Wind turbine generator converter igbt life prediction method based on operating condition

Abstract

The application relates to the field of converter IGBT life prediction, and provides a wind turbine converter IGBT life prediction method based on operation conditions, which synchronously acquires low-frequency SCADA data and high-frequency fault recording data of a wind turbine, constructs a parallel analysis path of conventional condition fatigue damage and extreme event impact damage, constructs an interaction model of impact events and subsequent fatigue damage, generates a dynamic fatigue damage amplification coefficient time sequence which decays with time based on the intensity of the identified impact events, thereby correcting the fatigue damage accumulation under the conventional condition in real time, fusing the direct impact damage and the corrected fatigue damage to obtain total cumulative damage which comprehensively reflects the influence of the actual condition, and predicting the remaining service life based on the total cumulative damage and the running time of the wind turbine converter IGBT, so that the problem that the prediction result is seriously inaccurate due to the neglect of the impact-fatigue coupling effect in the prior art is solved, and the life prediction accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of converter IGBT life prediction technology, and in particular to a method for predicting the life of wind turbine converter IGBTs based on operating conditions. Background Technology

[0002] Wind power, as a crucial component of renewable energy, plays a central role in the global energy structure transformation. In wind turbine units, the converter is a key device for achieving power conversion and grid-friendly integration; its operational reliability directly affects the stability and power generation efficiency of the entire unit. The Insulated Gate Bipolar Transistor (IGBT) module, as the core power semiconductor device of the converter, endures complex electrothermal stress cycles caused by wind speed fluctuations and grid disturbances over long periods, making it one of the most vulnerable components with a high failure rate in the converter and even the entire wind turbine unit. IGBT module failure often leads to unplanned outages, resulting in significant power generation losses and maintenance costs.

[0003] Currently, commonly used IGBT lifetime prediction methods typically follow a predetermined technical path: First, acquire the operating data of the wind turbine and estimate the junction temperature variation curve of the IGBT using an electro-thermal model; then, process the junction temperature variation curve using algorithms such as rainflow counting to extract the amplitude and mean of a series of thermal cycles; finally, substitute these thermal cycle parameters into a lifetime formula based on models such as Coffin-Manson, and linearly superimpose the damage caused by different thermal cycles according to Miner's linear cumulative damage rule, thereby assessing the total damage and predicting the remaining lifetime. However, such methods mainly model the thermal fatigue damage mechanism under normal operating conditions, seriously neglecting the destructive impact of extreme operating conditions, especially grid fault events such as Low Voltage Ride Through (LVRT), on IGBT lifetime. The damage characteristic under normal operating conditions is fatigue caused by the accumulation of thousands of small-to-medium amplitude thermal cycles, while an LVRT event is a severe, asymmetrical thermal shock triggered within milliseconds due to the converter's need to urgently inject a large amount of reactive current into the grid. These two types of damage are fundamentally different in terms of physical processes, time scales, and damage mechanisms. The low-frequency SCADA data that existing lifetime prediction frameworks rely on cannot capture instantaneous events such as LVRT, and their core rainflow counting method and fatigue life model are not designed to quantify single massive thermal shock damage. This makes it impossible for existing IGBT lifetime prediction methods to accurately assess the massive instantaneous damage caused by LVRT events. Summary of the Invention

[0004] The present invention aims to solve at least one of the problems existing in the prior art and provide a method for predicting the life of IGBTs in wind turbine converters based on operating conditions.

[0005] One aspect of the present invention provides a method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions, the method comprising: Acquire SCADA data and fault recording data of wind turbine units; Data synchronization and event identification are performed on SCADA data and fault recording data to obtain routine operating condition data and event sets; Based on conventional working condition data, fatigue damage accumulation under conventional working conditions is performed to obtain total fatigue damage and fatigue damage time series. Extreme event impact damage quantification is performed on the event set to obtain the impact damage set; The total fatigue damage, the time series of fatigue damage, and the impact damage set are fused to obtain the total cumulative damage. The remaining useful life is predicted based on the total cumulative damage and the operating time of the wind turbine converter IGBT.

[0006] Optionally, SCADA data and fault recording data are synchronized and event identified to obtain routine operating condition data and event sets, including: Each fault recording data block in the fault recording data is input into a rule-based event recognizer to obtain an event set and a list of event time intervals; The SCADA data is bifurcated based on the event time interval list to obtain the normal operating condition data.

[0007] Optionally, the SCADA data is forked based on a list of event time intervals to obtain the routine operating condition data, including: Extract each event time interval from the list of event time intervals; Add a safety buffer margin to each event time interval to obtain an isolation window set; Iterate through each data point in the SCADA data and determine whether each data point falls within any isolation window in the isolation window set. If not, add the corresponding data point to the normal operating condition data.

[0008] Optionally, fatigue damage accumulation under normal operating conditions is performed based on normal operating condition data to obtain total fatigue damage and time series of fatigue damage, including: Power loss time series are obtained by searching for power loss at each data point in the normal operating condition data based on the power loss lookup table. Extracting the radiator temperature time series from routine operating condition data; The junction temperature profile is estimated based on the lumped thermal network model to obtain the IGBT time series by performing power loss time series and radiator temperature time series. Cumulative damage is calculated based on IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage.

[0009] Optionally, cumulative damage calculation is performed based on the IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage, including: The thermal cycling matrix is ​​obtained by extracting the thermal cycling matrix from the IGBT time series based on the rainflow counting method. The cumulative damage is calculated based on the life model and Miner's rule on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage.

[0010] Optionally, cumulative damage calculation based on a life model and Miner's rule is performed on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage, including: The failure lifetime vector is derived from the thermal cycling matrix based on the lifetime model. Extract the cycle count vector from the thermal cycle matrix; The failure lifetime vector and cycle count vector are quantized into a damage contribution vector based on Miner's rule to obtain the loss contribution vector. The fatigue damage is aggregated and calculated from the loss contribution vector to obtain the total fatigue damage; Extract the original list of cycles with timestamps from the hot cycle matrix; The time-resolved damage sequence is reconstructed based on the original circular list with timestamps to obtain the time series of the fatigue damage.

[0011] Optionally, total damage is fused from the total fatigue damage, the time series of fatigue damage, and the impact damage set to obtain the total cumulative damage, including: The total impact damage is obtained by directly aggregating the impact damage components of the impact damage set. Based on the impact damage set, a time series of fatigue damage amplification factor is generated; Based on the fatigue damage amplification factor time series, the fatigue damage time series and total fatigue damage are calculated using time-varying acceleration correction to obtain the corrected fatigue damage. The corrected fatigue damage and the total impact damage are combined to obtain the total cumulative damage.

[0012] Optionally, the impact damage set is directly aggregated into impact damage components to obtain the total impact damage, including: Extract the impact damage value corresponding to each event entry in the impact damage set; The total impact damage is obtained by summing up all the extracted impact damage values.

[0013] Optionally, based on the time series of fatigue damage amplification factor, a time-varying acceleration-based corrected fatigue damage calculation is performed on the time series of fatigue damage and the total fatigue damage to obtain the corrected fatigue damage, including: Based on the time series of fatigue damage amplification factor and fatigue damage, the time series of corrected damage increment is calculated. The corrected damage increment time series is accumulated into the total fatigue damage to obtain the corrected fatigue damage.

[0014] Optionally, the remaining useful life is predicted based on the total accumulated damage and the operating time of the wind turbine converter IGBTs, including: Divide the total accumulated damage by the operating time of the wind turbine converter IGBT to calculate the average damage accumulation rate of the wind turbine converter IGBT during the historical operation period. Subtract the current total accumulated damage from the preset damage threshold to obtain the remaining damage margin. Divide the remaining damage margin by the average damage accumulation rate to obtain the predicted remaining service life.

[0015] Compared with existing technologies, this invention provides a method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions. It simultaneously acquires low-frequency SCADA data and high-frequency fault recording data from the wind turbine, constructing a parallel analysis path for fatigue damage under normal operating conditions and impact damage from extreme events. This solves the problem of existing technologies being unable to accurately quantify the massive instantaneous impact damage caused by extreme events such as low-voltage ride-through due to a single data source. Furthermore, this invention constructs an interaction model between impact events and subsequent fatigue damage. Based on the severity of identified impact events, it generates a time series of dynamic fatigue damage amplification coefficients that decays over time, and uses this to correct the accumulation of fatigue damage under normal operating conditions in real time. Finally, by fusing the direct impact damage with the corrected fatigue damage, it obtains the total accumulated damage that comprehensively reflects the impact of actual operating conditions. Based on the total accumulated damage and the operating time of the wind turbine converter IGBTs, it predicts the remaining service life, solving the technical problem of existing technologies neglecting the impact-fatigue coupling effect, which leads to severely inaccurate prediction results and improves the accuracy of lifespan prediction. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with the corresponding pictures in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0017] Figure 1 This is a flowchart of a wind turbine converter IGBT life prediction method based on operating conditions according to an embodiment of the present invention; Figure 2 This is a data flow diagram of a wind turbine converter IGBT lifetime prediction method based on operating conditions according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions, according to an embodiment of the present invention, which involves accumulating fatigue damage under normal operating conditions based on normal operating condition data to obtain total fatigue damage and a time series of fatigue damage. Figure 4 A flowchart illustrating the cumulative damage calculation based on IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage, according to an embodiment of the present invention, of the wind turbine converter IGBT life prediction method based on operating conditions. Figure 5 This is a flowchart illustrating the process of fusing total fatigue damage, the time series of fatigue damage, and the set of impact damage to obtain total cumulative damage in the wind turbine converter IGBT life prediction method based on operating conditions according to an embodiment of the present invention. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0019] As indicated in the specification and claims of this invention, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] This invention uses flowcharts to illustrate the specific operational steps of the method according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously, as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0021] Existing methods for predicting the lifespan of IGBTs in wind turbine converters heavily rely on thermal fatigue accumulation models under conventional operating conditions, thus failing to accurately quantify the instantaneous impact damage caused by extreme events such as low-voltage ride-through (LVRT). More importantly, existing technologies neglect the crucial physical interaction that microscopic damage caused by massive thermal shocks accelerates subsequent fatigue accumulation, leading to predictions that significantly deviate from reality. Therefore, this invention proposes a dual-modal damage fusion prediction method that considers the impact-fatigue interaction, namely, a wind turbine converter IGBT lifespan prediction method based on operating conditions. This method first constructs two parallel damage analysis paths by simultaneously processing and identifying low-frequency SCADA data and high-frequency fault recording data from the wind turbine: First, based on conventional operating condition data, using mature thermal cycling analysis and fatigue damage accumulation models, the basic total fatigue damage and its time distribution sequence are calculated; second, for identified extreme events such as LVRT, the direct impact damage caused by them is accurately quantified using high-frequency recording data. The innovation of this invention lies in the construction of a dynamic damage fusion model. This model generates a fatigue damage amplification coefficient time series that decays over time based on the impact damage set. The fatigue damage amplification coefficient is then used to weight and correct the fatigue damage time series after the impact event, thereby quantifying the accelerating effect of the impact on subsequent fatigue. Finally, the direct impact damage component and the corrected fatigue damage component are fused to obtain a summary cumulative damage value that is closer to the physical reality. Based on this, the remaining service life is predicted, significantly improving the accuracy of the prediction.

[0022] Figure 1 This is a flowchart of a wind turbine converter IGBT lifetime prediction method based on operating conditions, according to an embodiment of the present invention. Figure 2 This is a data flow diagram illustrating the IGBT lifetime prediction method for wind turbine converters based on operating conditions, according to an embodiment of the present invention. (In conjunction with...) Figure 1 and Figure 2According to an embodiment of the present invention, a method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions includes: S100, acquiring SCADA data and fault recording data of the wind turbine; S200, performing data synchronization and event identification on the SCADA data and fault recording data to obtain normal operating condition data and an event set; S300, accumulating fatigue damage under normal operating conditions based on the normal operating condition data to obtain total fatigue damage and a time series of fatigue damage; S400, quantifying extreme event impact damage on the event set to obtain an impact damage set; S500, fusing the total fatigue damage, the time series of fatigue damage, and the impact damage set to obtain total accumulated damage; and S600, predicting the remaining service life based on the total accumulated damage and the operating time of the wind turbine converter IGBTs to obtain the remaining service life.

[0023] Specifically, in step S100, SCADA data and fault recording data of the wind turbine are acquired. It should be understood that the thermal stress experienced by the IGBTs of the wind turbine converter has dual-mode characteristics: quasi-periodic thermal fatigue under normal operating conditions and transient thermal shock under extreme conditions such as grid faults. These two stress characteristics differ in time scale and amplitude, and a single data source cannot fully represent their entirety. Therefore, in the technical solution of this invention, SCADA data and fault recording data of the wind turbine are acquired to construct a multi-source heterogeneous dataset that can simultaneously cover long-term macroscopic operating trends and millisecond-level microscopic transient responses. This provides a complete and necessary data foundation for subsequent quantification of conventional fatigue damage and extreme impact damage, ensuring the comprehensiveness of the input information for the life prediction model.

[0024] More specifically, in a specific example of the present invention, the process of acquiring data in step S100 includes the following operations: First, access the SCADA master control server of the wind turbine through the industrial Ethernet interface, and submit a data request containing the unique identifier of the wind turbine, the target time range, and required variables such as active power and radiator temperature, to obtain a time series data file containing data aggregated at a target time interval, such as ten minutes; then, retrieve all fault waveform data blocks triggered by a preset grid voltage drop threshold within the same target time range through the fieldbus protocol or by directly accessing the storage unit embedded in the converter controller. These data blocks record the IGBT current and voltage waveforms within a specific time window before and after the event at a high sampling rate; finally, integrate the acquired SCADA time series data file, i.e., SCADA data, with all fault waveform data blocks, i.e., fault waveform data, and transmit them to a unified data analysis platform as the raw input for subsequent processing.

[0025] Specifically, in step S200, SCADA data and fault recording data are synchronized and event identified to obtain normal operating condition data and event sets. It should be understood that the acquired SCADA data and fault recording data are heterogeneous in terms of time reference, sampling frequency, and recording method, and correspond to two different physical damage processes of IGBTs: normal fatigue and extreme impact, respectively, making direct unified damage quantification impossible. Therefore, in the technical solution of this invention, SCADA data and fault recording data are further synchronized and event identified to obtain normal operating condition data and event sets, thereby establishing a unified time coordinate. Based on the extreme event characteristics identified in the fault recording data, the mixed data sources are precisely divided into a continuous low-frequency dataset representing normal operating conditions and a discrete high-frequency event set representing power grid faults. This effectively separates the data streams corresponding to different damage mechanisms, providing clear, non-interfering data input for subsequent targeted fatigue damage accumulation and impact damage quantification using different models, which is a prerequisite for ensuring the accuracy of dual-modal analysis.

[0026] More specifically, in this embodiment of the invention, data synchronization and event identification are performed on SCADA data and fault waveform data to obtain normal operating condition data and event set, including: inputting each fault waveform data block in the fault waveform data into a rule-based event identifier to obtain an event set and an event time interval list; and performing data stream forking on the SCADA data based on the event time interval list to obtain the normal operating condition data.

[0027] Accordingly, each fault waveform data block in the fault waveform recording data is input into a rule-based event identifier to obtain an event set and an event time interval list. It should be understood that the original fault waveform recording data is merely raw waveform data segments recorded based on simple threshold triggering; it does not contain structured information about the nature of the event or its precise start and end times, and cannot be directly used for subsequent data stream separation and targeted damage quantification. Therefore, in the technical solution of this invention, each fault waveform data block in the fault waveform recording data is further input into a rule-based event identifier to obtain an event set and an event time interval list. This allows for the parsing and calibration of the raw waveform data, automatically identifying the specific extreme event type corresponding to each recording, and extracting its precise span on the time axis. In this way, unstructured waveform recording data can be transformed into a structured event set containing event types and precise timestamps that can be directly accessed in subsequent steps, providing the necessary basis for accurately removing corresponding time periods from conventional operating condition data.

[0028] More specifically, in a specific example of the present invention, the process of obtaining the event set and the event time interval list first involves pre-setting a set of digital signature rules for determining low-voltage ride-through events. These rules define a voltage drop threshold as the effective value of the grid voltage falling below 90% of the rated voltage, with a minimum duration of 20 milliseconds. At the start of processing, a fault waveform data block is read from the fault waveform data, and the grid voltage time series is extracted. Then, the grid voltage time series is traversed to find the first time point below the preset voltage drop threshold as the start timestamp of a potential event, and the process continues to search for the first time point to recover to the preset voltage drop threshold. The time point above the set voltage drop threshold is used as the end timestamp. Then, the time difference between the start and end timestamps is calculated, and it is determined whether this time difference is greater than a preset minimum duration threshold. If the time difference is greater than the preset minimum duration threshold, it is confirmed that the fault waveform data block has recorded a valid low-voltage ride-through event. Subsequently, the corresponding event type label, start timestamp, and current and voltage waveform data within the fault waveform data block are encapsulated into a structured object and stored in the event set. At the same time, the time interval formed by the start and end timestamps is recorded in the event time interval list. This operation is performed sequentially for all fault waveform data blocks.

[0029] Accordingly, the SCADA data is forked based on the event time interval list to obtain the conventional operating condition data. It should be understood that the original SCADA data stream completely records the entire operating history of the wind turbine, covering both the steady-state conventional operating phase and the severe dynamic response phase caused by grid faults. However, the cumulative model used to assess thermal fatigue damage is only suitable for analyzing conventional operating conditions. If data containing fault dynamic processes is input into this cumulative model, non-fatigue thermal stress will be introduced, thus interfering with the accuracy of fatigue damage calculation. Therefore, in the technical solution of this invention, the SCADA data is further forked based on the event time interval list to obtain the conventional operating condition data. This utilizes the precise event timestamps identified in the previous step to accurately segment and filter continuous SCADA data, stripping away data segments representing extreme operating conditions. This generates a clean dataset containing only conventional operating condition information, ensuring the homogeneity and effectiveness of subsequent fatigue damage assessment data input, laying the foundation for accurate calculation of fatigue damage components.

[0030] More specifically, in a specific example of the present invention, data stream forking of SCADA data based on an event time interval list to obtain the normal operating condition data includes: extracting each event time interval from the event time interval list; adding a safety buffer margin to each event time interval to obtain an isolation window set; traversing each data point in the SCADA data and determining whether each data point falls within any isolation window in the isolation window set; if not, adding the corresponding data point to the normal operating condition data. In other words, the data stream forking process first extracts the specific time intervals corresponding to each identified low-voltage ride-through event from the event time interval list as each event time interval. Then, a ten-minute safety buffer margin is applied before and after each event time interval, forming a wider-coverage isolation window. This margin is designed to exclude potential instability in the wind turbine system before the event and the thermodynamic and electrical recovery processes after the event, which are not part of normal steady-state operation. This operation is performed on all event time intervals in the event time interval list, ultimately generating a set of isolation windows. Finally, all data points in the SCADA data are traversed, and the timestamp of each data point is assigned. If the timestamp of a data point does not fall within any isolation window in the isolation window set, the data point and its contained information, such as active power and radiator temperature, are retained and added to the normal operating condition dataset. Conversely, if the timestamp of a data point in the SCADA data falls within any isolation window in the isolation window set, the data point is discarded. After traversing all SCADA data points, the final normal operating condition data used for fatigue damage analysis is obtained.

[0031] Specifically, in step S300, fatigue damage accumulation under normal operating conditions is performed based on normal operating condition data to obtain total fatigue damage and a time series of fatigue damage. It should be understood that during normal grid-connected power generation, the power loss of the IGBT module fluctuates with wind speed and output power, leading to repeated changes in internal junction temperature. This continuous thermal cycle is the main physical cause of fatigue aging of the module's bond wires and solder layer, forming the basis for its lifespan depletion. Therefore, in the technical solution of this invention, fatigue damage accumulation under normal operating conditions is further performed based on normal operating condition data to obtain total fatigue damage and a time series of fatigue damage. This allows for the application of a series of electrothermal conversion and damage accumulation models to transform the purified operating condition data into quantitative fatigue damage values ​​that reflect the normal losses of the device. In this way, not only can the overall damage benchmark caused by normal operation within the evaluation period be obtained, but also a damage increment sequence precisely corresponding to the time axis can be generated. This damage increment sequence is the key basis for subsequent damage correction considering impact fatigue interaction.

[0032] Figure 3 This is a flowchart illustrating the method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions, according to an embodiment of the present invention. It describes the accumulation of fatigue damage under normal operating conditions based on conventional operating condition data to obtain total fatigue damage and a time series of fatigue damage. Figure 3 As shown, step S300 includes: S310, performing a power loss lookup table on each data point in the normal operating condition data to obtain a power loss time series; S320, extracting the radiator temperature time series from the normal operating condition data; S330, estimating the junction temperature profile based on a lumped thermal network model on the power loss time series and the radiator temperature time series to obtain the IGBT time series; S340, performing cumulative damage calculation based on the IGBT time series to obtain the total fatigue damage and the fatigue damage time series.

[0033] In step S310, a power loss time series is obtained by performing a power loss lookup table on each data point in the conventional operating condition data. It should be understood that the junction temperature of the IGBT module is a direct physical quantity for assessing its thermal fatigue damage, and the increase in junction temperature originates from the power loss within the module. However, the power loss value cannot be directly obtained through the SCADA system, and its accurate online calculation involves complex electromagnetic transient processes, resulting in a huge computational burden. Therefore, in the technical solution of this invention, a power loss time series is further obtained by performing a power loss lookup table on each data point in the conventional operating condition data. This provides an efficient and engineering-practical method to map directly obtainable macroscopic operating parameters, such as the active power of a wind turbine, to the microscopic heating power of the IGBT module. In this way, the original operating condition data can be transformed into the direct heat source input required for the subsequent thermal network model, completing the key data conversion from the electrical domain to the thermal domain and providing the prerequisite for junction temperature estimation.

[0034] More specifically, in a specific example of the present invention, the process of obtaining the power loss time series first involves pre-constructing a two-dimensional power loss lookup table. The two input dimensions of this lookup table are active power and radiator temperature. The values ​​stored in the lookup table are the total power loss values ​​of the IGBTs under the corresponding operating conditions, obtained through offline simulation or experimental calibration. When processing normal operating condition data, an operating point is extracted from the data. This operating point contains an active power value and a radiator temperature value. These two values ​​are then used as index coordinates to query the pre-defined power loss lookup table. Since the actual operating condition value may lie between grid points in the lookup table, a bilinear interpolation algorithm is used to calculate the precise power loss value of the current operating point based on the power loss values ​​of the four nearest grid points surrounding it. Finally, the calculated power loss value is associated with the timestamp of the operating point and stored as a new operating point in the power loss time series. This operation iterates through all operating points in the normal operating condition data, thereby generating a complete power loss time series covering the entire evaluation period.

[0035] In step S320, the heat sink temperature time series is extracted from the normal operating condition data. It should be understood that the junction temperature of an IGBT module depends not only on its own power loss but also on the heat dissipation efficiency of the cooling system. The heat sink temperature is a directly measurable key physical quantity characterizing the real-time thermal state of the cooling system, constituting the thermal boundary condition of the junction temperature estimation model. Therefore, in the technical solution of this invention, the heat sink temperature time series is further extracted from the normal operating condition data to separate this important thermal boundary parameter from the multivariate original dataset, forming an independent data sequence strictly aligned with the power loss time series on the time axis. This provides the necessary reference temperature for calculating the junction temperature in the subsequent lumped thermal network model, ensuring that the junction temperature estimation result accurately reflects the influence of actual heat dissipation conditions.

[0036] More specifically, in a specific example of the present invention, the extraction process of the radiator temperature time series first takes a filtered normal operating condition dataset as input. The structure of this normal operating condition dataset is a data table containing multiple columns such as timestamps, active power, and radiator temperature. Next, a specific column recording the radiator temperature in the data table is locked according to a predefined variable name. Subsequently, all values ​​of this specific column and all timestamp values ​​in the same row are read out together. Finally, these pairs of timestamps and temperature values ​​are organized into a new time series data structure. Each item in this structure consists of a time point and the radiator temperature value corresponding to that time point, thereby forming a radiator temperature time series that can be directly called in subsequent steps.

[0037] In step S330, the junction temperature profile of the power loss time series and the heat sink temperature time series is estimated based on a lumped heat network model to obtain the IGBT time series. It should be understood that the junction temperature of the IGBT chip is the fundamental physical quantity that directly drives thermal fatigue damage to its internal materials. However, this junction temperature cannot be directly measured in actual operating wind turbine units and must be estimated through indirect modeling. Therefore, in the technical solution of this invention, the junction temperature profile of the power loss time series and the heat sink temperature time series is further estimated based on a lumped heat network model to obtain the IGBT time series. This utilizes known heat source input and thermal boundary conditions, and through a physical model characterizing the heat transfer path within the module, to solve and reconstruct the complete curve of the IGBT chip junction temperature changing over time, which cannot be directly perceived. In this way, a junction temperature profile can be generated throughout the entire evaluation period. This junction temperature profile is the core and direct data foundation for subsequent thermal cycle extraction and fatigue damage quantification.

[0038] More specifically, in a specific example of the present invention, the estimation process of the junction temperature profile employs a simplified steady-state lumped thermal network model. First, an equivalent junction-to-case thermal resistance parameter is consulted and set from the IGBT module's specifications. Then, during data processing, the power loss value and radiator temperature value corresponding to each timestamp are extracted from the power loss time series and the radiator temperature time series. Next, for each timestamp, the power loss value at the corresponding moment is multiplied by the junction-to-case thermal resistance parameter according to the heat conduction formula to obtain the temperature difference between the junction temperature and the case temperature. This temperature difference is then superimposed on the radiator temperature value at that moment to calculate the instantaneous junction temperature. This calculation operation is performed for each synchronous data point in the power loss time series and the radiator temperature time series. Finally, all the calculated junction temperature values ​​and their corresponding timestamps are combined to form an IGBT junction temperature time series covering the entire evaluation period, which serves as the IGBT time series.

[0039] In step S340, cumulative damage calculation is performed based on the IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage. It should be understood that IGBT fatigue damage is not directly determined by the absolute value of the junction temperature, but rather by the repeated fluctuations in junction temperature, i.e., thermal cycling. Therefore, the continuous junction temperature time series obtained in the previous step cannot directly characterize the magnitude of the cumulative damage; it must be transformed into a series of discrete, countable stress cycling events. Therefore, in the technical solution of this invention, cumulative damage calculation is further performed based on the IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage. This allows for the application of a cycle counting algorithm and a physical failure-based lifetime model to analyze the continuous temperature curve into structured damage events, and quantifies and linearly superimposes the damage caused by each event. In this way, the complex electrothermal stress history can ultimately be transformed into one or a set of clear quantitative indicators for characterizing the device's lifetime consumption under normal operating conditions, completing the calculation endpoint of the fatigue damage path.

[0040] Figure 4 This is a flowchart illustrating the cumulative damage calculation based on IGBT time series data to obtain the total fatigue damage and the time series data of the fatigue damage, according to an embodiment of the wind turbine converter IGBT life prediction method based on operating conditions according to an embodiment of the present invention. Figure 4 As shown, step S340 includes: S341, extracting the thermal cycling matrix from the IGBT time series based on the rainflow counting method to obtain the thermal cycling matrix; S342, calculating the cumulative damage from the thermal cycling matrix based on the life model and Miner's rule to obtain the total fatigue damage and the time series of the fatigue damage.

[0041] In step S341, a thermal cycling matrix is ​​extracted from the IGBT time series using the rainflow counting method to obtain the thermal cycling matrix. It should be understood that IGBT fatigue damage accumulates from repeated fluctuations in junction temperature. The original junction temperature time series is a complex non-periodic waveform that cannot be directly substituted into a lifetime model with discrete cycle counts as input. It must first be decomposed into a series of equivalent, countable standard thermal cycles. Therefore, in the technical solution of this invention, a thermal cycling matrix is ​​further extracted from the IGBT time series using the rainflow counting method to obtain the thermal cycling matrix. This employs an engineering-recognized cycle counting standard to transform the continuous temperature change history into a structured statistical matrix that comprehensively reflects the cycle amplitude, mean, and frequency. This simplifies the complex load spectrum into a standardized input suitable for lifetime model calculations, completing the crucial transformation from time-domain temperature signals to frequency-domain cycle statistics, and providing a direct data interface for subsequent damage quantification calculations.

[0042] More specifically, in a specific example of the present invention, the extraction process of the thermal cycle matrix first processes the IGBT junction temperature time series, i.e., the IGBT time series, to identify and extract all local peak points and valley points, forming a peak-valley alternating time series as a peak-valley sequence. Next, this peak-valley sequence is used as input for the rainflow counting method. The rainflow counting method compares and pairs four adjacent peak-valley points by simulating the flow and dripping rules of water along the sequence path to identify all closed temperature cycles. Then, for each identified temperature cycle, its temperature difference amplitude and average temperature value are calculated. Finally, a two-dimensional matrix is ​​predefined, whose rows and columns correspond to bin intervals of temperature difference amplitude and average temperature, respectively. Each temperature cycle is assigned to the corresponding bin in the two-dimensional matrix according to its parameters, i.e., temperature difference amplitude and average temperature value, and the counter in the corresponding bin is incremented by one. After processing all temperature cycles, the resulting two-dimensional counting matrix is ​​the final output thermal cycle matrix.

[0043] In step S342, cumulative damage calculation based on the lifetime model and Miner's rule is performed on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage. It should be understood that the thermal cycling matrix obtained in the previous step is only a statistical description of the thermal stress experienced by the IGBT; it is not a direct measure of damage. Different thermal cycles with different parameters have different effects on device lifetime, requiring a physical model that can correlate stress and lifetime for conversion. Therefore, in the technical solution of this invention, cumulative damage calculation based on the lifetime model and Miner's rule is further performed on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage. This allows for a quantitative assessment of the damage contribution of each type of thermal cycle in the thermal cycling matrix. The damage caused by all different types of thermal cycles is superimposed using the linear accumulation rule to obtain a total fatigue damage amount as the total fatigue damage. In this way, statistical stress cycling data can be finally transformed into normalized damage values ​​characterizing device lifetime consumption, and a damage sequence with time-series information necessary for subsequent damage fusion steps can be generated, completing the final quantitative step of conventional operating condition path analysis.

[0044] More specifically, in a specific example of the present invention, performing cumulative damage calculation on the thermal cycling matrix based on a lifetime model and Miner's rule to obtain the total fatigue damage and the time series of the fatigue damage includes: deriving a failure lifetime vector from the thermal cycling matrix based on a lifetime model to obtain a failure lifetime vector; extracting a cycle count vector from the thermal cycling matrix; quantizing the failure lifetime vector and the cycle count vector using a damage contribution vector based on Miner's rule to obtain a loss contribution vector; performing aggregate calculation of the loss contribution vector to obtain the total fatigue damage; extracting a timestamped original cycle list from the thermal cycling matrix; and reconstructing a time-resolved damage sequence based on the timestamped original cycle list to obtain the time series of the fatigue damage.

[0045] Accordingly, a failure lifetime vector is derived from the thermal cycling matrix based on the lifetime model to obtain the failure lifetime vector; the cycle count vector is extracted from the thermal cycling matrix. It should be understood that the core of Miner's cumulative damage rule, or Miner's rule, is to calculate the ratio of the actual number of load cycles to the number of failure cycles under that load. Therefore, before calculating the final damage contribution value, two core elements must be considered: the actual number of cycles n and the theoretical failure lifetime. The failure lifetime vector is explicitly separated and calculated from the thermal cycling matrix. Therefore, in the technical solution of this invention, the failure lifetime vector is further derived from the thermal cycling matrix based on the lifetime model, and the cycle count vector is extracted from the thermal cycling matrix. This is used to deconstruct the two types of information contained in the thermal cycling matrix: first, the stress level is converted into the device's load-bearing capacity under that stress through the lifetime model; second, the actual number of cycles under that stress level is directly extracted. In this way, two vectors with the same dimension and one-to-one correspondence of elements can be generated, namely the failure lifetime vector representing the damage denominator and the cycle count vector representing the damage numerator. This provides standardized and aligned data preparation for the next step of performing element-by-element division to quantify the damage contribution of various cycles.

[0046] More specifically, in a specific example of the present invention, the thermal cycling matrix is ​​first taken as input, and two empty vectors of the same size as the number of rows in the thermal cycling matrix are initialized. One empty vector is used as a failure lifetime vector to store the failure lifetime, and the other empty vector is used as a cycle count vector to store the cycle count. Then, the thermal cycling matrix is ​​traversed row by row. For the first row of the thermal cycling matrix... Okay, first extract the two stress parameters, the temperature difference amplitude and the average temperature, and then substitute them into the preset life model formula to calculate the number of failure cycles corresponding to this type of cycle. Then As the first The first element is stored in the failure lifetime vector; simultaneously, the first element is retrieved from the thermal cycle matrix. The number of times this type of loop occurred can be directly read from the count value column of the row. and take it as the first Each element is stored in the cycle count vector. This traversal process will continue until all rows of the thermal cycle matrix have been processed, and finally output a failure lifetime vector and a cycle count vector with equal dimensions and complete content.

[0047] Accordingly, the failure lifetime vector and cycle count vector are quantized using the Miner's rule to obtain the loss contribution vector. It should be understood that the previous step only prepared the numerator (actual cycle count) and denominator (failure cycle count) required for damage calculation, but did not combine them to form a physical quantity that directly reflects the proportion of lifetime consumption. Therefore, in the technical solution of this invention, the failure lifetime vector and cycle count vector are further quantized using the Miner's rule to obtain the loss contribution vector. This allows for the calculation of the normalized damage value caused by each type of thermal cycle within the evaluation period, strictly based on the linear cumulative damage theory. This generates a vector that quantifies the degree of damage from various stress cycles, where each element directly corresponds to the fraction of lifetime consumed by a particular thermal cycle, providing quantified sub-item data for the final calculation of total fatigue damage.

[0048] More specifically, in a specific example of the present invention, the quantization process of the loss contribution vector first uses the failure lifetime vector and the cycle count vector as input vectors to initialize a loss contribution vector with the same dimension as the input vectors for storing the results; then, it performs element-wise operations on the input vectors, starting from the 1st element of the cycle count vector... Number of times the bit fetch loop occurs and from the first of the failure lifetime vector The corresponding failure cycle number is retrieved. Then perform the division operation. , obtained the Damage contribution value of thermal cycling ; then The first of the loss contribution vectors is stored in the loss contribution vector. This operation will be applied sequentially to all corresponding elements of the two input vectors until all types of thermal cycles have calculated their damage contribution values ​​and stored them in the loss contribution vector; finally, the loss contribution vector filled with damage contribution values ​​is the output of the quantization process of the loss contribution vector.

[0049] Accordingly, fatigue damage aggregation calculations are performed on the loss contribution vector to obtain the total fatigue damage. It should be understood that the loss contribution vector obtained in the previous step is a quantified result of various thermal cycle damages. To obtain the overall fatigue life consumption caused by normal operating conditions within the evaluation period, these discrete, individual damage values ​​must be combined. Therefore, in the technical solution of this invention, fatigue damage aggregation calculations are further performed on the loss contribution vector to obtain the total fatigue damage. Based on the linear cumulative damage theory, the life consumption fractions caused by thermal cycles at all different stress levels are summed to obtain a scalar value that represents the total effect. In this way, the total fatigue damage caused by normal operation within the evaluation period can be finally determined, providing a clear benchmark damage value for subsequent integration with impact damage.

[0050] More specifically, in a specific example of the present invention, the fatigue damage aggregation calculation process of the loss contribution vector first takes the loss contribution vector as input and initializes an accumulator variable for storing the total fatigue damage, with an initial value of zero; then, all elements in the loss contribution vector are summed, that is, each element value in the loss contribution vector is sequentially added to the accumulator variable. After traversing all elements, the final value of the accumulator variable is the total fatigue damage.

[0051] Accordingly, a time-stamped original cycle list is extracted from the thermal cycling matrix; a time-resolved damage sequence is reconstructed based on the time-stamped original cycle list to obtain the time series of the fatigue damage. It should be understood that the total fatigue damage calculated in the previous steps is an aggregated scalar value within an evaluation period, which loses the temporal information during the damage occurrence process. To accurately account for the accelerating effect of impact events on subsequent fatigue accumulation, the basic fatigue damage increment at each moment must be known. Therefore, in the technical solution of this invention, a time-stamped original cycle list is further extracted from the thermal cycling matrix, and a time-resolved damage sequence is reconstructed based on this list to obtain the time series of the fatigue damage. This restores the statistical damage results and remaps them back onto the time axis, assigning each minute thermal cycling event its occurrence time coordinate and independent damage value. In this way, a time-strictly corresponding, discretized damage increment sequence can be generated, which is an indispensable data foundation for subsequent dot-multiplication operations with the time-varying fatigue acceleration profile to achieve damage correction.

[0052] More specifically, in a specific example of the present invention, the reconstruction process of the fatigue damage time series first involves, during the execution of the rainflow counting algorithm, before classifying and statistically analyzing each identified independent thermal cycle into the thermal cycle matrix, pre-reserving an original cycle list containing the physical parameters of each cycle and its occurrence timestamp; then, based on this original cycle list, the damage sequence is reconstructed. This process iterates through each original cycle entry in the original cycle list, and for any cycle in the original cycle list, extracts parameters such as its temperature difference amplitude and average temperature, and substitutes them into a preset IGBT life model to calculate the number of failure cycles corresponding to that single cycle. Then, divide 1 by the number of failed cycles. The damage increment caused by the single cycle is obtained. Finally, this damage increment is paired with the timestamp of the single cycle and stored as a data point in a new time series. After traversing all the original cycles in the original cycle list, the new time series obtained is the fatigue damage time series.

[0053] Specifically, in step S400, the event set is subjected to extreme event impact damage quantification to obtain an impact damage set. It should be understood that a severe thermal shock event can cause microscopic damage to the internal material structure of the IGBT module, reducing its ability to resist subsequent conventional thermal cycling stress, thereby accelerating the accumulation rate of fatigue damage. This is a key physical coupling effect that must be reflected in the damage model. Therefore, in the technical solution of this invention, the event set is further subjected to extreme event impact damage quantification to obtain an impact damage set, thereby quantifying the damage caused by each impact event.

[0054] Specifically, in step S500, the total fatigue damage, the time series of fatigue damage, and the impact damage set are fused to obtain the total cumulative damage. It should be understood that the final failure of an IGBT is the result of the combined and accumulated effects of conventional thermal fatigue damage and extreme event impact damage. Furthermore, there is a physical coupling between these two damage mechanisms; that is, a severe thermal shock may degrade device performance, thereby accelerating the accumulation rate of fatigue damage under subsequent conventional operating conditions. Therefore, a simple linear summation cannot accurately reflect the overall lifespan consumption. Thus, in the technical solution of this invention, the total fatigue damage, the time series of fatigue damage, and the impact damage set are further fused to obtain the total cumulative damage. This constructs a nonlinear damage superposition model that reflects the interaction between impact and fatigue. This model not only considers the independent contributions of the two types of damage but also quantifies the dynamic acceleration effect of the impact event on subsequent fatigue damage. In this way, a more accurate and physically realistic total cumulative damage value can be generated than the traditional linear superposition result, solving the problem of inaccurate lifespan prediction caused by neglecting the coupling effect between damage mechanisms in existing technologies.

[0055] Figure 5 This is a flowchart illustrating the method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions, according to an embodiment of the present invention. It describes the accumulation of fatigue damage under normal operating conditions based on conventional operating condition data to obtain total fatigue damage and a time series of fatigue damage. Figure 5 As shown, step S500 includes: S510, performing direct impact damage component aggregation on the impact damage set to obtain total impact damage; S520, generating a fatigue damage amplification factor time series based on the impact damage set; S530, performing time-varying acceleration-based corrected fatigue damage calculation on the fatigue damage time series and the total fatigue damage based on the fatigue damage amplification factor time series to obtain corrected fatigue damage; S540, fusing the corrected fatigue damage and the total impact damage to obtain the total cumulative damage.

[0056] In step S510, the impact damage set is aggregated using direct impact damage components to obtain the total impact damage. It should be understood that a wind turbine may experience one or more extreme events within an assessment period. The impact damage set generated in the previous step contains the damage components caused by each independent event. To obtain the total direct damage caused by all impact events, these dispersed damage values ​​need to be merged. Therefore, in the technical solution of this invention, the impact damage set is further aggregated using direct impact damage components to obtain the total impact damage. This linearly sums all independent impact damage values ​​in the set, thereby obtaining a scalar value that represents the sum of the direct destructive effects of all extreme events within the assessment period. This provides a clear, aggregated total impact damage term for the final total damage fusion calculation.

[0057] More specifically, in a specific example of the present invention, the direct impact damage component aggregation of the impact damage set to obtain the total impact damage includes: extracting the impact damage value corresponding to each event entry in the impact damage set; and accumulating all the extracted impact damage values ​​to obtain the total impact damage. For example, the process of direct impact damage component aggregation of the impact damage set first takes the impact damage set as input and initializes an accumulator variable for storing the total impact damage, with an initial value of zero; then, iterates through all event entries in the impact damage set, extracting the corresponding impact damage value from the current event entry in each iteration and accumulating this value into the accumulator variable; after iterating through all event entries in the set, the final value stored in the accumulator variable is the total impact damage.

[0058] In step S520, a fatigue damage amplification factor time series is generated based on the impact damage set. This is used to mathematically model the damage acceleration effect of each impact event. The model converts the severity of the impact into an initial amplification factor, and its effect gradually decays over time, thereby quantifying the dynamic impact of the impact on the fatigue accumulation rate at different subsequent times. In this way, a time-varying multiplier profile can be generated to correct the basic fatigue damage calculation, providing a core adjustment factor for accurately accounting for the impact-fatigue interaction.

[0059] More specifically, in a specific example of the present invention, the process of generating the fatigue damage amplification factor time series first defines an exponential decay function with time and impact severity as variables to describe the behavior of the amplification factor; then, a baseline amplification factor time series with all values ​​equal to 1 and a time resolution consistent with the fatigue damage time series is initialized; subsequently, each event entry in the impact damage set is traversed, and for any impact event, its timestamp and impact severity parameters are extracted. Based on the impact severity of the event, an initial amplification factor value is calculated. Then, at all time points after the timestamp of the event, the exponential decay function is applied to calculate the trajectory of the amplification factor decaying with time, and the value of the corresponding time period of the baseline amplification factor time series is updated with this trajectory; if a value greater than 1 already exists in the corresponding time period of the baseline amplification factor time series during the update, the larger of the two values ​​is taken. After traversing all impact events, the final result is the fatigue damage amplification factor time series.

[0060] In step S530, based on the fatigue damage amplification factor time series, a time-varying accelerated correction fatigue damage calculation is performed on the fatigue damage time series and the total fatigue damage to obtain the corrected fatigue damage. It should be understood that the previous steps have generated a fatigue damage time series characterizing the basic lifespan consumption rate and a time series characterizing the amplification factor time series characterizing the damage acceleration effect after the impact event. To actually apply this acceleration effect to the calculation of the basic damage, the two must be combined. Therefore, in the technical solution of this invention, a time-varying accelerated correction fatigue damage calculation is further performed on the fatigue damage time series and the total fatigue damage based on the fatigue damage amplification factor time series to obtain the corrected fatigue damage. This allows the basic fatigue damage increment to be multiplied point-by-point by the damage amplification factor at the same moment, and the corrected damage increment to be accumulated, thereby achieving a dynamic re-estimation of the fatigue damage accumulation process. In this way, a more accurate corrected total fatigue damage value that takes into account the impact fatigue interaction can be obtained, which truly reflects the accelerated impact of extreme events on the normal aging process of the device.

[0061] More specifically, in a specific example of the present invention, based on the fatigue damage amplification factor time series, a time-varying acceleration-based corrected fatigue damage calculation is performed on the fatigue damage time series and the total fatigue damage to obtain the corrected fatigue damage, including: calculating the corrected damage increment time series based on the fatigue damage amplification factor time series and the fatigue damage time series; accumulating the corrected damage increment time series to the total fatigue damage to obtain the corrected fatigue damage.

[0062] In other words, more specifically, the calculation process for correcting fatigue damage first involves calculating the time series of corrected damage increments. This process takes the time series of fatigue damage amplification factor and the time series of fatigue damage as inputs, and multiplies the corresponding values ​​of these two time series at each timestamp point by point. That is, the amplification factor value at each moment is multiplied by the basic fatigue damage increment value at that moment to obtain the corrected damage increment value at that moment. All these corrected damage increment values ​​and their corresponding timestamps together constitute a corrected damage increment time series. Next, the corrected damage increment time series is accumulated to obtain the corrected fatigue damage. This process initializes an accumulator variable to zero, and then iterates through each corrected damage increment value in the corrected damage increment time series, accumulating them sequentially into the accumulator variable. After the iteration is completed, the final value of the accumulator variable is the corrected fatigue damage.

[0063] In step S540, the corrected fatigue damage and the total impact damage are fused to obtain the total cumulative damage. It should be understood that the previous steps have calculated the corrected fatigue damage taking into account the acceleration effect and the total impact damage directly caused by the extreme event. To obtain a comprehensive reflection of the total lifetime consumption of the device under all operating conditions within this evaluation period, these two final damage components must be merged. Therefore, in the technical solution of this invention, the corrected fatigue damage and the total impact damage are further fused to obtain the total cumulative damage. This allows for the linear summation of the nonlinearly corrected fatigue damage portion and the direct impact damage portion, completing the final convergence of the dual-mode damage paths. In this way, a single quantitative indicator that most completely and accurately characterizes the true aging degree of the IGBT module within this evaluation period—the total cumulative damage—can be generated, providing a direct input for the final remaining lifetime prediction.

[0064] More specifically, in a specific example of the present invention, the process of fusing the corrected fatigue damage and the total impact damage takes the scalar value of the corrected fatigue damage and the scalar value of the total impact damage as inputs. This process performs an addition operation, that is, adds the value of the corrected fatigue damage to the value of the total impact damage. The result of this addition operation is the total cumulative damage, which is used as the final output of this step.

[0065] Specifically, in step S600, the remaining service life is predicted based on the total accumulated damage and the operating time of the wind turbine converter IGBTs to obtain the remaining service life. It should be understood that the total accumulated damage calculated in the previous steps is a normalized indicator characterizing the current degree of aging, while for wind farm operation and maintenance decisions, a remaining service life indicator with clear physical meaning and measured in time units is more valuable. Therefore, in the technical solution of this invention, the remaining service life is further predicted based on the total accumulated damage and the operating time of the wind turbine converter IGBTs to obtain the remaining service life. This allows for the inference of future aging trends based on the aging history already occurred, extrapolating the abstract damage score to a specific operable time. This generates an operable final output that directly serves predictive maintenance, providing a quantitative decision-making basis for wind turbine spare parts management, maintenance window planning, etc.

[0066] More specifically, in a specific example of the present invention, the process of predicting the remaining service life first takes the total cumulative damage value and the total operating time of the wind turbine converter IGBT module since commissioning, i.e., the operating time of the wind turbine converter IGBT, as inputs. Then, the total cumulative damage value is divided by the operating time of the wind turbine converter IGBT to calculate the average damage accumulation rate of the wind turbine converter IGBT during historical operation. Subsequently, the current total cumulative damage is subtracted from a preset damage threshold, such as 1, to obtain the remaining damage margin. Finally, the remaining damage margin is divided by the previously calculated average damage accumulation rate to obtain the predicted remaining service life. The unit of the remaining service life is consistent with the unit of the operating time.

[0067] In summary, the wind turbine converter IGBT life prediction method based on operating conditions according to embodiments of the present invention is explained. It constructs a parallel analysis path for fatigue damage under normal operating conditions and impact damage from extreme events by simultaneously acquiring low-frequency SCADA data and high-frequency fault recording data of the wind turbine. This solves the problem of existing technologies being unable to accurately quantify the massive instantaneous impact damage caused by extreme events such as low-voltage ride-through due to a single data source. Furthermore, it constructs an interaction model between impact events and subsequent fatigue damage. Based on the severity of identified impact events, it generates a time series of dynamic fatigue damage amplification coefficients that decay over time, and uses this to correct the accumulation of fatigue damage under normal operating conditions in real time. Finally, by fusing the direct impact damage with the corrected fatigue damage, it obtains the total accumulated damage that comprehensively reflects the impact of actual operating conditions. Based on the total accumulated damage and the operating time of the wind turbine converter IGBT, it predicts the remaining service life, solving the technical problem of existing technologies neglecting the impact-fatigue coupling effect, which leads to severely inaccurate prediction results, and improving the accuracy of life prediction.

[0068] The wind turbine converter IGBT life prediction method based on operating conditions according to embodiments of the present invention can be implemented in various industrial control and computing devices, such as local controllers for wind turbines, edge computing nodes in wind farms, central operation and maintenance servers, or industrial IoT cloud platforms. In one possible implementation, the wind turbine converter IGBT life prediction method based on operating conditions according to embodiments of the present invention can be integrated as a software module or hardware module into the intelligent operation and maintenance or predictive health management system of a wind farm. For example, the wind turbine converter IGBT lifetime prediction method based on operating conditions can be an independent condition assessment application running on a central server, a lifetime prediction function plug-in module of a wind farm SCADA system or equipment health management platform, or an application interface deployed on a cloud platform that provides health status assessment and maintenance decision suggestions to each distributed wind turbine through the network. Of course, the core transient thermal shock analysis and damage fusion calculation process in this wind turbine converter IGBT lifetime prediction method based on operating conditions can also be embedded in hardware such as converter control unit, application-specific integrated circuit or field-programmable gate array, as one of the real-time condition monitoring or fault early warning hardware modules of the wind farm intelligent operation and maintenance system.

[0069] Those skilled in the art will understand that the above embodiments are specific implementations of the present invention, and in practical applications, various changes can be made in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions, characterized in that, The wind turbine converter IGBT life prediction method based on operating conditions includes: Acquire SCADA data and fault recording data of wind turbine units; Data synchronization and event identification are performed on SCADA data and fault recording data to obtain routine operating condition data and event sets; Based on conventional working condition data, fatigue damage accumulation under conventional working conditions is performed to obtain total fatigue damage and fatigue damage time series. Extreme event impact damage quantification is performed on the event set to obtain the impact damage set; The total fatigue damage, the time series of fatigue damage, and the impact damage set are fused to obtain the total cumulative damage. The remaining useful life is predicted based on the total cumulative damage and the operating time of the wind turbine converter IGBT.

2. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 1, characterized in that, Data synchronization and event identification are performed on SCADA data and fault recording data to obtain routine operating condition data and event sets, including: Each fault recording data block in the fault recording data is input into a rule-based event recognizer to obtain an event set and a list of event time intervals; The SCADA data is bifurcated based on the event time interval list to obtain the normal operating condition data.

3. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 2, characterized in that, The SCADA data is bifurcated based on a list of event time intervals to obtain the routine operating condition data, including: Extract each event time interval from the list of event time intervals; Add a safety buffer margin to each event time interval to obtain an isolation window set; Iterate through each data point in the SCADA data and determine whether each data point falls within any isolation window in the isolation window set. If not, add the corresponding data point to the normal operating condition data.

4. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 1, characterized in that, Based on conventional working condition data, fatigue damage accumulation under conventional working conditions is performed to obtain total fatigue damage and fatigue damage time series, including: Power loss time series are obtained by searching for power loss at each data point in the normal operating condition data based on the power loss lookup table. Extracting the radiator temperature time series from routine operating condition data; The junction temperature profile is estimated based on the lumped thermal network model to obtain the IGBT time series by performing power loss time series and radiator temperature time series. Cumulative damage is calculated based on IGBT time series to obtain the total fatigue damage and the time series of the fatigue damage.

5. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 4, characterized in that, Cumulative damage calculation based on IGBT time series is performed to obtain the total fatigue damage and the time series of the fatigue damage, including: The thermal cycling matrix is ​​obtained by extracting the thermal cycling matrix from the IGBT time series based on the rainflow counting method. The cumulative damage is calculated based on the life model and Miner's rule on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage.

6. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 5, characterized in that, The cumulative damage is calculated based on the life model and Miner's rule on the thermal cycling matrix to obtain the total fatigue damage and the time series of the fatigue damage, including: The failure lifetime vector is derived from the thermal cycling matrix based on the lifetime model. Extract the cycle count vector from the thermal cycle matrix; The failure lifetime vector and cycle count vector are quantized into a damage contribution vector based on Miner's rule to obtain the loss contribution vector. The fatigue damage is aggregated and calculated from the loss contribution vector to obtain the total fatigue damage; Extract the original list of cycles with timestamps from the hot cycle matrix; The time-resolved damage sequence is reconstructed based on the original circular list with timestamps to obtain the time series of the fatigue damage.

7. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 1, characterized in that, The total cumulative damage is obtained by fusing the total fatigue damage, the time series of fatigue damage, and the impact damage set, including: The total impact damage is obtained by directly aggregating the impact damage components of the impact damage set. Based on the impact damage set, a time series of fatigue damage amplification factor is generated; Based on the fatigue damage amplification factor time series, the fatigue damage time series and total fatigue damage are calculated using time-varying acceleration correction to obtain the corrected fatigue damage. The corrected fatigue damage and the total impact damage are combined to obtain the total cumulative damage.

8. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 7, characterized in that, The total impact damage is obtained by directly aggregating the impact damage components of the impact damage set, including: Extract the impact damage value corresponding to each event entry in the impact damage set; The total impact damage is obtained by summing up all the extracted impact damage values.

9. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 7, characterized in that, Based on the time series of fatigue damage amplification factor, time-varying acceleration-based corrected fatigue damage calculations are performed on the time series of fatigue damage and the total fatigue damage to obtain the corrected fatigue damage, including: Based on the time series of fatigue damage amplification factor and fatigue damage, the time series of corrected damage increment is calculated. The corrected damage increment time series is accumulated into the total fatigue damage to obtain the corrected fatigue damage.

10. The method for predicting the lifespan of IGBTs in wind turbine converters based on operating conditions according to claim 1, characterized in that, The remaining service life is predicted based on the total accumulated damage and the operating time of the wind turbine converter IGBTs, including: Divide the total accumulated damage by the operating time of the wind turbine converter IGBT to calculate the average damage accumulation rate of the wind turbine converter IGBT during the historical operation period. Subtract the current total accumulated damage from the preset damage threshold to obtain the remaining damage margin. Divide the remaining damage margin by the average damage accumulation rate to obtain the predicted remaining service life.

Images