A fault diagnosis method and system based on isolated kernel support vector machine
By integrating the isolated kernel support vector machine with the AdaBoost algorithm, the problems of long training time and high misclassification rate of traditional SVM in photovoltaic equipment fault diagnosis are solved, and higher fault diagnosis accuracy and generalization ability are achieved.
Patent Information
- Application Number
- CN202310876702.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-17
- Publication Date
- 2025-09-09
- Estimated Expiration
- 2043-07-17
AI Technical Summary
Traditional support vector machines have problems in photovoltaic equipment fault diagnosis, such as long training time, large computational complexity, and high error rate in abnormal data classification, making it difficult to effectively process large sample data.
The AdaBoost algorithm is used to integrate the isolated kernel support vector machine (ISO-SVM). Through multiple iterations, the sample weights are adjusted to increase the focus on abnormal data, build an integrated classifier, reduce the proportion of misclassified samples, and improve the accuracy of fault diagnosis.
The accuracy and generalization ability of fault diagnosis are improved, the misclassification rate of abnormal data is reduced, and it is suitable for fault diagnosis of photovoltaic equipment.
Smart Images
Figure CN116975727B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method and system based on an isolated kernel support vector machine. Background Art
[0002] The support vector machine (SVM) is an algorithm proposed by Vapnik et al. in the 1990s. It is based on the structural risk minimization principle in statistical learning theory (SLT). SVM is suitable for analyzing small sample problems and effectively addresses issues such as overfitting and underfitting. When the dimensionality of a data sample is too large or linearly inseparable, the SVM can use kernel techniques to map the data to higher dimensions, thereby avoiding the curse of dimensionality. The SVM utilizes kernel functions to separably process the data, mapping linearly inseparable data to higher dimensions. This kernel function eliminates numerous complex mathematical operations, allowing simple kernel function calculations to achieve the same effect. When processing large amounts of data, even though the kernel function avoids complex mathematical operations, it consumes significant memory and time, making it inadequate for large sample sizes and more suitable for small sample sizes.
[0003] Support vector machines have achieved remarkable success in many fields, such as image recognition, text analysis, regression estimation, and fault diagnosis. Photovoltaic equipment diagnosis is particularly important in this area. Photovoltaic equipment converts solar energy into electricity, generating clean energy. However, over time, photovoltaic equipment may experience various faults, leading to reduced power output, decreased efficiency, and even malfunction. Therefore, photovoltaic equipment fault diagnosis is crucial for ensuring the reliability and efficiency of photovoltaic power generation systems. Data-driven methods such as machine learning and deep learning can analyze and process large amounts of photovoltaic equipment data, enabling automated fault diagnosis. This approach can learn patterns and characteristics of equipment failures from data, improving diagnostic accuracy and efficiency.
[0004] The feasibility of the traditional SVM algorithm is explored in photovoltaic equipment fault diagnosis. By using the sensor data and label data of photovoltaic equipment, the SVM model is trained to classify faults and achieve fault diagnosis. Currently, the main development directions of SVM include theory, algorithm level, and practical application. Theoretical aspects mainly focus on: the selection of kernel functions and the adjustment of specific parameters in SVM; the use of SVM for multi-class analysis to handle multi-class data classification problems. Regarding the SVM algorithm level, the training process of SVM has problems such as slow training time, the large number of parameters involved in the SVM algorithm, and the large amount of computation required to be addressed. However, due to the insufficient attention paid to abnormal data during support vector machine classification, the probability of misclassifying abnormal data as normal data is high. Summary of the Invention
[0005] The purpose of the present invention is to provide a fault diagnosis method and system based on isolated kernel support vector machine, thereby improving the accuracy of fault diagnosis.
[0006] To achieve the above object, the present invention provides the following solutions:
[0007] A fault diagnosis method based on isolated kernel support vector machine, comprising:
[0008] Collect test data of the equipment to be diagnosed;
[0009] The detection data of the device to be diagnosed is input into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; the base classifier used by the AdaBoost algorithm is an isolated kernel support vector machine; the fault identification result is whether a fault has occurred or not.
[0010] Optionally, the multiple base classifier training process includes:
[0011] Construct a training set; the sample data in the training set includes input data and label data, the input data is sample detection data, and the label data is the fault result corresponding to the sample detection data;
[0012] Initialize a weight for each sample data, the weight is 1 / N, N represents the number of sample data in the training set;
[0013] Initialize M base classifiers, and use the training set to train each base classifier in turn to obtain M trained base classifiers:
[0014] After the mth base classifier is trained using the training set, the classification error of the trained mth base classifier is calculated based on the weight distribution of the current training set; m ranges from 1 to M;
[0015] Calculate the classifier weight of the mth base classifier after training according to the classification error of the mth base classifier after training;
[0016] The weight distribution of the current training set is updated based on the classifier weight of the mth base classifier.
[0017] Optionally, the classification error of the mth base classifier after training is calculated as follows:
[0018]
[0019] Among them, ε m Represents the classification error of the mth base classifier after training, p() classification error calculation function, G m (x i ) represents the i-th sample data x i Input the classification result of the mth base classifier after training, y i Represents the label data corresponding to the i-th sample data, I() represents the loss function, w mi Represents the weight of the i-th sample data when training the m-th base classifier.
[0020] Optionally, the calculation formula of the classifier weight of the m-th base classifier after training is expressed as:
[0021]
[0022] Among them, α m represents the classifier weight of the m-th base classifier after training.
[0023] Optionally, the weight distribution of the current training set is updated based on the classifier weight of the m-th base classifier, specifically including:
[0024] The weight distribution update formula of the current training set is expressed as:
[0025] D m+1 =(w m+11 ,w m+12 ,w m+13 ,…w m+1N );
[0026]
[0027] Among them, D m+1 Indicates the updated weights, w m+1i It represents the updated value of the weight of the i-th sample data after training the m-th base classifier, Z m is the normalization parameter.
[0028] Optionally, the ensemble classifier is expressed as:
[0029]
[0030] Among them, G(x) represents the integrated classifier, f(x) represents the integration of the mth base classifier after training, M represents the number of base classifiers, α m represents the classifier weight of the mth base classifier after training, G m (x) represents the classification result of the mth base classifier after the detection data x is input and trained.
[0031] Optionally, the device to be diagnosed is a photovoltaic device, and the detection data of the photovoltaic device includes the current of the photovoltaic cell array, the voltage of the photovoltaic cell array, the light intensity, the temperature of the photovoltaic cell array, and the power output of the photovoltaic cell array.
[0032] The present invention discloses a fault diagnosis system based on an isolated kernel support vector machine, comprising:
[0033] A detection data acquisition module is used to collect detection data of the device to be diagnosed;
[0034] The fault diagnosis module is used to input the detection data of the device to be diagnosed into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; the base classifier used by the AdaBoost algorithm is an isolated kernel support vector machine; the fault identification result is whether a fault has occurred or not.
[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0036] The integrated classifier of the present invention is a classifier integrated using the AdaBoost algorithm, wherein the base classifier used in the AdaBoost algorithm is an isolated kernel support vector machine, which improves the defect of too high misclassification ratio of abnormal data, improves the accuracy of fault diagnosis, and also enhances the generalization ability of the final fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0038] Figure 1 A schematic flow chart of a fault diagnosis method based on an isolated kernel support vector machine provided in an embodiment of the present invention;
[0039] Figure 2A schematic diagram of the process of constructing an integrated classifier provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0040] 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0041] The purpose of the present invention is to provide a fault diagnosis method and system based on isolated kernel support vector machine, thereby improving the accuracy of fault diagnosis.
[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] The present invention integrates isolated kernel support vector machine classifiers, constructing a higher-performance ensemble learner using the isolated kernel support vector machine as a base classifier to improve the classifier's fault detection performance. The ensemble learner has stronger generalization capabilities than individual member learners. When using different kernel functions, SVMs have different data sensitivities and classification accuracies. By changing the specific kernel function, an appropriate kernel function can be selected to train the classifier. The present invention employs a new kernel function—the isolated kernel—which relies solely on the data distribution and does not require class information or explicit learning to function as a classifier. In contrast, existing data-dependent kernels rely heavily on class information and explicit learning to generate classifiers. Under a uniform density distribution, the isolated kernel approximates a data-independent kernel function, namely the Laplace kernel. The isolated kernel can be viewed as a data-dependent kernel that adapts a data-independent kernel to the structure of the data set. The AdaBoost algorithm is introduced to address the SVM's sensitivity to partial support vector data. The AdaBoost algorithm improves the classification accuracy of the overall learner through multiple integrations, each time increasing the weight of misclassified data. AdaBoost integration is performed with SVM as the base classifier to increase its sensitivity to misclassified samples, effectively improving the performance of the learner. The weight function of AdaBoost is adjusted to make it better compatible with ISO-SVM (isolated kernel support vector machine).
[0044] Example 1
[0045] like Figure 1 As shown, this embodiment provides a fault diagnosis method based on an isolated kernel support vector machine, which includes the following steps.
[0046] Step 101: Collect detection data of the device to be diagnosed.
[0047] Step 102: Input the detection data of the device to be diagnosed into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; the base classifier used by the AdaBoost algorithm is an isolated kernel support vector machine; the fault identification result is whether a fault has occurred or not.
[0048] like Figure 2 As shown, the process of training multiple base classifiers to form an integrated classifier includes:
[0049] Construct a training set; the sample data in the training set includes input data and label data, the input data is sample test data, and the label data is the fault result corresponding to the sample test data. The fault result is whether a fault has occurred or not.
[0050] A weight is initialized for each sample data, and the weight is 1 / N, that is, the initial weight distribution of the training set is obtained, and N represents the number of sample data in the training set.
[0051] Initialize M base classifiers (determine the number of iterations M), the base classifier adopts an isolated kernel support vector machine, and use the training set to train each base classifier in turn to obtain M trained base classifiers.
[0052] The weight distribution of training samples when training the mth base classifier is expressed as D m (m=1,2,3,4,…M) indicates w mi Then it is the weight of the i-th data in the training sample of the m-th base classifier. Assume that the weight of each data in the training sample is: D1(w 1i ,w 2i ,w 3i ,…w 1N ),
[0053] Use D m (m=1,2,3,4,…M) training set training to obtain the base classifier G m (x):χ→-1, 1. χ represents the weight of each base classifier.
[0054] After the mth base classifier is trained using the training set, the classification error of the trained mth base classifier is calculated based on the weight distribution of the current training set; the value range of m is 1 to M.
[0055] Calculate the classifier weight of the trained m-th base classifier according to the classification error of the trained m-th base classifier.
[0056] The weight distribution of the current training set is updated based on the classifier weight of the mth base classifier. When updating the weight distribution, it is necessary to give larger weights to the misclassified samples. By increasing the attention paid to the misclassified samples, the accuracy of the next (m+1th) base classifier can be improved.
[0057] The weights of the training set are updated according to the calculated classification weight samples to train the next base classifier.
[0058] After training reaches the specified number of iterations (the number of base classifiers M), the weight of each base classifier is calculated according to the error rate, and they are combined into the final ensemble classifier.
[0059] The calculation formula of the classification error of the mth base classifier after training is expressed as:
[0060]
[0061] Among them, ε m Represents the classification error of the mth base classifier after training, p() classification error calculation function, G m (x i ) represents the i-th sample data x i Input the classification result of the mth base classifier after training, y i represents the label data corresponding to the i-th sample data (the actual category to which it belongs), I() represents the loss function, and w mi Represents the weight of the i-th sample data when training the m-th base classifier.
[0062] is a 0, 1 loss function, and x' indicates whether the predicted category is the same as the actual category.
[0063] The calculation formula of the classifier weight of the mth base classifier after training is expressed as:
[0064]
[0065] Among them, α m Represents the classifier weight of the mth base classifier after training. The base classifier needs to ensure that it has a higher than minimum classification accuracy. After training, the base classifier with higher accuracy has a larger weight, and the base classifier with lower accuracy will have a smaller weight.
[0066] Update the weight distribution of the current training set based on the classifier weight of the mth base classifier, specifically including:
[0067] The weight distribution update formula of the current training set is expressed as:
[0068] D m+1 =(wm+11 ,w m+12 ,w m+13 ,…w m+1N );
[0069]
[0070] w m+1i =w m+1i Z m =w mi exp(-α m y i G m (x i ));
[0071] Among them, D m+1 Indicates the updated weights, w m+1i It represents the updated value of the weight of the i-th sample data after training the m-th base classifier, Z m is the normalization parameter. m Will change D m+1 This sample is not misclassified. When training the next base class, the weight of this sample will be increased to gain higher attention. Otherwise, the weight will be decreased:
[0072]
[0073] According to the weight of each base classifier, an integrated classifier is constructed, and the final result is obtained through the activation function. The integrated classifier is expressed as:
[0074]
[0075]
[0076] Among them, G(x) represents the integrated classifier, f(x) represents the integration of the mth base classifier after training, M represents the number of base classifiers, α m represents the classifier weight of the mth base classifier after training, G m (x) represents the classification result of the mth base classifier after the detection data x is input and trained.
[0077] The above is the ISO-SVM-AdaBoost algorithm process. Table 1 shows the ISO-SVM-AdaBoost algorithm pseudocode. The ISO-SVM-AdaBoost algorithm adjusts the training dataset based on the results of the previous base classifier, increasing its focus on misclassified samples, thereby improving the classification accuracy of the next base classifier on these error-prone data. By training multiple base classifiers, the proportion of misclassified samples and abnormal data is reduced. Finally, ISO-SVM-AdaBoost determines the weight of each base classifier based on its classification accuracy and combines all base classifiers into the final ensemble classifier.
[0078] Table 1 ISO-SVM-AdaBoost algorithm pseudo code
[0079]
[0080] In Table 1, (x1, y1), (x2, y2), ..., (x N ,y N ) represents sample data.
[0081] The device to be diagnosed is a photovoltaic device, and the detection data of the photovoltaic device includes the current of the photovoltaic cell array, the voltage of the photovoltaic cell array, the light intensity, the temperature of the photovoltaic cell array and the power output of the photovoltaic cell array; the fault is caused by dust coverage, partial obstruction, hot spot effect, inverter failure or connection failure.
[0082] For photovoltaic equipment, various data can be collected to describe its operating status and performance characteristics. This data may include the following: 1. PV array current and voltage: This data collects current and voltage data for each photovoltaic cell in the array to understand the overall operating status of the array. 2. Light intensity: This data measures light intensity to determine solar energy input. 3. Temperature: This data monitors the temperature of the photovoltaic equipment, including the temperature of components such as the photovoltaic cell array and inverter. 4. Power output: This data records the power output of the photovoltaic equipment. Additionally, this data may include data from other sensors, such as wind speed, humidity, and other parameters related to the photovoltaic equipment.
[0083] For training sample sets used for photovoltaic equipment fault diagnosis, each sample must be labeled with the corresponding fault type. Common fault types in photovoltaic equipment fault diagnosis include: 1. Dust cover: Large amounts of dust or dirt accumulate on the surface of the photovoltaic cell array, affecting the efficiency of solar energy absorption. 2. Partial shading: Parts of the photovoltaic cell are shaded, resulting in reduced battery pack power output. 3. Hot spot effect: Hot spots appear in photovoltaic cells, which may be caused by local damage or uneven operation of the battery pack. 4. Inverter failure: The inverter fails to operate properly, resulting in the photovoltaic device being unable to convert direct current to alternating current. 5. Connection failure: There is a fault or disconnection within the photovoltaic cell array or between the battery array and the inverter. These fault types can be manually labeled based on actual conditions.
[0084] In order for the AdaBoost method to improve algorithm performance, the selected base classifiers should not be too strong. Instead, they should be weak classifiers with relatively weak classification performance, and the classification accuracy of each base classifier should be below a certain threshold. Using ISO-SVM can circumvent the problem of lacking data feature information and effectively improve classification accuracy. However, the accuracy of ISO-SVM in this case is not high, and it is still a weak learner. In this paper, ISO-SVM is used as the base classifier of the AdaBoost algorithm.
[0085] The present invention has the following features: i) The AdaBoost algorithm, using ISO-SVM as the base classifier, can effectively improve the attention paid to a small number of fault samples. ii) The ISO-SVM-AdaBoost algorithm can significantly reduce the false positive rate of the algorithm after multiple iterations.
[0086] Example 2
[0087] This embodiment provides a fault diagnosis system based on an isolated kernel support vector machine, including:
[0088] The detection data acquisition module is used to collect the detection data of the device to be diagnosed.
[0089] The fault diagnosis module is used to input the detection data of the device to be diagnosed into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; the base classifier used by the AdaBoost algorithm is an isolated kernel support vector machine; the fault identification result is whether a fault has occurred or not.
[0090] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.
[0091] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.
Claims
1. A fault diagnosis method based on isolated kernel support vector machine, characterized in that: include: Collecting detection data of the device to be diagnosed; the device to be diagnosed is a photovoltaic device, and the detection data of the photovoltaic device includes the current of the photovoltaic cell array, the voltage of the photovoltaic cell array, the light intensity, the temperature of the photovoltaic cell array and the power output of the photovoltaic cell array; Inputting the detection data of the device to be diagnosed into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; The base classifier used in the AdaBoost algorithm is the isolated kernel support vector machine; The fault identification result is that a fault has occurred or no fault has occurred; The training process of multiple base classifiers includes: Construct a training set; the sample data in the training set includes input data and label data, the input data is sample detection data, and the label data is the fault result corresponding to the sample detection data; Initialize a weight for each sample data, the weight is 1 / N, N represents the number of sample data in the training set; Initialize M base classifiers, and use the training set to train each base classifier in turn to obtain M trained base classifiers: After the mth base classifier is trained using the training set, the classification error of the trained mth base classifier is calculated based on the weight distribution of the current training set; m ranges from 1 to M; Calculate the classifier weight of the mth base classifier after training according to the classification error of the mth base classifier after training; The weight distribution of the current training set is updated based on the classifier weight of the mth base classifier.
2. The fault diagnosis method based on isolated kernel support vector machine according to claim 1, characterized in that: The calculation formula of the classification error of the mth base classifier after training is expressed as: Among them, ε m Represents the classification error of the mth base classifier after training, p() classification error calculation function, G m (x i ) represents the i-th sample data x i Input the classification result of the mth base classifier after training, y i Represents the label data corresponding to the i-th sample data, I() represents the loss function, w mi Represents the weight of the i-th sample data when training the m-th base classifier.
3. The fault diagnosis method based on isolated kernel support vector machine according to claim 2, characterized in that: The calculation formula of the classifier weight of the mth base classifier after training is expressed as: Among them, α m represents the classifier weight of the m-th base classifier after training.
4. The fault diagnosis method based on isolated kernel support vector machine according to claim 2, characterized in that: Update the weight distribution of the current training set based on the classifier weight of the mth base classifier, specifically including: The weight distribution update formula of the current training set is expressed as: D m+1 =(w m+11 ,w m+12 ,w m+13 ,…w m+1N ); Among them, D m+1 Indicates the updated weights, w m+1i It represents the updated value of the weight of the i-th sample data after training the m-th base classifier, Z m is the normalization parameter.
5. The fault diagnosis method based on isolated kernel support vector machine according to claim 1, characterized in that: The ensemble classifier is expressed as: Among them, G(x) represents the integrated classifier, f(x) represents the integration of the mth base classifier after training, M represents the number of base classifiers, α m represents the classifier weight of the mth base classifier after training, G m (x) represents the classification result of the mth base classifier after the detection data x is input and trained.
6. A fault diagnosis system based on isolated kernel support vector machine, characterized in that: include: A detection data acquisition module is used to collect detection data of the device to be diagnosed; the device to be diagnosed is a photovoltaic device, and the detection data of the photovoltaic device includes the current of the photovoltaic cell array, the voltage of the photovoltaic cell array, the light intensity, the temperature of the photovoltaic cell array, and the power output of the photovoltaic cell array; A fault diagnosis module is used to input the detection data of the device to be diagnosed into an integrated classifier for fault classification to obtain a fault identification result; the integrated classifier is a classifier integrated using the AdaBoost algorithm; The base classifier used in the AdaBoost algorithm is the isolated kernel support vector machine; The fault identification result is that a fault has occurred or no fault has occurred; The training process of multiple base classifiers includes: Construct a training set; the sample data in the training set includes input data and label data, the input data is sample detection data, and the label data is the fault result corresponding to the sample detection data; Initialize a weight for each sample data, the weight is 1 / N, N represents the number of sample data in the training set; Initialize M base classifiers, and use the training set to train each base classifier in turn to obtain M trained base classifiers: After the mth base classifier is trained using the training set, the classification error of the trained mth base classifier is calculated based on the weight distribution of the current training set; m ranges from 1 to M; Calculate the classifier weight of the mth base classifier after training according to the classification error of the mth base classifier after training; The weight distribution of the current training set is updated based on the classifier weight of the mth base classifier.