using the health reference state. The semi-discrete optimal transmission is performed by Oliker-Prussner. Laguerre cells are constructed by node weights to obtain the initial health reference home state. The Laguerre cell migration is tracked based on the initial health reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is subtracted to obtain the same operating condition return residual state. Based on the return residual status under the same working conditions, the source quality transportation qualification is divided to obtain the quality ownership status, which includes fixed ownership quality and adjustable quality. By utilizing the quality attribution state, the fixed attribution quality is separated from the quality balance, and the node weights and Laguerre cells are updated with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The health reference state is reconstructed by modifying the health reference attribution state, and irreversible aging is identified by combining target monitoring data and quality attribution state, and the inverter aging monitoring results are output.
2. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The specific methods for obtaining the inverter monitoring status include: Acquire health operation data and target monitoring data, perform time alignment and monitoring window division on the corresponding monitoring quantities, and organize the data within the same monitoring window according to the collection order to form a status alignment sequence; A health baseline is determined for the health operation data based on the state alignment sequence. The relative state changes of the health operation data and the target monitoring data are characterized according to the health baseline to form a state characterization sequence. The operating condition changes and health response changes are extracted from the state characterization sequence, and a correspondence is established between the operating condition changes and health response changes within the same monitoring window. The state is then organized according to the continuous monitoring time sequence to obtain the inverter monitoring state.
3. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The acquisition of the health reference status specifically includes: Extract the operating condition status and health response status corresponding to the health operation data from the inverter monitoring status, divide the health reference area according to the distribution location of the operating condition status, determine the representative operating condition location in each health reference area and associate it with the corresponding health response status to form a set of health reference nodes. Call the set of health reference nodes, extract the health operation status with continuous operation relationship in the health reference area corresponding to each health reference node, determine the operating condition change and health response change of the health operation status relative to the corresponding health reference node, establish the correspondence between the operating condition change and the health response change belonging to the same operation change process, and perform local fitting on the correspondence to form a health response mapping relationship. Associat the set of health reference nodes with the mapping relationship of health response, organize the corresponding relationship according to the working condition position of the health reference nodes, and generate the health reference status.
4. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The initial health reference attribution status is obtained specifically through: The mapping relationship between the health reference nodes and health response in the health reference status is invoked to determine the direction and degree of change of the operating condition status of the target monitoring data relative to each health reference node, and the association between the health reference node and the corresponding operating condition change is retained to form the node operating condition offset status. The node operating condition offset status is processed by using the health response mapping relationship corresponding to each health reference node, and the sensitivity of different operating condition change directions to normal health response changes under healthy operating conditions is determined. The node-level scale recalibration is performed on the operating condition change direction corresponding to each health reference node according to the sensitivity, and the transportation cost of correcting the target operating condition status to the health response of the corresponding health reference node is calculated using the recalibrated operating condition offset. The health response-corrected transportation cost is used as the transportation cost of the Oliker-Prussner method. Node weights are configured for each health reference node. The target operating condition is assigned to the health reference node with the minimum transportation result after the health response-corrected transportation cost is adjusted according to the corresponding node weights. Laguerre cells are constructed, and the node-level scale recalibration relationship determined by the health operation data remains unchanged during the node weight adjustment process, thus forming the initial Laguerre state. The source quality carried by each Laguerre cell in the initial Laguerre state is statistically analyzed. The source quality is compared with the discrete target quality of the corresponding healthy reference node. The node weight is adjusted node by node according to the quality deviation. After each adjustment of the node weight, the unchanging healthy response is called to correct the transportation cost and redetermine the Laguerre cell boundary and source quality. The node weight adjustment is repeated until the quality balance condition is reached, and the balanced Laguerre state is obtained. Read the balanced Laguerre state, assign the operating conditions in the target monitoring data to the health reference node according to the corresponding Laguerre unit, organize the attribution relationship between the operating conditions, Laguerre units and health reference nodes along the continuous acquisition time sequence of the target monitoring data, and obtain the initial health reference attribution state.
5. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The specific steps to obtain the residual state under the same working condition include: Read the initial health reference home status, track the Laguerre unit migration corresponding to the operating condition status according to the continuous acquisition time sequence of the target monitoring data, identify the return process of the target inverter after leaving and re-entering the same Laguerre unit, associate the home position before leaving and the home position after returning to form the same unit return status. For the same unit return status, compare the working condition status of the home position before leaving and the home position after returning, determine the working condition differences that still exist after re-entering the same Laguerre unit, call the health response mapping relationship of the corresponding health reference node to determine the normal health response change corresponding to the working condition difference, and correspond it with the actual health response change of the two home positions to form the return response status. Subtract the corresponding normal health response changes from the actual health response changes in the return response state to obtain the return residuals that cannot be explained by the difference in return operating conditions. Collect the return residuals formed during the continuous return process according to the corresponding Laguerre unit and health reference node, and organize the change direction and continuity relationship of the return residuals along the acquisition time sequence of the target monitoring data to obtain the return residual state under the same operating conditions.
6. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The acquisition of the quality attribution status, which includes both fixed and adjustable quality, specifically includes: Read the status of the return residual under the same working condition, collect the return residual generated by continuous return according to the same Laguerre cell and healthy reference node, compare the change direction, residual amplitude and recovery trend of the continuous return residual, and associate the return residual that still maintains the same direction of offset after repeated return and has not recovered to the healthy operating fluctuation range with the corresponding source quality to determine the continuous offset source quality. Determine the Laguerre cell and healthy reference node corresponding to the persistent offset source quality, revoke the retransport eligibility of the persistent offset source quality with other healthy reference nodes, maintain the affiliation relationship between the persistent offset source quality and the current healthy reference node and determine it as a fixed affiliation quality, determine the source quality that does not meet the persistent offset conditions as an adjustable quality, and form a source quality transport eligibility status. Based on the health reference node and the Laguerre unit's source quality transportation eligibility status, a fixed association is established between the fixed assigned quality and the corresponding return residual status under the same working condition. The adjustable quality is then associated with the current Laguerre unit's assignment, thus obtaining the quality assignment status.
7. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The specific steps for obtaining the corrected health reference attribution status include: Read the quality attribution status, calculate the fixed attribution quality and adjustable quality according to the healthy reference nodes, separate the fixed attribution quality from the Oliker-Prussner quality balance object, and deduct the quality share already occupied by the fixed attribution quality from the discrete target quality of the corresponding healthy reference node, determine the remaining target quality of each healthy reference node borne by the adjustable quality, and at the same time correct the remaining target quality of each healthy reference node according to the total amount of all adjustable quality, forming an adjustable quality balance state; Compare the adjustable quality of the current Laguerre cell carried by each healthy reference node with the corresponding remaining target quality to determine the adjustable quality difference. Adjust the Oliker-Prussner node weights according to the adjustable quality difference. When the adjustable quality exceeds the corresponding remaining target quality, adjust the node weights in the direction of shrinking the carrying range of the corresponding Laguerre cell. When the adjustable quality is lower than the corresponding remaining target quality, adjust the node weights in the direction of expanding the carrying range of the corresponding Laguerre cell. After each node weight adjustment, keep the fixed assigned quality and the corresponding healthy reference assignment unchanged. Only for the adjustable quality, call the health response to correct the transportation cost and re-divide the Laguerre cells. Repeatedly execute adjustable quality statistics, quality difference comparison, node weight adjustment, and Laguerre cell update until the difference between the adjustable quality of each healthy reference node and the corresponding remaining target quality falls within the allowable deviation range. Retain the node weights and Laguerre cell partitioning results after quality balancing is completed. Merge the health reference attribution of the fixed attribution quality with the health reference attribution of the rebalanced adjustable quality. Organize the correspondence according to the continuous acquisition time sequence of the target monitoring data to form the corrected health reference attribution status.
8. The inverter aging monitoring data processing method based on big data according to claim 1, characterized in that, The results of the inverter aging monitoring are obtained specifically in the following ways: Read the corrected health reference attribution status, call the corresponding health response mapping relationship according to the updated health reference node attribution, reconstruct the normal health response corresponding to the target operating condition status, and form the reconstructed health reference status. By comparing the actual health response of the target monitoring data with the normal health response of the reconstructed health reference state, and combining the quality attribution state to screen for health response deviations, the irreversible aging state is obtained. The irreversible aging state is organized according to the acquisition time sequence of the target monitoring data, the health reference attribution, duration range and degree of change corresponding to irreversible aging are determined, and the inverter aging monitoring results are output.
9. A big data-based inverter aging monitoring data processing system, comprising executing the big data-based inverter aging monitoring data processing method according to any one of claims 1 to 8, characterized in that, include: The data status characterization module is used to acquire the inverter's health operation data and target monitoring data, perform status characterization, and obtain the inverter's monitoring status. The health reference construction module is used to build health reference nodes for inverter monitoring status, establish a mapping relationship between changes in operating conditions and changes in health response, and generate health reference status. The transportation cost optimization module is used to correct the transportation cost between the working condition and the health reference node using the health reference state. It uses Oliker-Prussner to perform semi-discrete optimal transmission, constructs Laguerre cells through node weights, and obtains the initial health reference home state. The return residual extraction module is used to track the migration of the Laguerre cell in the initial healthy reference home state. When the target inverter returns to the previous Laguerre cell, the change in operating condition response is deducted to obtain the return residual state under the same operating condition. The transportation qualification classification module is used to classify the source quality transportation qualification based on the return residual status under the same working conditions, and obtain the quality classification status including fixed quality and adjustable quality. The quality balance correction module is used to separate the fixed-attribute quality from the quality balance using the quality attribution state, and update the node weights and Laguerre cells with the quality difference of the adjustable quality to form a corrected healthy reference attribution state. The aging status identification module is used to reconstruct the health reference status by using the corrected health reference attribution status, identify irreversible aging by combining target monitoring data and quality attribution status, and output the inverter aging monitoring results.
