A wind power converter IGBT monitoring system

Abstract

The invention discloses a wind power converter IGBT monitoring system, which specifically relates to the field of wind power converter IGBT module monitoring, and is used to solve the problem of inaccurate fault prediction and early warning in long-term operation, and is achieved through cycle division, parameter screening, index evaluation, dynamic adjustment and multi-stage early warning collaboration. Cycle division is used to identify the aging characteristics of IGBT modules in different life cycles, and screen out monitoring parameters that are highly correlated with the state of the bonding wire. Through a multi-dimensional dynamic fusion model, the uniqueness of the monitoring parameters is evaluated to ensure the accuracy of fault prediction. The dynamic adjustment module dynamically optimizes the weights of the monitoring parameters according to cycle changes and environmental variables, thereby enhancing the adaptability and real-time performance of the monitoring. The multi-stage early warning mechanism generates early warning signals of different levels according to the changes in key parameters within the cycle, and warns of potential faults in advance. The reliability of the operation of the wind power converter is improved, the risk of sudden failures is reduced, and the safe and stable operation of the system is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of wind power converter IGBT module monitoring, and more specifically, to a wind power converter IGBT monitoring system. Background Art

[0002] During the long-term operation of wind power converters, IGBT modules, as one of the core components, bear the dual pressure of high-frequency current and complex loads. The bonding wire structure inside the module is very prone to irreversible fatigue damage under drastic temperature changes, frequent current switching and long-term mechanical vibration. This damage will directly lead to a gradual increase in bonding resistance, accompanied by an increase in thermal resistance, which will eventually affect the overall performance of the IGBT module and may even cause serious system failures. Traditional IGBT monitoring methods mainly rely on temperature monitoring, but due to the strong short-term temperature fluctuations, it is impossible to accurately reflect the local aging or micro-damage inside the module, and it is difficult to provide effective fault warnings. In addition, a single temperature parameter is difficult to capture complex fault causes, which makes the monitoring system of the existing technology incapable of dealing with diverse aging modes.

[0003] The main shortcoming of the existing technology is that it cannot accurately distinguish the aging characteristics of the IGBT module at different operating stages, and it relies too much on a single monitoring indicator and ignores the multi-stage operating characteristics of the module. As the most vulnerable part of the IGBT module, the damage process of the bonding wire is affected by the external environment and operating conditions, and shows obvious differences in different cycles. However, the traditional monitoring system fails to effectively distinguish these stages, resulting in a single monitoring indicator and the inability to dynamically adapt to changes in operating conditions. In addition, the existing fault warning system lacks flexibility in parameter evaluation and adjustment mechanisms, and cannot dynamically optimize and adjust monitoring parameters according to real-time status, resulting in low accuracy of fault warnings and difficulty in effectively preventing sudden failures in advance.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an IGBT monitoring system for a wind power converter, which accurately identifies the aging characteristics of the IGBT module in different life cycles through the collaborative work of cycle division, parameter screening, index evaluation, dynamic adjustment and multi-stage early warning. Through cycle division, the monitoring focus in different stages is determined, and the monitoring parameters highly related to the bonding wire state are screened out to ensure that the monitoring indicators can reflect the actual operating status of the equipment. The index evaluation module evaluates the uniqueness of the monitoring parameters based on the multi-dimensional dynamic fusion model to ensure the accuracy of fault prediction. The dynamic adjustment module automatically optimizes the weights of the monitoring parameters according to the cycle changes and environmental variables, so that the monitoring process has a high degree of adaptability and real-time performance. The multi-stage early warning mechanism combines the changes in key parameters within the cycle to generate early warning signals of different levels, and warns of the potential failure risks of the IGBT module in advance. The invention significantly improves the operating reliability of the wind power converter, reduces the risk of downtime caused by sudden failures, and ensures the safe and stable operation of the wind power system to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A wind power converter IGBT monitoring system, comprising: a cycle identification module, a parameter screening module, an index evaluation module, a dynamic adjustment module and an early warning judgment module;

[0008] Cycle identification module: Use multi-dimensional parameter clustering method to divide the operating state of IGBT into multiple cycles, calculate the cycle characteristic function based on key variables, determine the cycle boundary and dynamically divide the state characteristics; output the cycle characteristic data set to the parameter screening module;

[0009] Parameter screening module: Perform sensitivity analysis on the parameters in each cycle, use correlation analysis to calculate the correlation strength between each parameter and the bond wire state, screen out the parameters related to the cycle state change, generate a monitoring parameter set; output the monitoring parameter set to the indicator evaluation module;

[0010] Index evaluation module: Use a multi-factor fusion model to evaluate the uniqueness of monitoring indicators in each cycle to ensure that a single monitoring indicator has the ability to predict faults within the cycle; if the evaluation fails, re-combine sensitive parameters to construct a monitoring indicator set; output the evaluated monitoring indicator set to the dynamic adjustment module;

[0011] Dynamic adjustment module: monitors the operating status in real time, adjusts the weight of the monitoring parameter set according to periodic changes and environmental variables, and uses an adaptive adjustment algorithm to optimize parameter adaptation in different periods; outputs the adjusted monitoring parameters and weights to the early warning judgment module;

[0012] Early warning judgment module: Build a multi-stage early warning model, trigger early warnings based on monitoring results within a period, set early warning levels and record historical data, and generate corresponding early warning signals based on periodic monitoring priorities.

[0013] In a preferred embodiment, the operation process of the cycle identification module includes the following contents:

[0014] A1, obtain data during the operation of IGBT in real time through sensors, and perform data cleaning and preprocessing.

[0015] A2, based on the preprocessed data, calculates the time series characteristics of each key variable, including the temperature fluctuation amplitude, load fluctuation rate and the growth rate of the operating time; through wavelet transform and time-frequency analysis, extracts the characteristic information of each variable at different frequencies to form a feature matrix of the time-frequency dual dimension.

[0016] A3, using the time-frequency dual-dimensional feature matrix to construct a periodic feature function.

[0017] A4, using multi-dimensional parameter clustering method, cluster analysis is performed on the output results of the periodic characteristic function; using clustering algorithm, the IGBT operation data under different states are divided into multiple clusters, each cluster corresponding to an operation cycle.

[0018] A5, through the cluster analysis results, determine the boundaries of each operation cycle; based on the center points and boundary points of different clusters, the application results of the boundary fuzzy set function generate a periodic state feature set containing precise boundaries and fuzzy transition zones.

[0019] A6, outputs the state features of each cycle to form a cycle feature data set.

[0020] In a preferred embodiment, the operation process of the parameter screening module includes the following contents:

[0021] B1, conduct sensitivity analysis on the parameters in each cycle; adopt the global sensitivity analysis method to evaluate the influence of each parameter on the change of cycle state by calculating the contribution of each parameter; use the Sobol index to calculate the global sensitivity value of each parameter and determine the key parameters.

[0022] B2, perform correlation analysis on the selected key parameters; use the Pearson correlation coefficient to calculate the correlation strength between each parameter and the bonding line state; mark the parameters with correlation strength higher than the corresponding threshold as highly sensitive parameters, otherwise mark them as low-sensitive parameters.

[0023] B3. Generate a dynamic monitoring parameter set based on the results of sensitivity analysis and correlation analysis; screen the parameters in each cycle and include highly sensitive parameters in the monitoring parameter set.

[0024] B4, outputs the filtered dynamic parameter set.

[0025] In a preferred embodiment, the operation process of the indicator evaluation module includes the following contents:

[0026] C1, from the dynamic monitoring parameter set output by the parameter screening module, first evaluate the uniqueness of the single monitoring indicator in each cycle; use the multi-factor fusion model to analyze each monitoring indicator and calculate its predictive ability for periodic state changes; the evaluation indicators include the response sensitivity index S final , Fault trend discrimination rate T f , parameter steady-state persistence coefficient P s , combined with the output of the periodic characteristic function, a multi-factor fusion model is constructed to output a uniqueness score; if the uniqueness score is greater than or equal to the corresponding threshold, it indicates that the corresponding single monitoring indicator has the ability to independently predict faults, and the corresponding indicator passes the evaluation.

[0027] C2, if the uniqueness score of a single monitoring indicator is less than the corresponding threshold, it means that it has failed the uniqueness assessment, and it will be reorganized; low-sensitivity indicators in the same period are combined with other high-correlation parameters to construct multidimensional monitoring indicators.

[0028] C3, model optimization is performed for single monitoring indicators and combined multi-dimensional monitoring indicators that have passed the uniqueness evaluation.

[0029] C4, after completing the evaluation and optimization of single and multi-dimensional monitoring indicators, generates the final set of monitoring indicators.

[0030] In a preferred embodiment, C1.1, the processing of the response sensitivity index:

[0031] C1.1.1, first, obtain the monitoring parameters in each cycle from the monitoring parameter set output by the parameter screening module, and use its change data in each cycle as input for preprocessing.

[0032] C1.1.2, for each parameter in each cycle, calculate the sensitivity of the monitoring parameter to the change of the bond wire state; let the monitoring parameter be P(t) and the bond wire state be B(t). Within a certain cycle T, the sensitivity change rate ΔS represents the response rate of the monitoring parameter to the change of the bond wire state, and the formula is:

[0033] C1.1.3, to capture the overall response sensitivity within the cycle, calculate the cumulative response deviation R acc , the formula is: in, is the acceleration of the bond wire state, which indicates the rate at which the bond wire state changes.

[0034] C1.1.4, Calculate the time-varying response sensitivity index S r (t), the formula is:

[0035] C1.1.5, combining the cumulative response deviation and time-varying response sensitivity within the cycle, the response sensitivity index S within the cycle is obtained final , the formula is:

[0036] In a preferred embodiment, C1.2, the processing process of the fault tendency discrimination rate:

[0037] C1.2.1, obtain the monitoring parameters P(t) and bonding wire status data B(t) within the cycle from the parameter screening module; at the same time, collect the operating load L(t) and current fluctuation I(t) within the cycle as supplementary parameters to form a complete data set.

[0038] C1.2.2, analyze the bonding wire status data in each cycle to find the critical point where the bonding wire causes failure; the failure trigger point t f is the moment when the bond wire state change rate reaches a certain threshold, and the formula is: in, is the rate of change of the bond wire state, and ∈1 is the preset fault judgment threshold.

[0039] C1.2.3, Fault trend discrimination rate requires analysis of the correlation between monitoring parameters and fault trigger points within the cycle; first calculate the fault change rate F r (t), the formula is:

[0040] C1.2.4, in order to capture the accelerated change of fault trend, calculate the fault trend acceleration A f (t), the formula is:

[0041] C1.2.5, based on the fault change rate and fault trend acceleration, calculate the fault trend discrimination rate T f , the formula is: Among them, t0 is the start time of the cycle, t f is the fault triggering moment.

[0042] In a preferred embodiment, C1.3, the processing process of the parameter steady-state persistence coefficient is:

[0043] C1.3.1, obtain the monitoring parameters P(t) and period characteristic function C(t) within the cycle from the parameter screening module, and extract the stable interval of the monitoring parameters within each cycle in combination with the period division results provided by the period identification module to ensure that the monitoring data presents a relatively stable change trend within the cycle.

[0044] C1.3.2, in order to capture the steady-state interval within the cycle, calculate the rate of change of the monitored parameter; the steady-state interval [t s ,t e The identification formula of ] is: Among them, t s and t e are the start and end time of the steady-state interval, respectively; δ is the threshold of the steady-state change rate, which is used to determine whether the monitored parameter is in a stable state.

[0045] C1.3.3, for the identified steady-state interval, calculate the duration T within the steady-state interval s , the formula is: T s =t e -t s .

[0046] C1.3.4, define steady-state deviation as the deviation of the monitoring parameter from the periodic characteristic function, deviation D s The calculation formula is:

[0047] C1.3.5, in order to quantify the parameter stability in the steady-state interval, calculate the steady-state energy density E s , the formula is: in, is the rate of change of the periodic characteristic function.

[0048] C1.3.6, combine the steady-state duration, steady-state deviation and steady-state energy density to calculate the parameter steady-state persistence coefficient P s , the formula is:

[0049] In a preferred embodiment, the operation of the dynamic adjustment module includes the following:

[0050] D1, obtains the operation status data from the cycle identification module and the environmental monitoring system in real time, combines the operation status data with the monitoring indicator set output by the indicator evaluation module, and builds a real-time monitoring data stream to ensure the synchronous tracking of the operation status and environmental changes.

[0051] D2, based on real-time monitoring data, sets the triggering conditions for adaptive adjustment; when certain key variables in the monitoring data, such as temperature fluctuation amplitude or load change rate, exceed the preset threshold, the adaptive adjustment mechanism is triggered; the triggering condition is set as: |P(t)-P ref (t)|>∈2; where P(t) is the current monitoring parameter value, P ref (t) is the historical reference value, and ∈2 is the set adjustment threshold.

[0052] D3. For the monitoring parameters that trigger adjustment, perform parameter adaptability evaluation; compare the historical cycle data with the current cycle status, and calculate the adaptability of the monitoring parameters in different cycles. The formula is: Among them, A i (t) is the adaptability score of parameter i in the current cycle, P i (t) is the monitoring parameter, C(t) is the cycle characteristic function, T now and T past are the current and historical cycle intervals respectively. D4. According to the parameter adaptability score, dynamically adjust the weight of the monitoring parameter. The weight adjustment formula is: Among them, ω i (t) is the weight of parameter i at time t, ω i (t - 1) is the weight at the previous moment.

[0053] In a preferred embodiment, the operation of the early warning judgment module includes the following:

[0054] E1. Establish a multi-stage early warning model, and set multiple early warning thresholds according to the monitoring results in different cycles; the model combines the change rate of the cycle characteristic function with the response sensitivity index S of the monitoring parameter final and divides it into three early warning levels: normal, abnormal early warning, and emergency early warning; the calculation formula for the early warning level is: Among them, E(t) is the early warning signal intensity, and α1 and β1 are the dynamically adjusted weight coefficients.

[0055] E2. Determine the corresponding early warning level according to the magnitude of the early warning signal intensity; the early warning level is distinguished by the signal intensity threshold, which is set to the following three states: normal: E(t) ≤ θ1; abnormal early warning: θ1 < E(t) ≤ θ2; emergency early warning: E(t) > θ2; where θ1 and θ2 are the preset multi-stage early warning thresholds.

[0056] E3. When the early warning signal is triggered, record the corresponding historical monitoring data.

[0057] E4. Generate corresponding early warning signals according to the current cycle monitoring focus.

[0058] The technical effects and advantages of the IGBT monitoring system for a wind power converter of the present invention:

[0059] 1. Through the combination of cycle identification module and parameter screening module, the accuracy and flexibility of wind power converter IGBT monitoring system can be effectively improved, and the problems of unclear cycle division and inaccurate parameter screening in the prior art are solved. The cycle identification module accurately identifies the cycle changes in the operation process of IGBT through multi-dimensional parameter clustering method and boundary fuzzy set function, avoids the mutation phenomenon between cycles, and realizes the smooth transition of cycle boundaries, ensuring the dynamic adaptability of cycle division. Based on sensitivity analysis and correlation analysis, the parameter screening module accurately selects monitoring parameters with strong correlation for different cycle states, ensuring that the selected parameters can fully reflect the changes in the bond wire state, avoiding the problem of parameter redundancy or invalidity in traditional monitoring systems. The introduction of dynamic weights enables low-sensitivity parameters to still have a certain predictive value within a specific cycle, ensuring the comprehensiveness and robustness of monitoring. Through the close combination of cycle identification and parameter screening, the dynamic optimization and adjustment of cycle characteristics and monitoring parameters are realized, the accuracy and real-time performance of fault prediction are improved, and the bond wire state can be accurately monitored in different operation stages, preventing the occurrence of potential faults, and ensuring the long-term stable operation of the wind power converter system.

[0060] 2. Through the organic combination of the index evaluation module, the dynamic adjustment module and the early warning judgment module, the system realizes accurate monitoring, flexible adjustment and efficient early warning under complex periodic changes. The index evaluation module comprehensively evaluates the uniqueness of the monitoring indicators through a multi-dimensional dynamic fusion model, ensuring that the monitoring parameters in each cycle can effectively characterize the fault characteristics, thereby improving the accuracy and reliability of fault prediction. The dynamic adjustment module dynamically optimizes the weights of the monitoring parameters based on real-time monitoring data through an adaptive adjustment mechanism and parameter adaptability evaluation, so that the monitoring system can quickly respond to changes in the cycle and environment and ensure the continuity of monitoring accuracy. The early warning judgment module uses a multi-stage early warning model, combined with the response sensitivity of the periodic characteristic function and key parameters, to accurately judge the abnormal situation within the cycle and issue corresponding early warning signals in time, thereby effectively preventing the occurrence of faults. The system as a whole realizes efficient coordination of monitoring, adjustment and early warning, so that the IGBT monitoring of wind power converters has the characteristics of high sensitivity, strong adaptability and high reliability, and can provide stable fault early warning capabilities in a dynamic environment to ensure the long-term safe operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The present invention is a schematic structural diagram of a wind power converter IGBT monitoring system.

[0062] Figure 2 The present invention is a schematic diagram of the steps of a cycle identification module of a wind power converter IGBT monitoring system. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Example 1

[0065] Figure 1 The invention provides a wind power converter IGBT monitoring system, comprising: a cycle identification module, a parameter screening module, an index evaluation module, a dynamic adjustment module and an early warning judgment module.

[0066] Cycle identification module: Use the multi-dimensional parameter clustering method to divide the operating state of the IGBT into multiple cycles, calculate the cycle characteristic function based on key variables, determine the cycle boundary and dynamically divide the state characteristics; output the cycle characteristic data set to the parameter screening module.

[0067] Parameter screening module: Perform sensitivity analysis on the parameters within each cycle, use correlation analysis to calculate the correlation strength between each parameter and the bond wire state, screen out parameters related to cycle state changes, generate a monitoring parameter set; output the monitoring parameter set to the indicator evaluation module.

[0068] Index evaluation module: Use a multi-factor fusion model to evaluate the uniqueness of monitoring indicators within each cycle to ensure that a single monitoring indicator has the ability to predict faults within the cycle; if the evaluation fails, re-combine sensitive parameters to construct a monitoring indicator set; output the evaluated monitoring indicator set to the dynamic adjustment module.

[0069] Dynamic adjustment module: monitors the operating status in real time, adjusts the weights of the monitoring parameter set according to periodic changes and environmental variables, and uses an adaptive adjustment algorithm to optimize parameter adaptation in different periods; outputs the adjusted monitoring parameters and weights to the early warning judgment module.

[0070] Early warning judgment module: Build a multi-stage early warning model, trigger early warnings based on monitoring results within a period, set early warning levels and record historical data, and generate corresponding early warning signals based on periodic monitoring priorities.

[0071] During the long-term operation of wind power converters, IGBT modules, as one of the core components, face harsh working environments and complex current conversion tasks. The bonding wire structure inside the IGBT module is very prone to fatigue aging under the action of high-frequency current, high-power load and temperature changes. As the bonding wire gradually ages, the on-resistance will gradually increase, and the thermal resistance will also increase, which will affect the working performance of the entire IGBT module and even cause converter system failure. Traditional IGBT monitoring technology mainly relies on temperature monitoring, but due to the large short-term fluctuations in temperature and the inability to accurately reflect the subtle damage of internal devices, it is difficult to provide effective early warning. Therefore, it is necessary to introduce more sensitive monitoring parameters that can truly reflect the health status of the device.

[0072] As a relatively fragile and irreversible part of the IGBT module, the bonding wire has relatively clear characteristic changes in its damage process. Especially during the aging process, the change in bond resistance is highly correlated with its degree of damage. This makes the bonding wire state an important parameter to characterize the health of the IGBT module, and it can warn of module failure in advance by monitoring its aging state. However, the damage of the bonding wire does not progress linearly, but is affected by a variety of external factors and operating conditions. Its state changes show significant differences in different operating cycles. In order to accurately capture these changes, the most representative monitoring indicators must be selected for different cycles to meet the monitoring needs of different periods.

[0073] The introduction of cycle division is mainly based on the multi-stage characteristics of IGBT modules in long-term operation. The operating state of IGBT is affected by factors such as ambient temperature, load changes, and operating time, resulting in different operating characteristics at different stages of its life cycle. In some stages, temperature and current fluctuations have a more significant impact on the bonding wires, while in other stages, stress and vibration may be the main aging drivers. Therefore, through cycle division, the specific aging trends and damage mechanisms of IGBT modules at each stage can be reflected in more detail. The purpose of cycle identification is to effectively distinguish these stages, ensure the use of adaptive monitoring indicators in different cycles, avoid over-reliance on a single indicator, and ignore key fault warning signals.

[0074] By screening parameters after cycle division, the monitoring parameters with the strongest correlation with the bond wire status can be effectively identified. In some cycles, bond resistance may be the core monitoring indicator, but in other cycles, other temperature-sensitive or stress-sensitive parameters may need to be introduced to enhance the accuracy of fault prediction. Such a multi-stage, multi-parameter monitoring mechanism can dynamically adapt to changes in the operating environment of the IGBT module, ensuring that key aging signals can be effectively captured in each cycle.

[0075] Indicator evaluation further ensures the effectiveness of monitoring indicators. In each cycle, by performing a weighted evaluation on the selected parameters, it can be determined whether a single indicator has sufficient predictive power or needs to be combined with other parameters to form a multi-dimensional monitoring indicator set. This step can avoid misjudgments due to the limitations of a single indicator and improve the accuracy and adaptability of the early warning system. The dynamic adjustment module ensures that the monitoring parameters are automatically adapted as the cycle changes, so that the entire monitoring system can adaptively track the operating status of the IGBT module and dynamically adjust according to real-time environmental variables to ensure the continuity and accuracy of the monitoring process.

[0076] The final early warning judgment generates a multi-stage early warning signal by integrating the monitoring data in each cycle and combining it with the dynamically adjusted parameter set. This early warning mechanism can not only give an early warning before the IGBT module is about to fail, but also set different levels of early warning signals according to the focus of periodic monitoring, providing operation and maintenance personnel with a more accurate basis for fault prediction. This multi-level monitoring and early warning system makes the operation of the IGBT module of the wind power converter more reliable, reduces the risk of system downtime caused by sudden failures, and improves the operating efficiency and safety of the entire wind power system.

[0077] like Figure 2 As shown, the operation process of the cycle identification module includes the following:

[0078] A1, real-time data of IGBT operation is obtained through sensors, including key variables such as temperature, load, current, and operation time, and data cleaning and preprocessing are performed. The sliding window method is used to smooth instantaneous fluctuations, eliminate abnormal data, and normalize each variable to ensure the consistency of scales between different variables, which is convenient for subsequent cluster analysis.

[0079] A2, based on the preprocessed data, calculates the time series characteristics of each key variable, including the temperature fluctuation amplitude, load fluctuation rate and the growth rate of running time. Through wavelet transform and time-frequency analysis, the characteristic information of each variable at different frequencies is extracted to form a time-frequency dual-dimensional feature matrix as the basic data for cycle identification.

[0080] The time-frequency dimension information in the feature matrix is ​​obtained through wavelet transform. The frequency domain data is used to capture the high-frequency fluctuation characteristics in a short period of time, while the time domain data provides trend changes in a long period of time. This dual-dimensional feature enables the periodic characteristic function to accurately characterize the operating state changes of the IGBT in different time periods and frequency ranges. Through the dynamically adjusted periodic characteristic function, the operating cycle characteristics of different stages can be accurately identified, ensuring that the cycle identification module can still maintain high-precision cycle division capabilities under complex working conditions.

[0081] A3, using the time-frequency dual-dimensional feature matrix, constructs the periodic characteristic function. The characteristic values ​​of variables such as temperature and load are used as input, and the multi-dimensional features are combined through weighting coefficients to form the periodic characteristic function formula. The weight of each variable is dynamically adjusted according to its influence on the IGBT state to ensure that the periodic characteristic function can accurately characterize the changes in the operating state at different stages.

[0082] The time-frequency dual-dimensional feature matrix obtained through preprocessing and feature extraction selects the characteristic values ​​of variables such as temperature, load, current fluctuation, and operating time as the input data of the periodic characteristic function. In each cycle, the amplitude of temperature fluctuation, the rate of change of load, and the instantaneous fluctuation frequency of current are input as important characteristic parameters. The basic model of the periodic characteristic function is constructed by using the multivariate linear regression method combined with the multidimensional data of the feature matrix.

[0083] To ensure that the periodic characteristic function can dynamically adapt to different operating states, the weight of each variable is adjusted through a weighting coefficient. The calculation of the weighting coefficient is based on the different impacts of each characteristic parameter on its state during the operation of the IGBT. Through historical data analysis, the correlation between the IGBT failure rate and each variable is used to determine the initial weight coefficient of each variable. Subsequently, based on the real-time operating state, an adaptive optimization algorithm is used to dynamically adjust the weight coefficient so that the parameters with greater impact on the IGBT state in each cycle receive higher weights to improve the accuracy of the periodic characteristic function.

[0084] A4, using multi-dimensional parameter clustering method, cluster analysis is performed on the output results of the periodic characteristic function. Using appropriate clustering algorithms (such as K-means or hierarchical clustering), the IGBT operating data under different states are divided into multiple clusters, each cluster corresponding to an operating cycle. The accuracy of the period division is ensured by evaluating the compactness within the cluster and the separation between clusters.

[0085] A5, through the cluster analysis results, determine the boundaries of each operating cycle. According to the center points and boundary points of different clusters, use the boundary fuzzy set function to optimize the transition area of ​​the cycle boundary to ensure smooth transition between cycles without obvious faults. Dynamically correct the cycle boundary to adapt to the real-time changes in the operating status.

[0086] The specific process of using boundary fuzzy set function to optimize the transition area of ​​periodic boundary is as follows:

[0087] In the cycle identification module, the cycle boundary obtained by multidimensional parameter clustering method may be mutated, especially between different cycles, the state change does not occur immediately, and there is usually a certain transition area. In order to optimize these transition areas, the boundary fuzzy set function is used to further process the cycle boundary, making the cycle division smoother and more accurate.

[0088] First, the initial boundary points of each cycle after cluster analysis, including the cycle start point and end point, are extracted to form the initial cycle boundary set. By analyzing the continuous change of the periodic characteristic function, the state change trend in each cycle boundary area is identified, and the initial interval of the boundary fuzzy area is generated. The setting of this interval is based on the fluctuation amplitude and frequency of the characteristic parameters near the cycle boundary, and the time-frequency dual-dimensional information in the characteristic matrix is ​​used to evaluate the fuzziness of the cycle boundary.

[0089] Next, the bounding fuzzy set function is applied to process these fuzzy intervals. The fuzzy set function takes the following form: Among them, x represents the value of the characteristic parameter in the boundary area, and θ is the central value of the boundary fuzzy set, that is, the interval where the value of the periodic characteristic function changes most dramatically. Through this function, the characteristic parameter value of the fuzzy boundary area is gradually transitioned and smoothed, so that in the periodic boundary area, the change of the periodic characteristic function no longer presents abrupt changes, but presents a smooth transition curve.

[0090] The parameters of the fuzzy set function are dynamically adjusted based on actual operating data to ensure that the transition area between different cycles is optimized according to the real-time working conditions. For characteristic parameters in the transition area, such as temperature and load, the weighted average method is used to further correct their fluctuations within the fuzzy boundary to reduce unnecessary transition interval lengths. In this way, the cycle division is not only more accurate, but also avoids misjudgment and delay during cycle switching, and improves the overall response speed and accuracy of the system.

[0091] Finally, the application result of the boundary fuzzy set function generates a periodic state feature set containing precise boundaries and fuzzy transition zones. This feature set is used to output the final result of the period identification module, ensuring smooth and accurate period division while guaranteeing operational stability under complex working conditions.

[0092] A6 outputs the state characteristics of each cycle, including the characteristic values ​​of key variables in each cycle (characteristic parameters such as temperature, load, current fluctuation and running time), cycle length and boundary point position, and historical data of bond wire status, including key parameters such as bond resistance and bond wire detachment rate, to form a cycle characteristic data set. The cycle characteristic data set is passed to the parameter screening module as the cycle division result and serves as the basic data for subsequent monitoring and analysis.

[0093] The operation process of the parameter screening module includes the following:

[0094] From the cycle characteristic data set output by the cycle identification module, characteristic parameters such as temperature, load, current fluctuation and running time in each cycle are obtained. At the same time, historical data of the bonding wire status is obtained, including key parameters such as bonding resistance and bonding wire shedding rate, forming a complete cycle parameter data set to provide a basis for subsequent analysis.

[0095] B1, perform sensitivity analysis on the parameters in each cycle. Use the global sensitivity analysis method to evaluate the degree of influence of each parameter on the cycle state change by calculating the contribution of each parameter. The core of sensitivity analysis is to clarify which parameters play a dominant role in bonding wire aging and cycle state change by separating the independent effect and interaction effect of each parameter. Use the Sobol index or variance decomposition method to calculate the global sensitivity value of each parameter and determine the key parameters that have a greater impact on the cycle state.

[0096] B2, performing correlation analysis on the selected key parameters, using the Pearson correlation coefficient or the Spearman rank correlation coefficient to calculate the correlation strength between each parameter and the bonding wire state (such as bonding resistance, bonding wire fall-off rate).

[0097] The parameters with correlation strength higher than the corresponding threshold are marked as high-sensitivity parameters, and those with correlation strength lower than the threshold are marked as low-sensitivity parameters.

[0098] B3, based on the results of sensitivity analysis and correlation analysis, generate a dynamic monitoring parameter set. Screen the parameters in each cycle and include highly sensitive parameters in the monitoring parameter set. For low-sensitivity parameters, use a weighted algorithm to assign dynamic weights and keep them in the parameter set to ensure the comprehensiveness of the prediction.

[0099] For low-sensitivity parameters, the specific weight is calculated based on the correlation coefficient of the correlation analysis, and the formula is: Among them, w o is the weight of the oth low-sensitivity parameter, r o is the correlation strength between the corresponding parameter and the bond wire state. This formula ensures that the weight of the low-sensitivity parameter in a specific period matches its correlation. This method not only retains the predictive value of the low-sensitivity parameter under specific conditions, but also avoids excessive interference of the low-sensitivity parameter on the overall monitoring results.

[0100] B4, outputs the screened dynamic parameter set as the basic data for subsequent evaluation and prediction.

[0101] The combination of the cycle identification module and the parameter screening module can effectively improve the accuracy and flexibility of the wind power converter IGBT monitoring system, and solve the problems of unclear cycle division and inaccurate parameter screening in the prior art. The cycle identification module accurately identifies the cycle changes during the operation of the IGBT through multi-dimensional parameter clustering method and boundary fuzzy set function, avoiding the mutation phenomenon between cycles, and realizing the smooth transition of the cycle boundary to ensure the dynamic adaptability of the cycle division. Based on sensitivity analysis and correlation analysis, the parameter screening module accurately selects monitoring parameters with strong correlation for different cycle states, ensuring that the selected parameters can fully reflect the changes in the bond wire state, avoiding the problem of parameter redundancy or invalidity in the traditional monitoring system. The introduction of dynamic weights enables low-sensitivity parameters to still have a certain predictive value within a specific cycle, ensuring the comprehensiveness and robustness of monitoring. Through the close combination of cycle identification and parameter screening, the dynamic optimization and adjustment of cycle characteristics and monitoring parameters are realized, the accuracy and real-time performance of fault prediction are improved, and the bond wire state can be accurately monitored at different operation stages to prevent the occurrence of potential faults and ensure the long-term stable operation of the wind power converter system.

[0102] The operation process of the indicator evaluation module includes the following:

[0103] C1, from the dynamic monitoring parameter set output by the parameter screening module, first evaluate the uniqueness of the single monitoring indicator in each cycle. Use the multi-factor fusion model to analyze each monitoring indicator and calculate its predictive ability for periodic state changes. The evaluation indicators include the response sensitivity index S final , Fault trend discrimination rate T f , parameter steady-state persistence coefficient P s , combined with the output of the periodic characteristic function, for example, by the formula: Among them, U i is the uniqueness score of a single monitoring indicator i, α is the exponential coefficient of the dynamic adjustment of the cycle, and the flexibility of the evaluation model is adjusted according to the characteristics of the cycle change to improve the adaptability to different cycle states. If the uniqueness score reaches the preset threshold, it indicates that the single monitoring indicator has the ability to independently predict faults, and the indicator passes the evaluation.

[0104] If the uniqueness score is greater than or equal to the corresponding threshold, it indicates that the single monitoring indicator has the ability to independently predict faults, and the indicator passes the evaluation.

[0105] C2, if the uniqueness score of a single monitoring indicator is less than the corresponding threshold, it means that it has failed the uniqueness assessment, and it is reorganized. Combine low-sensitivity indicators with other high-correlation parameters in the same period to construct multidimensional monitoring indicators. Through multivariate regression analysis, calculate the synergy between sensitive parameters to form a multidimensional monitoring indicator. For example, through the formula: Among them, M is a multidimensional monitoring indicator, ω v is the sensitive parameter involved in the combination, mg v The weights are determined through sensitivity analysis and correlation analysis. This combination ensures that the monitoring indicators have comprehensive fault prediction capabilities within the cycle, making up for the prediction deviation caused by the insufficiency of a single indicator.

[0106] C3, further optimizes the model for single monitoring indicators and combined multi-dimensional monitoring indicators that have passed the uniqueness evaluation. According to the output of the periodic characteristic function of each period, test the dynamic adaptability of the indicator, adjust its weight distribution in different periods, and ensure that the monitoring indicator can maintain its predictive ability during the periodic change process. The robustness and generalization ability of the optimized monitoring indicator set are verified by cross-validation.

[0107] C4, after completing the evaluation and optimization of single and multi-dimensional monitoring indicators, generates the final monitoring indicator set. This indicator set has the ability to predict faults within a cycle and can adapt to cycle changes. The evaluated monitoring indicator set is output to the dynamic adjustment module as the final result to provide data support for the subsequent adaptive adjustment of the monitoring system.

[0108] C1.1, Processing of response sensitivity index:

[0109] C1.1.1, first, obtain the monitoring parameters in each cycle from the monitoring parameter set output by the parameter screening module, such as temperature fluctuation amplitude, load change rate, current fluctuation, etc. The change data of these parameters in each cycle are used as input for preprocessing to ensure the smoothness and consistency of the data and eliminate noise and abnormal points.

[0110] C1.1.2, for each parameter in each cycle, calculate the sensitivity of the monitoring parameter to the change of the bond wire state. Let the monitoring parameter be P(t), the bond wire state be B(t), and within a certain cycle T, the sensitivity change rate ΔS represents the response rate of the monitoring parameter to the change of the bond wire state, and the formula is: By calculating the change speed of the monitoring parameter relative to the bond wire state at time t, the instant response of the monitoring parameter to the change of the bond wire state is reflected.

[0111] C1.1.3, in order to capture the overall response sensitivity within the cycle, further calculate the cumulative response deviation R acc , which reflects the overall responsiveness of the monitoring parameters to the change of the bonding wire state during the cycle. The formula is: in, is the acceleration of the bond wire state, indicating the rate of change of the bond wire state. This formula can quantify the sensitive response of the monitoring parameter to the bond wire state within the cycle by accumulating the response deviation of the monitoring parameter within each cycle.

[0112] C1.1.4, in order to ensure that the response sensitivity can be adaptively adjusted as the periodic state changes, calculate the time-varying response sensitivity index S r (t), the formula is: By combining the instant response rate of the monitoring parameters and the change rate of the bond wire state The response sensitivity is dynamically combined with the periodic state changes to achieve time-varying adjustment of the sensitivity and ensure the real-time adaptation of the monitoring indicators to the state changes within the cycle.

[0113] C1.1.5, combining the cumulative response deviation and time-varying response sensitivity within the cycle, the response sensitivity index S within the cycle is obtained final , the formula is: By combining the time-varying sensitivity and the cumulative response deviation, the response sensitivity of the monitoring parameter within the cycle is comprehensively evaluated. This index can quantify the sensitivity of the monitoring parameter to the change of the bond wire state in each cycle and serve as an important parameter for evaluating the uniqueness of the monitoring indicator.

[0114] C1.2, Processing of fault trend discrimination rate:

[0115] C1.2.1, obtain the monitoring parameters P(t) and bonding wire status data B(t) within the cycle from the parameter screening module. At the same time, collect the operating load L(t) and current fluctuation I(t) within the cycle as supplementary parameters to form a complete data set. Eliminate noise and abnormal data through data preprocessing to ensure the continuity and accuracy of monitoring data.

[0116] C1.2.2, analyze the bonding wire status data in each cycle to find the critical point where the bonding wire may cause failure. Failure trigger point t f is the moment when the bond wire state change rate reaches a certain threshold, and the formula is: in, is the rate of change of the bond wire state, and ∈1 is the preset fault judgment threshold. The goal of this step is to quickly locate the trigger point of potential faults within the cycle and provide a time reference for the subsequent discrimination rate calculation.

[0117] C1.2.3, Fault trend discrimination rate requires analysis of the correlation between monitoring parameters and fault trigger points within a cycle. First, calculate the fault change rate F r (t), which reflects the change speed of the monitoring parameter relative to the fault trigger point, and the formula is: This formula calculates the change trend of the monitoring parameter when it is close to the fault trigger point, reflecting the change slope of the monitoring parameter as it approaches the fault over time. The fault change rate is used to quantify the fluctuation degree of the monitoring parameter before and after the fault within the cycle and evaluate its response speed to the fault occurrence.

[0118] C1.2.4, in order to capture the accelerated change of fault trend, further calculate the fault trend acceleration A f (t). This acceleration reflects the acceleration of the change of the monitoring parameter relative to the fault trigger point, and the formula is: Through this formula, the change acceleration of the monitoring parameters is calculated to ensure that the fault trend change not only includes the change rate before and after the fault, but also captures its acceleration trend. The fault trend acceleration can effectively reveal the abnormal acceleration of the monitoring parameters within the cycle and reflect the proximity of the critical point of system failure.

[0119] C1.2.5, based on the fault change rate and fault trend acceleration, comprehensively calculate the fault trend discrimination rate T f The discrimination rate quantifies the overall correlation strength between the monitoring parameters and the fault trigger point within the period, and the formula is: Among them, t0 is the start time of the cycle, t f is the fault triggering moment. This formula integrates the fault change rate and the fault trend acceleration, accumulates the fault trend change process of the monitoring parameters in the entire cycle, and determines the fault indication ability of the parameters in the cycle.

[0120] The fault trend discrimination rate reflects the fault prediction ability of the monitoring parameters within the cycle. The larger the value, the more sensitive the monitoring parameters are to faults. The final fault prediction result is generated by calculating the fault trend discrimination rate, and it is used as a key parameter in the uniqueness evaluation of the monitoring index, providing data support for subsequent index optimization and dynamic adjustment.

[0121] C1.3, Processing process of parameter steady-state persistence coefficient:

[0122] C1.3.1, obtain the monitoring parameters P(t) and period characteristic function C(t) within the cycle from the parameter screening module, and extract the stable interval of the monitoring parameters within each cycle in combination with the period division results provided by the period identification module to ensure that the monitoring data presents a relatively stable change trend within the cycle. Preprocess the data to ensure that it is smooth and noise-free for the accuracy of subsequent calculations.

[0123] C1.3.2, in order to capture the steady-state interval within the cycle, calculate the rate of change of the monitoring parameter. Define the steady-state interval as the time period when the rate of change of the monitoring parameter is lower than the preset threshold. s ,t eThe identification formula of ] is: Among them, t s and t e are the start and end time of the steady-state interval, respectively; δ is the threshold of the steady-state change rate, which is used to determine whether the monitored parameter is in a stable state.

[0124] C1.3.3, for the identified steady-state interval, calculate the duration T within the steady-state interval s , the formula is: T s =t e -t s ; This formula simply and clearly expresses the duration of the steady-state interval and is one of the foundations of the steady-state persistence coefficient. The length of the steady-state interval within the cycle reflects the stability of the monitored parameters.

[0125] C1.3.4, within the steady-state interval, the deviation of the monitoring parameters can affect the assessment of steady-state persistence. Steady-state deviation is defined as the degree of deviation of the monitoring parameters relative to the periodic characteristic function. The deviation D s The calculation formula is: By integrating the absolute deviation of the monitoring parameter from the periodic characteristic function in the steady-state interval, the fluctuation range of the parameter in the steady-state interval is reflected. The smaller the deviation, the higher the stability of the monitoring parameter in the steady-state interval.

[0126] C1.3.5, in order to further quantify the parameter stability in the steady-state interval, calculate the steady-state energy density E s , the formula is: in, is the rate of change of the periodic characteristic function, which indicates the impact of the change of the periodic state on the monitoring parameter. This formula evaluates the response intensity of the parameter in the steady-state interval by calculating the energy density in the steady-state interval, and measures the steady-state performance of the parameter in the cycle by combining the steady-state duration.

[0127] C1.3.6, combine the steady-state duration, steady-state deviation and steady-state energy density to calculate the parameter steady-state persistence coefficient P s , the formula is: Combined with the steady-state duration and deviation, the length of time in the steady-state interval and the degree of parameter deviation from the periodic characteristics are weighed, and then combined with the energy density to reflect the stability of the parameter. The parameter steady-state persistence coefficient can effectively quantify the steady-state performance of the monitoring parameter within the cycle, and as a key factor in the uniqueness evaluation of the monitoring indicator, ensure that the monitoring parameters within the cycle have high stability and persistence.

[0128] The operation of the dynamic adjustment module includes the following:

[0129] D1, obtains real-time operation status data from the cycle identification module and environmental monitoring system, including temperature, load, current fluctuations within the cycle, and external environmental variables such as wind speed, humidity, etc. These data are combined with the monitoring indicator set output by the indicator evaluation module to build a real-time monitoring data stream to ensure the synchronous tracking of operation status and environmental changes.

[0130] D2, based on real-time monitoring data, sets the triggering conditions for adaptive adjustment. When certain key variables in the monitoring data, such as temperature fluctuation or load change rate, exceed the preset threshold, the adaptive adjustment mechanism is triggered. The triggering condition is set as: |P(t)-P ref (t)|>∈2; where P(t) is the current monitoring parameter value, P ref (t) is the historical reference value, and ∈2 is the set adjustment threshold. This mechanism ensures that the monitoring parameter set can be adjusted in time when the system operation fluctuates.

[0131] D3, for the monitoring parameters that trigger the adjustment, the parameter adaptability is evaluated. The historical cycle data is compared with the current cycle status to calculate the adaptability of the monitoring parameters in different cycles. The formula is: Among them, A i (t) is the adaptability score of parameter i in the current cycle, P i (t) is the monitoring parameter, C(t) is the periodic characteristic function, T now and T past are the current and historical period intervals respectively. This formula evaluates the adaptability of the parameters in the current period by integrating the difference between the current and historical periods.

[0132] D4, dynamically adjust the weight of the monitoring parameter according to the parameter adaptability score. If a parameter has a high adaptability in the current cycle, the weight will increase; if the adaptability is low, the weight will decrease. The weight adjustment formula is: Among them, ω i (t) is the weight of parameter i at time t, ω i (t-1) is the weight of the previous moment, and the adaptability score A i (t) The weight adjustment is normalized to ensure that the total weight remains stable.

[0133] D5, optimizes the monitoring parameter set after weight adjustment to ensure that it adapts to the changes in the current cycle and environmental variables. The optimized monitoring parameter set completes the re-matching of cycle characteristics through the adaptive adjustment mechanism, and is output to the early warning judgment module as the final result, providing accurate data support for subsequent fault prediction and early warning.

[0134] The operation of the early warning judgment module includes the following:

[0135] Receive monitoring parameters and weight data from the dynamic adjustment module, and analyze monitoring results according to the current cycle characteristics provided by the cycle identification module. Classify and process monitoring parameters according to different characteristics within the cycle, distinguish key parameters from secondary parameters, focus on monitoring key parameters that are highly correlated with cycle changes, and determine their impact on early warning judgment based on parameter weights.

[0136] E1, establish a multi-stage early warning model, and set multiple early warning thresholds according to the monitoring results in different periods. The model takes the rate of change of the periodic characteristic function as The response sensitivity index S of the monitoring parameter final Combined with the above, it is divided into three warning levels: normal, abnormal warning, and emergency warning. The calculation formula for the warning level is: Among them, E(t) is the warning signal strength, α1 and β1 are dynamically adjusted weight coefficients, which are adjusted according to the monitoring focus of different periods. If the warning signal strength exceeds the preset threshold, the corresponding warning signal is triggered.

[0137] E2, according to the strength of the warning signal, determine the corresponding warning level. The warning level is distinguished by the signal strength threshold and is set to the following three states:

[0138] Normal: E(t)≤θ1.

[0139] Abnormal warning: θ1 <E(t)≤θ2。

[0140] Emergency warning: E(t)>θ2.

[0141] Among them, θ1 and θ2 are preset multi-stage warning thresholds, which are dynamically adjusted according to historical monitoring data and current cycle status. By determining the warning level, cycle abnormal states of different severity can be distinguished.

[0142] E3, when the warning signal is triggered, record the corresponding historical monitoring data, including the time when the warning is triggered, monitoring parameters, cycle characteristics and environmental variables. Compare these data with the previous historical cycle data, analyze the change trend of the warning signal, and provide a basis for the subsequent cycle status monitoring and warning model optimization. The recording and analysis of historical data helps to monitor the cycle trend in the long term and identify potential long-term fault trends.

[0143] E4, generates corresponding early warning signals based on the current cycle monitoring focus. For normal status, a green signal is output, and no further action is required; for abnormal early warning, a yellow signal is output, indicating that the system may have potential risks and records them; for emergency early warning, a red signal is output, requiring immediate emergency measures, and the signal is transmitted to the upper control system to issue an alarm and emergency response instructions.

[0144] Through the organic combination of the index evaluation module, the dynamic adjustment module and the early warning judgment module, the system realizes accurate monitoring, flexible adjustment and efficient early warning under complex periodic changes. The index evaluation module comprehensively evaluates the uniqueness of the monitoring indicators through a multi-dimensional dynamic fusion model, ensuring that the monitoring parameters in each cycle can effectively characterize the fault characteristics, thereby improving the accuracy and reliability of fault prediction. The dynamic adjustment module dynamically optimizes the weights of the monitoring parameters based on real-time monitoring data through an adaptive adjustment mechanism and parameter adaptability evaluation, so that the monitoring system can quickly respond to changes in the cycle and environment and ensure the continuity of monitoring accuracy. The early warning judgment module uses a multi-stage early warning model, combined with the response sensitivity of the periodic characteristic function and key parameters, to accurately judge the abnormal situation within the cycle and issue corresponding early warning signals in time, thereby effectively preventing the occurrence of faults. The system as a whole realizes efficient coordination of monitoring, adjustment and early warning, so that the IGBT monitoring of wind power converters has the characteristics of high sensitivity, strong adaptability and high reliability, and can provide stable fault early warning capabilities in a dynamic environment to ensure the long-term safe operation of the system.

[0145] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0146] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0147] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0148] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A wind power converter IGBT monitoring system, characterized in that: include: Cycle identification module, parameter screening module, indicator evaluation module, dynamic adjustment module and early warning judgment module; Cycle identification module: Use multi-dimensional parameter clustering method to divide the operating state of IGBT into multiple cycles, calculate the cycle characteristic function based on key variables, determine the cycle boundary and dynamically divide the state characteristics; Output the periodic feature data set to the parameter screening module; Parameter screening module: Perform sensitivity analysis on the parameters in each cycle, use correlation analysis to calculate the correlation strength between each parameter and the bond wire state, screen out the parameters related to the cycle state change, generate a monitoring parameter set; output the monitoring parameter set to the indicator evaluation module; Index evaluation module: Use a multi-factor fusion model to evaluate the uniqueness of monitoring indicators in each cycle to ensure that a single monitoring indicator has the ability to predict faults within the cycle; if the evaluation fails, re-combine sensitive parameters to construct a monitoring indicator set; output the evaluated monitoring indicator set to the dynamic adjustment module; Dynamic adjustment module: monitors the operating status in real time, adjusts the weight of the monitoring parameter set according to periodic changes and environmental variables, and uses adaptive adjustment algorithms to optimize parameter adaptation in different periods; Output the adjusted monitoring parameters and weights to the early warning judgment module; Early warning judgment module: build a multi-stage early warning model, trigger early warnings based on the monitoring results within the cycle, set early warning levels and record historical data, and generate corresponding early warning signals based on the key points of periodic monitoring; The operation process of the indicator evaluation module includes the following: C1, from the dynamic monitoring parameter set output by the parameter screening module, first evaluate the uniqueness of the single monitoring indicator in each cycle; use the multi-factor fusion model to analyze each monitoring indicator and calculate its predictive ability for periodic state changes; the evaluation indicators include the response sensitivity index , Fault trend discrimination rate , parameter steady-state persistence coefficient , combined with the output of the periodic characteristic function, a multi-factor fusion model is constructed to output a uniqueness score; if the uniqueness score is greater than or equal to the corresponding threshold, it indicates that the corresponding single monitoring indicator has the ability to independently predict faults, and the corresponding indicator passes the evaluation; C2, if the uniqueness score of a single monitoring indicator is less than the corresponding threshold, it means that it has failed the uniqueness assessment, and it will be reorganized; low-sensitivity indicators in the same period are combined with other high-correlation parameters to construct multidimensional monitoring indicators; C3, model optimization for single monitoring indicators and combined multi-dimensional monitoring indicators that have passed the uniqueness evaluation; C4, after completing the evaluation and optimization of single and multi-dimensional monitoring indicators, the final set of monitoring indicators is generated; C1.1, Processing of response sensitivity index: C1.1.1, first, obtain the monitoring parameters in each cycle from the monitoring parameter set output by the parameter screening module, and use its change data in each cycle as input for preprocessing; C1.1.2, for each parameter in each cycle, calculate the sensitivity of the monitoring parameter to the change in the bond wire state; let the monitoring parameter be , the bonding wire state is , in a certain cycle Sensitivity change rate It indicates the response rate of the monitoring parameter to the change of the bonding wire state. The formula is: ; C1.1.3, to capture the overall response sensitivity within the cycle, calculate the cumulative response deviation , the formula is: ;in, is the acceleration of the bond wire state, indicating the rate of change of the bond wire state; C1.1.4, Calculation of Time-Varying Response Sensitivity Index , the formula is: ; C1.1.5, combining the cumulative response deviation and time-varying response sensitivity within the cycle, the response sensitivity index within the cycle is obtained , the formula is: ; C1.2, Processing of fault trend discrimination rate: C1.2.1, obtain the monitoring parameters within the cycle from the parameter screening module and bond wire status data ; At the same time, collect the operating load during the cycle and current fluctuations As supplementary parameters, a complete data set is formed; C1.2.2, analyze the bonding wire status data in each cycle to find the critical point where the bonding wire causes failure; the failure trigger point is the moment when the bond wire state change rate reaches a certain threshold, and the formula is: ;in, is the rate of change of the bond wire state, is the preset fault judgment threshold; C1.2.3, Fault trend discrimination rate requires analysis of the correlation between monitoring parameters and fault trigger points within a cycle; first calculate the fault change rate , the formula is: ; C1.2.4, in order to capture the accelerated change of fault trend, calculate the fault trend acceleration , the formula is: ; C1.2.5, Calculate the fault trend discrimination rate based on the fault change rate and fault trend acceleration , the formula is: ;in, is the cycle start time, is the fault triggering moment; C1.3, Processing process of parameter steady-state persistence coefficient: C1.3.1, obtain the monitoring parameters within the cycle from the parameter screening module and periodic characteristic functions , and combined with the period division results provided by the period identification module, the stable interval of the monitoring parameters in each period is extracted to ensure that the monitoring data presents a relatively stable change trend within the period; C1.3.2, in order to capture the steady-state interval within the cycle, calculate the rate of change of the monitored parameter; steady-state interval The identification formula is: ;in, and are the start and end time of the steady-state interval, is the threshold of the steady-state change rate, which is used to determine whether the monitoring parameter is in a stable state; C1.3.

3. For the identified steady-state interval, calculate the duration of the steady-state interval. , the formula is: ; C1.3.4, define steady-state deviation as the deviation of the monitoring parameter from the periodic characteristic function. The calculation formula is: ; C1.3.5, in order to quantify the parameter stability in the steady-state interval, calculate the steady-state energy density , the formula is: ;in, is the rate of change of the periodic characteristic function; C1.3.6, combine steady-state duration, steady-state deviation and steady-state energy density to calculate the parameter steady-state persistence coefficient , the formula is: .

2. A wind power converter IGBT monitoring system according to claim 1, characterized in that: The operation process of the cycle identification module includes the following: A1, obtain data during IGBT operation in real time through sensors, and perform data cleaning and preprocessing; A2, based on the preprocessed data, calculate the time series characteristics of each key variable, including the temperature fluctuation amplitude, load fluctuation rate and the growth rate of the running time; through wavelet transform and time-frequency analysis, extract the characteristic information of each variable at different frequencies to form a characteristic matrix of time-frequency dual dimensions; A3, using the time-frequency dual-dimensional feature matrix to construct a periodic feature function; A4, using multi-dimensional parameter clustering method, cluster analysis is performed on the output results of the periodic characteristic function; Using a clustering algorithm, the IGBT operating data in different states is divided into multiple clusters, each cluster corresponds to an operating cycle; A5, through the cluster analysis results, determine the boundaries of each operation cycle; based on the center points and boundary points of different clusters, the application results of the boundary fuzzy set function generate a periodic state feature set containing precise boundaries and fuzzy transition zones; A6, outputs the state features of each cycle to form a cycle feature data set.

3. A wind power converter IGBT monitoring system according to claim 2, characterized in that: The operation process of the parameter screening module includes the following: B1, conduct sensitivity analysis on the parameters in each cycle; adopt the global sensitivity analysis method to evaluate the influence of each parameter on the change of cycle state by calculating the contribution of each parameter; Use the Sobol index to calculate the global sensitivity value of each parameter and determine the key parameters; B2, conduct correlation analysis on the selected key parameters; use the Pearson correlation coefficient to calculate the correlation strength between each parameter and the bond line state; The parameters with correlation strength higher than the corresponding threshold are marked as high-sensitivity parameters, and those with correlation strength lower than the threshold are marked as low-sensitivity parameters; B3, based on the results of sensitivity analysis and correlation analysis, generate a dynamic monitoring parameter set; screen the parameters in each cycle and include highly sensitive parameters in the monitoring parameter set; B4, outputs the filtered dynamic parameter set.

4. The wind power converter IGBT monitoring system according to claim 1, characterized in that: The operation of the dynamic adjustment module includes the following: D1, obtains the operation status data from the cycle identification module and the environmental monitoring system in real time, combines the operation status data with the monitoring indicator set output by the indicator evaluation module, and builds a real-time monitoring data stream to ensure the synchronous tracking of the operation status and environmental changes; D2, based on real-time monitoring data, sets the trigger conditions for adaptive adjustment; when certain key variables in the monitoring data, such as temperature fluctuation amplitude or load change rate, exceed the preset threshold, the adaptive adjustment mechanism is triggered; the trigger conditions are set as: ;in, is the current monitoring parameter value, is a historical reference value. is the set adjustment threshold; D3, for the monitoring parameters that trigger adjustment, parameter adaptability evaluation is performed; historical cycle data is compared with the current cycle status to calculate the adaptability of the monitoring parameters in different cycles. The formula is: ;in, For parameters The suitability score in the current cycle, To monitor the parameters, is a periodic characteristic function, and They are the current and historical cycle intervals respectively; D4, dynamically adjust the weight of the monitoring parameters according to the parameter adaptability score. The weight adjustment formula is: ;in, For parameters In time The weight of is the weight at the previous moment.

5. A wind power converter IGBT monitoring system according to claim 4, characterized in that: The operation of the early warning judgment module includes the following: E1, establish a multi-stage early warning model, set multiple early warning thresholds according to the monitoring results in different periods; the model takes the rate of change of the periodic characteristic function as Response sensitivity index to monitoring parameters Combined, it is divided into three warning levels: normal, abnormal warning, and emergency warning; The calculation formula for the warning level is: ;in, is the warning signal strength, and is the weight coefficient after dynamic adjustment; E2, according to the strength of the warning signal, determines the corresponding warning level; the warning level is distinguished by the signal strength threshold and is set to the following three states: normal: ; Abnormal warning: ; Emergency warning: ; in, and It is a preset multi-stage warning threshold; E3, when the early warning signal is triggered, the corresponding historical monitoring data is recorded; E4, generates corresponding early warning signals based on the current cycle monitoring focus.

Images