High-power MOS transistor monitoring method and system and storage medium

Abstract

The invention relates to the technical field of MOS transistor monitoring, and discloses a high-power MOS transistor monitoring method and system and a storage medium, and the method comprises the steps: S1, collecting key parameter data in a working process of a high-power MOS transistor; s2, performing gradient calculation on all parameter data in the key parameter data to obtain a parameter gradient feature set; s3, performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set to obtain abnormal behavior data, constructing a behavior analysis rule, and inputting the behavior data into the behavior analysis rule to obtain a risk behavior mode of the abnormal behavior data; s4, performing behavior score calculation based on the risk behavior mode to obtain a behavior score, and judging whether the behavior score is smaller than a set state intervention threshold value or not; by calculating the behavior score of each segment and comparing the behavior score with the threshold value, the behavior abnormity of the MOS transistor can be detected in the early stage, so that measures can be taken in time to avoid potential problems.

Description

Technical Field

[0001] The present invention relates to the technical field of MOS tube monitoring, and in particular to a high-power MOS tube monitoring method, system and storage medium. Background Art

[0002] High-power MOS tubes are widely used in power converters, motor drives, inverters and other fields. As key switching devices, the performance of MOS tubes directly affects the efficiency, reliability and safety of the entire power electronic system. Since high-power MOS tubes often work in high voltage and high current environments and are easily affected by load mutations, it is particularly important to monitor their working status in real time. Timely detection of abnormal behavior of MOS tubes can prevent faults, reduce maintenance costs and improve system reliability.

[0003] At present, when a high-power MOS tube experiences a sudden load change, it will be subjected to an instantaneous high current shock, causing key parameters inside the MOS tube to change. The abnormal parameter change will affect the normal operation of the MOS tube, and further affect the normal operation of other electronic components. If its behavior pattern is not monitored, the stability of the MOS tube will be reduced when an abnormality occurs. Therefore, a high-power MOS tube monitoring method, system and storage medium are proposed herein. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art that abnormal parameter changes when experiencing a momentary high current impact will affect the normal operation of the MOS tube, and to propose a high-power MOS tube monitoring method, system and storage medium.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A high-power MOS tube monitoring method, the MOS tube monitoring steps are:

[0007] S1: Collect key parameter data of high-power MOS tube during operation;

[0008] S2: Perform gradient calculation on all parameter data in the key parameter data to obtain a parameter gradient feature set;

[0009] S3: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing a behavior analysis rule, and obtaining a risk behavior pattern of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rule;

[0010] S4: Based on the risk behavior pattern, a behavior score is calculated to obtain a behavior score, and it is determined whether the behavior score is less than a set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

[0011] The step of acquiring the parameter gradient feature set comprises:

[0012] Extraction of each gradient data feature in key parameter data;

[0013] Construct a parameter gradient feature set based on each gradient data feature;

[0014] The gradient data features include current data gradient features, temperature data gradient features, voltage data gradient features, on-resistance data gradient features and gate drive voltage data gradient features;

[0015] The steps for extracting each gradient data feature in the key parameter data include:

[0016] Calculate the gradient data of the current changing with time according to a preset gradient calculation method to obtain the gradient characteristics of the current data;

[0017] Calculate the gradient data of temperature changing with time and obtain the gradient characteristics of temperature data;

[0018] Calculate the gradient data of voltage changing with time to obtain the gradient characteristics of voltage data;

[0019] Calculate the gradient data of the on-resistance changing with time to obtain the gradient characteristics of the on-resistance data;

[0020] Calculating the gradient data of the gate drive voltage data changing with time to obtain the gradient characteristics of the gate drive voltage data;

[0021] The preset gradient calculation method includes the steps of:

[0022] Get the data series of all key parameter data changing with time, where two consecutive time points are t i and t i+1 , calculate the two adjacent time points t i and t i+1 The current data difference is divided by the time difference to obtain the current data gradient feature, and the current data gradient is calculated at two adjacent time points t i and t i+1 The temperature data difference is divided by the time difference to obtain the temperature data gradient feature, and the temperature gradient between two adjacent time points t is calculated. i and t i+1 The voltage data difference is divided by the time difference to obtain the voltage data gradient feature, and the voltage data gradient is calculated at two adjacent time points ti and t i+1 The on-resistance data difference is divided by the time difference to obtain the on-resistance data gradient feature, and the on-resistance data gradient is calculated at two adjacent time points t i and t i+1 The gate driving voltage data gradient characteristic is obtained by dividing the gate driving voltage data difference by the time difference.

[0023] The step of acquiring abnormal behavior data comprises:

[0024] Calculate the Pearson correlation coefficient of two random parameter gradient features in the parameter gradient feature set. The Pearson correlation coefficient between the two parameter gradient features is Among them, x and y are any two different parameter gradient features in the parameter gradient feature set, is the observed mean of x, is the observed average value of y, calculate the Pearson correlation coefficient of all two random different parameter gradient features, and obtain a Pearson correlation coefficient set. Select the largest Pearson correlation coefficient from the Pearson correlation coefficient set. The two parameter gradient features corresponding to the largest Pearson correlation coefficient are parameter gradient feature a and parameter gradient feature b;

[0025] The parameter gradient feature a and the parameter gradient feature b are expressed as non-sinusoidal signal forms, and the non-sinusoidal signal form expression method includes the steps of:

[0026] Based on the formula: Where A is the DC component, n is the harmonic order, when n=1 it is the fundamental wave, when n>1 it is the higher harmonic, T is the periodic representation, t is the time variable, representing a certain moment on the time axis, and f(t) represents the non-sinusoidal signal value at time t;

[0027] The step of obtaining the DC component A comprises:

[0028] The cosine coefficient a n The steps to obtain include: Among them, a n It represents the coefficient of the nth cosine term;

[0029] The sine coefficient b n The steps to obtain include: Among them, b n It represents the coefficient of the nth sine term;

[0030] Among them, the periodicity of parameter gradient feature a and parameter gradient feature b is expressed as T a and T b ;

[0031] Use T aand T b Substitute T and T in the replacement formula a After replacing T, the non-sinusoidal signal form f(t a ), T b After replacing T, the non-sinusoidal signal form f(t b ), where f(t a ) represents the non-sinusoidal signal of parameter gradient feature a at time t, f(t b ) represents the non-sinusoidal signal of parameter gradient feature b at time t;

[0032] For non-sinusoidal signal f(t a ) and f(t b ) is weighted to obtain a periodic signal, and the new periodic signal is S, S = w a f(t a )+w b f(t b ), where w a represents the weight corresponding to the parameter gradient feature a, w b Represents the weight corresponding to the parameter gradient feature b;

[0033] The harmonic content rate of the periodic signal S is calculated, and the total harmonic content rate is T, then Among them, a1 is the cosine function of the fundamental wave, b1 is the sine function of the fundamental wave, It represents the sum of squares of all high-order harmonics. The total harmonic content rate is T to represent the MOS tube status information. The normal range of the total harmonic content rate T is from M to N. When it exceeds this range, the timestamp data of the parameter data corresponding to the abnormal total harmonic content rate T and the total harmonic content rate T are recorded to obtain abnormal behavior data.

[0034] The steps of obtaining the risk behavior pattern include:

[0035] Use a data structure to store the timestamp data and total harmonic content rate T of the abnormal behavior data, create a list of abnormal behavior data points, including the timestamp data and total harmonic content rate T of the abnormal data points, obtain time series data based on the abnormal behavior data point list, divide the time series data into W segments, calculate the average total harmonic content rate T for each segment, and obtain the segmented sequence. The segmented sequence is η={T ，1 , T ，2 ,…,T ，W}, T ，W It represents the total harmonic content T value of the Wth segment. The slope of the segmented sequence is m. Where i represents the index, and 1 <i<W, represents the average value of all segment indices, T ，1 Indicates the total harmonic content T value of the i-th segment, Represents the average value of the total harmonic distortion rate T for all segments;

[0036] Based on the slope m, an abnormal behavior threshold θ is preset. When ∣m∣>θ and m>0, the time series data shows a positive growth trend over time, indicating a risk. When ∣m∣>θ and m<0, the time series data shows a negative decreasing trend over time, indicating a risk. When the absolute value of the slope m is less than the threshold θ or the slope is close to zero, the change trend of the time series data is stable, that is, there is no risk. Obtain the time series data within the ranges of ∣m∣>θ and m>0, and ∣m∣>θ and m<0 to obtain the risk behavior pattern.

[0037] The steps for obtaining the number of segments W include:

[0038] The time series data X = [δ1, δ2, …, δ n obtained through the abnormal behavior data point list, where n is the number of data points δ. Each segment in the segmented time series data is represented as S, then: where S i represents the i-th segment, represents the error between the i-th data point within each segment and the average value of its segment, δ i represents the i-th data point in segment S, represents the average value of all data points in segment S;

[0039] Use a two-dimensional array C to store the total error after segmentation. The size of array C is N×(W + 1), where N is the total number of data points and W is the maximum number of segments. Initialize the total error of the entire data set as The preset number of segments is k, 1 < k < W. The splitting point for each preset number of segments k is Then where represents the error of the segment from the j-th data point to the i-th data point, is for all splitting points among the values, and obtain The array corresponding to the minimum value of is represented as C[N][W], where W is the optimal number of segments, and C[N][W] contains the minimum total error when splitting the entire data set using W segments.

[0040] The steps for obtaining the behavior score are as follows:

[0041] Use each segment in the risk behavior pattern to form a state space

[0042] The space state includes the monitoring action of the MOS tube, and the setting includes 3 actions, which are expressed as A={a1, a2, a3}, where a1 represents monitoring the MOS tube, a2 represents adjusting the working parameters of the MOS tube, and a3 represents issuing an alarm;

[0043] Based on state space and space state A, the transition probability is Indicates that in state S i Take action a i After transfer to state The probability of , where i is the index;

[0044] Among them, the status The error is state The maximum error is The minimum error is

[0045] Based on reward function in, For state S i The error, when near When the reward is close to 0, the state S i In normal state, when near When the reward is close to -1, the state state S i is an abnormal state;

[0046] Using the state value function V(S i ) and the action-value function Q(S i ,a i ), V(S i ) indicates that in state S i The expected return of taking the next action, Q(S i ,a i ) indicates that in state S i Take action a i The expected return of all states is initialized to 0, V(S i )=0, for each state S i , calculated in state S i Take all actions a i The action value function Q(S i ,a i ), Among them, γ is the discount factor, and its value range is between [0, 1]. Status The expected return under update state S i The value function V(Si ) for all actions a i The maximum Q(S i ,a i ),but in, Indicates that in state S i All possible actions a i The maximum value of the action value function in ;

[0047] A convergence threshold ∈ is preset, and the convergence stops when the value function change of all states is less than ∈, which can be expressed as: For each state S i , choose to make Q(S i , a i ) H As the optimal strategy, denoted as π(S i ),based on Calculate the action value function Q(S) under the optimal action i , a H ), Q(S i , a H ) represents the behavior score, expressed as F, and sets the abnormal behavior threshold F Q , when F>F Q When the risk behavior pattern of the corresponding segment is abnormal, the time series data corresponding to the abnormal risk behavior pattern is retrieved to intervene in the MOS management status.

[0048] The monitoring system comprises:

[0049] Status information acquisition module: used to collect key parameter data during the working process of high-power MOS tube;

[0050] Feature data acquisition module: used to perform gradient calculation on all parameter data in the key parameter data to obtain parameter gradient feature set;

[0051] Risk analysis module: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing behavior analysis rules, and obtaining risk behavior patterns of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rules;

[0052] Abnormal behavior decision module: Based on the risk behavior pattern, the behavior score is calculated to obtain the behavior score, and it is determined whether the behavior score is less than the set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

[0053] The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-power MOS tube monitoring method according to any one of claims 1 to 7 is implemented.

[0054] The present invention has the following beneficial effects:

[0055] 1. In the present invention, by performing feature selection based on feature pairs corresponding to the maximum Pearson correlation coefficient, the number of features can be reduced and the multicollinearity problem can be avoided. At the same time, by reducing the number of features based on the maximum Pearson correlation coefficient, the complexity of subsequent decisions can be reduced and the decision-making speed can be improved. At the same time, by acquiring abnormal behavior data to generate a risk behavior pattern, a quantitative indicator is provided to measure the difference between the behavior of the MOS tube and the expectation, which is helpful to evaluate the degree of abnormality. The behavior score calculated through the risk behavior pattern can guide engineers to take corresponding intervention measures for abnormal behavior in a specific state, such as adjusting the gate voltage or changing the working environment. By automatically calculating the behavior score and comparing it with the threshold, the behavior of the MOS tube can be monitored in real time, reducing the need for manual intervention. By calculating the behavior score of each segment and comparing it with the threshold, the behavior abnormality of the MOS tube can be detected at an early stage, so that measures can be taken in time to avoid potential problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A method step diagram of a high-power MOS tube monitoring method, system and storage medium proposed by the present invention;

[0057] Figure 2 This is a system block diagram of a high-power MOS tube monitoring method, system and storage medium proposed by the present invention. DETAILED DESCRIPTION

[0058] 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.

[0059] Embodiment 1

[0060] like Figure 1 As shown, the present invention proposes a high-power MOS tube monitoring method, and the MOS tube monitoring steps are:

[0061] S1: Collect key parameter data of high-power MOS tube during operation;

[0062] S2: Perform gradient calculation on all parameter data in the key parameter data to obtain a parameter gradient feature set;

[0063] S3: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing a behavior analysis rule, and obtaining a risk behavior pattern of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rule;

[0064] S4: Based on the risk behavior pattern, a behavior score is calculated to obtain a behavior score, and it is determined whether the behavior score is less than a set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

[0065] Key parameter data include current data, temperature data, voltage data, on-resistance data and gate drive voltage. Among them, current data is used to monitor the instantaneous current of the MOS tube, especially the peak current; temperature data is used to monitor the operating temperature of the MOS tube, especially the temperature change during high current shock; voltage data is used to monitor the drain-source voltage and gate-source voltage of the MOS tube to evaluate the working state of the MOS tube; on-resistance data is used to monitor the on-resistance of the MOS tube, which is an important indicator for evaluating the health status of the MOS tube; gate drive voltage is used to monitor the gate drive voltage to ensure that the MOS tube receives the correct drive signal.

[0066] The steps for obtaining the parameter gradient feature set include:

[0067] Extraction of each gradient data feature in key parameter data;

[0068] Construct a parameter gradient feature set based on each gradient data feature;

[0069] Based on the key parameter data, extract the gradient data features of current data, temperature data, voltage data, on-resistance data and gate drive voltage data in the key parameter data;

[0070] The method for extracting each gradient data feature in the key parameter data includes the following steps:

[0071] For current data, calculate the gradient data of current changing with time to obtain the current data gradient feature; for temperature data, calculate the gradient data of temperature changing with time to obtain the temperature data gradient feature; for voltage data, calculate the gradient data of voltage changing with time to obtain the voltage data gradient feature; for on-resistance data, calculate the gradient data of on-resistance changing with time to obtain the on-resistance data gradient feature; for gate drive voltage data, calculate the gradient data of gate drive voltage data changing with time to obtain the gate drive voltage data gradient feature;

[0072] The preset gradient calculation method includes the steps of:

[0073] Get the data series of all key parameter data changing with time, where two consecutive time points t and t i+1 The current data gradient feature at time point t is calculated by calculating the current data gradient feature of two adjacent time points t i and t i+1 The current difference is divided by the time difference to get;

[0074] The temperature data gradient feature at time point t is calculated by calculating the temperature gradient feature of two adjacent time points t i and t i+1 The temperature difference is divided by the time difference;

[0075] The voltage data gradient feature at time point t is calculated by calculating the voltage gradient feature of two adjacent time points t i and t i+1 The voltage difference is divided by the time difference;

[0076] The on-resistance data gradient feature at time point t is calculated by calculating the on-resistance data gradient feature of two adjacent time points t i and t i+1 The on-resistance difference is divided by the time difference;

[0077] The gate drive voltage data gradient feature at time point t is calculated by calculating the gate drive voltage data gradient feature at two adjacent time points t i and t i+1 The gate drive voltage difference is divided by the time difference;

[0078] A parameter gradient feature set is constructed based on the current data gradient feature, the temperature data gradient feature, the voltage data gradient feature, the on-resistance data gradient feature and the gate drive voltage data gradient feature.

[0079] The steps for obtaining abnormal behavior data include:

[0080] Calculate the Pearson correlation coefficient of two random parameter gradient features in the parameter gradient feature set. The Pearson correlation coefficient between the two parameter gradient features is Among them, x and y are any two different parameter gradient features in the parameter gradient feature set, is the observed mean of x, is the observed average value of y, calculate the Pearson correlation coefficient of all two random different parameter gradient features, and obtain a Pearson correlation coefficient set. Select the largest Pearson correlation coefficient from the Pearson correlation coefficient set. The two parameter gradient features corresponding to the largest Pearson correlation coefficient are parameter gradient feature a and parameter gradient feature b;

[0081] The parameter gradient feature a and the parameter gradient feature b are expressed as non-sinusoidal signal forms, and the non-sinusoidal signal form expression method includes the steps of:

[0082] Based on the formula: Where A is the DC component, n is the harmonic order, when n=1 it is the fundamental wave, when n>1 it is the higher harmonic, T is the periodic representation, t is the time variable, representing a certain moment on the time axis, and f(t) represents the non-sinusoidal signal value at time t;

[0083] The step of obtaining the DC component A comprises:

[0084] The cosine coefficient a n The steps to obtain include: Among them, a n represents the coefficient of the nth cosine term, which is obtained by dividing the function f(t) by The coefficient obtained by multiplying and integrating within one cycle reflects the amplitude of the nth harmonic in the signal;

[0085] The sine coefficient b n The steps to obtain include: Among them, b n represents the coefficient of the nth sine term, which is obtained by dividing the function f(t) by The coefficient obtained by multiplying and integrating within one cycle reflects the amplitude of the nth harmonic in the signal;

[0086] Among them, the periodicity of parameter gradient feature a and parameter gradient feature b is expressed as T a and T b ;

[0087] Using T a and T b Substitute T, T in the replacement formula a After replacing T, the non-sinusoidal signal form of the parameter gradient feature a is obtained, which is expressed as f(t a ), T b After replacing T, the non-sinusoidal signal form of the parameter gradient feature b is obtained, which is expressed as f(t b ), where f(t a ) represents the non-sinusoidal signal of parameter gradient feature a at time t, f(t b ) represents the non-sinusoidal signal of parameter gradient feature b at time t;

[0088] For non-sinusoidal signal f(t a ) and f(t b ) is weighted to obtain a periodic signal, and the new periodic signal is S, S = w a f(t a )+wb f(t b ), where w a represents the weight corresponding to the parameter gradient feature a, w b Represents the weight corresponding to the parameter gradient feature b;

[0089] The harmonic content rate of the periodic signal S is calculated, and the total harmonic content rate is T, then Among them, a1 is the cosine function of the fundamental wave, b1 is the sine function of the fundamental wave, It represents the sum of squares of all higher harmonics (i.e., all harmonics except the fundamental wave). Furthermore, the total harmonic content rate T represents the ratio of the power of all higher harmonics to the power of the fundamental wave, and is often used to evaluate the quality of the signal. A lower T indicates that the signal contains less harmonics, which usually means better signal quality. A higher T indicates that the signal contains more harmonics, which will affect the accuracy and reliability of the signal. The total harmonic content rate T represents the MOS tube status information. The normal range of the total harmonic content rate T is 0.5% to 3.5%. When it exceeds this range, the timestamp data of the parameter data corresponding to the abnormal total harmonic content rate T and the total harmonic content rate T are recorded to obtain abnormal behavior data.

[0090] The steps of obtaining the risk behavior pattern include:

[0091] Use a data structure to store the timestamp data and total harmonic content rate T of abnormal behavior data, create a list of abnormal behavior data points, which contains the timestamp data and total harmonic content rate T of abnormal data points, obtain time series data based on the list of abnormal behavior data points, divide the time series data into W segments, calculate the average total harmonic content rate T for each segment, and obtain the segmented sequence, which is {T ，1 , T ，2 ,…,T ，W}, the slope of the segmented sequence is m, Where i represents the index, and 1 <i<W, represents the average value of all segment indices, T ，i The total harmonic content rate of the i-th segment is T value, Indicates the average value of the total harmonic content rate of all segments is T;

[0092] Based on the slope m, an abnormal behavior threshold θ is preset. The behavior analysis rule is that when ∣m∣>θ and m>0, it indicates that the time series data shows a positive growth trend over time and there is a risk. When ∣m∣>θ and m<0, it indicates that the time series data shows a negative decreasing trend over time and there is a risk. If the absolute value of the slope m is less than the threshold θ, or the slope is close to zero, then the change trend of the time series data is considered stable, that is, there is no risk. Obtain a series of time series data in the ranges of ∣m∣>θ and m>0, ∣m∣>θ and m<0 to obtain the risk behavior pattern.

[0093] Based on historical data, determine a normal operation range, such as the change range of current intensity. If the change exceeds this range, it is considered abnormal.

[0094] The steps for obtaining the number of segments W include:

[0095] The time series data obtained through the abnormal behavior data point list is expressed as X = [δ1, δ2, …, δ n , where n is the number of data points δ. Each segment in the segmented time series data is expressed as S, then: Among them, S i represents the i-th segment, represents the error between the data points within each segment and the average value of its segment. δ i represents the i-th data point in segment S, represents the average value of all data points in segment S;

[0096] Use a two-dimensional array C to store the total error after segmentation. The size of the array C is N×(W + 1), where N is the total number of data points and W is the maximum number of segments. Initialize the total error of the entire data set as The preset number of segments is k, 1 < k < W. The splitting points for each preset number of segments k are Then Among them represents the error of the segment from the j-th data point to the i-th data point, is for all splitting points in the value of, and obtain The array corresponding to the minimum value of is expressed as C[N][W], where W is the optimal number of segments, and C[N][W] contains the minimum total error when splitting the entire data set using W segments.

[0097] The steps for obtaining the behavior score are:

[0098] Use each segment in the risk behavior pattern to form the state space

[0099] The spatial state includes the monitoring action of the MOS tube, which is set to include 3 actions, represented by A = {a1, a2, a3}, where a1 represents monitoring the MOS tube to collect more data about its behavior, a2 represents adjusting the working parameters of the MOS tube, and a3 represents issuing an alarm to notify the operator that the MOS tube has abnormal behavior;

[0100] Based on state space and space state A, the transition probability is Indicates that in state S i Take action a i After transfer to state The probability of , where i is the index;

[0101] Among them, the status The error is state The maximum error is The minimum error is Reward Function in, For state S i The error, when near When the reward is close to 0, it means that the state S i Closer to normal, when near When the reward is close to -1, it means that the state S i Closer to abnormal state;

[0102] Using the state value function V(S i ) and the action-value function Q(S i ,a i ), V(S i ) indicates that in state S i The expected return of taking the next action, Q(S i ,a i ) indicates that in state S i Take action a i The expected return of all states is initialized to a certain initial value, usually 0, V(S i )=0, for each state S i , calculated in state S i Take all possible actions a i The action value function Q(S i ,a i ), Among them, γ is the discount factor, and its value range is between [0, 1]. Status The expected return under update state S i The value function V(S i ) is the maximum Q(S i ,a i ),but in, Indicates that in state S i All possible actions a i The maximum value of the action value function in ;

[0103] A convergence threshold ∈ is preset, and the convergence stops when the value function change of all states is less than ∈, which can be expressed as: For each state S i , choose to make Q(S i , a i ) The largest action a H As the optimal strategy, denoted as π(S i ),but Calculate the action value function Q(S) under the optimal action i , a H ), Q(S i , a H ) represents the behavior score, expressed as F, and sets the abnormal behavior threshold F Q , when F>F Q When the risk behavior pattern of the corresponding segment is abnormal, a series of time series data corresponding to the abnormal risk behavior pattern is retrieved to intervene in the MOS management status;

[0104] Furthermore, if the basic data of a series of time series data corresponding to the abnormal risk behavior pattern is gradient data, then the abnormal segment may represent that the behavior of the MOS tube under these conditions deviates from the expected behavior pattern. According to the abnormal behavior trend, if the specific numerical value has a large impact and is too high or too low, the working state of the MOS tube is adjusted according to this situation.

[0105] Embodiment 2

[0106] Based on Example 1, please refer to Figure 2 As shown, the part not described in detail in this embodiment can be found in the description of embodiment 1, and a high-power MOS tube monitoring system is provided, including a module for monitoring the MOS tube:

[0107] Status information acquisition module: used to collect key parameter data during the working process of high-power MOS tubes;

[0108] Feature data acquisition module: used to perform gradient calculation on all parameter data in the key parameter data to obtain parameter gradient feature set;

[0109] Risk analysis module: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing behavior analysis rules, and obtaining risk behavior patterns of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rules;

[0110] Abnormal behavior decision module: Based on the risk behavior pattern, the behavior score is calculated to obtain the behavior score, and it is determined whether the behavior score is less than the set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

[0111] Embodiment 3

[0112] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0113] S1: Collect key parameter data of high-power MOS tube during operation;

[0114] S2: Perform gradient calculation on all parameter data in the key parameter data to obtain a parameter gradient feature set;

[0115] S3: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing a behavior analysis rule, and obtaining a risk behavior pattern of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rule;

[0116] S4: Based on the risk behavior pattern, a behavior score is calculated to obtain a behavior score, and it is determined whether the behavior score is less than a set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

[0117] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device can be referred to in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-power MOS tube monitoring method, characterized in that: The MOS tube monitoring steps are as follows: S1: Collect key parameter data of high-power MOS tube during operation; S2: Perform gradient calculation on all parameter data in the key parameter data to obtain a parameter gradient feature set; S3: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing a behavior analysis rule, and obtaining a risk behavior pattern of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rule; S4: Based on the risk behavior pattern, a behavior score is calculated to obtain a behavior score, and it is determined whether the behavior score is less than a set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

2. A high-power MOS tube monitoring method according to claim 1, characterized in that: The step of acquiring the parameter gradient feature set comprises: Extraction of each gradient data feature in key parameter data; Construct a parameter gradient feature set based on each gradient data feature; The gradient data features include current data gradient features, temperature data gradient features, voltage data gradient features, on-resistance data gradient features and gate drive voltage data gradient features; The steps for extracting each gradient data feature in the key parameter data include: Calculate the gradient data of the current changing with time according to a preset gradient calculation method to obtain the gradient characteristics of the current data; Calculate the gradient data of temperature changing with time and obtain the gradient characteristics of temperature data; Calculate the gradient data of voltage changing with time to obtain the gradient characteristics of voltage data; Calculate the gradient data of the on-resistance changing with time to obtain the gradient characteristics of the on-resistance data; Calculating the gradient data of the gate drive voltage data changing with time to obtain the gradient characteristics of the gate drive voltage data; The preset gradient calculation method includes the steps of: Get the data series of all key parameter data changing with time, where two consecutive time points are t i and t i+1 , calculate the two adjacent time points t i and t i+1 The current data difference is divided by the time difference to obtain the current data gradient feature, and the current data gradient is calculated at two adjacent time points t i and t i+1 The temperature data difference is divided by the time difference to obtain the temperature data gradient feature, and the temperature gradient between two adjacent time points t is calculated. i and t i+1 The voltage data difference is divided by the time difference to obtain the voltage data gradient feature, and the voltage data gradient is calculated at two adjacent time points t i and t i+1 The on-resistance data difference is divided by the time difference to obtain the on-resistance data gradient feature, and the on-resistance data gradient is calculated at two adjacent time points t i and t i+1 The gate driving voltage data gradient characteristic is obtained by dividing the gate driving voltage data difference by the time difference.

3. A high-power MOS tube monitoring method according to claim 1, characterized in that: The step of acquiring abnormal behavior data comprises: Calculate the Pearson correlation coefficient of two random parameter gradient features in the parameter gradient feature set. The Pearson correlation coefficient between the two parameter gradient features is Among them, x and y are any two different parameter gradient features in the parameter gradient feature set, is the observed mean of x, is the observed average value of y, calculate the Pearson correlation coefficient of all two random different parameter gradient features, and obtain a Pearson correlation coefficient set. Select the largest Pearson correlation coefficient from the Pearson correlation coefficient set. The two parameter gradient features corresponding to the largest Pearson correlation coefficient are parameter gradient feature a and parameter gradient feature b; The parameter gradient feature a and the parameter gradient feature b are expressed as non-sinusoidal signal forms, and the non-sinusoidal signal form expression method includes the steps of: Based on the formula: Where A is the DC component, n is the harmonic order, when n=1 it is the fundamental wave, when n>1 it is the higher harmonic, T is the periodic representation, t is the time variable, representing a certain moment on the time axis, and f(t) represents the non-sinusoidal signal value at time t; The step of obtaining the DC component A comprises: The cosine coefficient a n The steps to obtain include: Among them, a n It represents the coefficient of the nth cosine term; The sine coefficient b n The steps to obtain include: Among them, b n It represents the coefficient of the nth sine term; Among them, the periodicity of parameter gradient feature a and parameter gradient feature b is expressed as T a and T b ; Use T a and T b Substitute T and T in the replacement formula a After replacing T, the non-sinusoidal signal form f(t a ), T b After replacing T, the non-sinusoidal signal form f(t b ), where f(t a ) represents the non-sinusoidal signal of parameter gradient feature a at time t, f(t b ) represents the non-sinusoidal signal of parameter gradient feature b at time t; For non-sinusoidal signal f(t a ) and f(t b ) is weighted to obtain a periodic signal, and the new periodic signal is S, S = w a f(t a )+w b f(t b ), where w a represents the weight corresponding to the parameter gradient feature a, w b Represents the weight corresponding to the parameter gradient feature b; The harmonic content rate of the periodic signal S is calculated, and the total harmonic content rate is T, then Among them, a1 is the cosine function of the fundamental wave, b1 is the sine function of the fundamental wave, It represents the sum of squares of all high-order harmonics. The total harmonic content rate is T to represent the MOS tube status information. The normal range of the total harmonic content rate T is from M to N. When it exceeds this range, the timestamp data of the parameter data corresponding to the abnormal total harmonic content rate T and the total harmonic content rate T are recorded to obtain abnormal behavior data.

4. A high-power MOS tube monitoring method according to claim 1, characterized in that: The steps of obtaining the risk behavior pattern include: Use a data structure to store the timestamp data and total harmonic content rate T of the abnormal behavior data, create a list of abnormal behavior data points, including the timestamp data and total harmonic content rate T of the abnormal data points, obtain time series data based on the abnormal behavior data point list, divide the time series data into W segments, calculate the average total harmonic content rate T for each segment, and obtain the segmented sequence, which is η={T ，1 , T ，2 , …, T ，W }, T ，W It represents the total harmonic content T value of the Wth segment. The slope of the segmented sequence is m. Where i represents the index, and 1 <i<W, represents the average value of all segment indices, T ，i Indicates the total harmonic content T value of the i-th segment, It represents the average value of the total harmonic content rate T of all segments; Based on the slope m, an abnormal behavior threshold θ is preset. When |m|>θ and m>0, the time series data shows a positive increasing trend over time, and there is a risk. When |m|>θ and m<0, the time series data shows a negative decreasing trend over time, and there is a risk. When the absolute value of the slope m is less than the threshold θ, or the slope is close to zero, the change trend of the time series data is stable, that is, there is no risk. Obtain time series data in the range of |m|>θ and m>0, |m|>θ and m<0, and obtain the risk behavior pattern.

5. A high-power MOS tube monitoring method according to claim 5, characterized in that: The step of obtaining the number of segments W comprises: The time series data X = [δ1, δ2, ..., δ n ], where n is the number of data points δ, and each segment in the segmented time series data is represented by S, then: Among them, S i represents the i-th segment, represents the error between the i-th data point in each segment and its segment mean, δ i represents the i-th data point in segment S, represents the average value of all data points in segment S; Use a two-dimensional array C to store the total error after segmentation. The size of the array C is N×(W + 1), where N is the total number of data points and W is the maximum number of segments. Initialize the total error of the entire data set as The preset number of segments is k, where 1 < k < W. The segmentation points for each preset number of segments k are Then where represents the error of the segment from the j-th data point to the i-th data point, for all segmentation points in the value of, and obtain The array corresponding to the minimum value of is denoted as C[N][W], where W is the optimal number of segments, and C[N][W] contains the minimum total error when the entire data set is segmented using W segments.

6. A high-power MOS tube monitoring method according to claim 1, characterized in that: The steps for obtaining the behavior score are as follows: Use the segments in the risk behavior model to form a state space The space state includes the monitoring action of the MOS tube, and the setting includes 3 actions, which are expressed as A={a1, a2, a3}, where a1 represents monitoring the MOS tube, a2 represents adjusting the working parameters of the MOS tube, and a3 represents issuing an alarm; Based on state space and space state A, the transition probability is Indicates that in state S i Take action a i After transfer to state The probability of , where i is the index; Among them, the status The error is state The maximum error is The minimum error is Based on reward function in, For state S i The error, when near When the reward is close to 0, the state S i In normal state, when near When the reward is close to -1, the state state S i is an abnormal state; Using the state value function V(S i ) and the action-value function Q(S i ,a i ), V(S i ) indicates that in state S i The expected return of taking the next action, Q(S i ,a i ) indicates that in state S i Take action a i The expected return of all states is initialized to 0, V(S i )=0, for each state S i , calculated in state S i Take all actions a i The action value function Q(S i ,a i ), Among them, γ is the discount factor, and its value range is between [0, 1]. Status The expected return under update state S i The value function V(S i ) for all actions a i The maximum Q(S i ,a i ),but in, Indicates that in state S i All possible actions a i The maximum value of the action value function in ; A convergence threshold ∈ is preset, and the convergence stops when the value function change of all states is less than ∈, which can be expressed as: For each state S i , choose to make Q(S i , a i ) is the largest action a H As the optimal strategy, denoted as π(S i ),based on Calculate the action value function Q(S) under the optimal action i , a H ), Q(S i , a H ) represents the behavior score, expressed as F, and sets the abnormal behavior threshold F Q , when F>F Q When the risk behavior pattern of the corresponding segment is abnormal, the time series data corresponding to the abnormal risk behavior pattern is retrieved to intervene in the MOS management status.

7. A high-power MOS tube monitoring system, characterized in that: The monitoring system using the high-power MOS tube monitoring method according to any one of claims 1 to 7 comprises: Status information acquisition module: used to collect key parameter data during the working process of high-power MOS tube; Feature data acquisition module: used to perform gradient calculation on all parameter data in the key parameter data to obtain parameter gradient feature set; Risk analysis module: performing abnormal behavior analysis on each parameter gradient feature in the parameter gradient feature set, obtaining abnormal behavior data, constructing behavior analysis rules, and obtaining risk behavior patterns of the abnormal behavior data after inputting the abnormal behavior data into the behavior analysis rules; Abnormal behavior decision module: Based on the risk behavior pattern, the behavior score is calculated to obtain the behavior score, and it is determined whether the behavior score is less than the set state intervention threshold; if it is greater than the state intervention threshold, the MOS tube state is considered normal and monitoring continues; if it is less than the state intervention threshold, the MOS tube state is considered abnormal and MOS tube state intervention is performed.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-power MOS tube monitoring method according to any one of claims 1 to 7 is implemented.

Images