Inverter fault diagnosis methods, devices, equipment and storage media

Abstract

This invention provides a method, apparatus, device, and storage medium for inverter fault diagnosis, relating to the field of new energy power electronic equipment technology. The method includes: determining dynamic control limits for the squared prediction error statistic based on a principal component analysis (PCA) model; the PCA model is constructed based on historical data of the inverter during normal operation; calculating the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the PCA model, and comparing the real-time squared prediction error statistic with the dynamic control limits; and then determining whether the inverter has a fault or anomaly based on the comparison result. This invention solves the problems of scarce fault samples and poor real-time performance, achieving unsupervised and efficient inverter fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of new energy power electronic equipment technology, and in particular to an inverter fault diagnosis method, device, equipment and storage medium. Background Technology

[0002] Inverters, as the core interface equipment between photovoltaic / wind power generation systems and the power grid / load, undertake critical functions such as DC-AC power conversion and power quality regulation. Their reliability directly determines power generation efficiency and grid stability. Inverter faults, such as open / short circuits in insulated-gate bipolar transistors (IGBTs), aging of DC bus capacitors, and drive circuit failures, often manifest as abnormal fluctuations in multiple parameters such as voltage, current, and temperature. Current mainstream fault diagnosis methods have the following limitations: (1) Threshold and feature engineering-based methods: rely on expert experience to set fixed thresholds or extract features such as current / voltage amplitude and phase, which are difficult to adapt to dynamic changes under complex working conditions (such as sudden changes in light intensity and load fluctuations), resulting in high false alarm / false alarm rates; and require manual intervention in feature screening, which has weak generalization ability.

[0003] (2) Supervised learning methods: A large number of labeled fault samples are required to train the classifier, such as support vector machine (SVM) and random forest. However, in practice, inverter fault data is scarce (especially early minor faults), the training cost is high and the model is prone to overfitting.

[0004] (3) Deep learning methods: Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN) are used to extract time-series features. Although they can handle complex patterns, they have high computational complexity and are difficult to deploy in embedded diagnostic terminals. Moreover, the model is a "black box" and lacks physical interpretability of fault features.

[0005] Existing methods either rely on fault samples, are computationally complex, or lack interpretability. Therefore, there is an urgent need to provide an effective technical solution to address the aforementioned technical problems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an inverter fault diagnosis method, apparatus, device, and storage medium, which solves the problems of scarce fault samples, poor real-time performance, and difficulty in fault location, and achieves unsupervised and efficient inverter fault diagnosis.

[0007] In a first aspect, the present invention provides an inverter fault diagnosis method, the method comprising the following steps: The dynamic control limit of the squared prediction error statistic is determined based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter under normal operating conditions. The real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic is compared with the dynamic control limit. The inverter is determined to have any faults or abnormalities based on the comparison results.

[0008] According to the inverter fault diagnosis method provided by the present invention, the step of determining the dynamic control limit of the squared prediction error statistic based on the principal component analysis model includes: Multiple historical data points are collected as training samples to construct a training dataset; the historical data are multi-source parameter data. The training dataset is preprocessed to obtain a preprocessed training dataset; Based on the preprocessed training dataset, principal components whose cumulative variance contribution rate meets the preset conditions are selected to form a projection matrix to construct the principal component analysis model. Based on the principal component analysis model and the preprocessed training dataset, the squared prediction error statistic for each training sample is calculated, and the dynamic control limit is determined by the kernel density estimation method.

[0009] According to the inverter fault diagnosis method provided by the present invention, the step of preprocessing the training dataset to obtain a preprocessed training dataset includes: The training dataset is standardized and denoised to obtain the preprocessed training dataset.

[0010] According to the inverter fault diagnosis method provided by the present invention, the step of calculating the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model includes: Based on the principal component analysis model, the reconstructed values ​​of the real-time operating data of the inverter in the principal component space are determined; The square of the Euclidean distance between the real-time operating data of the inverter and the reconstructed value of the real-time operating data of the inverter in the principal component space is determined as the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter.

[0011] According to the present invention, an inverter fault diagnosis method is provided, wherein determining whether the inverter has a fault or abnormality based on a comparison result includes: If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is greater than the dynamic control limit, it is determined that the inverter has a fault or abnormality. If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is less than or equal to the dynamic control limit, it is determined that the inverter has no fault or abnormality.

[0012] According to the present invention, an inverter fault diagnosis method further includes: In the event that the inverter is found to be faulty or abnormal, the contribution of each variable in the real-time operating data of the inverter to the squared prediction error statistic is determined. The source of the inverter's fault is located based on the variable corresponding to the maximum contribution.

[0013] Secondly, the present invention also provides an inverter fault diagnosis device, which includes the following modules: The modeling module is used to determine the dynamic control limits of the squared prediction error statistic based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter under normal operating conditions. The fault diagnosis module is used to calculate the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model, and compare the real-time squared prediction error statistic with the dynamic control limit; The inverter is determined to have any faults or abnormalities based on the comparison results.

[0014] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the inverter fault diagnosis method as described above.

[0015] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the inverter fault diagnosis method as described above.

[0016] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the inverter fault diagnosis method as described above.

[0017] This invention provides an inverter fault diagnosis method, apparatus, device, and storage medium. First, a dynamic control limit for the squared prediction error statistic under normal operating conditions is determined based on a principal component analysis model (PCA). The PCA model is constructed based on historical data of the inverter under the normal operating conditions. Then, the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the PCA model, and the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is compared with the dynamic control limit. Finally, the inverter is judged to have any faults or abnormalities based on the comparison results.

[0018] This invention uses principal component analysis (PCA) to model inverter faults, requiring only normal historical data to build the model without needing fault samples. Furthermore, it determines dynamic control limits for the squared prediction error statistic based on the PCA model, allowing for adaptive adjustment of anomaly criteria. The invention then compares the squared prediction error statistic corresponding to the inverter's real-time operating data with the dynamic control limits, determining whether the inverter has any faults or anomalies based on the comparison results. This invention solves the problems of scarce fault samples and poor real-time performance, achieving unsupervised and efficient inverter fault diagnosis. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts of the inverter fault diagnosis method provided by the present invention.

[0021] Figure 2 This is a schematic diagram of calculating dynamic control limits using the kernel density estimation method provided by the present invention.

[0022] Figure 3 This is a schematic diagram illustrating the average contribution of each variable to anomaly detection provided by the present invention.

[0023] Figure 4 This is the second flowchart of the inverter fault diagnosis method provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the inverter fault diagnosis device provided by the present invention.

[0025] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first node can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0028] This invention addresses the shortcomings of existing inverter fault diagnosis technologies by providing an unsupervised, efficient, and fault-source-locating inverter fault diagnosis technology. Specifically, it provides an unsupervised inverter anomaly detection method and system based on Principal Component Analysis (PCA) and Squared Prediction Error (SPE). By using PCA modeling adapted to inverter parameters, dynamic SPE thresholding, and quantified feature contribution rate localization, it solves the problems of scarce fault samples, poor real-time performance, and difficulty in localization, thus overcoming the deficiencies of existing technologies.

[0029] The following is combined Figures 1 to 6 The present invention describes an inverter fault diagnosis method, apparatus, device, and storage medium.

[0030] Figure 1 This is one of the flowcharts of the inverter fault diagnosis method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Determine the dynamic control limits of the squared prediction error statistic based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter under normal operating conditions.

[0031] In this embodiment, the execution subject is an electronic device, which is used to achieve unsupervised and efficient inverter fault diagnosis.

[0032] This invention comprises two phases: an offline modeling phase and an online monitoring and diagnostic phase. The offline modeling phase involves constructing a principal component analysis (PCA) model based on historical data from the inverter during normal operation and determining the dynamic control limits for the squared prediction error statistic. The online monitoring and diagnostic phase involves projecting real-time collected operational data samples onto the constructed PCA model and combining this with the dynamic control limits for efficient and accurate fault diagnosis.

[0033] The offline modeling phase includes the following: First, historical data of the inverter during normal operation is acquired. This historical data consists of multi-source parameter data. Specifically, for example, during normal inverter operation, the DC-side voltage (…) is collected. AC side current () ), IGBT junction temperature ( ), capacitor ripple current ( ), ambient temperature ( The training dataset is composed of multiple source parameter data, including those from various sources. (n is the number of samples, p is the number of variables).

[0034] Further, a Principal Component Analysis (PCA) model is constructed based on the training dataset. This PCA model is specific to a particular type or model of inverter because the "normal" parameter relationships differ between inverters of different models and installed environments. The PCA model is a set of mathematical structures and rules, "trained" or "learned" from historical normal data, used to describe the baseline of the inverter's healthy operating state. The PCA model consists of the following parts: 1. Core: Projection Matrix This is the most important part of the PCA model, consisting of the first k principal component direction vectors. It comprises a set of "principal component direction vectors" arranged in descending order of importance (variance contribution). Each principal component direction is a representation of the original multivariable space (e.g., […]). , , , , A new coordinate axis is introduced in the 5-dimensional space (composed of the data). The direction of this new coordinate axis is the direction in which the variation (covariance) is most significant in the historical normal data.

[0035] For example, the first principal component may represent the direction of change of the "overall power level of the system" (voltage and current change in the same direction); the second principal component may represent the coupling relationship between "heat dissipation and electrical efficiency" (temperature and current change in a certain specific ratio).

[0036] 2. Decision parameters: Number of principal components The model selects the k most important principal components (variables) based on a criterion (e.g., "cumulative variance contribution rate ≥ 95%)". This is equivalent to defining a low-dimensional health state subspace. Fluctuations within the k-dimensional subspace are considered controllable changes under normal operating conditions; while fluctuations outside the k-dimensional subspace are considered abnormal.

[0037] 3. Benchmark parameters: mean and standard deviation Before building the model, the data needs to be Z-score standardized. The baseline parameters are the mean (μ) and standard deviation (σ) of each variable used in the Z-score standardization, which defines the baseline for data transformation and ensures that the online data and the offline model are compared on the same scale.

[0038] Furthermore, after obtaining the projection matrix, the dynamic control limits of the squared prediction error statistic (SPE) are determined based on the projection matrix of the principal component analysis model and the training data (multiple samples). For example, the SPE statistic for each sample is first calculated, and then the 99% confidence interval control limits of the SPE statistic are calculated using the kernel density estimation method. In other words, dynamic control limits are determined, and samples that exceed the dynamic control limits are identified as abnormal samples, that is, they have a fault, thereby ensuring high sensitivity detection of anomalies.

[0039] The squared prediction error (SPE) statistic for each sample is calculated using the following formula: in, This represents the SPE statistic for the i-th training sample. This represents the i-th training sample. This represents the reconstruction value of the i-th training sample in the principal component space.

[0040] The reconstructed value refers to the value obtained by "compressing" (dimensionality reduction) a sample data point using a trained PCA model and then "restoring" (reconstructing) it back to the original data space. This reconstructed value represents "what this sample should look like if it perfectly conforms to the historical normal pattern." The reconstructed value represents the value that the sample "should" present under the current PCA model (i.e., the historical normal state). The reconstructed value filters out noise and minor fluctuations, retaining only the main part that conforms to the normal coupling relationship. The original value is the true value actually measured by the sensor. The difference between the two is the reconstruction error. If the sample is completely normal, the error is small; if a fault occurs, disrupting the normal coupling relationship between parameters, this error will increase significantly. The SPE statistic is the sum of squares of all elements in this reconstruction error vector; it quantifies the overall degree to which the sample deviates from the "normal pattern." Furthermore, by analyzing the proportion of the squared error of each variable (such as voltage and current) in the SPE (i.e., contribution rate), it is possible to pinpoint which parameter(s) first and most severely disrupted the normal mode, thereby accurately identifying the source of the fault (for example, if the IGBT junction temperature has the highest contribution rate, it points to heat dissipation or chip failure).

[0041] In summary, the principal component analysis model can be viewed as the mean and standard deviation of historical normal data, the principal component projection matrix calculated from these data, and the number of principal components k determined based on the contribution rate.

[0042] Step 102: Calculate the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model, and compare the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter with the dynamic control limit.

[0043] Specifically, after obtaining the principal component analysis model, the real-time operating data of the inverter is further projected onto the principal component analysis model to calculate the squared prediction error statistic corresponding to the real-time operating data of the inverter.

[0044] The steps for calculating the squared prediction error statistic corresponding to the real-time operating data of the inverter are similar to those for calculating the squared prediction error statistic corresponding to the training samples, and will not be repeated here.

[0045] Furthermore, the squared prediction error statistic corresponding to the real-time operating data of the inverter is... With dynamic control limits By making comparisons, it is possible to determine whether the inverter has any faults or abnormalities based on the comparison results.

[0046] Step 103: Determine whether there is a fault or abnormality in the inverter based on the comparison results.

[0047] Specifically, if the squared prediction error statistic corresponds to the real-time operating data of the inverter Exceeding dynamic control limits If so, it is determined that the inverter has a fault or abnormality.

[0048] The core logic of this invention for fault detection is not to directly match fault modes, but to detect "the disruption of the normal mode." This is the essence of "unsupervised" learning; it requires knowing "what a fault looks like," but only "what a healthy state looks like." Any deviation from the healthy state is considered suspicious.

[0049] The method provided in this embodiment first determines the dynamic control limit of the squared prediction error statistic based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter during normal operation; then, the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is compared with the dynamic control limit; then, the inverter is judged to have any faults or abnormalities based on the comparison results.

[0050] This invention uses principal component analysis (PCA) to model inverter faults, requiring only normal historical data to build the model without needing fault samples. Furthermore, it determines dynamic control limits for the squared prediction error statistic based on the PCA model, allowing for adaptive adjustment of anomaly criteria. The invention then compares the squared prediction error statistic corresponding to the inverter's real-time operating data with the dynamic control limits, determining whether the inverter has any faults or anomalies based on the comparison results. This invention solves the problems of scarce fault samples and poor real-time performance, achieving unsupervised and efficient inverter fault diagnosis.

[0051] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.

[0052] According to the present invention, an inverter fault diagnosis method is provided, which determines the dynamic control limit of the squared prediction error statistic based on a principal component analysis model, including: Multiple historical data points were collected as training samples to construct a training dataset; the historical data consisted of multi-source parameter data. The training dataset is preprocessed to obtain the preprocessed training dataset; Based on the preprocessed training dataset, principal components whose cumulative variance contribution rate meets the preset conditions are selected to form a projection matrix in order to construct a principal component analysis model. Based on the principal component analysis model and the preprocessed training dataset, the squared prediction error statistic for each training sample is calculated, and the dynamic control limits are determined by the kernel density estimation method.

[0053] Specifically, in some embodiments, step 102 can be implemented through the following steps: First, multiple historical data sets were collected as training samples to construct the training dataset. (n is the number of samples, p is the number of variables). Then, the training dataset is preprocessed, such as by standardization and denoising, to obtain a preprocessed training dataset, which facilitates the subsequent construction of a principal component analysis model with unified dimensions and the determination of the dynamic control limits of the squared prediction error statistic.

[0054] A principal component analysis model is further constructed. Specifically, the covariance matrix of the standardized data is calculated based on the preprocessed training dataset, and eigenvalues ​​are obtained through eigenvalue decomposition. and corresponding feature vectors, The first eigenvalue, The second eigenvalue, For the first Each eigenvalue is used. Then, the top k principal component vectors with a cumulative variance contribution rate ≥ 95% are selected to construct the projection matrix. , This is the first principal component vector. This is the second principal component vector. This represents the k-th principal component vector. This completes the construction of the principal component analysis model.

[0055] Furthermore, based on the principal component analysis model, the squared prediction error statistic for each training sample is calculated, and the dynamic control limits are determined by the kernel density estimation method.

[0056] Specifically, first calculate the reconstruction value of each training sample in the principal component space. The process of calculating the reconstructed value includes the following steps: Step 1: Projection (Compression): Project the normalized original sample vector Z onto the principal component space (by left-multiplying the normalized original sample vector Z by the projection matrix P). This matrix multiplication maps the high-dimensional Z to the low-dimensional "principal component space," resulting in a k-dimensional "score" vector T. T represents the coordinates of the sample in the most significant directions of change, which can be considered as its core feature encoding. PCA dimensionality reduction reduces computational complexity from... Down to The number of principal component vectors, k, is much smaller than the number of original variables, p. SPE calculation is completed in milliseconds, making it suitable for embedded terminal deployment and meeting real-time monitoring requirements.

[0057] Step 2: Backprojection (Reconstruction): Multiply the compressed principal component scores T (K-dimensional, core features) on the left by the transpose of the projection matrix P. This operation "decodes" the low-dimensional features T back into the original p-dimensional space. Since information from (pk) minor components has been discarded during the projection process, this reconstructed value can recover the content that the PCA model considers to be the "principal part" or "normal pattern".

[0058] Then, the reconstructed values ​​of each training sample in the principal component space are combined. The squared prediction error statistic for each training sample is calculated using the following formula: in, This represents the SPE statistic for the i-th training sample. This represents the i-th training sample. This represents the reconstruction value of the i-th training sample in the principal component space.

[0059] Furthermore, the dynamic control limits are determined by the kernel density estimation method. That is, based on all SPE statistics calculated during the training phase, the kernel density estimation method is used to fit their probability distribution.

[0060] Kernel density estimation (KD) is a nonparametric probability density estimation method. It does not require assumptions that the data follows a specific distribution (such as a normal distribution), but rather estimates the probability density function directly from the data itself. Simply put, it places a smooth "kernel function" (such as a Gaussian bell curve) at each data point, and then superimposes all these kernel functions to obtain the overall probability density curve of the data. Specifically, using the Gaussian function as the kernel, the optimal bandwidth is determined through the Silverman rule, constructing a nonparametric probability density function for the SPE value. Subsequently, the cumulative distribution function is obtained through numerical integration, and the SPE value corresponding to the cumulative distribution function reaching 99% is determined as the dynamic control limit. This method does not require prior assumptions about the statistical distribution of SPE, and the obtained control limits can accurately reflect the statistical characteristics of SPE fluctuations under normal operating conditions, providing adaptive and statistically significant anomaly criteria for subsequent online monitoring.

[0061] See Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the calculation of dynamic control limits using the kernel density estimation method provided by the present invention. The horizontal axis represents the sample number, and the vertical axis represents the squared prediction error. Figure 2 The red dashed line in the diagram represents the 99% control limit determined by the kernel density estimation method (KDE). Samples below the red line are considered normal samples, while samples above the red line are considered abnormal samples.

[0062] The method provided in this embodiment can build a model with only normal historical data and no fault samples are required, which perfectly solves the problem of scarce inverter fault data and supports full life cycle monitoring. Based on kernel density estimation, the 99% confidence limit of the SPE statistic is calculated, and the anomaly criteria are adaptively adjusted to avoid false alarms under dynamic operating conditions with fixed thresholds.

[0063] According to the inverter fault diagnosis method provided by the present invention, the training dataset is preprocessed to obtain the preprocessed training dataset, including: The training dataset is standardized and denoised to obtain the preprocessed training dataset.

[0064] Specifically, in some embodiments, preprocessing the training dataset to obtain a preprocessed training dataset is achieved through the following steps: The training dataset is standardized and denoised to obtain preprocessed training samples.

[0065] Denoising preprocessing refers to removing noise from the samples. For example, median filtering is used to denoise the training samples.

[0066] The inverter parameters exhibit vastly different units and numerical ranges (e.g., voltage is measured in volts, with values ​​ranging from hundreds of volts; current is measured in amperes, with values ​​ranging from tens of amperes; temperature is measured in degrees Celsius, with values ​​ranging from tens to over one hundred degrees Celsius). Without proper handling, variables with large numerical ranges (such as voltage) will dominate the PCA model, obscuring the changes in smaller but potentially crucial variables (such as certain ripple currents). Standardization, such as Z-score standardization of the training samples, helps eliminate the influence of dimensions.

[0067] Specifically, when performing Z-score standardization on the training sample dataset X (an n×p matrix, where n is the number of samples and p is the number of variables), the core idea is to process each variable (each column) independently to make it conform to a standard normal distribution with a mean of 0 and a standard deviation of 1. The specific steps are as follows: Step 1: Calculate the mean and standard deviation for each variable. For the j-th variable in the dataset (e.g., all data in the "DC-side voltage" column), calculate the arithmetic mean μ of all its samples. j At the same time, calculate the population standard deviation σ of this variable. j .

[0068] Step 2: Transform each sample value For the original value x of the j-th variable in the i-th sample ij Standardize using the following formula: in: It is the standardized value, μ j σ represents the arithmetic mean of the j-th variable in all samples. j This represents the population standard deviation of the j-th variable. This indicates data centralization (mean removal), divided by This indicates that the data will be scaled (except for the standard deviation).

[0069] Step 3: Generate the standardized dataset Repeat the above calculation for all samples (n rows) of all variables (p columns) to obtain a new standardized dataset Z. In this new dataset, the mean of each column (each variable) is 0 and the standard deviation is 1.

[0070] The method provided in this embodiment performs Z-score standardization on the data to eliminate the influence of dimensions and performs median filtering for noise reduction. This data preprocessing is a fundamental key technology for realizing the invention's key features such as "unsupervised learning," "multi-source parameter fusion analysis," and "dynamic threshold adaptation." This ensures that the principal component analysis model can extract features reflecting the true health status of the equipment from heterogeneous inverter parameters, thereby supporting subsequent high-accuracy, low-false-alarm fault detection and location.

[0071] According to the present invention, an inverter fault diagnosis method calculates the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on a principal component analysis model, including: Based on the principal component analysis model, the reconstructed values ​​of the real-time operating data of the inverter in the principal component space are determined. The square of the Euclidean distance between the real-time operating data of the inverter and the reconstructed value of the real-time operating data of the inverter in the principal component space is determined as the squared prediction error statistic corresponding to the real-time operating data of the inverter.

[0072] Specifically, step 103 is implemented through the following steps: First, based on the principal component analysis model, the reconstructed values ​​of the inverter's real-time operating data in the principal component space are determined. That is, after the projection (compression) in step one, the high-dimensional real-time samples are mapped to the low-dimensional "principal component space," resulting in a k-dimensional "score" vector T. Then, in step two, the back projection (reconstruction) multiplies the compressed score T by the transpose of the projection matrix P, "decoding" the low-dimensional features T back to the original p-dimensional space, outputting the reconstructed values ​​of the real-time samples.

[0073] Furthermore, the square of the Euclidean distance between the real-time operating data of the inverter and its reconstructed value in the principal component space is determined as the squared prediction error statistic (SPE) corresponding to the real-time operating data of the inverter. The SPE statistic is essentially a reconstruction error; it measures not whether a single parameter exceeds its absolute range, but whether the overall coupling relationship between all parameters is disrupted. Early inverter faults (such as slight IGBT aging or capacitor degradation) often manifest as subtle imbalances in the coordination between parameters, rather than obvious exceedances of any single parameter. Traditional threshold methods are insensitive to this.

[0074] This invention does not rely on any fault sample labels. The system only needs to compare the real-time SPE value with dynamic control limits learned from historical normal data to make an objective judgment of "normal" or "abnormal". This fundamentally solves the industry problem of scarce fault data and high labeling costs.

[0075] The method provided in this embodiment achieves a crucial leap from "model" to "diagnosis" by calculating real-time SPE statistics. It is the core detection action in the entire online fault diagnosis process, transforming the high-dimensional and complex inverter operating status (a set of parameters) into a single, quantifiable anomaly indicator (SPE value). By calculating the overall deviation between the "actual measured value" and the "normal value predicted by the model," SPE can sensitively capture such weak correlation anomalies, thereby enabling early warning.

[0076] According to the present invention, an inverter fault diagnosis method is provided, which determines whether the inverter has a fault or abnormality based on a comparison result, including: If the comparison results indicate that the squared prediction error statistic corresponding to the real-time operating data of the inverter is greater than the dynamic control limit, it is determined that the inverter has a fault or abnormality. If the squared prediction error statistic corresponding to the real-time operating data of the inverter is less than or equal to the dynamic control limit, it is determined that the inverter has no fault or abnormality.

[0077] Specifically, if the squared prediction error statistic SPE corresponding to the real-time operating data of the inverter is greater than the dynamic control limit... If the inverter does not exhibit any faults or abnormalities, it is determined that the inverter has a fault or abnormality. Conversely, if it does not exhibit any faults or abnormalities, it is determined that the inverter does not have a fault or abnormality.

[0078] The method provided in this embodiment compares the judgment benchmark with a dynamic control limit learned from historical normal operating conditions, rather than a fixed threshold. This allows the judgment benchmark to adapt to the normal fluctuation range of different inverters or the same inverter under different seasons and loads. This effectively overcomes the high false alarm rate of traditional methods under dynamic operating conditions (such as sudden changes in sunlight or load switching). The system can better distinguish between "normal environmental fluctuations" and "genuine equipment failures."

[0079] According to the present invention, an inverter fault diagnosis method further includes: In the event that the inverter is faulty or abnormal, determine the contribution of each variable in the real-time operating data of the inverter to the squared prediction error statistic. Locate the source of inverter failure based on the variable corresponding to the maximum contribution.

[0080] In some embodiments, the method further includes locating the source of the fault, specifically including the following steps: When an inverter malfunction is confirmed, determine the contribution of each variable in the inverter's real-time operating data to the Squared Prediction Error (SPE) statistic. For example, the contribution of each original variable in the outlier sample to the SPE statistic can be calculated using the following formula: in, This represents the contribution of the j-th original variable in the outlier samples to the SPE statistic of the squared prediction error. Let i be the i-th sample (anomaly sample). The reconstructed values ​​of the samples in the principal component space. Indicates the number of variables. Let SPE represent the squared prediction error statistic for the i-th sample (abnormal sample).

[0081] Furthermore, based on the variable corresponding to the largest contribution, the source of the inverter fault is located. For example, the first variable (DC-side voltage) If the component that contributes the most to the squared prediction error (SPE) statistic is identified as having a fault origin at the DC-side voltage level, then the fault source is determined to be the component at the DC-side voltage level with the highest contribution rate. Indicates a rectifier unit malfunction.

[0082] For example, Figure 3 This is a schematic diagram illustrating the average contribution of each variable to anomaly detection, as provided by this invention. Specifically, it represents the calculation of the average contribution of each variable to anomaly detection when the SPE statistic exceeds the dynamic control limit during a certain fault occurrence in the online monitoring phase. The horizontal axis represents the average contribution ratio of (each variable) to the squared prediction error (e.g., the contribution of the j-th original variable to the squared prediction error statistic), and the vertical axis represents each variable, such as... Figure 3 As shown, the total power factor has the largest average power ratio to the squared prediction error, so the fault source can be further located based on the total power factor.

[0083] The method provided in this embodiment quantifies the contribution of each parameter to the anomaly by calculating the contribution rate of feature variables, providing physically interpretable diagnostic results and directly guiding fault diagnosis. Feature contribution rate analysis reduces the average fault diagnosis time by 60%, improves operational efficiency, and enables rapid response and location.

[0084] Figure 4 This is the second flowchart of the inverter fault diagnosis method provided by the present invention, as shown below. Figure 4 As shown, the method includes: Offline training phase begins: Data processing: Acquisition of V dc / I ac Historical data → forming a training sample set X → standardization + denoising; PCA training and modeling: Constructing the covariance matrix → Eigenvalue decomposition → Selecting the first k principal components to construct the projection matrix P; SPE Calculation: Calculate the training SPE (calculate the SPE for each training sample) → Obtain the dynamic control limit SPE using kernel density estimation. limit ; Model deployment is ready; Model input, online monitoring phase begins: Real-time data: Sample x of real-time running data collected new Standardize using offline parameters; SPE detection: This involves analyzing samples of real-time running data. new Projected PCA is used to calculate real-time SPE → to determine if the real-time SPE exceeds the SPE. limit ; Determine if SPE > SPE limit If yes, then an anomaly is determined, a fault alarm is triggered → contribution rate is calculated → fault source is located; if no, then normal is determined, and monitoring continues. Finish.

[0085] The inverter fault diagnosis device provided by the present invention is described below. The inverter fault diagnosis device described below and the inverter fault diagnosis method described above can be referred to in correspondence.

[0086] Figure 5 This is a schematic diagram of the inverter fault diagnosis device provided by the present invention, as shown below. Figure 5 As shown, the inverter fault diagnosis device 500 includes the following modules: Modeling module 510 is used to determine the dynamic control limit of the squared prediction error statistic based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter under normal operating conditions; The fault diagnosis module 520 is used to calculate the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model, and compare the real-time squared prediction error statistic with the dynamic control limit; The inverter is determined to have any faults or abnormalities based on the comparison results.

[0087] The apparatus provided in this embodiment of the invention includes a modeling module 510, which is used to determine the dynamic control limit of the squared prediction error statistic based on a principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter during the normal operation state; and a fault diagnosis module 520, which is used to calculate the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model, and compare the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter with the dynamic control limit; and then, determine whether there is a fault or abnormality in the inverter based on the comparison result.

[0088] This invention uses principal component analysis (PCA) to model inverter faults, requiring only normal historical data to build the model without needing fault samples. Furthermore, it determines dynamic control limits for the squared prediction error statistic based on the PCA model, allowing for adaptive adjustment of anomaly criteria. The invention then compares the squared prediction error statistic corresponding to the inverter's real-time operating data with the dynamic control limits, determining whether the inverter has any faults or anomalies based on the comparison results. This invention solves the problems of scarce fault samples and poor real-time performance, achieving unsupervised and efficient inverter fault diagnosis.

[0089] According to the present invention, an inverter fault diagnosis device 500 is provided, wherein the modeling module 510 is specifically used for: Multiple historical data points are collected as training samples to construct a training dataset; the historical data are multi-source parameter data. The training dataset is preprocessed to obtain a preprocessed training dataset; Based on the preprocessed training dataset, principal components whose cumulative variance contribution rate meets the preset conditions are selected to form a projection matrix to construct the principal component analysis model. Based on the principal component analysis model and the preprocessed training dataset, the squared prediction error statistic for each training sample is calculated, and the dynamic control limit is determined by the kernel density estimation method.

[0090] According to the inverter fault diagnosis device 500 provided by the present invention, the modeling module 510 is further used for: The training dataset is standardized and denoised to obtain the preprocessed training dataset.

[0091] According to the present invention, an inverter fault diagnosis device 500 is provided, wherein the fault diagnosis module 520 is specifically used for: Based on the principal component analysis model, the reconstructed values ​​of the real-time operating data of the inverter in the principal component space are determined; The square of the Euclidean distance between the real-time operating data of the inverter and the reconstructed value of the real-time operating data of the inverter in the principal component space is determined as the squared prediction error statistic corresponding to the real-time operating data of the inverter.

[0092] According to the present invention, an inverter fault diagnosis device 500 is provided, wherein the fault diagnosis module 520 is further used for: If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is greater than the dynamic control limit, it is determined that the inverter has a fault or abnormality. If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is less than or equal to the dynamic control limit, it is determined that the inverter has no fault or abnormality.

[0093] According to the present invention, an inverter fault diagnosis device 500 is provided, the device further comprising a fault location module; The fault location module is used for: In the event that the inverter is found to be faulty or abnormal, the contribution of each variable in the real-time operating data of the inverter to the squared prediction error statistic is determined. The source of the inverter's fault is located based on the variable corresponding to the maximum contribution.

[0094] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 can call logic instructions in the memory 630 to execute an inverter fault diagnosis method, which includes: The dynamic control limit of the squared prediction error statistic is determined based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter during normal operation. The real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic is compared with the dynamic control limit. The inverter is determined to have any faults or abnormalities based on the comparison results.

[0095] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the inverter fault diagnosis method provided by the above methods, the method comprising: The dynamic control limit of the squared prediction error statistic is determined based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter during normal operation. The real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic is compared with the dynamic control limit. The inverter is determined to have any faults or abnormalities based on the comparison results.

[0097] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the inverter fault diagnosis method provided by the methods described above, the method comprising: The dynamic control limit of the squared prediction error statistic is determined based on the principal component analysis model; the principal component analysis model is constructed based on historical data of the inverter during normal operation. The real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic is compared with the dynamic control limit. The inverter is determined to have any faults or abnormalities based on the comparison results.

[0098] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing inverter faults, characterized in that, include: Determine the dynamic control limits of the squared prediction error statistic based on the principal component analysis model; The principal component analysis model is constructed based on historical data of the inverter during normal operation. The real-time squared prediction error statistic corresponding to the real-time operating data of the inverter is calculated based on the principal component analysis model, and the real-time squared prediction error statistic is compared with the dynamic control limit. The inverter is determined to have any faults or abnormalities based on the comparison results.

2. The inverter fault diagnosis method according to claim 1, characterized in that, The dynamic control limits for determining the squared prediction error statistic based on the principal component analysis model include: Multiple historical data points are collected as training samples to construct a training dataset; the historical data are multi-source parameter data. The training dataset is preprocessed to obtain a preprocessed training dataset; Based on the preprocessed training dataset, principal components whose cumulative variance contribution rate meets the preset conditions are selected to form a projection matrix to construct the principal component analysis model. Based on the principal component analysis model and the preprocessed training dataset, the squared prediction error statistic for each training sample is calculated, and the dynamic control limit is determined by the kernel density estimation method.

3. The inverter fault diagnosis method according to claim 2, characterized in that, The preprocessing of the training dataset to obtain the preprocessed training dataset includes: The training dataset is standardized and denoised to obtain the preprocessed training dataset.

4. The inverter fault diagnosis method according to any one of claims 1-3, characterized in that, The calculation of the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model includes: Based on the principal component analysis model, the reconstructed values ​​of the real-time operating data of the inverter in the principal component space are determined; The square of the Euclidean distance between the real-time operating data of the inverter and the reconstructed value of the real-time operating data of the inverter in the principal component space is determined as the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter.

5. The inverter fault diagnosis method according to any one of claims 1-3, characterized in that, The step of determining whether the inverter has a fault or abnormality based on the comparison result includes: If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is greater than the dynamic control limit, it is determined that the inverter has a fault or abnormality. If the comparison result indicates that the squared prediction error statistic corresponding to the real-time operating data of the inverter is less than or equal to the dynamic control limit, it is determined that the inverter has no fault or abnormality.

6. The inverter fault diagnosis method according to claim 1, characterized in that, The method further includes: In the event that the inverter is found to be faulty or abnormal, the contribution of each variable in the real-time operating data of the inverter to the squared prediction error statistic is determined. The source of the inverter's fault is located based on the variable corresponding to the maximum contribution.

7. An inverter fault diagnosis device, characterized in that, include: The modeling module is used to determine the dynamic control limits of the squared prediction error statistic based on the principal component analysis model. The principal component analysis model is constructed based on historical data of the inverter during normal operation. The fault diagnosis module is used to calculate the real-time squared prediction error statistic corresponding to the real-time operating data of the inverter based on the principal component analysis model, and compare the real-time squared prediction error statistic with the dynamic control limit; The inverter is determined to have any faults or abnormalities based on the comparison results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the inverter fault diagnosis method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the inverter fault diagnosis method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the inverter fault diagnosis method as described in any one of claims 1 to 6.

Images