A high-side drive circuit fault diagnosis method and system

Abstract

The present application relates to the technical field of circuit fault diagnosis, and particularly relates to a high-side drive circuit fault diagnosis method and system, which comprises the following steps: firstly, using an unsupervised anomaly detection algorithm to process collected power data to obtain an initial anomaly score; calculating a power fluctuation factor reflecting the current power deviation from its historical average level, and combining the power fluctuation accumulation degree in the future time window to jointly generate a correction factor; correcting the initial anomaly score with the correction factor to obtain a target anomaly score; and determining the fault according to whether the target anomaly score meets a preset condition. The present application combines statistical models with physical feature analysis, effectively suppresses noise interference, and improves the accuracy and reliability of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of circuit fault diagnosis technology, and in particular to a method and system for diagnosing faults in high-side drive circuits. Background Technology

[0002] High-side drive circuits are key components in power electronic and control systems, and their stable operation is crucial to the safety and performance of these systems. Therefore, developing efficient and accurate online fault diagnosis methods to promptly detect and locate potential faults in circuits has significant engineering application value.

[0003] Currently, methods for intelligent fault diagnosis based on circuit operation data are gaining increasing attention. Unsupervised anomaly detection algorithms, such as Isolation Forest, have been applied in practice due to their ability to automatically discover abnormal patterns from data without requiring the pre-labeling of a large number of fault samples. The Isolation Forest algorithm isolates data points by constructing random trees and determines whether a data point is an anomaly based on the average path length required to isolate it. Typically, anomalies are easier to isolate due to their scarcity and unique characteristics, resulting in shorter path lengths.

[0004] However, when applied to scenarios involving strong electromagnetic interference and complex operating conditions, such as high-side drive circuits, these algorithms, which rely purely on statistical data distribution, reveal their inherent limitations: while the Isolation Forest algorithm can effectively identify outliers in the data stream, it cannot physically distinguish between instantaneous high-amplitude noise spikes and persistent power anomalies representing real faults. This leads to the algorithm easily misjudging normal transient disturbances as faults when faced with complex noise interference, resulting in a high false alarm rate and affecting the reliability and practicality of the diagnostic results. Summary of the Invention

[0005] To address the technical problem that unsupervised anomaly detection algorithms such as isolated forests cannot distinguish between transient high-amplitude noise and persistent power anomalies of real faults, and thus easily misjudge normal transient disturbances as faults, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for diagnosing faults in a high-side drive circuit, the method comprising the steps of:

[0007] Power data of the high-side drive circuit is collected at a preset sampling frequency for several time periods. An unsupervised anomaly detection algorithm is used to process the power data at each sampling time to obtain an initial anomaly score for each sampling time. For any given sampling time, a power fluctuation factor is calculated based on the power data at that sampling time and the total energy transmitted by the circuit within its historical neighborhood window. A correction factor is calculated for that sampling time. The correction factor is positively correlated with both the power fluctuation factor at that sampling time and the power fluctuation degree of the future neighborhood window at that sampling time. The initial anomaly score is corrected using the correction factor at that sampling time to obtain a target anomaly score. The target anomaly score is positively correlated with both the initial anomaly score at that sampling time and its correction factor. The target anomaly score is calculated for each sampling time. If the target anomaly score meets a preset judgment condition, a high-side drive circuit fault is determined.

[0008] This invention first obtains an initial anomaly score using an unsupervised algorithm, then introduces a power fluctuation factor and a future neighborhood window power fluctuation degree. The former quantifies the degree of instantaneous power mutation, while the latter characterizes the persistence of anomalies. Based on these two factors, a correction factor is constructed to dynamically correct the initial anomaly score. This dual verification mechanism, combining statistical models and physical characteristics, can distinguish between two types of signals: one is a real fault with a long duration and significant energy anomaly, and the other is a noise spike with a short duration and low total energy. Ultimately, while ensuring high sensitivity to real faults, it effectively reduces the false alarm rate and improves the accuracy of the diagnostic system.

[0009] Preferably, the step of calculating the power fluctuation factor based on the power data at the sampling time and the total energy transmitted by the circuit within its historical neighborhood window includes: obtaining the power data at the sampling time and the average power value within its historical neighborhood window; and performing normalization processing based on the deviation between the power data at the sampling time and the average power data, combined with the rated power value of the circuit, to obtain the power fluctuation factor.

[0010] This invention standardizes the measurement of power fluctuations by calculating the deviation between the current power value and the historical average power and performing normalization processing, which provides stable and reliable basic data for subsequent calculation of correction factors.

[0011] Preferably, the power fluctuation factor satisfies the following relationship:

[0012] ;

[0013] in, It is the first Power fluctuation factor at the sampling time; It is the first Power data at the sampling time, It is the length of the historical neighborhood window. It is the first Within the historical neighborhood window of the sampling time Power data at any given time; It is the absolute value symbol; This is the rated power value of the circuit; It is the index value at the sampling time.

[0014] This invention quantifies the degree of power fluctuation by calculating the absolute deviation between the current sampling power value and the average power within a historical neighborhood window, and then normalizing it using the circuit's rated power. On one hand, the absolute value calculation ensures that deviations in both positive and negative directions are considered, avoiding misjudgments caused by offsetting deviations. On the other hand, the normalization operation by dividing by the rated power eliminates differences between circuits of different power levels, providing a unified measurement scale for the fluctuation factor, making it applicable to high-side drive circuits of different specifications. Simultaneously, the introduction of the historical neighborhood window allows the assessment of current power fluctuations to be based on a reference to recent stable states, improving the accuracy of identifying abnormal fluctuations and providing standardized basic parameters for subsequent correction factor calculations.

[0015] Preferably, obtaining the power fluctuation level of the future neighborhood window at the sampling time includes: obtaining the average value of the power data within the future neighborhood window; calculating the absolute value of the difference between each power data within the future neighborhood window and the average value; and integrating the absolute value within the future neighborhood window to obtain the power fluctuation level.

[0016] This invention effectively quantifies the sustained fluctuation characteristics after an anomaly occurs by first calculating the average power within a future neighborhood window, then calculating and integrating the absolute deviation of the power from the average value at each time point. This method establishes a baseline state within the window through the average value, and accumulates the total fluctuation within the entire window through absolute deviation and integration, thus fully reflecting the persistence and severity of the anomaly. This calculation method can effectively distinguish between instantaneous noise and genuine faults.

[0017] Preferably, the correction factor satisfies the following relationship:

[0018] ;

[0019] in, It is the first Correction factor at sampling time; It is the first Power fluctuation factor at the sampling time; It is the index value at the sampling time; It is the length of the future neighborhood window; It is the first Within the future neighborhood window at the sampling time, the first Power data at any given time; It is the first The average power value within a future neighborhood window at the sampling time; It is the absolute value symbol; It is the standard normalization function.

[0020] This invention multiplies a power fluctuation factor that reflects the degree of the current instantaneous change with a future neighborhood window power fluctuation integral term that reflects the continuous impact of the anomaly. This allows it to capture both the intensity of the instantaneous change and the continuous effect of the anomalous state, thus avoiding the one-sidedness of a single-dimensional assessment.

[0021] Preferably, the step of correcting the initial anomaly score using the correction factor at the sampling time to obtain the target anomaly score includes: multiplying the initial anomaly score at the sampling time by its correction factor to obtain the target anomaly score.

[0022] Preferably, the preset judgment condition is that the target abnormal score is greater than a preset fault threshold.

[0023] Preferably, the unsupervised anomaly detection algorithm is the Isolation Forest algorithm.

[0024] Preferably, the historical neighborhood window and the future neighborhood window have the same length.

[0025] In a second aspect, the present invention provides a high-side drive circuit fault diagnosis system, which includes a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the high-side drive circuit fault diagnosis method of the first aspect of the present invention is implemented.

[0026] By adopting the above technical solution, a computer program for diagnosing faults in a high-side drive circuit according to the first aspect of the present invention is generated and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of a terminal device based on the memory and the processor, making it convenient to use.

[0027] The beneficial effects of this invention are as follows: First, an unsupervised anomaly detection algorithm is used to process the collected power data to obtain an initial anomaly score. Then, a correction mechanism based on physical characteristics is introduced: a correction factor is generated by calculating a power fluctuation factor reflecting the deviation of the current power from its historical average level and combining this with the cumulative degree of power fluctuation within a future neighborhood window. This correction factor is used to correct the initial anomaly score, resulting in a target anomaly score that better reflects the true probability of a fault. Finally, a fault is determined based on whether the target anomaly score meets preset conditions. This invention combines statistical models with physical feature analysis, effectively suppressing noise interference and improving the accuracy and reliability of fault diagnosis. Attached Figure Description

[0028] Figure 1 A flowchart of a high-side drive circuit fault diagnosis method provided in an embodiment of the present invention;

[0029] Figure 2 A graph showing the original power data of the high-side drive circuit provided in an embodiment of the present invention;

[0030] Figure 3 The initial anomaly score curves at each sampling time provided in the embodiments of the present invention;

[0031] Figure 4 The target anomaly score curves at various sampling times provided in this embodiment of the invention;

[0032] Figure 5 This is a structural block diagram of a high-side drive circuit fault diagnosis system provided in an embodiment of the present invention. Detailed Implementation

[0033] The first aspect of this invention provides a method for diagnosing faults in high-side drive circuits, such as... Figure 1 As shown, the method includes steps S100-S600:

[0034] Step S100: Collect power data of the high-side drive circuit at several moments according to the preset sampling frequency.

[0035] It should be noted that the power signal of the high-side drive circuit directly reflects its operating status. When the circuit experiences faults such as short circuits, open circuits, or component aging, the power consumption will exhibit significantly different characteristics from the normal state. Therefore, acquiring accurate power data is a prerequisite for subsequent accurate fault diagnosis.

[0036] Specifically, power sensors are deployed at key nodes of the target high-side drive circuit, at either the power input or output end. The power input end collects the overall power consumption of the circuit, reflecting the matching status between the global power supply and the load. The output end collects the load-side power, directly correlated with the fault characteristics of the driven object. Real-time power data during circuit operation is continuously collected at a preset sampling frequency, ultimately forming a power data time series that includes the time dimension.

[0037] As a preferred implementation, the preset sampling frequency can be set to 1kHz. The selection of this parameter is based on a balance between accuracy and cost: too low a sampling frequency may miss transient fault characteristics such as momentary short circuits, which typically last for more than 1ms; too high a frequency will significantly increase the data processing burden and storage costs. A 1kHz sampling frequency corresponds to a 1ms sampling interval, which can cover the feature capture requirements of most common transient faults while avoiding unnecessary resource consumption.

[0038] At this point, the power data of the high-side drive circuit, including several sampling times, has been obtained.

[0039] Step S200: Use an unsupervised anomaly detection algorithm to process the power data at each sampling time to obtain the initial anomaly score at each sampling time.

[0040] It should be noted that after obtaining the raw power data, the data first needs to be preliminarily screened to identify potential anomalies. Unsupervised anomaly detection algorithms are well-suited for online real-time monitoring scenarios because they do not require pre-labeling of fault samples and can directly detect anomalies from the distribution characteristics of the data itself. Therefore, this type of algorithm was chosen to complete the data processing in this step.

[0041] In this embodiment, the unsupervised anomaly detection algorithm is preferably the Isolation Forest algorithm. The Isolation Forest algorithm isolates data points by randomly constructing multiple decision trees (i.e., a forest). Its basic idea is that outliers, due to their sparse and distinct characteristics, are usually closer to the root node in the decision tree than normal points, meaning they can be isolated using a shorter path. When the algorithm runs, it calculates the average path length required to isolate each data point, thereby quantifying the degree of anomaly and ultimately obtaining an initial anomaly score. For example... Figures 2-3 As shown, Figure 2 This is a graph of the raw power data of the high-side drive circuit. The horizontal axis represents time in seconds, and the vertical axis represents power in watts. Figure 2 It shows the changes in the raw power data at each sampling time. Figure 3 This is a graph showing the initial anomaly scores at each sampling time after processing with the Isolation Forest algorithm. The horizontal axis also represents time, in seconds, and the vertical axis represents the initial anomaly score, which is used to reflect the degree of anomaly in the data at each time point after processing with the Isolation Forest algorithm.

[0042] As a feasible implementation method, the number of trees in the isolated forest algorithm can be set to 100, and the number of subsamples in each tree can be set to 256. These hyperparameters can be set empirically according to the data characteristics of the actual application scenario, or they can be optimized through cross-validation to adapt to the anomaly detection needs in different scenarios.

[0043] In addition, other unsupervised anomaly detection algorithms can also be applied to this invention to obtain initial anomaly scores, such as local anomaly factor, single-class support vector machine, etc. The above unsupervised anomaly detection algorithms are all existing technologies and will not be elaborated here.

[0044] At this point, the initial anomaly score for each sampling time has been obtained.

[0045] Step S300: For any sampling time, calculate the power fluctuation factor based on the power data at the sampling time and the total energy transmitted by the circuit within its historical neighborhood window.

[0046] It should be noted that the initial anomaly score is evaluated only from a statistical perspective and does not consider the dynamic changes in circuit power data over time. In real circuit faults, power anomalies are usually not isolated but related to the state at a previous moment. Therefore, this step introduces a power fluctuation factor to quantify the degree of deviation of the current power value from its nearest historical state, supplementing the dynamic dimension of anomaly judgment.

[0047] Specifically, for the first At the sampling time, first define a length of A historical neighborhood window, which contains data from the first... Sampling time up to the All power data at the sampling time, that is, window coverage A continuous historical sampling point, excluding the first one. The sampling time itself. Among these, the window length... The settings need to be determined in conjunction with the circuit's operating cycle characteristics and sampling frequency. The setting principle is to cover the typical stable fluctuation range during normal circuit operation, while avoiding a decrease in sensitivity to recent anomalies due to an excessively long window, or misjudging normal high-frequency fluctuations due to an excessively short window.

[0048] For example, if the sampling frequency is 1kHz, meaning one data point is collected every 1ms, and the power stabilization period of the high-side drive circuit during normal operation is 50ms-100ms, then... It can be preferentially set to 50-100; if the circuit load is of the high-frequency fluctuation type, and the normal fluctuation period is as short as 10ms-20ms, then It can be adjusted to 10-20 to ensure that normal high-frequency fluctuations and abnormal mutations can be distinguished.

[0049] During circuit operation, the greater the deviation between the current power value and the average power data within the historical neighborhood window, the more drastic the power fluctuation at the current moment is, and the more likely it is to be accompanied by an abnormal state.

[0050] Based on the above logic, the power fluctuation factor satisfies the following relationship:

[0051] ;

[0052] in, It is the first Power fluctuation factor at the sampling time; It is the first Power data at the sampling time, It is the length of the historical neighborhood window. It is the first Within the historical neighborhood window of the sampling time Power data at any given time; It is the absolute value symbol; This is the rated power value of the circuit; It is the index value at the sampling time.

[0053] In this relation, Indicates the first The total energy transmitted by the circuit in the historical neighborhood window at a given moment, this value is related to Multiplication represents the first The average power within the historical neighborhood window at time n. When the th moment... The power value at time t deviates significantly from its previous value. When the average power at time step is reached, the value of the molecule increases, which in turn leads to a corresponding increase in the value of the power fluctuation factor. This indicates that at the ... A significant power surge occurred at the sampling time, and such surges are an important indicator of potential circuit anomalies.

[0054] At this point, the power fluctuation factor corresponding to each sampling moment is obtained.

[0055] Step S400: Calculate the correction factor for the sampling time.

[0056] It should be noted that the core of this step lies in distinguishing the physical characteristics of real faults from transient noise. Real circuit faults are usually accompanied by continuous abnormal energy dissipation; while noise pulses, although their instantaneous power may be very high, have an extremely short duration and a limited overall abnormal energy scale. Therefore, by analyzing the energy scale characteristics within a short period after the anomaly occurs, a correction factor can be constructed to adjust the reliability of the initial anomaly score.

[0057] Specifically, for the first At the sampling time, define another length as A future neighborhood window that contains from arrive Power data at any given time. The length of the future neighborhood window needs to be consistent with the historical neighborhood window length to maintain... and Consistency ensures that the stable benchmark of historical data and the continuous fluctuation of future data are judged on the same scale over time, avoiding the imbalance between the standards for judging historical fluctuations and judging future persistence due to differences in window length, and improving the objectivity of the correction factor.

[0058] The severity of an anomalous event depends not only on the instantaneous impact intensity at the moment of its occurrence but also on the subsequent sustained power disturbances. Therefore, the correction factor should be positively correlated with both the current power fluctuation factor and the power fluctuation degree of the future neighborhood window. The power fluctuation degree of the future neighborhood window can be quantified by calculating the cumulative deviation of the power at each moment within that window from the window's average power.

[0059] Based on the above logic, the correction factor satisfies the following relation:

[0060] ;

[0061] in, It is the first Correction factor at sampling time; It is the first Power fluctuation factor at the sampling time; It is the index value at the sampling time; It is the length of the future neighborhood window; It is the first Within the future neighborhood window at the sampling time, the first Power data at any given time; It is the first The average power value within a future neighborhood window at the sampling time; It is the absolute value symbol; These are standard normalization functions, such as min-max normalization, used to map computation results to... The interval is used to facilitate subsequent processing.

[0062] This relation consists of two parts, the first part... Reflects the degree of change at the current moment; Part Two This reflects the degree of sustained power fluctuation over a future period. When transient noise occurs, although... The value may be large, but because the power will quickly return to normal, the fluctuation within the future neighborhood window will be small, resulting in a small final correction factor. Conversely, when a real fault occurs, The impact will be significant, and due to the persistence of the fault, the power within the neighborhood window will continue to fluctuate abnormally in the future, causing... It is also very large, which leads to a significant increase in the final correction factor.

[0063] At this point, the correction factors for each sampling time have been obtained.

[0064] Step S500: Correct the initial anomaly score using the correction factor at the sampling time to obtain the target anomaly score.

[0065] It should be noted that this step aims to integrate the dynamic characteristics of the time series with the static statistical characteristics of the data, thereby obtaining a more reliable comprehensive evaluation index that better reflects the actual probability of failure.

[0066] Specifically, the target anomaly score is obtained by directly multiplying the initial anomaly score corresponding to the sampling time by the correction factor. The core function of this calculation logic is: when a persistent anomaly is detected, the correction factor... A larger correction factor can amplify the initial anomaly score, making it easier for the final target anomaly score to exceed the fault determination threshold, ensuring that real faults are not missed; conversely, when only transient disturbances such as noise are detected, the correction factor is smaller. The smaller value can suppress the initial anomaly score, keeping it below the fault determination threshold and effectively avoiding false alarms caused by noise.

[0067] like Figure 4 The image shows the target anomaly score at each sampling time after correction. The horizontal axis represents time in seconds, and the vertical axis represents the target anomaly score, used to measure the degree of anomaly in the data at each time point. (Comparison) Figure 3 It is evident that the target anomaly score obtained after adjustment by the correction factor can more accurately identify the anomaly corresponding to the real fault, reduce misjudgment of normal disturbances such as instantaneous high-amplitude noise, and make the fault detection results more reliable.

[0068] At this point, the target anomaly scores for each sampling time have been obtained.

[0069] Step S600: Calculate the target anomaly score for each sampling time. If the target anomaly score meets the preset judgment condition, determine that the high-side driving circuit is faulty.

[0070] It should be noted that this step is the final decision-making stage of fault diagnosis. This step compares a target anomaly score that can comprehensively reflect the degree and persistence of the anomaly with a preset threshold to achieve the final determination of the circuit status.

[0071] Specifically, the preset judgment condition is first defined: when the target anomaly score at a certain sampling time is greater than the preset fault threshold, the fault judgment condition is met. Based on this, the target anomaly scores at all the acquired sampling times are judged one by one: the target anomaly score at each sampling time is compared with the fault threshold. If the target anomaly score at a certain sampling time is greater than the threshold, it is determined that the high-side drive circuit has failed at that time.

[0072] The setting of the fault threshold can be determined by conducting offline test analysis on a large amount of historical normal and fault data, aiming to achieve the best balance between the false alarm rate and the false alarm rate.

[0073] For example, to determine the fault threshold, 1,000 normal samples and 100 known fault samples were collected, and their target anomaly scores were calculated. It was found that the scores of normal samples were concentrated in the range of 0.1-0.5, while the scores of fault samples were all higher than 0.8. In order to achieve a balance between the false negative rate and the false positive rate, we set the fault threshold to 0.85.

[0074] Once a fault is detected, the system can immediately generate an early warning signal to notify maintenance personnel to carry out repairs, thereby ensuring the safe and stable operation of the power electronic system.

[0075] The second aspect of this embodiment provides a high-side drive circuit fault diagnosis system, such as Figure 5 As shown, the high-side drive circuit fault diagnosis system includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the first aspect of the present invention: a high-side drive circuit fault diagnosis method.

[0076] The high-side drive circuit fault diagnosis system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their setup and functions are known in the art and will not be described in detail here.

[0077] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0078] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A high-side driver circuit failure diagnosis method characterized by, The method comprises the steps of: collecting power data of the high-side drive circuit at several time points according to a preset sampling frequency; processing the power data at each sampling time point using an unsupervised anomaly detection algorithm to obtain an initial anomaly score at each sampling time point; for any sampling time point, calculating a power fluctuation factor according to the power data at the sampling time point and the total energy transmitted by the circuit in a historical neighborhood window of the sampling time point; ； in, It is the first Power fluctuation factor at the sampling time; It is the first Power data at the sampling time, It is the length of the historical neighborhood window. It is the first Within the historical neighborhood window of the sampling time Power data at any given time; It is the absolute value symbol; This is the rated power value of the circuit; It is the index value at the sampling time; calculating a correction factor of the sampling time point; the correction factor is positively correlated with the power fluctuation factor of the sampling time point and the power fluctuation degree of a future neighborhood window of the sampling time point, and the specific formula is: ； in, It is the first Correction factor at sampling time; It is the first Power fluctuation factor at the sampling time; It is the index value at the sampling time; It is the length of the future neighborhood window; It is the first Within the future neighborhood window at the sampling time, the first Power data at any given time; It is the first The average power value within a future neighborhood window at the sampling time; It is the absolute value symbol; It is a standard normalized function; correcting the initial anomaly score of the sampling time point using the correction factor to obtain a target anomaly score; the target anomaly score is positively correlated with the initial anomaly score of the sampling time point and the correction factor thereof; calculating the target anomaly score of all sampling time points one by one, and determining that the high-side drive circuit is faulty in response to the target anomaly score meeting a preset determination condition.

2. The high-side driver circuit fault diagnosis method of claim 1, wherein The power fluctuation degree of the future neighborhood window of the sampling time point is obtained by: obtaining the average value of the power data in the future neighborhood window; calculating the absolute value of the difference between each power data in the future neighborhood window and the average value; integrating the absolute value in the future neighborhood window to obtain the power fluctuation degree.

3. The high-side driver circuit fault diagnosis method of claim 1, wherein The correction factor of the sampling time point is obtained by: calculating the average value of the power data in the historical neighborhood window of the sampling time point; 4. The high-side driver circuit fault diagnosis method of claim 1, wherein calculating the absolute value of the difference between each power data in the historical neighborhood window and the average value; 5. The high-side driver circuit fault diagnosis method of claim 1, wherein integrating the absolute value in the historical neighborhood window to obtain the power fluctuation degree.

6. The high-side driver circuit fault diagnosis method of claim 1, wherein The correction factor of the sampling time point is obtained by:

7. A high-side driver circuit fault diagnosis system characterized by comprising: calculating the average value of the power data in the historical neighborhood window of the sampling time point; calculating the absolute value of the difference between each power data in the historical neighborhood window and the average value; integrating the absolute value in the historical neighborhood window to obtain the power fluctuation degree. The correction factor of the sampling time point is obtained by: calculating the average value of the power data in the historical neighborhood window of the sampling time point; calculating the absolute value of the difference between each power data in the historical neighborhood window and the average value; integrating the absolute value in the historical neighborhood window to obtain the power fluctuation degree. The correction factor of the sampling time point is obtained by: calculating the average value of the power data in the historical neighborhood window of the sampling time point; calculating the absolute value of the difference between each power data in the historical neighborhood window and the average value; integrating the absolute value in the historical neighborhood window to obtain the power fluctuation degree. The correction factor of the sampling time point is obtained by: calculating the average value of the power data in the historical neighborhood window of the sampling time point; calculating the absolute value of the difference between each power data in the historical neighborhood window and the average value; integrating the absolute value in the historical neighborhood window to obtain the power fluctuation degree. The preset determination condition is that the target anomaly score is greater than a preset fault threshold. The unsupervised anomaly detection algorithm is an isolation forest algorithm. The lengths of the historical neighborhood window and the future neighborhood window are the same. The high-side drive circuit fault diagnosis system comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the high-side drive circuit fault diagnosis method according to any one of claims 1-6.

Images