WO2019214268A1 - 一种基于复合信息的光伏阵列故障诊断方法 - Google Patents

一种基于复合信息的光伏阵列故障诊断方法 Download PDF

Info

Publication number
WO2019214268A1
WO2019214268A1 PCT/CN2019/000095 CN2019000095W WO2019214268A1 WO 2019214268 A1 WO2019214268 A1 WO 2019214268A1 CN 2019000095 W CN2019000095 W CN 2019000095W WO 2019214268 A1 WO2019214268 A1 WO 2019214268A1
Authority
WO
WIPO (PCT)
Prior art keywords
working state
fault
model
array
image
Prior art date
Application number
PCT/CN2019/000095
Other languages
English (en)
French (fr)
Inventor
邓方
梁泽浪
丁宁
樊欣宇
高欣
蔡烨芸
陈杰
Original Assignee
北京理工大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京理工大学 filed Critical 北京理工大学
Publication of WO2019214268A1 publication Critical patent/WO2019214268A1/zh
Priority to US17/093,244 priority Critical patent/US20210135625A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • H02S50/10—Testing of PV devices, e.g. of PV modules or single PV cells
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/809—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of classification results, e.g. where the classifiers operate on the same input data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10048—Infrared image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/50—Photovoltaic [PV] energy

Definitions

  • the invention relates to the technical field of fault diagnosis, in particular to a method for fault diagnosis of a photovoltaic array based on composite information.
  • the fault diagnosis methods for these faults are sometimes domain reflection method, intelligent algorithm, power comparison method, electrical characteristic detection method and infrared image detection method.
  • the time domain reflection method is similar to the radar detection method, and uses the input signal to enter the input line. When an impedance mismatch occurs, a reflected signal is generated, and the input signal is compared to the reflected signal to detect the fault; the intelligent algorithm collects a large amount of fault data into intelligence. The algorithm provides training, although the effect is better, but the data acquisition has become its biggest obstacle.
  • the power comparison method is simple, it can not locate the fault, and can only judge whether the fault is generated.
  • the electrical characteristic detection method uses the voltage and current sensor pair.
  • the signal is analyzed to achieve fault diagnosis, which requires a large number of sensors to achieve signal collection, so there are significant limitations.
  • the infrared image diagnosis method is based on the fact that there is a certain temperature difference between the normal and abnormal state after the failure of the solar panel, and the infrared image can reflect the temperature difference characteristic of the panel. At the same time, the infrared image can not only realize the fixed point detection of the fault but also the infrared image. It is easy to collect, but the infrared image can only judge whether the solar panel is faulty. There is no way to identify the infrared image of the fault type. In summary, whether it is based on infrared images or text data based on current and voltage, the use of fault information has certain limitations, is not comprehensive enough, and has low accuracy.
  • the present invention provides a photovoltaic array fault diagnosis method based on composite information, which can respectively establish a fault classification model for image data and text data, and combine the two to obtain a PV array fault diagnosis model based on composite information.
  • the comprehensive utilization of fault information is realized, and the accuracy of fault diagnosis is greatly improved.
  • a method for fault diagnosis of photovoltaic array based on composite information comprising:
  • the working state composite information data includes the PV array working state image data and the PV array working state text data.
  • the pre-established fault classification model based on support vector machine is trained by using the PV array working state text data. After the training is completed, the text fault classification model is obtained.
  • the image fault classification model and the text fault classification model are merged by logistic regression algorithm to obtain the fusion model, and the fusion model is trained by using the PV array working state composite information data.
  • the training is completed to obtain the PV array fault diagnosis model based on the composite information. .
  • the working state of the photovoltaic array includes: a normal working state, a hot spot fault, an open circuit fault, and a short circuit fault; and a corresponding label is set for each working state.
  • the photovoltaic array operational state image data includes an infrared image of the photovoltaic array and a photovoltaic array operational status label.
  • the PV array operating state text data includes the open circuit voltage of the photovoltaic component, the short circuit current, the maximum power point voltage, the maximum power point current, the ambient light, the temperature, and a label describing the operational state of the photovoltaic array.
  • the PV array working state composite information data is preprocessed, including preprocessing the volt array working state image data and preprocessing the PV array working state text data.
  • Preprocessing the volt array working state image data includes:
  • the array working state image data is converted into an RGB image, and data normalization processing is performed.
  • the normalized processed array working state image data is processed by the principal component analysis PCA whitening operation.
  • the preprocessing of the PV array working state text data includes: normalizing the data of the PV array working state text data.
  • the depth convolutional neural network fault classification model pre-established by using the image data is trained, and the image fault classification model is obtained after the training is completed, which specifically includes:
  • a deep convolutional neural network fault classification model is established in advance, including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer.
  • the PV array working state image data is used as the training sample image, and a plurality of training sample images are randomly sampled to form a minimum batch file mini-batch input to the input layer of the deep convolutional neural network fault classification model.
  • the input layer inputs the training sample image in the mini-batch of the minimum batch file to the convolutional layer.
  • the convolutional layer has n convolution kernels, n is a set value, and the images entering the convolutional layer are convolutionally filtered by n convolution kernels to extract n neighborhood feature maps.
  • the neighborhood feature map enters the pooling layer, and the pooling layer uses the maximum pooling technique to sample the neighborhood feature map to obtain a new feature map.
  • the deep convolutional neural network fault classification model has a set number of convolutional layers and pooling layers, and the new feature map obtained by the pooling layer enters the next convolutional layer or enters the fully connected layer.
  • the fully connected layer expands the new feature map into it into a one-dimensional feature vector, which enters the output layer as a training sample image feature.
  • the output layer is a sofimax classifier, which uses the training sample image features and the PV array working state tags in the training sample image to train the softmax classifier, and uses the back propagation algorithm to adjust the deep convolutional neural network model until deep convolution The neural network model satisfies the accuracy threshold or reaches the preset maximum number of iterations to complete the training. After the training is completed, the image fault classification model is obtained.
  • activation function used in the convolution filtering process of the convolutional layer is function linear correction units ReLUs.
  • a logistic regression algorithm to obtain a fusion model, which is specifically:
  • the output of the image fault classification model is The output of the text fault classification model is Both constitute the input of the fusion model
  • the training data set of the fusion model is
  • the weight of the fusion model is trained by the training data set T 3 of the fusion model, and the fusion model after training is used as the fault diagnosis model of the photovoltaic array based on the composite information.
  • the fault diagnosis method based on composite information of photovoltaic arrays proposed by the present invention respectively establishes a fault classification model based on image data and text data, performs fault classification of image data through a deep convolutional neural network, and uses a support vector machine to perform voltage and current.
  • the text data fault classification is represented by the following; finally, the two models are merged by using the logistic regression algorithm, and finally the photovoltaic array fault diagnosis method based on the composite information is realized; the invention can simultaneously classify the fault with the image data and the text data, compared with the traditional
  • the fault diagnosis method utilizes a single type of fault information for fault diagnosis analysis, and the invention can realize the comprehensive utilization of the fault information, breaking the limitation of the traditional technology; the sensitivity and type of the image fault classification model and the text fault classification model for the data Different, the two types of models are merged to increase the robustness of the fault diagnosis model, reduce the dependence on the domain expert knowledge, and improve the accuracy of fault diagnosis.
  • the present invention proposes a photovoltaic array fault diagnosis method based on a deep convolution network and a support vector machine. This method is different from the traditional image processing method for fault diagnosis, but utilizes the powerful feature extraction capability of the deep convolutional neural network. A large number of infrared images are processed, which greatly reduces the expert experience. For the processing of text data, the support vector machine algorithm is also used for efficient fault classification.
  • FIG. 1 is a flowchart of a method for fault diagnosis of a photovoltaic array based on composite information according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a basic structure of a deep convolutional neural network according to an embodiment of the present invention.
  • the present invention provides a method for diagnosing a fault of a photovoltaic array based on composite information.
  • the flow of the method is as shown in FIG. 1 , and specifically includes:
  • the working state composite information data includes the PV array working state image data and the PV array working state text data.
  • the working state of the photovoltaic array includes: a normal working state, a hot spot fault, an open circuit fault, and a short circuit fault; and a corresponding label is set for each working state.
  • the photovoltaic array operational state image data includes an infrared image of the photovoltaic array and a photovoltaic array operational status label.
  • the PV array operating state text data includes the open circuit voltage of the photovoltaic component, the short circuit current, the maximum power point voltage, the maximum power point current, the ambient light, the temperature, and a label describing the operational state of the photovoltaic array.
  • the composite information of the PV array working state is preprocessed, including preprocessing the volt array working state image data and preprocessing the PV array working state text data.
  • the preprocessing of the PV array working state image data includes:
  • S101 Converting the PV array working state image data into an RGB image, and performing data normalization processing; in the embodiment of the present invention, the PV array working state image is an infrared image, and the infrared image is converted into a three-dimensional RGB image, and the image pixel is 160 ⁇ 120, in order to eliminate the dimension effect between feature quantities, data standardization processing is needed to solve the comparability between data indicators. After the original data is processed by data standardization, each indicator is in the same order of magnitude, which is suitable for comprehensive comparative evaluation.
  • the embodiment of the present invention uses the Z-score standardization method as shown in the following equation:
  • the image representing the working state of the original photovoltaic array can be expressed in the form of an image pixel matrix; ⁇ A and ⁇ A respectively represent the mean and standard deviation of the image of the working state of the photovoltaic array.
  • the processed data conforms to the standard normal distribution, ie the mean is 0 and the standard deviation is 1.
  • the preprocessing of the PV array working state text data includes: normalizing the data of the PV array working state text data.
  • the pre-processed text data is pre-processed because the dimensional dimension of voltage, current, temperature, illumination, etc. is not uniform.
  • data standardization processing is needed to resolve data indicators. Comparability. After the original data is processed by data standardization, each indicator is in the same order of magnitude, which is suitable for comprehensive comparative evaluation.
  • the present invention uses the Z-score standardization method as shown in the following equation:
  • the pre-established fault classification model based on support vector machine is trained by using the PV array working state text data. After the training is completed, the text fault classification model is obtained.
  • the training for the image fault classification model adopts the following steps:
  • a deep convolutional neural network fault classification model including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer.
  • the pre-established deep convolutional neural network fault classification model adopts a depth in the embodiment of the present invention.
  • the basic structure of the convolutional neural network is shown in Figure 2, which consists of an input layer, three convolutional layers, two pooling layers, a fully connected layer and an output layer.
  • the convolution layer and the pooling layer are spaced apart, and the convolution layer and the pooling layer have a step size of 2.
  • S22 uses the PV array working state image data as the training sample image, and randomly samples the plurality of training sample images to form a minimum batch file mini-batch input to the input layer of the deep convolutional neural network fault classification model;
  • the input layer inputs the training sample image in the mini-batch of the minimum batch file to the convolution layer;
  • the convolutional layer has n convolution kernels, n is a set value, and the images entering the convolutional layer are convolutionally filtered by n convolution kernels to extract n neighborhood feature maps.
  • the image entering the convolution layer may be a training sample image entered from the input layer, or may be a new feature image entered through the pooling layer of the upper layer.
  • the number of convolution kernels set by different convolution layers is different, and the more convolution kernels are obtained, the more the feature maps are obtained, and the better the learning ability of the model is, the better the recognition effect is, but the convolution kernels are More will increase the complexity of the network, increase the complexity of the calculation, the sample size used in this example is small, so the selected convolution kernel is larger.
  • the activation function used in convolution filtering is ReLUs (function linear correction unit), which is nonlinear, unsaturated and unilaterally suppressed, relatively wide excitatory boundary and sparse relative to sigmoid and tanh functions.
  • the activated features allow them to achieve better results in training.
  • the neighborhood feature map enters the pooling layer, and the pooling layer uses the maximum pooling technique to sample the neighborhood feature map to obtain a new feature map.
  • the role of the pool layer is mainly to reduce the feature dimension by sampling according to the invariance of the image neighborhood feature. The number, and so that the sampled features can maintain some invariance (rotation, translation, scaling, etc.), can effectively reduce the computational complexity and prevent overfitting (the activation function of the pooling layer still uses ReLUs).
  • the deep convolutional neural network fault classification model has a set number of convolutional layers and pooling layers, and the new feature map obtained by the pooling layer enters the next convolutional layer or enters the fully connected layer.
  • the fully connected layer expands the new feature map into a one-dimensional feature vector, which enters the output layer as the training sample image feature. Since the fully connected layer nodes are often large, in order to prevent over-fitting, use dropout in the fully connected layer.
  • the way to make the partial implicit node not work that is, in each iteration, the implicit node selects the partial node as the working node with the probability p, and does not update the implicit node of the dropout when the back propagation update weight is used.
  • the activation function still uses ReLUs.
  • the output layer is a softmax classifier, which uses the training sample image features and the PV array working state tags in the training sample image to train the soffmax classifier, and uses the back propagation algorithm to adjust the deep convolutional neural network model until deep convolution The neural network model satisfies the accuracy threshold or reaches the preset maximum number of iterations to complete the training. After the training is completed, the image fault classification model is obtained.
  • the invention proposes a photovoltaic array fault diagnosis method based on a deep convolution network and a support vector machine, which is different from the traditional image processing method for fault diagnosis, and utilizes the powerful feature extraction capability of the deep convolutional neural network for a large amount of infrared
  • the image is processed, which greatly reduces the expert experience.
  • the support vector machine algorithm is also used for efficient fault classification.
  • a nonlinear fault support vector machine learning algorithm is used in the art to establish a text fault classification model.
  • a specific implementation form of a common nonlinear support vector machine learning algorithm is presented to prove the algorithm. It is possible to do so without limiting the invention. Specifically:
  • the pre-processed training data set is:
  • K(x,z) is a positive definite kernel function
  • the above formula is a convex quadratic programming problem, and the solution exists
  • ⁇ * and b * are the parameters of the classification decision function, that is, the weight of the text data classification model.
  • the image fault classification model and the text fault classification model are merged by logistic regression algorithm to obtain the fusion model, and the fusion model is trained by using the PV array working state composite information data.
  • the training is completed to obtain the PV array fault diagnosis model based on the composite information. .
  • S3 is specifically:
  • the output of the image fault classification model is The output of the text fault classification model is Both constitute the input of the fusion model
  • the training data set of the fusion model is
  • the weight of the fusion model is trained by the training data set T 3 of the fusion model, and the fusion model after training is used as the fault diagnosis model of the photovoltaic array based on the composite information.
  • the fault diagnosis method based on composite information of photovoltaic arrays proposed by the invention respectively establishes a fault classification model based on image data and text data, and performs fault classification of image data through a deep convolutional neural network, and uses a support vector machine to represent voltage and current. Text data fault classification; finally, the two models are merged by logistic regression algorithm, and finally the photovoltaic array fault diagnosis method based on composite information is realized; the invention can simultaneously classify faults for image data and text data, compared with traditional fault diagnosis.
  • the method utilizes a single type of fault information for fault diagnosis analysis.
  • the present invention can fully utilize the fault information and breaks the limitation of the conventional technology; since the image fault classification model and the text fault classification model are different in sensitivity and type to the data, The fusion of the two types of models increases the robustness of the fault diagnosis model, reduces the dependence on the domain expert knowledge, and improves the accuracy of fault diagnosis.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种基于复合信息的光伏阵列故障诊断方法，属于故障诊断技术领域。该方法包括：采集光伏阵列工作状态复合信息数据并进行预处理，工作状态复合信息数据包括光伏阵列工作状态图像数据以及光伏阵列工作状态文本数据；利用光伏阵列工作状态图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型；利用光伏阵列工作状态文本数据训练预先建立的基于支持向量机的故障分类模型，训练完成后得到文本故障分类模型；将图像故障分类模型和文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，并利用光伏阵列工作状态复合信息数据对融合模型进行训练，训练完成得到基于复合信息的光伏阵列故障诊断模型。

Description

一种基于复合信息的光伏阵列故障诊断方法 技术领域
本发明涉及故障诊断技术领域，具体涉及一种基于复合信息的光伏阵列故障诊断方法。
背景技术
随着煤炭，石油和天然气等化石燃料的急剧增加，不可再生资源的总有一天会耗尽，而且，化石燃料的燃烧会产生大量的有害气体，对人类生存环境有很大的危害。因此太阳能作为一种可再生资源，因其取之不尽，用之不竭，清洁环保，不受地域因素限制，成为了目前最理想的可再生能源。
太阳能光伏技术的发展带来了巨大的经济效益，但是在实际应用中太阳能光伏阵列由于制造或者是环境的原因会产生各种类型的故障，目前光伏阵列的故障主要有三种：太阳能电池板的裂片问题，长期使用导致的老化问题以及光伏阵列的热斑现象。
目前针对这些故障的故障诊断方法有时域反射法，智能算法，功率对比法，电特性检测法以及红外图像检测法。时域反射法类似于雷达检测方法，利用输入信号进入输入线，当出现阻抗不匹配是会产生反射信号，通过比较输入信号于反射信号来检测故障；智能算法是通过采集大量的故障数据为智能算法提供训练，虽然效果比较好，但是数据的采集成为了其最大的阻碍；功率对比法虽然方法简单，但是其无法定位故障，只能判断故障是否产生；电特性检测法是利用电压电流传感器对信号进行分析来实现故障诊断，其需要大量的传感器才能实现信号的收集，所以有很大的局限性。红外图像诊断法是根据太阳能电池板发生故障后正常与非正常状态下会存在一定的温差，而红外图像恰好能反映 电池板的温差特性，同时，红外图像不仅可以实现故障的定点检测而且红外图像易于采集，但是红外图像只能对于太阳能电池板是否发生故障做出判断，对于故障类型的判断红外图像没有办法识别。综上，不管是基于红外图像还是基于电流电压等文本数据，对于故障信息的利用都都有一定的局限性、不够全面，准确率低。
发明内容
有鉴于此，本发明提供了一种基于复合信息的光伏阵列故障诊断方法，能够针对图像数据和文本数据分别建立故障分类模型，并将二者融合，获得基于复合信息的光伏阵列故障诊断模型，实现了对故障信息的全面利用，大大提高了故障诊断准确率。
为达到上述目的，本发明的技术方案为：
一种基于复合信息的光伏阵列故障诊断方法，该方法包括：
S1、采集光伏阵列工作状态复合信息数据并进行预处理，工作状态复合信息数据包括光伏阵列工作状态图像数据以及光伏阵列工作状态文本数据。
S2、利用光伏阵列工作状态图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型。
利用光伏阵列工作状态文本数据训练预先建立的基于支持向量机的故障分类模型，训练完成后得到文本故障分类模型。
S3、将图像故障分类模型和文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，并利用光伏阵列工作状态复合信息数据对融合模型进行训练，训练完成得到基于复合信息的光伏阵列故障诊断模型。
进一步地，光伏阵列工作状态包括：正常工作状态、热斑故障、开路故障以及短路故障；为每个工作状态设置相应标签。
光伏阵列工作状态图像数据包括光伏阵列的红外图像以及光伏阵列工作状态标签。
光伏阵列工作状态文本数据包括光伏组件的开路电压、短路电流、最大功率点电压、最大功率点电流、环境光照、温度以及描述光伏阵列工作状态的标签。
进一步地，对光伏阵列工作状态复合信息数据进行预处理，包括对伏阵列工作状态图像数据进行预处理以及对光伏阵列工作状态文本数据进行预处理。
对伏阵列工作状态图像数据进行预处理包括：
将阵列工作状态图像数据转换为RGB图像，并进行数据标准化处理。
采用主成分分析法PCA白化操作对标准化处理后的阵列工作状态图像数据进行处理。
对光伏阵列工作状态文本数据进行预处理包括：将光伏阵列工作状态文本数据进行数据标准化处理。
进一步地，述利用图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型，具体包括：
预先建立深度卷积神经网络故障分类模型，包括输入层、卷积层、池化层、全连接层和输出层。
以光伏阵列工作状态图像数据作为训练样本图像，随机采样多个训练样本图像构成一个最小批处理文件mini-batch输入至深度卷积神经网络故障分类模型的输入层。
输入层将最小批处理文件mini-batch中的训练样本图像输入至卷积层。
卷积层中具有n个卷积核，n为设定数值，利用n个卷积核对进入卷积层的图像进行卷积滤波提取到n个邻域特征图。
邻域特征图进入池化层，池化层采用最大池化技术对邻域特征图进行采样 获得新特征图。
深度卷积神经网络故障分类模型中具备设定数量的卷积层和池化层，池化层得到的新特征图进入下一卷积层或者进入全连接层。
全连接层将进入其中的新特征图展开为一维特征向量，作为训练样本图像特征进入输出层。
输出层为sofimax分类器，利用训练样本图像特征以及训练样本图像中的光伏阵列工作状态标签对softmax分类器进行训练，并利用反向传播算法对深度卷积神经网络模型进行调整，直到深度卷积神经网络模型满足准确率阈值或者达到预设的最大迭代次数完成训练，训练完成后得到图像故障分类模型。
进一步地，卷积层的卷积滤波过程中、池化层的最大池化技术中、以及全连接层中采用的激活函数均为函数线性修正单元ReLUs。
进一步地，将图像故障分类模型和文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，具体为：
图像故障分类模型的输出结果为
Figure PCTCN2019000095-appb-000001
文本故障分类模型的输出结果为
Figure PCTCN2019000095-appb-000002
两者构成融合模型的输入
Figure PCTCN2019000095-appb-000003
其中S1采集的光伏阵列工作状态复合信息数据数量为N；i＝1，2，...，N。
假设
Figure PCTCN2019000095-appb-000004
y i∈{0，1，2，3}，y i等于0时为正常工作状态，yi等于1时为热斑故障，y i等于2时为开路故障，y i等于3时为短路故障。
融合模型的训练数据集为
Figure PCTCN2019000095-appb-000005
采用逻辑回归算法建立多项逻辑回归模型，作为融合模型：
Figure PCTCN2019000095-appb-000006
其中k＝1，2，...K-1，K＝4，x∈R n+1，w k∈R n+1，w k为融合模型的权值。
利用融合模型的训练数据集T 3对融合模型的权值进行训练，训练完成后的 融合模型作为基于复合信息的光伏阵列故障诊断模型。
有益效果：
(1)本发明提出的基于光伏阵列复合信息的故障诊断方法，分别建立基于图像数据、文本数据的故障分类模型，通过深度卷积神经网络进行图像数据的故障分类，利用支持向量机进行电压电流为代表的文本数据故障分类；最后利用逻辑回归算法对两个模型进行融合，最终实现基于复合信息的光伏阵列故障诊断方法；本发明能够针对图像数据和文本数据同时进行故障分类，相比于传统的故障诊断方法利用单一类型故障信息进行故障诊断分析，本发明能够实现对故障信息的全面利用，打破了传统技术的局限性；由于图像故障分类模型和文本故障分类模型对于数据的敏感程度和类型不相同，把两类模型进行融合，加大故障诊断模型的鲁棒性，减少了对领域专家知识的依赖，提高了故障诊断的准确率。
(2)本发明提出基于深度卷积网络和支持向量机的光伏阵列故障诊断方法，该方法不同于传统的利用图像处理方法实现故障诊断，而是利用深度卷积神经网络强大的特征提取能力对大量的红外图像进行处理，从而大大减少了对专家经验，对于文本数据的处理也利用支持向量机算法进行高效的故障分类。
附图说明
图1为本发明实施例提出的一种基于复合信息的光伏阵列故障诊断方法流程图；
图2为本发明实施例提供的深度卷积神经网络基本结构示意图。
具体实施方式
下面结合附图并举实施例，对本发明进行详细描述。
本发明提供了一种基于复合信息的光伏阵列故障诊断方法，该方法流程如图1所示，具体包括：
S1、采集光伏阵列工作状态复合信息数据并进行预处理，工作状态复合信息数据包括光伏阵列工作状态图像数据以及光伏阵列工作状态文本数据。
本发明实施例中，光伏阵列工作状态包括：正常工作状态、热斑故障、开路故障以及短路故障；为每个工作状态设置相应标签。
光伏阵列工作状态图像数据包括光伏阵列的红外图像以及光伏阵列工作状态标签。
光伏阵列工作状态文本数据包括光伏组件的开路电压、短路电流、最大功率点电压、最大功率点电流、环境光照、温度以及描述光伏阵列工作状态的标签。
本发明实施例中针对光伏阵列工作状态复合信息数据进行预处理，包括对伏阵列工作状态图像数据进行预处理以及对光伏阵列工作状态文本数据进行预处理。
其中对光伏阵列工作状态图像数据进行预处理包括：
S101、将光伏阵列工作状态图像数据转换为RGB图像，并进行数据标准化处理；本发明实施例中采集到的光伏阵列工作状态图像为红外图像，将红外图像转换为三维的RGB图像，图片像素为160×120，为了消除特征量之间的量纲影响，需要进行数据标准化处理，以解决数据指标之间的可比性。原始数据经过数据标准化处理后，各指标处于同一数量级，适合进行综合对比评价。本发明实施例使用Z-score标准化方法，如下式所示：
Figure PCTCN2019000095-appb-000007
式中
Figure PCTCN2019000095-appb-000008
表示预处理后的光伏阵列工作状态图像；
Figure PCTCN2019000095-appb-000009
表示原始光伏阵列工作 状态图像，可以表示成图像像素矩阵的形式；μ A，σ A分别表示的是光伏阵列工作状态图像的均值和标准差。经过处理后的数据符合标准正态分布，即均值为0，标准差为1。
S102、采用主成分分析法PCA白化操作对标准化处理后的阵列工作状态图像数据进行处理；
对光伏阵列工作状态文本数据进行预处理包括：将光伏阵列工作状态文本数据进行数据标准化处理。
对采集到的文本数据进行预处理，因为电压、电流、温度、光照等数据量纲量纲不统一，为了消除特征量之间的量纲影响，需要进行数据标准化处理，以解决数据指标之间的可比性。原始数据经过数据标准化处理后，各指标处于同一数量级，适合进行综合对比评价。本发明使用Z-score标准化方法，如下式所示：
Figure PCTCN2019000095-appb-000010
式中
Figure PCTCN2019000095-appb-000011
表示预处理后的文本数据，
Figure PCTCN2019000095-appb-000012
表示原始文本数据，μ B，σ B分别表示的是原始文本数据的均值和标准差。经过处理后的文本数据符合标准正态分布，即均值为0，标准差为1。
S2、利用光伏阵列工作状态图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型。
利用光伏阵列工作状态文本数据训练预先建立的基于支持向量机的故障分类模型，训练完成后得到文本故障分类模型。
本发明实施例中，针对图像故障分类模型的训练采用如下步骤：
S21、预先建立深度卷积神经网络故障分类模型，包括输入层、卷积层、池化层、全连接层和输出层；本发明实施例中预先建立的深度卷积神经网络故障 分类模型采用深度卷积神经网络的基本构造如图2所示，即包括一个输入层，三个卷积层，两个池化层，一个全连接层和一个输出层。其中卷积层和池化层间隔设置，且卷积层和池化层步长均为2。
S22以光伏阵列工作状态图像数据作为训练样本图像，随机采样多个训练样本图像构成一个最小批处理文件mini-batch输入至深度卷积神经网络故障分类模型的输入层；
输入层将最小批处理文件mini-batch中的训练样本图像输入至卷积层；
卷积层中具有n个卷积核，n为设定数值，利用n个卷积核对进入卷积层的图像进行卷积滤波提取到n个邻域特征图。本发明实施例中，进入卷积层的图像可能是从输入层进入的训练样本图像，也可能是通过上一层的池化层进入的新特征图。本发明实施例中不同的卷积层设置的卷积核个数不同，卷积核越多获得的特征图就越多，模型的学习能力就越强，识别效果越好，但是卷积核过多会增大网络的复杂度，加大了计算的复杂度，本实例中使用的样本量较小，因此选取的卷积核较大。
本发明实施例中在进行卷积滤波时所用得激活函数为ReLUs(函数线性修正单元)，相对于sigmoid和tanh函数来说其非线性，非饱和以及单侧抑制，相对宽阔的兴奋边界以及稀疏激活的特性都使得其能在训练中获得更好的效果。
邻域特征图进入池化层，池化层采用最大池化技术对邻域特征图进行采样获得新特征图；池化层的作用主要是根据图像邻域特征的不变性，利用采样降低特征维数，并使得采样后的特征能都保持某些不变性(旋转、平移、伸缩等)，能够有效地降低计算的复杂度以及防止过拟合(池化层的激活函数依旧使用ReLUs)。
深度卷积神经网络故障分类模型中具备设定数量的卷积层和池化层，池化层得到的新特征图进入下一卷积层或者进入全连接层。
全连接层将进入其中的新特征图展开为一维特征向量，作为训练样本图像特征进入输出层；由于全连接层节点往往很大，因此为了防止过拟合的出现，在全连接层使用dropout方式来使得部分隐含节点不工作，即在每次迭代中使得隐含节点以概率p选择部分节点作为工作的节点，并且在反向传播更新权值时不再更新dropout掉的隐含节点。激活函数仍然使用ReLUs。
输出层为softmax分类器，利用训练样本图像特征以及训练样本图像中的光伏阵列工作状态标签对soffmax分类器进行训练，并利用反向传播算法对深度卷积神经网络模型进行调整，直到深度卷积神经网络模型满足准确率阈值或者达到预设的最大迭代次数完成训练，训练完成后得到图像故障分类模型。
本发明提出基于深度卷积网络和支持向量机的光伏阵列故障诊断方法，该方法不同于传统的利用图像处理方法实现故障诊断，而是利用深度卷积神经网络强大的特征提取能力对大量的红外图像进行处理，从而大大减少了对专家经验，对于文本数据的处理也利用支持向量机算法进行高效的故障分类。
本发明实施例中，采用本领域常规的非线性支持向量机学习算法，建立文本故障分类模型，本发明中给出一种常见的非线性支持向量机学习算法的具体实施形式，以证明该算法可行，并不是为限定本发明。具体为：
S201、预处理后训练数据集为：
Figure PCTCN2019000095-appb-000013
其中
Figure PCTCN2019000095-appb-000014
向量，y i∈Y＝{0，1，2，3}，i＝1，2，...，N，N为文本数据总数；
S202、建立分类决策函数作为文本数据分类模型：
Figure PCTCN2019000095-appb-000015
当K(x，z)是正定核函数时，上式是凸二次规划问题，解是存在的；α *和b *为分类决策函数的参数，即文本数据分类模型的权值。
选取适当的核函数K(x，z)和适当的参数C，构造并求解最优化问题
Figure PCTCN2019000095-appb-000016
Figure PCTCN2019000095-appb-000017
其中0≤α≤C文本分类器模型的权值向量α＝(α 1，α 2，...，α N) T，i＝1，2，...N；求得最优解
Figure PCTCN2019000095-appb-000018
选择α *的一个正分量
Figure PCTCN2019000095-appb-000019
计算
Figure PCTCN2019000095-appb-000020
S3、将图像故障分类模型和文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，并利用光伏阵列工作状态复合信息数据对融合模型进行训练，训练完成得到基于复合信息的光伏阵列故障诊断模型。
S3具体为：
图像故障分类模型的输出结果为
Figure PCTCN2019000095-appb-000021
文本故障分类模型的输出结果为
Figure PCTCN2019000095-appb-000022
两者构成融合模型的输入
Figure PCTCN2019000095-appb-000023
其中S1采集的光伏阵列工作状态复合信息数据数量为N；i＝1，2，...，N；
假设
Figure PCTCN2019000095-appb-000024
y i∈{0，1，2，3}，y i等于0时为正常工作状态，y i等于1时为热斑故障，y i等于2时为开路故障，y i等于3时为短路故障，
融合模型的训练数据集为
Figure PCTCN2019000095-appb-000025
采用逻辑回归算法建立多项逻辑回归模型，作为融合模型：
Figure PCTCN2019000095-appb-000026
其中k＝1，2，...K-1，K＝4，x∈R n+1，w k∈R n+1，w k为融合模型的权值；
利用融合模型的训练数据集T 3对融合模型的权值进行训练，训练完成后的融合模型作为基于复合信息的光伏阵列故障诊断模型。
本发明提出的基于光伏阵列复合信息的故障诊断方法，分别建立基于图像数据、文本数据的故障分类模型，通过深度卷积神经网络进行图像数据的故障 分类，利用支持向量机进行电压电流为代表的文本数据故障分类；最后利用逻辑回归算法对两个模型进行融合，最终实现基于复合信息的光伏阵列故障诊断方法；本发明能够针对图像数据和文本数据同时进行故障分类，相比于传统的故障诊断方法利用单一类型故障信息进行故障诊断分析，本发明能够实现对故障信息的全面利用，打破了传统技术的局限性；由于图像故障分类模型和文本故障分类模型对于数据的敏感程度和类型不相同，把两类模型进行融合，加大故障诊断模型的鲁棒性，减少了对领域专家知识的依赖，提高了故障诊断的准确率。
综上所述，以上仅为本发明的较佳实施例而已，并非用于限定本发明的保护范围。凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (6)

  1. 一种基于复合信息的光伏阵列故障诊断方法，其特征在于，该方法包括：
    S1、采集光伏阵列工作状态复合信息数据并进行预处理，所述工作状态复合信息数据包括光伏阵列工作状态图像数据以及光伏阵列工作状态文本数据；
    S2、利用所述光伏阵列工作状态图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型；
    利用所述光伏阵列工作状态文本数据训练预先建立的基于支持向量机的故障分类模型，训练完成后得到文本故障分类模型；
    S3、将所述图像故障分类模型和所述文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，并利用所述光伏阵列工作状态复合信息数据对所述融合模型进行训练，训练完成得到基于复合信息的光伏阵列故障诊断模型。
  2. 如权利要求1所述的方法，其特征在于，所述光伏阵列工作状态包括：正常工作状态、热斑故障、开路故障以及短路故障；为每个工作状态设置相应标签；
    所述光伏阵列工作状态图像数据包括所述光伏阵列的红外图像以及所述光伏阵列工作状态标签；
    所述光伏阵列工作状态文本数据包括光伏组件的开路电压、短路电流、最大功率点电压、最大功率点电流、环境光照、温度以及描述所述光伏阵列工作状态的标签。
  3. 如权利要求1或2所述的方法，其特征在于，对所述光伏阵列工作状态复合信息数据进行预处理，包括对所述伏阵列工作状态图像数据进行预处理以及对所述光伏阵列工作状态文本数据进行预处理；
    对所述光伏阵列工作状态图像数据进行预处理包括：
    将所述阵列工作状态图像数据转换为RGB图像，并进行数据标准化处理；
    采用主成分分析法PCA白化操作对标准化处理后的所述阵列工作状态图像数据进行处理；
    所述对所述光伏阵列工作状态文本数据进行预处理包括：将所述光伏阵列工作状态文本数据进行数据标准化处理。
  4. 如权利要求3所述的方法，其特征在于，所述利用所述图像数据进行训练预先建立的深度卷积神经网络故障分类模型，训练完成后得到图像故障分类模型，具体包括：
    预先建立深度卷积神经网络故障分类模型，包括输入层、卷积层、池化层、全连接层和输出层；
    以所述光伏阵列工作状态图像数据作为训练样本图像，随机采样多个训练样本图像构成一个最小批处理文件mini-batch输入至所述深度卷积神经网络故障分类模型的输入层；
    所述输入层将所述最小批处理文件mini-batch中的训练样本图像输入至所述卷积层；
    所述卷积层中具有n个卷积核，n为设定数值，利用所述n个卷积核对进入所述卷积层的图像进行卷积滤波提取到n个邻域特征图；
    所述邻域特征图进入所述池化层，所述池化层采用最大池化技术对所述邻域特征图进行采样获得新特征图；
    所述深度卷积神经网络故障分类模型中具备设定数量的卷积层和池化层，池化层得到的所述新特征图进入下一卷积层或者进入全连接层；
    所述全连接层将进入其中的新特征图展开为一维特征向量，作为训练样本图像特征进入所述输出层；
    所述输出层为softmax分类器，利用所述训练样本图像特征以及训练样本图像中的所述光伏阵列工作状态标签对所述softmax分类器进行训练，并利用反向 传播算法对所述深度卷积神经网络模型进行调整，直到所述深度卷积神经网络模型满足准确率阈值或者达到预设的最大迭代次数完成训练，训练完成后得到图像故障分类模型。
  5. 如权利要求4所述的方法，其特征在于，所述卷积层的卷积滤波过程中、所述池化层的最大池化技术中、以及所述全连接层中采用的激活函数均为函数线性修正单元ReLUs。
  6. 如权利要求3所述的方法，其特征在于，所述将所述图像故障分类模型和所述文本故障分类模型采用逻辑回归算法进行融合，得到融合模型，具体为：
    所述图像故障分类模型的输出结果为
    Figure PCTCN2019000095-appb-100001
    所述文本故障分类模型的输出结果为
    Figure PCTCN2019000095-appb-100002
    两者构成融合模型的输入
    Figure PCTCN2019000095-appb-100003
    其中S1采集的光伏阵列工作状态复合信息数据数量为N；i＝1，2，...，N；
    假设
    Figure PCTCN2019000095-appb-100004
    y i∈{0，1，2，3}，y i等于0时为正常工作状态，y i等于1时为热斑故障，y i等于2时为开路故障，y i等于3时为短路故障；
    融合模型的训练数据集为
    Figure PCTCN2019000095-appb-100005
    采用逻辑回归算法建立多项逻辑回归模型，作为所述融合模型：
    Figure PCTCN2019000095-appb-100006
    其中k＝1，2，...K-1，K＝4，x∈R n+1，w k∈R n+1，w k为所述融合模型的权值；
    利用所述融合模型的训练数据集T 3对所述融合模型的权值进行训练，训练完成后的所述融合模型作为所述基于复合信息的光伏阵列故障诊断模型。
PCT/CN2019/000095 2018-05-09 2019-05-07 一种基于复合信息的光伏阵列故障诊断方法 WO2019214268A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/093,244 US20210135625A1 (en) 2018-05-09 2020-11-09 Photovoltaic array fault diagnosis method based on composite information

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810439521.5 2018-05-09
CN201810439521.5A CN108647716B (zh) 2018-05-09 2018-05-09 一种基于复合信息的光伏阵列故障诊断方法

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/093,244 Continuation US20210135625A1 (en) 2018-05-09 2020-11-09 Photovoltaic array fault diagnosis method based on composite information

Publications (1)

Publication Number Publication Date
WO2019214268A1 true WO2019214268A1 (zh) 2019-11-14

Family

ID=63753856

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/000095 WO2019214268A1 (zh) 2018-05-09 2019-05-07 一种基于复合信息的光伏阵列故障诊断方法

Country Status (3)

Country Link
US (1) US20210135625A1 (zh)
CN (1) CN108647716B (zh)
WO (1) WO2019214268A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111026870A (zh) * 2019-12-11 2020-04-17 华北电力大学 一种综合文本分类和图像识别的ict系统故障分析方法
CN111242914A (zh) * 2020-01-09 2020-06-05 武汉博晟信息科技有限公司 基于窗格检测和线性回归算法光伏组件热斑缺陷定位方法
CN111461298A (zh) * 2020-03-26 2020-07-28 广西电网有限责任公司电力科学研究院 一种用于断路器故障识别的卷积神经网络及方法
CN111815560A (zh) * 2020-06-09 2020-10-23 理光软件研究所（北京）有限公司 光伏电站故障检测方法、装置、便携检测设备及存储介质
CN111860184A (zh) * 2020-06-23 2020-10-30 广东省特种设备检测研究院珠海检测院 一种自动扶梯机械故障诊断方法和系统
CN112085071A (zh) * 2020-08-21 2020-12-15 广东电网有限责任公司广州供电局 一种基于边缘计算的配电房设备故障分析预判方法及装置
CN112184711A (zh) * 2020-11-05 2021-01-05 杭州青枭科技有限公司 一种光伏组件缺陷检测和定位方法及系统
CN113504240A (zh) * 2021-07-07 2021-10-15 广东电力信息科技有限公司 基于分层编码和视觉感知的电力设备故障检测方法
CN115498956A (zh) * 2022-09-22 2022-12-20 福州大学 一种光伏阵列串联电弧故障诊断方法
CN116310859A (zh) * 2023-01-16 2023-06-23 自然资源部国土卫星遥感应用中心 一种基于多源遥感数据的光伏阵列故障智能检测方法

Families Citing this family (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108647716B (zh) * 2018-05-09 2021-08-06 北京理工大学 一种基于复合信息的光伏阵列故障诊断方法
KR101958941B1 (ko) * 2018-11-27 2019-03-18 주식회사 나눔에너지 머신러닝 기반 태양광 발전 제어 시스템 및 방법
KR101958474B1 (ko) * 2018-11-27 2019-03-15 주식회사 나눔에너지 태양광 발전 밸런싱 제어 시스템 및 방법
CN109359702A (zh) * 2018-12-14 2019-02-19 福州大学 基于卷积神经网络的光伏阵列故障诊断方法
CN109640335B (zh) * 2019-02-28 2022-02-08 福建师范大学 基于卷积神经网络的无线传感器故障诊断方法
CN110059377B (zh) * 2019-04-02 2022-07-05 西南交通大学 一种基于深度卷积神经网络的燃料电池寿命预测方法
CN110535435B (zh) * 2019-08-12 2021-12-28 华为数字能源技术有限公司 一种光伏电站的电池片检测方法、装置及系统
CN110736547A (zh) * 2019-10-17 2020-01-31 华能海南发电股份有限公司 基于红外成像技术的光伏板故障智能诊断系统
CN111091141B (zh) * 2019-11-21 2022-05-17 浙江大学 一种基于分层Softmax的光伏背板故障诊断方法
CN111191733B (zh) * 2020-01-02 2020-09-29 平安科技（深圳）有限公司 多数据来源的数据融合方法、装置、电子设备及存储介质
CN112232339B (zh) * 2020-10-15 2023-04-07 中国民航大学 基于卷积神经网络的航显设备故障检测方法及监控装置
CN113326881B (zh) * 2021-05-31 2023-02-14 西安思安云创科技有限公司 一种电力变压器故障诊断方法
CN113221839B (zh) * 2021-06-02 2022-02-01 哈尔滨市科佳通用机电股份有限公司 一种货车图像自动识别方法及识别系统
TWI797806B (zh) * 2021-11-01 2023-04-01 中原大學 使用3維卷積神經網路進行光伏陣列的故障分類裝置
CN114266720B (zh) * 2021-11-08 2024-04-23 国网河南省电力公司电力科学研究院 基于混合神经网络的电力导线故障检测方法及系统
CN114137358B (zh) * 2021-11-17 2023-04-28 国网天津市电力公司信息通信公司 一种基于图卷积神经网络的输电线路故障诊断方法
CN115001941B (zh) * 2022-04-20 2023-10-13 广东省电信规划设计院有限公司 一种通信网管故障的确定方法及装置
CN115272850A (zh) * 2022-07-20 2022-11-01 哈尔滨市科佳通用机电股份有限公司 一种铁路货车bab型闸调器拉杆头折断故障识别方法
KR102650607B1 (ko) * 2022-09-29 2024-03-26 김동완 발전량의 증대를 위한 오토인코더 기반 딥러닝 신경망 모델을 이용한 하이브리드 병렬형 전력변환시스템의 이상징후 감지 방법
CN115455730B (zh) * 2022-09-30 2023-06-20 南京工业大学 一种基于完全邻域保持嵌入的光伏组件热斑故障诊断方法
CN115940809B (zh) * 2023-03-09 2023-06-09 深圳市迪晟能源技术有限公司 基于功率数据及视觉分析的太阳能板故障检测方法
CN116205892B (zh) * 2023-03-13 2024-04-05 北京智盟信通科技有限公司 一种评估光伏组串工况的方法及装置
CN116051949B (zh) * 2023-03-28 2023-10-03 苏芯物联技术(南京)有限公司 一种融合图像检测的焊接异常实时诊断方法
CN116541772B (zh) * 2023-05-05 2023-10-10 兰州理工大学 基于多源融合残差网络的级联h桥逆变器故障诊断方法
CN116232222B (zh) * 2023-05-10 2023-09-08 山东科技大学 一种分布式光伏系统云边协同积灰程度监测方法及系统
CN116738323B (zh) * 2023-08-08 2023-10-27 北京全路通信信号研究设计院集团有限公司 铁路信号设备的故障诊断方法、装置、设备及介质
CN116996008B (zh) * 2023-08-10 2024-02-09 杭州市能源集团工程科技有限公司 光伏发电系统的安装方法及其系统
CN117520924B (zh) * 2023-12-29 2024-04-12 国网浙江省电力有限公司舟山供电公司 基于多模态数据的海岛光伏运维故障原因分析方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012166713A2 (en) * 2011-05-31 2012-12-06 Solar Power Technologies, Inc. Automatic monitoring and adjustment of a solar panel array
CN104753461A (zh) * 2015-04-10 2015-07-01 福州大学 基于粒子群优化支持向量机的光伏发电阵列故障诊断与分类方法
CN105955148A (zh) * 2016-07-01 2016-09-21 福州大学 基于ZigBee的光伏组件在线监测与故障诊断方法
CN106230377A (zh) * 2016-07-01 2016-12-14 重庆大学 一种光伏电池板热斑故障检测方法
CN107276531A (zh) * 2017-06-08 2017-10-20 南昌航空大学 一种光伏组件在线故障分级诊断系统及方法
CN108647716A (zh) * 2018-05-09 2018-10-12 北京理工大学 一种基于复合信息的光伏阵列故障诊断方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140278108A1 (en) * 2013-03-13 2014-09-18 Locus Energy, Llc Methods and Systems for Optical Flow Modeling Applications for Wind and Solar Irradiance Forecasting
CN105141255A (zh) * 2015-08-12 2015-12-09 河海大学常州校区 一种光伏阵列故障诊断方法
US11262718B2 (en) * 2017-06-30 2022-03-01 Merit Si, Llc Method and system for managing microgrid assets
US10949676B2 (en) * 2018-04-09 2021-03-16 Accenture Global Solutions Limited Aerial monitoring system and method for identifying and locating object features

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012166713A2 (en) * 2011-05-31 2012-12-06 Solar Power Technologies, Inc. Automatic monitoring and adjustment of a solar panel array
CN104753461A (zh) * 2015-04-10 2015-07-01 福州大学 基于粒子群优化支持向量机的光伏发电阵列故障诊断与分类方法
CN105955148A (zh) * 2016-07-01 2016-09-21 福州大学 基于ZigBee的光伏组件在线监测与故障诊断方法
CN106230377A (zh) * 2016-07-01 2016-12-14 重庆大学 一种光伏电池板热斑故障检测方法
CN107276531A (zh) * 2017-06-08 2017-10-20 南昌航空大学 一种光伏组件在线故障分级诊断系统及方法
CN108647716A (zh) * 2018-05-09 2018-10-12 北京理工大学 一种基于复合信息的光伏阵列故障诊断方法

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111026870A (zh) * 2019-12-11 2020-04-17 华北电力大学 一种综合文本分类和图像识别的ict系统故障分析方法
CN111242914B (zh) * 2020-01-09 2023-09-08 武汉赛摩博晟信息科技有限公司 基于窗格检测和线性回归算法光伏组件热斑缺陷定位方法
CN111242914A (zh) * 2020-01-09 2020-06-05 武汉博晟信息科技有限公司 基于窗格检测和线性回归算法光伏组件热斑缺陷定位方法
CN111461298A (zh) * 2020-03-26 2020-07-28 广西电网有限责任公司电力科学研究院 一种用于断路器故障识别的卷积神经网络及方法
CN111815560A (zh) * 2020-06-09 2020-10-23 理光软件研究所（北京）有限公司 光伏电站故障检测方法、装置、便携检测设备及存储介质
CN111815560B (zh) * 2020-06-09 2024-04-05 理光软件研究所（北京）有限公司 光伏电站故障检测方法、装置、便携检测设备及存储介质
CN111860184A (zh) * 2020-06-23 2020-10-30 广东省特种设备检测研究院珠海检测院 一种自动扶梯机械故障诊断方法和系统
CN111860184B (zh) * 2020-06-23 2024-04-16 广东省特种设备检测研究院珠海检测院 一种自动扶梯机械故障诊断方法和系统
CN112085071A (zh) * 2020-08-21 2020-12-15 广东电网有限责任公司广州供电局 一种基于边缘计算的配电房设备故障分析预判方法及装置
CN112184711A (zh) * 2020-11-05 2021-01-05 杭州青枭科技有限公司 一种光伏组件缺陷检测和定位方法及系统
CN112184711B (zh) * 2020-11-05 2024-04-02 杭州青枭科技有限公司 一种光伏组件缺陷检测和定位方法及系统
CN113504240B (zh) * 2021-07-07 2022-06-21 广东电力信息科技有限公司 基于分层编码和视觉感知的电力设备故障检测方法
CN113504240A (zh) * 2021-07-07 2021-10-15 广东电力信息科技有限公司 基于分层编码和视觉感知的电力设备故障检测方法
CN115498956A (zh) * 2022-09-22 2022-12-20 福州大学 一种光伏阵列串联电弧故障诊断方法
CN116310859B (zh) * 2023-01-16 2023-09-12 自然资源部国土卫星遥感应用中心 一种基于多源遥感数据的光伏阵列故障智能检测方法
CN116310859A (zh) * 2023-01-16 2023-06-23 自然资源部国土卫星遥感应用中心 一种基于多源遥感数据的光伏阵列故障智能检测方法

Also Published As

Publication number Publication date
CN108647716A (zh) 2018-10-12
US20210135625A1 (en) 2021-05-06
CN108647716B (zh) 2021-08-06

Similar Documents

Publication Publication Date Title
WO2019214268A1 (zh) 一种基于复合信息的光伏阵列故障诊断方法
Su et al. Deep learning-based solar-cell manufacturing defect detection with complementary attention network
Zhu et al. Fault diagnosis approach for photovoltaic arrays based on unsupervised sample clustering and probabilistic neural network model
Korkmaz et al. An efficient fault classification method in solar photovoltaic modules using transfer learning and multi-scale convolutional neural network
Gao et al. A novel fault identification method for photovoltaic array via convolutional neural network and residual gated recurrent unit
CN109660206B (zh) 一种基于Wasserstein GAN的光伏阵列故障诊断方法
Juan et al. Photovoltaic cell defect detection model based-on extracted electroluminescence images using SVM classifier
CN113837994B (zh) 一种基于边缘检测卷积神经网络的光伏板缺陷诊断方法
Bakır et al. Automatic detection of deteriorated photovoltaic modules using IRT images and deep learning (CNN, LSTM) strategies
Yu et al. Convolutional neural network with feature reconstruction for monitoring mismatched photovoltaic systems
Et-taleby et al. Applications of machine learning algorithms for photovoltaic fault detection: a review
Sohail et al. Fault detection and computation of power in PV cells under faulty conditions using deep-learning
Mellit et al. TinyML for fault diagnosis of photovoltaic modules using edge impulse platform
Pa et al. A Fault Detection Scheme Utilizing Convolutional Neural Network for PV Solar Panels with High Accuracy
Adhya et al. Stacking ensemble based fault diagnosis approach for improved operation of photovoltaic arrays
Mazlan et al. Convolution Neural Network (CNN) Architectures Analysis for Photovoltaic (PV) Module Defect Images Classification
Fan et al. Hot Spot Detection of Photovoltaic Module Infrared Near-field Image based on Convolutional Neural Network
Al-Otum Deep learning-based automated defect classification in Electroluminescence images of solar panels
Adhya et al. Diagnosis of PV array faults using RUSBoost
Bharathi et al. Enhanced Classification of Faults of Photovoltaic Module Through Generative Adversarial Network
Sandeep et al. Monitoring of PV modules and hotspot detection using TensorFlow
Venkatesh et al. A comparative study on bayes classifier for detecting photovoltaic module visual faults using deep learning features
Gu et al. YOLOv3-MSSA based hot spot defect detection for photovoltaic power stations
Huang et al. Transmission equipment image recognition based on Ensemble Learning
Vo Realization and Verification of Deep Learning Models for Fault Detection and Diagnosis of Photovoltaic Modules

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19799613

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19799613

Country of ref document: EP

Kind code of ref document: A1