CN-121299361-B - Intelligent fault identification method and control system of photovoltaic grid-connected system

CN121299361BCN 121299361 BCN121299361 BCN 121299361BCN-121299361-B

Abstract

The invention relates to an intelligent fault identification method and a control system of a photovoltaic grid-connected system, wherein multichannel signals are collected and processed to obtain a sample set, an optimal parameter of variation modal decomposition is found by a frost and ice optimization algorithm, modal components of the processed multichannel signals are decomposed by an improved fuzzy multichannel variation modal decomposition strategy, characteristics are extracted from the multichannel signals decomposed by the modal components and then input into a fault classification network based on a Transformer-BiLSTM to obtain a fault identification result, a receiving unit receives a fault category, a control unit invokes a preset regulation strategy corresponding to the fault category, and an output unit outputs control parameters of an energy storage system and an inverter corresponding to the control strategy. The method improves the accuracy and the robustness of subsequent identification, remarkably improves the stability and the characteristic separability of modal decomposition, still maintains high stability and generalization capability under complex grid-connected disturbance, and improves the reliability and the electric energy quality of the system.

Inventors

ZHANG YOUBING
WANG JIANJUN
WU HAO
YUAN JIANXIANG
WENG GUOQING
WANG GUOFENG
WANG LICHENG
LIU QIANGQIANG
DU CHAO
BAO JIANFEI
ZHANG WEIJIE
WANG ZEZHOU
SHI JIANXUN
ZHOU AIHUA
GAO HUIXIN
XIE LUYAO
Shao changzheng
SHEN KEJU

Assignees

浙江工业大学
国网浙江省电力有限公司海盐县供电公司
浙江涵普电力科技有限公司
中国电力科学研究院有限公司
重庆大学
正泰集团股份有限公司
国网浙江省电力有限公司慈溪市供电公司
国网浙江省电力有限公司绍兴市上虞区供电公司

Dates

Publication Date: 20260505
Application Date: 20251212

Claims (6)

1. The intelligent fault identification method of the photovoltaic grid-connected system is characterized by comprising the steps of collecting multichannel signals and processing the multichannel signals to obtain a sample set; Order the The fitness function is defined as, , Wherein K is the initial target of the frost ice optimization algorithm, In order to blur the number of clusters, In order to penalize the coefficients, An error is reconstructed for the signal and, N is the total number of data points of the signal, The signal is reconstructed and the signal is then processed, In order to decompose the mode frequency uniformity index, , For each modal centre frequency Difference between Is set in the standard deviation of (2), Penalty items for modality availability; for the weighting coefficient, satisfy ; Updating the population by using a frost ice optimization algorithm to obtain an optimal solution and obtain an optimal parameter of variation modal decomposition; Dynamic setting of adaptive decomposition modal number of each cluster based on fuzzy clustering Comprising the following steps: Performing VMD on the signal of each channel with the optimal parameters of the variational modal decomposition, and obtaining a group of initial modal components including the center frequency of the c-th channel and the K-th modal component by using the initial modal number K optimized by the frost-ice algorithm ; Defining a frequency-energy joint feature vector based on the initial modal component , , Wherein, the ; Calculating membership matrix by adopting fuzzy C-means clustering algorithm , , Wherein, the Is the first A cluster-like center, wherein the cluster-like center, Is that Is used for the indexing of (a), As a result of the blurring factor, ; Adaptive decomposition modality number per cluster The method can be used for solving the problems that, , Wherein, the Is a membership threshold; summing all initial modes judged to belong to the same cluster by fuzzy clustering in the channels thereof through mode recombination, synthesizing a virtual channel signal representing the cluster characteristics, executing MVMD on the virtual channel to adaptively decompose the mode number As a constraint, output And ; Improved fuzzy multichannel variational modal decomposition strategy for processing multichannel signals Decompose into a plurality of modal components with different center frequencies so that the first Individual channel signals The method can be used for solving the problems that, , In order to blur the number of clusters, For the number of decomposition modes per cluster, And Respectively is And Is used for the indexing of (a), Is the first Fuzzy cluster, the first Signal channels, the first The optimized objective function of the improved fuzzy multichannel variational modal decomposition strategy is, , Wherein, the Is the first Fuzzy cluster, the first Signal channels, the first The center frequency of the modal components, C is the total number of signal channels, C is the index of C, In order to penalize the coefficients, For time of Is used for the partial derivative of (a), Is a dirac delta function; And extracting features from the multi-channel signals after the modal component decomposition, and inputting the extracted features into a fault classification network based on a transducer-BiLSTM to obtain a fault recognition result.
2. The intelligent fault identification method of the photovoltaic grid-connected system according to claim 1, wherein the multichannel signals are operation electric signals of a photovoltaic array, an inverter and a power grid side of the photovoltaic grid-connected system, and a signal matrix with a uniform format is obtained as a sample set through 3 sigma filtering, normalization and downsampling.
3. The intelligent fault identification method of the photovoltaic grid-connected system according to claim 1, wherein the characteristics extracted from the multichannel signals after the modal component decomposition comprise IMF energy characteristics, time domain statistical characteristics and frequency domain characteristics; and performing dimension reduction treatment on all the features to obtain a new dimension-reduced feature matrix.
4. The intelligent fault identification method of the photovoltaic grid-connected system according to claim 1, wherein the fault classification network based on the converter-BiLSTM comprises a position embedding layer, a bidirectional LSTM layer, a multi-head self-attention layer, a Dropout layer, a full-connection layer, an activation function and an output layer which are sequentially arranged; output layer output failure category 。
5. A control system of an intelligent fault identification method based on a photovoltaic grid-connected system as set forth in any one of claims 1 to 4, wherein the system comprises: a receiving unit for receiving the fault category ; The control unit is used for calling a preset regulation strategy corresponding to the fault type; and the output unit is used for outputting control parameters of the energy storage system and the inverter corresponding to the control strategy.
6. The control system of claim 5, wherein the system further comprises a dynamic feedback unit for collecting power quality feedback indexes, and dynamically adjusting the control parameters output by the output unit, wherein the power quality feedback indexes comprise DC bus voltage stability, current total harmonic distortion and three-phase current imbalance.

Description

Intelligent fault identification method and control system of photovoltaic grid-connected system Technical Field The invention relates to the technical field of measuring electric variables and magnetic variables, in particular to an intelligent fault identification method and a control system of a photovoltaic grid-connected system in the fields of photovoltaic power generation technology and intelligent power grid control. Background With the continuous improvement of the duty ratio of the photovoltaic power generation in the energy structure, the stable operation and the electric energy quality guarantee of the photovoltaic grid-connected system become key problems of industry attention. Photovoltaic systems are typically nonlinear, dynamic and strongly coupled in nature, and their operating states are susceptible to multiple factors such as environmental changes, device aging, and grid disturbances, exhibiting significant time-variability and uncertainty. In practical conditions, the system may face multiple types of disturbances from the photovoltaic array, the direct-current side circuit, the power converter and the power grid side at the same time, and these disturbances are often superimposed on each other to form a complex fault mode, which poses serious challenges to the reliability and operation safety of the system. Currently, fault diagnosis research of a photovoltaic system mainly focuses on three types of methods, namely signal processing, machine learning and deep learning. The traditional signal processing technology, such as empirical mode decomposition, is easy to generate mode aliasing when dealing with non-stationary signals, and has poor adaptability due to expert experience, and the machine learning methods such as Principal Component Analysis (PCA), support Vector Machine (SVM) and the like have certain diagnosis capability, but have limited capture of time sequence characteristics and insufficient generalization performance under complex operation environments, and the deep learning method has superior characteristic extraction, but has complex model structure, high calculation resource consumption and difficulty in meeting real-time requirements. More importantly, the prior researches generally treat fault diagnosis and power quality control as two links independent from each other. Such a disjoint results in the diagnosed fault class not being used to guide the execution of a particular control strategy, such that the system does not respond in time in the face of multiple disturbance concurrency or fault evolution situations, with limited suppression and compensation effects. Therefore, a fault identification and regulation method integrating multichannel signal decomposition, multi-scale feature extraction, intelligent parameter optimization and deep learning classification identification is needed, and is used for improving the fault identification accuracy, classification decision precision and power quality regulation capability of a photovoltaic grid-connected system under various disturbance conditions. Disclosure of Invention The invention solves the problems existing in the prior art and provides an intelligent fault identification method and a control system of a photovoltaic grid-connected system. The technical scheme adopted by the invention is that an intelligent fault identification method of a photovoltaic grid-connected system is adopted, and the method is used for collecting and processing multichannel signals to obtain a sample set; After searching the optimal parameters of variation modal decomposition by a frost ice optimization algorithm, decomposing modal components of the processed multichannel signals by an improved fuzzy multichannel variation modal decomposition strategy; And extracting features from the multi-channel signals after the modal component decomposition, and inputting the extracted features into a fault classification network based on a transducer-BiLSTM to obtain a fault recognition result. Preferably, the multichannel signal is a photovoltaic array, an inverter and an operation electric signal at a power grid side of a photovoltaic grid-connected system, and a signal matrix with a uniform format is obtained as a sample set through 3 sigma filtering, normalization and downsampling processing. Preferably, let theThe fitness function is defined as, Wherein K is the initial target of the frost ice optimization algorithm,In order to blur the number of clusters,To penalty coefficients, in general, K,,;An error is reconstructed for the signal and,N is the total data point number of the signal, x i is the original signal,The signal is reconstructed and the signal is then processed,In order to decompose the mode frequency uniformity index,,For each modal centre frequencyDifference betweenIs set in the standard deviation of (2),Penalty items for modality availability; for the weighting coefficient, satisfy ; Updating the population by using a frost ic