Three-phase inverter direct current side grounding weak fault type detection method and system

Abstract

The invention relates to the technical field of photovoltaic three-phase grid-connected inverter fault detection, and discloses a three-phase inverter direct current side grounding weak fault type detection method and system, and the method comprises the steps: S1, carrying out the preprocessing of collected leakage current data, and extracting leakage current mechanism features and leakage current statistical features; s2, constructing a fault feature system based on leakage current mechanism features and leakage current statistical features obtained through preprocessing, and generating a fault classification model; and S3, classifying the test set data by using the fault classification model, and outputting a fault detection result. The method has high identification precision, strong generalization capability and short calculation time, and provides theoretical and technical support for rapid diagnosis of grid-connected inverter direct current side grounding weak faults in practical engineering.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault detection, and in particular relates to a method and system for detecting a weak ground fault type on the DC side of a three-phase inverter. Background Art

[0002] Grid-connected inverters are key devices that convert DC power generated by power generation systems into AC power. Within power generation systems, various factors, such as insulation damage, environmental influences, and temperature, lead to diverse types of DC ground faults. This not only increases potential safety hazards but also increases the risk of electric shock or equipment damage. Therefore, detecting weak DC ground faults is crucial to ensuring safe operation and optimized performance of power generation systems.

[0003] Traditional fault detection for photovoltaic grid-connected inverters primarily focuses on open-circuit and short-circuit faults in the inverter's passive components, AC grid faults on the inverter side, and power switch faults within the inverter. Faults are often identified through current, voltage, and ground resistance monitoring. However, these methods suffer from slow response speeds, high false alarm and missed alarm rates, and poor flexibility, making them difficult to handle intricate or sudden fault situations. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology and provide a method and system for detecting the type of weak grounding fault on the DC side of a three-phase inverter. Through machine learning and data driving, it provides theoretical and technical support for the rapid investigation and identification of weak grounding faults on the DC side of photovoltaic systems in actual projects.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for detecting a weak ground fault on the DC side of a three-phase grid-connected inverter comprises the following steps:

[0007] S1: Preprocess the collected leakage current data to extract leakage current mechanism characteristics and leakage current statistical characteristics;

[0008] S2: Build a fault feature system based on the leakage current mechanism characteristics and leakage current statistical characteristics obtained by preprocessing, and generate a fault classification model;

[0009] S3: Use the fault classification model to classify the test set data and output the fault detection results.

[0010] Furthermore, step S1 includes:

[0011] S11: collecting leakage current data, and performing denoising, normalization and state encoding preprocessing on the leakage current data;

[0012] S12: Reconstruct the effective fault signal using wavelet packet transform to extract wavelet energy and wavelet Shannon entropy;

[0013] S13: extracting leakage current mechanism characteristics through FFT analysis, wherein the leakage current mechanism characteristics include a DC component, an AC component, and a phase difference between current and voltage of the leakage current;

[0014] S14: Extracting leakage current statistical features through time domain analysis, wherein the leakage current statistical features include a mean value, a standard deviation, a maximum value, and a minimum value of the leakage current.

[0015] Furthermore, the specific steps of wavelet packet transform in S12 are as follows:

[0016] S121: Selecting a suitable wavelet basis function for the residual current data and selecting a suitable decomposition layer number according to the current sampling frequency;

[0017] S122: performing wavelet packet decomposition on the residual current data to obtain a wavelet packet binary tree, that is, obtaining a node coefficient graph at different current frequencies;

[0018] S123: reconstructing the node coefficient graphs at different frequencies of the current by wavelet packets to obtain the wavelet energy and wavelet Shannon entropy at the nodes of the current at different frequencies.

[0019] Furthermore, step S2 includes:

[0020] S21: Construct a fault discrimination feature system, divide the data set into training set and test set in proportion, select multi-dimensional feature vectors as model input, and use the state codes of different fault types as fault classification model output;

[0021] S22: Preliminarily set hyperparameters based on the fault classification model's performance for pre-training. Based on the feature splitting and leaf-wise splitting strategies of the histogram algorithm, optimize the loss function through multiple rounds of iterations to build a fault classification model.

[0022] S23: Multiple hyperparameters of the fault classification model are optimized simultaneously through the Bayesian optimization algorithm, and the optimal hyperparameter value is output based on whether the model training loss value reaches the minimum as the termination condition.

[0023] Furthermore, step S3 includes:

[0024] S31: Use the test set to test the fault classification model and output the test classification results;

[0025] S32: Based on the test classification results obtained in S31, the fault detection results are obtained in combination with the actual labels of the original data sample set.

[0026] A three-phase grid-connected inverter DC side weak ground fault detection system, comprising:

[0027] A data preprocessing unit is used to preprocess the collected leakage current data and extract the leakage current mechanism characteristics and leakage current statistical characteristics;

[0028] A model training unit is used to build a fault feature system based on the leakage current mechanism characteristics and leakage current statistical characteristics obtained through preprocessing, and generate a fault classification model;

[0029] The fault identification unit is used to classify the test set data using the fault classification model and output the fault detection results.

[0030] Furthermore, the data preprocessing unit is specifically used to:

[0031] Collect leakage current data and perform denoising, normalization and state encoding preprocessing on the leakage current data;

[0032] Wavelet packet transform is used to reconstruct the effective fault signal and extract the wavelet energy and wavelet Shannon entropy;

[0033] Extracting leakage current mechanism characteristics through FFT analysis, wherein the leakage current mechanism characteristics include the leakage current DC component, AC component and the phase difference between current and voltage;

[0034] The leakage current statistical features are extracted through time domain analysis, and the leakage current statistical features include the mean value, standard deviation, maximum value and minimum value of the leakage current.

[0035] Furthermore, the specific steps of wavelet packet transform are as follows:

[0036] Select appropriate wavelet basis functions for residual current data and select appropriate decomposition levels according to current sampling frequency;

[0037] The residual current data is decomposed by wavelet packets to obtain a wavelet packet binary tree, that is, a node coefficient graph at different current frequencies;

[0038] The node coefficient graphs at different frequencies of current are reconstructed by wavelet packets to obtain the wavelet energy and wavelet Shannon entropy at the nodes of current with different frequencies.

[0039] Furthermore, the model training unit is specifically used to:

[0040] Construct a fault discrimination feature system, divide the data set into training set and test set in proportion, select multi-dimensional feature vectors as model input, and use the state codes of different fault types as fault classification model output;

[0041] Preliminary hyperparameters are set based on the performance of the fault classification model for pre-training. Based on the feature splitting and leaf-wise splitting strategies of the histogram algorithm, the loss function is optimized through multiple rounds of iterations to build a fault classification model.

[0042] The Bayesian optimization algorithm is used to simultaneously optimize multiple hyperparameters of the fault classification model, and the optimal hyperparameter value is output based on whether the model training loss value reaches the minimum as the termination condition.

[0043] Furthermore, the fault identification unit is specifically configured to:

[0044] Use the test set to test the fault classification model and output the test classification results;

[0045] Based on the obtained test classification results, the fault detection results are obtained by combining the actual labels of the original data sample set.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention achieves accurate detection of multiple types of weak ground faults on the DC side of photovoltaic grid-connected inverters by introducing machine learning technology and efficient fault feature extraction and classification algorithms. Compared with traditional passive detection systems that rely on current, voltage and ground resistance monitoring, the present invention has significant advantages such as fast response speed, high detection accuracy, low false alarm and missed alarm rates, and strong ability to adapt to complex fault scenarios. By combining wavelet packet transform, FFT analysis and time domain feature extraction, a more comprehensive fault feature system is constructed; combined with the LGBM model and Bayesian optimization algorithm, the classification performance and model efficiency are greatly improved. The present invention not only improves the reliability and flexibility of fault detection, but also provides technical support for the stable operation of complex photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a structural diagram of a system for detecting weak ground faults on the DC side of a three-phase grid-connected inverter according to an embodiment of the present invention;

[0049] Figure 2 This is a flow chart of a method for detecting a weak ground fault on the DC side of a three-phase grid-connected inverter according to an embodiment of the present invention;

[0050] Figure 3 The test confusion matrix and receiver operating characteristic curve of the LGBM1 fault recognition model are shown;

[0051] Figure 4 Test confusion matrix and receiver operating characteristic curve for the LGBM2 fault discrimination model. DETAILED DESCRIPTION

[0052] The present invention will be described in detail below with reference to the following embodiments. It should be noted that the embodiments and features of the embodiments may be combined unless they conflict. For ease of description, the words "upper," "lower," "left," and "right" appearing below merely indicate the directions of upper, lower, left, and right relative to the accompanying drawings and do not limit the structure.

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] Example

[0055] An embodiment of the present invention proposes a system for detecting weak grounding faults on the DC side of a three-phase grid-connected inverter. One of its purposes is to improve the detection accuracy of multiple types of weak grounding faults on the DC side of a photovoltaic grid-connected inverter and reduce false alarms and missed alarms by constructing a comprehensive fault feature system and an optimized machine learning algorithm. A second purpose is to utilize innovative feature extraction and classification technologies to enhance the fault detection system's adaptability to complex scenarios and sudden faults, thereby ensuring the stable operation of the photovoltaic system.

[0056] Due to factors such as the photovoltaic panel's capacitance to ground and insulation damage, the ground impedance of the inverter's DC-side leakage circuit may include multiple operating conditions such as resistance, capacitance, and their series-parallel combinations, such as pure resistance fault leakage, pure capacitance fault leakage, resistance and capacitance series fault leakage, resistance and capacitance parallel fault leakage, etc. The present invention uses the above four typical fault types as examples to analyze their leakage characteristics. Figure 1 This embodiment involves a three-phase grid-connected inverter DC side weak ground fault detection system, including a data preprocessing unit, a model training unit and a fault identification unit. The simulation parameters of this embodiment are shown in Table 1:

[0057] Table 1 Simulation model parameters

[0058]

[0059] Figure 2This is a flow chart of a method for detecting weak ground faults on the DC side of a three-phase grid-connected inverter of the present invention. First, a data preprocessing unit is responsible for preprocessing the collected leakage current data to ensure data quality and validity. Specifically, this includes data denoising, normalization, and state encoding operations to unify the data format. The mechanism characteristics of the fault signal are extracted through wavelet packet transform and FFT analysis. The statistical characteristics of the data are extracted through time domain analysis to provide comprehensive data support for subsequent model training. Then, a model training unit constructs a fault feature system based on the preprocessed data, taking the feature vector as input and the state code of the fault type as output. The system is divided into a training set and a test set in proportion, and pre-trained using the LGBM model. The model performance is gradually optimized using histogram feature splitting and leaf-wise leaf splitting strategies. The hyperparameters are adjusted using the Bayesian optimization algorithm to ultimately generate an efficient and accurate fault classification model. The hyperparameter optimization table is shown in Table 2:

[0060] Table 2 Hyperparameter optimization range

[0061]

[0062] Finally, the fault identification unit uses the trained LGBM model to classify the test data and output the fault detection results. The model performance is verified by combining the actual labels of the original data and calculating the classification accuracy to ensure that the system can accurately identify different types of weak ground faults.

[0063] This embodiment designs two fault diagnosis models: LGBM1 is used to diagnose whether a leakage fault has occurred, and LGBM2 is used to diagnose the type of leakage fault in the inverter and identify the current fault type of the inverter. The feature inputs of LGBM1 and LGBM2 are both 8 mechanism features and 4 statistical features. LGBM1 outputs the diagnosis result, with the target label being 1 for normal and 0 for fault. LGBM2 outputs four fault types, with target labels of 1, 2, 3, and 4, respectively, as shown in Table 3:

[0064] Table 3 Sample distribution and target output

[0065]

[0066] Too few or too many samples can cause model overfitting or compromise performance, reducing diagnostic and classification accuracy. Therefore, in this example, a total of 1250 sample data points were selected for learning and training the fault identification system. These 1250 samples fully reflect the characteristics of all fault types. The training and test sample ratios were as follows: 80% training samples and 20% test samples. Since the test samples are completely independent of network training, they can be used to verify that the training results of the LGBM fault identification system meet expectations.

[0067] Figure 3 The test confusion matrix and receiver operating characteristic curve of the LGBM1 fault recognition model are shown; Figure 4 The test confusion matrix and receiver operating characteristic curve of the LGBM2 fault discrimination model are shown in Figure 2. Figure 3 It shows that in the 250 test sets, the accuracy of fault identification can reach 100%, and the ROC curve is perfectly close to the upper left corner, indicating that the model has a very high accuracy in fault identification. Figure 4 The results show that there was only one misclassification in the 200 test sets, achieving a fault classification accuracy of 99.5%. This indicates that the combination of all features provides sufficient information to correctly distinguish each fault type. The ROC curve approaches the upper left corner, indicating that the classifier has strong predictive power for each category and very low error. This indicates that the combination of statistical and mechanistic features provides the most information for fault identification and demonstrates the best classification results.

[0068] The present invention solves the shortcomings of traditional photovoltaic grid-connected inverter fault detection by introducing innovative machine learning technologies, and accurately and efficiently realizes the detection of multiple types of weak grounding faults on the DC side. First, through the data preprocessing step, the leakage current data is denoised, normalized and state-encoded, and the fault mechanism characteristics are extracted using wavelet packet transform and FFT analysis, and statistical features are extracted in combination with time domain analysis, thereby forming a comprehensive feature data set. Secondly, in the model training stage, based on the constructed fault feature system, the LGBM model is adopted, and the classification performance is optimized through histogram feature splitting and Leaf-wise splitting strategies, and the hyperparameters are adjusted in combination with the Bayesian optimization algorithm to construct an efficient classification model. Finally, in the fault identification stage, the performance of the classification model is verified by the test set, and the classification accuracy is calculated in combination with the actual labels to achieve accurate classification and identification of complex faults. The whole process has a rapid response, high accuracy and strong flexibility, which effectively overcomes the shortcomings of traditional methods.

[0069] Traditional photovoltaic grid-connected inverter fault detection systems mainly rely on current, voltage and ground resistance monitoring, focusing on solving open circuit and short circuit faults of passive components, AC grid faults on the inverter side, and power switch faults. These methods mostly use fixed threshold judgments, and have problems such as slow response speed, high false alarm and missed alarm rates, and poor adaptability to complex fault scenarios. However, this embodiment is based on the LGBM algorithm. By constructing a comprehensive fault feature system and combining wavelet packet transform, FFT analysis and time domain statistical feature extraction, it achieves accurate classification and identification of multiple types of weak ground faults on the DC side. Compared with traditional technologies, this method makes full use of the adaptive ability and optimization performance of machine learning, significantly improving the accuracy, flexibility and robustness of fault detection, especially showing higher efficiency and reliability when dealing with complex or sudden faults.

[0070] As photovoltaic systems continue to grow in size and complexity, traditional fault detection systems have struggled to accurately identify various types of weak ground faults on the DC side. This embodiment, through the introduction of machine learning technology, not only addresses the slow response speed and high false alarm and missed alarm rates of traditional methods, but also improves the detection capabilities for complex operating conditions and weak faults. This method can quickly and efficiently identify complex fault types, ensuring the stable operation of photovoltaic grid-connected systems and reducing the impact of faults on system performance, providing technical support for the efficient management and safe operation of photovoltaic systems.

[0071] 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 it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A three-phase grid-connected inverter DC side weak ground fault detection system, characterized in that: The steps include: S1: Preprocess the collected leakage current data to extract leakage current mechanism characteristics and leakage current statistical characteristics; S2: Build a fault feature system based on the leakage current mechanism characteristics and leakage current statistical characteristics obtained by preprocessing, and generate a fault classification model; S3: Use the fault classification model to classify the test set data and output the fault detection results.

2. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 1, characterized in that: Step S1 includes: S11: collecting leakage current data, and performing denoising, normalization and state encoding preprocessing on the leakage current data; S12: Reconstruct the effective fault signal using wavelet packet transform to extract wavelet energy and wavelet Shannon entropy; S13: extracting leakage current mechanism characteristics through FFT analysis, wherein the leakage current mechanism characteristics include a DC component, an AC component, and a phase difference between current and voltage of the leakage current; S14: Extracting leakage current statistical features through time domain analysis, wherein the leakage current statistical features include a mean value, a standard deviation, a maximum value, and a minimum value of the leakage current.

3. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 2, characterized in that: The specific steps of wavelet packet transform in S12 are as follows: S121: Selecting a suitable wavelet basis function for the residual current data and selecting a suitable decomposition layer number according to the current sampling frequency; S122: performing wavelet packet decomposition on the residual current data to obtain a wavelet packet binary tree, that is, obtaining a node coefficient graph at different current frequencies; S123: reconstructing the node coefficient graphs at different frequencies of the current by wavelet packets to obtain the wavelet energy and wavelet Shannon entropy at the nodes of the current at different frequencies.

4. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 1, characterized in that: Step S2 includes: S21: Construct a fault discrimination feature system, divide the data set into training set and test set in proportion, select multi-dimensional feature vectors as model input, and use the state codes of different fault types as fault classification model output; S22: Preliminarily set hyperparameters based on the fault classification model's performance for pre-training. Based on the feature splitting and leaf-wise splitting strategies of the histogram algorithm, optimize the loss function through multiple rounds of iterations to build a fault classification model. S23: Multiple hyperparameters of the fault classification model are optimized simultaneously through the Bayesian optimization algorithm, and the optimal hyperparameter value is output based on whether the model training loss value reaches the minimum as the termination condition.

5. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 1, characterized in that: Step S3 includes: S31: Use the test set to test the fault classification model and output the test classification results; S32: Based on the test classification results obtained in S31, the fault detection results are obtained in combination with the actual labels of the original data sample set.

6. A three-phase grid-connected inverter DC side weak ground fault detection system, characterized in that: include: A data preprocessing unit is used to preprocess the collected leakage current data and extract the leakage current mechanism characteristics and leakage current statistical characteristics; A model training unit is used to build a fault feature system based on the leakage current mechanism characteristics and leakage current statistical characteristics obtained through preprocessing, and generate a fault classification model; The fault identification unit is used to classify the test set data using the fault classification model and output the fault detection results.

7. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 6, characterized in that: The data preprocessing unit is specifically used to: Collect leakage current data and perform denoising, normalization and state encoding preprocessing on the leakage current data; Wavelet packet transform is used to reconstruct the effective fault signal and extract the wavelet energy and wavelet Shannon entropy; Extracting leakage current mechanism characteristics through FFT analysis, wherein the leakage current mechanism characteristics include the leakage current DC component, AC component and the phase difference between current and voltage; The leakage current statistical features are extracted through time domain analysis, and the leakage current statistical features include the mean value, standard deviation, maximum value and minimum value of the leakage current.

8. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 7, characterized in that: The specific steps of wavelet packet transform are as follows: Select appropriate wavelet basis functions for residual current data and select appropriate decomposition levels according to current sampling frequency; The residual current data is decomposed by wavelet packets to obtain a wavelet packet binary tree, that is, a node coefficient graph at different current frequencies; The node coefficient graphs at different frequencies of current are reconstructed by wavelet packets to obtain the wavelet energy and wavelet Shannon entropy at the nodes of current with different frequencies.

9. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 6, characterized in that: The model training unit is specifically used to: Construct a fault discrimination feature system, divide the data set into training set and test set in proportion, select multi-dimensional feature vectors as model input, and use the state codes of different fault types as fault classification model output; Preliminary hyperparameters are set based on the performance of the fault classification model for pre-training. Based on the feature splitting and leaf-wise splitting strategies of the histogram algorithm, the loss function is optimized through multiple rounds of iterations to build a fault classification model. The Bayesian optimization algorithm is used to simultaneously optimize multiple hyperparameters of the fault classification model, and the optimal hyperparameter value is output based on whether the model training loss value reaches the minimum as the termination condition.

10. The three-phase grid-connected inverter DC side weak ground fault detection system according to claim 6, characterized in that: The fault identification unit is specifically used to: Use the test set to test the fault classification model and output the test classification results; Based on the obtained test classification results, the fault detection results are obtained by combining the actual labels of the original data sample set.

Images