An online monitoring and state evaluation system for photovoltaic grid-connected power generation under transient interference

Abstract

The application discloses a kind of photovoltaic grid-connected power generation on-line monitoring and state evaluation system under transient interference, it is related to grid-connected power generation evaluation technical field, it solves the technical problem that it is difficult to accurately analyze the various nodes of photovoltaic grid-connected power generation under transient interference by ignoring the mutual influence characteristics between regional nodes;By integrating analysis of topological structure, electrical distance, impedance matrix, transient characteristics and other multi-source data, the limitations of single data source are overcome, and it is convenient to dynamically adjust regional division and detection strategy with system topology changes. The sub-region is dynamically divided by the transient characteristics and power characteristics of the initial regional node. Nodes with similar transient characteristics and closely coupled power can be grouped into the same sub-region, avoiding the masking of local characteristics due to the averaging of excessively large regions. By integrating node and regional data, a deep learning evaluation model is trained to facilitate more accurate analysis and evaluation of the status of nodes in the region.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of grid-connected power generation evaluation, and particularly relates to an online monitoring and state evaluation system for photovoltaic grid-connected power generation under transient disturbance. BACKGROUND

[0002] The photovoltaic grid-connected power generation system converts solar energy into direct current through the photoelectric effect of semiconductor materials, and then converts the direct current into alternating current with the same frequency and phase as the power grid through an inverter to realize power grid connection. The system mainly comprises photovoltaic components, an inverter and a power grid access device. When there is no energy storage design, the excess power can be fed back to the power grid. With the rapid increase of photovoltaic penetration, the problem of transient disturbance faced by the grid-connected system is increasingly prominent, which poses a serious challenge to the safe and stable operation of the power grid. Under this background, the online monitoring and state evaluation system for photovoltaic grid-connected power generation under transient disturbance emerges as the times require, and becomes a key technical support for ensuring the reliability of the new power system.

[0003] The existing online monitoring and evaluation technology for photovoltaic grid-connected power generation usually only focuses on a single indicator of power quality or operating efficiency, only processes a single type of data, ignores the mutual influence characteristics between regional nodes, and is difficult to accurately analyze multiple nodes of photovoltaic grid-connected power generation under transient disturbance. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art; for this purpose, the present application proposes an online monitoring and state evaluation system for photovoltaic grid-connected power generation under transient disturbance, which is used to solve the technical problem that the mutual influence characteristics between regional nodes are ignored and it is difficult to accurately analyze multiple nodes of photovoltaic grid-connected power generation under transient disturbance.

[0005] To solve the above problems, the first aspect of the present application provides an online monitoring and state evaluation system for photovoltaic grid-connected power generation under transient disturbance, comprising:

[0006] The data acquisition layer sets photovoltaic grid-connected nodes in the photovoltaic grid-connected power generation area, and real-time monitors the electrical data and environmental data of the grid-connected nodes by the high-precision synchronous measurement sensors deployed at the photovoltaic grid-connected nodes;

[0007] The edge computing layer comprises:

[0008] The photovoltaic grid-connected power generation area is equivalent to a multi-feed line-multi-bus topology structure, and the photovoltaic grid-connected nodes are labeled at the buses of the topology structure. The electrical distance between nodes is quantified according to the data collected in the acquisition layer;

[0009] The spectral clustering algorithm is used to divide the initial area with the node impedance matrix as the input, the voltage sag events of each node in the initial area are detected according to the electromagnetic compatibility standard, and the transient characteristics of each node are recorded;

[0010] According to the transient characteristics and power characteristics of the initial regional nodes, the initial region is divided into sub-regions;

[0011] For the divided sub-regions, by constructing a regional analysis model, the power quality, operation efficiency and environmental adaptability of the sub-regions when a transient event occurs are evaluated.

[0012] The cloud analysis layer: according to the transient characteristics of the nodes and the power quality, operation efficiency and environmental adaptability of the sub-regions when a transient event occurs, an integrated multi-source data fusion algorithm and a deep learning evaluation model are constructed to evaluate the state of the nodes.

[0013] Optionally, in one example of the above aspect, the data acquisition layer comprises:

[0014] The electrical quantity acquisition unit: real-time monitoring of the three-phase voltage / current of the grid-connected point, the DC side voltage / current and the inverter output power;

[0015] The environmental perception unit: monitoring the illumination intensity and temperature of the grid-connected point;

[0016] The transient event triggering unit: setting the triggering threshold through a hardware comparator to detect the transient characteristics in real time, such as voltage sag / surge amplitude, frequency offset and harmonic distortion.

[0017] Optionally, in one example of the above aspect, the photovoltaic grid-connected power generation region is equivalent to a multi-feed line-multi-bus topology structure, and the photovoltaic grid-connected nodes are labeled at the buses of the topology structure. According to the data of the acquisition layer, the electrical distance between nodes is quantified, including the following steps:

[0018] By equivalent the line connecting the photovoltaic array and the grid-connected point to a feed line, and the photovoltaic branch convergence point or the grid-connected point to a bus, the photovoltaic grid-connected power generation region is equivalent to a multi-feed line-multi-bus topology structure;

[0019] The photovoltaic grid-connected node comprises:

[0020] The photovoltaic injection node: located at the bus and the AC side of the inverter;

[0021] The grid connection node: located at the bus and the connection of the upper grid;

[0022] The load node: located at the bus and the connection of the power structure;

[0023] The photovoltaic injection node, the grid connection node and the grid connection node are labeled at the buses of the topology structure, and the nodes are numbered according to the node type;

[0024] According to the data of the acquisition layer, the node impedance matrix is used to quantify the electrical distance between nodes:

[0025]

[0026] wherein, Dij is the electrical distance between node i and node j, Zii is the self-impedance of node i, Zjj is the self-impedance of node j, and Zij is the mutual impedance between node i and j.

[0027] Optionally, in one example of the above aspect, the initial region is divided by using a spectral clustering algorithm with the node impedance matrix as input, including the following steps:

[0028] The node admittance matrix Ybus is formed, and then Zbus=Ybus-1 is obtained by matrix inversion, and the node impedance matrix Zbus is established by the complex impedance between nodes;

[0029] The similarity between nodes is calculated by using a Gaussian kernel function:

[0030]

[0031] wherein, Wij is the similarity between node i and node j, |Zij| is the mutual impedance between node i and j, and σ is a similarity decay coefficient;

[0032] A similarity matrix W is constructed according to the similarity Wij between nodes;

[0033] The Laplacian matrix L=D-W is calculated, wherein D is a degree matrix, and the diagonal elements of the degree matrix are equal to the degrees of the corresponding nodes;

[0034] The eigenvectors corresponding to the first k smallest eigenvalues of L are calculated to form an eigenvector matrix U;

[0035] The eigenvectors corresponding to each node of the eigenvector matrix U are divided into k clusters by using a K-means clustering algorithm, and the initial region division is completed.

[0036] Optionally, in one example of the above aspect, the voltage sag events of each node in the initial region are detected according to the electromagnetic compatibility standard, and the transient characteristics of each node are recorded, including the following steps:

[0037] The voltage sag events of each node in the initial region are detected according to the electromagnetic compatibility standard;

[0038] The transient energy distribution is extracted by using a WPT wavelet packet transform method:

[0039]

[0040] wherein, dj,k is the coefficient of the jth node in the kth subband, Ej is the concentration degree of the transient energy of node j in the frequency domain, k∈(1,2,…,N), and N is the total number of subbands;

[0041] The transient characteristics of each node are recorded by counting the concentration of the transient energy of the node in the frequency domain.

[0042] Optionally, in one example of the above aspect, the initial region is divided into sub-regions according to the transient characteristics and power characteristics of the initial region nodes, including the following steps:

[0043] The nodes of the initial region are divided into sub-branches according to the transient energy distribution of the region nodes and the correlation of the transient characteristics between the nodes of the initial region;

[0044] The historical output data of each sub-branch are collected, and the DTW dynamic time warping distance is calculated, and the branches with similar power characteristics are merged into sub-regions by hierarchical clustering using the DTW distance as input.

[0045] Optionally, in one example of the above aspect, the inner nodes of the initial region are divided into sub-branches according to the transient energy distribution of the region nodes and the correlation of the transient characteristics between the nodes of the initial region, including the following steps:

[0046] The node transient signal is decomposed by wavelet, and the maximum decomposition layer is set, in each layer, the energy value of each frequency band in the layer is calculated, and the energy value of each frequency band is taken as an element of the node energy feature vector to construct the node energy feature vector;

[0047] The vector correlation between two nodes is calculated by calculating the ratio of the covariance of the energy feature vectors of the two nodes and the product of the standard deviations:

[0048] Where ρij is the vector correlation between nodes, Cov(Eni, Enj) is the covariance of the energy feature vector Eni of node i and the energy feature vector Enj of node j, and σEni and σEnj are the standard deviations between the energy feature vector Eni of node i and the energy feature vector Enj of node j.

[0049] The vector correlation and the cosine similarity are weighted and averaged to obtain the correlation between the nodes by calculating the cosine similarity of the energy feature vectors of the two nodes.

[0050] Each node is regarded as an independent cluster by a clustering algorithm.

[0051] The correlation between all clusters is calculated by the correlation calculation results between nodes, and the two clusters with the smallest correlation are merged, and the step is repeated until all nodes are merged into k clusters.

[0052] Each cluster corresponds to a sub-branch, and the nodes in the cluster are members of the sub-branch, forming the secondary divided sub-branch.

[0053] Optionally, in one example of the above aspect, the historical output data of each sub-branch is collected, the DTW dynamic time warping distance is calculated, the sub-regions with similar power characteristics are combined by hierarchical clustering using the DTW distance as input, and the following steps are included:

[0054] By collecting the historical output data of each photovoltaic branch, the transient characteristics of each photovoltaic branch are obtained according to the historical output data;

[0055] The transient characteristics are mapped to the interval [-1, 1] by Min-Max normalization algorithm;

[0056] The power fluctuation correlation analysis is performed by the time series X=(x1,x2,…,xn) and Y=(y1,y2,…,ym) of the transient characteristics between nodes, and the DTW dynamic time warping distance is calculated:

[0057]

[0058] Wherein, DTW(X,Y) is the DTW distance between nodes, X and Y are the power time series of two branches, d(xi,yj) is the Euclidean distance between nodes, n is the total number of time series of transient characteristics of node i, and m is the total number of time series of transient characteristics of node j;

[0059] The DTW distance between all branches is calculated, and a symmetric distance matrix is formed;

[0060] By hierarchical clustering method, each branch is initialized as an independent cluster;

[0061] Iterative merging is performed to find the two clusters Cp and Cq with the smallest current distance, merge Cp and Cq, update the distance matrix, and calculate the distance of the new cluster using the average linkage strategy;

[0062] The termination condition includes: preset cluster number, maximum cluster distance threshold (in this embodiment, set to 0.2-0.5);

[0063] The electrical constraint verification is set: the electrical distance check is performed, and the electrical distance between nodes in the merged sub-region is less than 0.3pu.

[0064] Optionally, in one example of the above aspect, for the divided sub-regions, the power quality, operation efficiency and environmental adaptability of the sub-regions when the transient event occurs are evaluated by constructing a regional analysis model, and the following steps are included:

[0065] A regional analysis model for evaluating the power quality of the sub-region is constructed:

[0066] By analyzing the harmonic distortion, combined with the depth of the transient, duration and frequency of occurrence, the power quality coefficient of the sub-region at the time of the transient event is calculated;

[0067] By constructing a regional analysis model for evaluating the operation efficiency of the sub-region:

[0068] By constructing a BCC model to calculate the operation efficiency of the sub-region power grid, the efficiency value obtained is used as the operation efficiency coefficient of the sub-region at the time of the transient event;

[0069] By constructing a regional analysis model for evaluating the environmental adaptability of the sub-region:

[0070] By analyzing the influence of extreme climate on the power grid, the environmental adaptability coefficient of the sub-region at the time of the transient event is calculated.

[0071] Optionally, in one example of the above aspect, according to the transient characteristics of the node and the power quality, operation efficiency and environmental adaptability of the sub-region at the time of the transient event, an integrated multi-source data fusion algorithm and deep learning evaluation model is constructed to evaluate the state of the node, including the following steps:

[0072] Obtain the transient characteristic data in the historical output data of the node when the transient event occurs and in the preset time period before the transient event occurs, and calculate the corresponding power quality, operation efficiency and environmental adaptability at the time of the transient event;

[0073] According to whether the node fails in the preset time period before and after the transient event, the corresponding transient characteristic data, and the power quality, operation efficiency and environmental adaptability at the time of the transient event, the labels of potential fault data or normal data are added;

[0074] An integrated multi-source data fusion algorithm and deep learning evaluation model is constructed, the weights of the power quality, operation efficiency and environmental adaptability at the time of the transient event are dynamically calculated based on the entropy weight method, and weighted fusion is performed;

[0075] The transient characteristic data and the weighted fused data are input into the LSTM-Attention model for training, and the model output node data is potential fault data or normal data;

[0076] After weighting and fusing the latest monitored node data of the corresponding sub-region at the time of the transient event, the transient characteristic data is input into the trained LSTM-Attention model to identify the data as potential fault data or normal data.

[0077] Compared with the prior art, the beneficial effects of the present application are:

[0078] The application overcomes the limitation of a single data source by integrating and analyzing multiple source data such as topological structure, electrical distance, impedance matrix, and transient characteristics, thereby providing a more comprehensive perspective for system state assessment.

[0079] The application dynamically divides sub-regions according to the transient characteristics and power characteristics of initial regional nodes. In a photovoltaic grid-connected system, nodes with similar transient characteristics and close power coupling can be classified into the same sub-region, thereby avoiding the average of local characteristics caused by an excessively large region.

[0080] The application analyzes node transient characteristic data, and power quality, operation efficiency, and environmental adaptability data of the corresponding region, trains a deep learning evaluation model by comprehensively analyzing node and region data, and facilitates more accurate analysis and evaluation of the state of nodes in the region. BRIEF DESCRIPTION OF DRAWINGS

[0081] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on these drawings.

[0082] Fig. 1 The figure is a schematic diagram of the system framework of the present application.

[0083] Fig. 2 The figure is a schematic diagram of the data acquisition layer framework of the present application. DETAILED DESCRIPTION

[0084] The technical solutions of the present application will be described below in conjunction with the embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0085] Please refer to Figs. 1-2 The first aspect embodiment of the present application provides a photovoltaic grid-connected power generation online monitoring and state assessment system under transient interference, which comprises:

[0086] Data acquisition layer: photovoltaic grid-connected nodes are set in the photovoltaic grid-connected region, and high-precision synchronous measurement sensors are deployed at the photovoltaic grid-connected nodes to monitor the electrical data and environmental data of the grid-connected nodes in real time.

[0087] Edge computing layer:

[0088] The photovoltaic grid-connected power generation area is equivalent to a multi-feed line-multi-bus topological structure, and photovoltaic grid-connected nodes are marked at the buses of the topological structure, and the electrical distance between nodes is quantified according to the data collected in the layer;

[0089] With the node impedance matrix as input, the spectral clustering algorithm is used to divide the initial area, and the voltage sag events of each node in the initial area are detected according to the electromagnetic compatibility standard, and the transient characteristics of each node are recorded;

[0090] According to the transient characteristics and power characteristics of the nodes in the initial area, the initial area is divided into sub-areas;

[0091] For the divided sub-areas, the power quality, operation efficiency and environmental adaptability of the sub-areas when the transient event occurs are evaluated by constructing a regional analysis model;

[0092] The cloud analysis layer: according to the transient characteristics of the nodes and the power quality, operation efficiency and environmental adaptability of the sub-areas when the transient event occurs, an integrated multi-source data fusion algorithm and a deep learning evaluation model are constructed to evaluate the state of the nodes.

[0093] Specifically, in the present embodiment, the complex photovoltaic grid-connected system is abstracted into a structured network through equivalent modeling, and the connection relationship between each feed line and bus is clear, providing a clear physical framework for subsequent analysis. Based on the data collected in the layer, the electrical distance between nodes is calculated by impedance matrix or related algorithm, which overcomes the limitations of traditional methods that only rely on geographical distance or simple topological relationship, and more accurately reflects the electrical coupling strength between nodes.

[0094] According to the electromagnetic compatibility standard (such as IEC 61000-4-11, etc.), the threshold value, duration and other parameters of voltage sag are determined to ensure the standardization and comparability of the detection results. The existing technology may miss detection or misjudge due to non-uniform standards or rough detection methods. While detecting voltage sag, the transient characteristics of the nodes are recorded to provide a basis for in-depth analysis of the causes of voltage sag, while traditional methods may only focus on the amplitude and duration of voltage sag.

[0095] Integrating topological structure, electrical distance, impedance matrix, transient characteristics and other multi-source data for analysis overcomes the limitations of a single data source, providing a more comprehensive perspective for system state evaluation. It is convenient to dynamically adjust the region division and detection strategy with the change of system topology, while traditional methods may need to re-model or parameterize.

[0096] The existing technology relies on static topology or fixed geographical boundaries to divide regions, while the present method dynamically divides sub-areas based on the transient characteristics and power characteristics of the nodes in the initial area. For example, in a photovoltaic grid-connected system, nodes with similar transient characteristics and tight power coupling can be grouped into the same sub-area, avoiding the averaging of local characteristics due to the large size of the region.

[0097] The prior art usually only focuses on a single indicator of power quality or operation efficiency, while the regional analysis model constructed by the method simultaneously evaluates power quality, operation efficiency and environmental adaptability. For example, in an industrial park power grid, a sub-region with poor power quality but high operation efficiency under transient events can be identified, providing a contradiction balance point for optimal scheduling.

[0098] The prior art processes only a single type of data, ignoring the mutual influence characteristics between regional nodes, while the method comprehensively analyzes node transient characteristic data and the power quality, operation efficiency and environmental adaptability data of the region corresponding to the node, trains a deep learning evaluation model through comprehensive node and regional data, and facilitates more accurate analysis and evaluation of the state of nodes in the region.

[0099] In one embodiment of the present application, the data acquisition layer comprises:

[0100] The electrical quantity acquisition unit: real-time monitors the three-phase voltage / current of the grid-connected point, the DC side voltage / current and the inverter output power, and supports the IEC 61850 standard communication protocol;

[0101] The environmental perception unit: integrates an illumination intensity sensor and a temperature sensor (resolution 0.1℃) to monitor the illumination intensity and temperature of the grid-connected point;

[0102] The transient event triggering unit: sets a triggering threshold through a hardware comparator to real-time detect transient characteristics; the voltage sudden drop / sudden rise amplitude, frequency deviation and harmonic distortion, and the triggering threshold is configurable and can be set to start recording when the voltage drops to 85%Un.

[0103] In one embodiment of the present application, the photovoltaic grid-connected power generation region is equivalent to a multi-feed line-multi-bus topology structure, and photovoltaic grid-connected nodes are labeled at the buses of the topology structure. According to the data of the acquisition layer, the electrical distance between nodes is quantified, including the following steps:

[0104] The photovoltaic grid-connected power generation region is equivalent to a multi-feed line-multi-bus topology structure by equating the line connecting the photovoltaic array and the grid-connected point to a feed line and equating the photovoltaic branch junction point or the grid-connected point to a bus;

[0105] The photovoltaic grid-connected node comprises:

[0106] The photovoltaic injection node: located at the bus and the AC side of the inverter, directly reflects the position of the photovoltaic power injection into the grid;

[0107] The grid connection node: located at the connection between the bus and the upper-level grid, used for monitoring parameters such as grid-connected power, voltage and frequency;

[0108] The load node: located at the connection of the bus and the power consumption structure, the load node needs to be labeled to distinguish between power generation and power consumption behavior;

[0109] Labeling photovoltaic injection nodes, grid connection nodes and grid connection nodes at the bus of the topology, and numbering the nodes according to the node type;

[0110] According to the acquisition layer data, the electrical distance between nodes is quantified by using a node impedance matrix:

[0111]

[0112] Wherein, Dij is the electrical distance between node i and node j, Zii is the self-impedance of node i, the diagonal element of the impedance matrix, Zjj is the self-impedance of node j, Zij is the mutual impedance between node i and j, the non-diagonal element of the impedance matrix.

[0113] In one embodiment of the application, the spectral clustering algorithm is used to divide the initial area with the node impedance matrix as the input, including the following steps:

[0114] Forming a node admittance matrix Ybus, and then obtaining Zbus=Ybus-1 through matrix inversion, and establishing a node impedance matrix Zbus through the complex impedance between nodes;

[0115] The similarity between nodes is calculated by using a Gaussian kernel function:

[0116]

[0117] Wherein, Wij is the similarity between node i and j, |Zij| is the mutual impedance between node i and j, and sigma is the similarity decay coefficient;

[0118] According to the similarity Wij between nodes, a similarity matrix W is constructed;

[0119] The Laplacian matrix L=D-W is calculated, wherein D is a degree matrix, the diagonal elements of the degree matrix are equal to the degrees of the corresponding nodes, reflecting the connection strength of the nodes with other nodes;

[0120] The eigenvectors corresponding to the first k smallest eigenvalues of L are calculated to form an eigenvector matrix U;

[0121] The eigenvectors corresponding to each node of the eigenvector matrix U are divided into k clusters by using the K-means clustering algorithm, and the initial area division is completed.

[0122] In one embodiment of the application, according to the electromagnetic compatibility standard, the voltage sag events of each node in the initial area are detected, and the transient characteristics of each node are recorded, including the following steps:

[0123] According to the electromagnetic compatibility standard, the voltage sag events of each node in the initial area are detected;

[0124] The transient energy distribution is extracted by the WPT wavelet packet transform method:

[0125]

[0126] wherein dj,k is the coefficient of the jth node in the kth subband, Ej is the concentration degree of the transient energy of node j in the frequency domain, k∈(1,2,…,N), and N is the total number of subbands;

[0127] The concentration degree of the transient energy of the node in the frequency domain is counted, and recorded into the transient characteristics of each node.

[0128] In the embodiment, a detailed model of the photovoltaic power station is built in PSCAD / EMTDC, the IEC 61000-4-34 electromagnetic compatibility standard is injected, and the voltage sag event defined in the standard is recorded, with a depth of 20% and a duration of 100 ms, and the transient voltage / current waveforms of each node are recorded.

[0129] The WPT wavelet packet transform method is used to extract the transient energy distribution.

[0130] In one of the embodiments of the application, the initial region is divided into sub-regions according to the transient characteristics and power characteristics of the nodes in the initial region, including the following steps:

[0131] The nodes in the initial region are divided into sub-branches according to the transient energy distribution of the nodes in the region and the correlation of the transient characteristics between the nodes in the initial region;

[0132] The historical output data of each sub-branch are collected, the DTW dynamic time warping distance is calculated, and the branches with similar power characteristics are merged into sub-regions by using hierarchical clustering with the DTW distance as the input.

[0133] In one of the embodiments of the application, the inner nodes in the initial region are divided into sub-branches according to the transient energy distribution of the nodes in the region and the correlation of the transient characteristics between the nodes in the initial region, including the following steps:

[0134] The node transient signal is decomposed by wavelet, and the maximum decomposition layer is set, in each layer, the energy value of each frequency band in the layer is calculated, and each frequency band energy value is taken as an element of the node energy feature vector to construct the node energy feature vector;

[0135] The vector correlation is calculated by calculating the ratio of the covariance of the energy feature vectors of two nodes to the product of the standard deviations:

[0136] wherein, ρij is the vector correlation between nodes, Cov(Eni, Enj) is the covariance of the energy feature vector Eni of node i and the energy feature vector Enj of node j, σEni and σEnj are the standard deviations between the energy feature vector Eni of node i and the energy feature vector Enj of node j;

[0137] The correlation between nodes is obtained by weighting and averaging the vector correlation and the cosine similarity through calculating the cosine similarity of the energy feature vectors of two nodes;

[0138] Each node is regarded as an independent cluster through a clustering algorithm;

[0139] The correlation between all clusters is calculated through the correlation calculation results between nodes, and the two clusters with the minimum correlation are merged, and the step is repeated until all nodes are merged into k clusters;

[0140] Each cluster corresponds to a sub-branch, and the nodes in the cluster are members of the sub-branch, forming a sub-branch of secondary division.

[0141] In one embodiment of the present application, historical output data of each sub-branch is collected, and a DTW dynamic time warping distance is calculated, and the DTW distance is taken as input to merge the branches with similar power characteristics into a sub-region by using hierarchical clustering, including the following steps:

[0142] The transient characteristics of each photovoltaic branch are obtained according to the historical output data by collecting historical output data of each photovoltaic branch;

[0143] The transient characteristics are mapped to the interval [-1, 1] through Min-Max normalization algorithm;

[0144] The power fluctuation correlation analysis is performed through the time series X=(x1, x2, …, xn) and Y=(y1, y2, …, ym) of the transient characteristics between nodes, and the DTW dynamic time warping distance is calculated:

[0145]

[0146] wherein, DTW(X, Y) is the DTW distance between nodes, X and Y are the power time series of two branches, d(xi, yj) is the Euclidean distance between nodes, n is the total number of the time series of the transient characteristics of node i, and m is the total number of the time series of the transient characteristics of node j;

[0147] The DTW distances between all branches are calculated, and a symmetric distance matrix is formed;

[0148] Each branch is initialized as an independent cluster through hierarchical clustering method;

[0149] Iterative merging is performed to find the two clusters Cp and Cq with the minimum current distance, merge Cp and Cq, update the distance matrix, and calculate the distance of the new cluster by using the average linkage strategy;

[0150] The termination condition includes a preset cluster number and a maximum cluster distance threshold (in this embodiment, the threshold is set to 0.2-0.5);

[0151] The electrical constraint verification is set: The electrical distance between nodes in the merged sub-region is less than 0.3 pu, and the strong coupling relationship is ensured.

[0152] In one of the embodiments of the application, for the divided sub-region, a region analysis model is constructed to evaluate the power quality, operation efficiency and environmental adaptability of the sub-region when a transient event occurs, including the following steps:

[0153] The region analysis model for evaluating the power quality of the sub-region is constructed:

[0154] The power quality coefficient of the sub-region when a transient event occurs is calculated by analyzing the harmonic distortion, combining the depth of sag, duration and occurrence frequency:

[0155]

[0156] wherein TE is the power quality coefficient of the sub-region, Uh is the effective value of the harmonic voltage of the hth node of the sub-region, U0 is the effective value of the fundamental voltage of the hth node of the sub-region; h∈(1, 2, …, H), H is the total number of nodes of the sub-region, wh is a dynamic weight calculated by the entropy weight method; Vch is the residual voltage of the hth node of the sub-region during the sag, V0h is the rated voltage of the hth node of the sub-region; th is the average value of the voltage sag duration of the hth node of the sub-region, T is the statistical period; λh is the correction coefficient of the sag occurrence frequency of the hth node of the sub-region, fitted by the Poisson distribution; a1 and a2 are corresponding coefficients respectively;

[0157] The region analysis model for evaluating the operation efficiency of the sub-region is constructed:

[0158] The operation efficiency of the sub-region power grid is calculated by constructing the BCC model, and the obtained efficiency value is taken as the operation efficiency coefficient of the sub-region when a transient event occurs;

[0159]

[0160]

[0161]

[0162]

[0163] wherein, θ is an efficiency value (0≤θ≤1), is the efficiency value of the cth input of all nodes in the sub-region, is the rth output of all nodes in the sub-region, is a non-Archimedean infinitesimal; xch is the cth input of the hth node in the sub-region; yrh is the rth output of the hth node in the sub-region; sc - and sr + are slack variables, representing input redundancy and output deficiency, c∈(1,2,…,g), r∈(1,2,…,s), g is the total number of input items of all nodes in the sub-region, and s is the total number of output items of all nodes in the sub-region;

[0164] By constructing a regional analysis model for evaluating the environmental adaptability of the sub-region:

[0165] By analyzing the impact of extreme climate on the power grid, the environmental adaptability coefficient of the sub-region when a transient event occurs is calculated:

[0166]

[0167] wherein, Re is the environmental adaptability coefficient of the sub-region, Pb(h,t) is the device failure probability of the hth node in the sub-region in the tth time monitoring interval, f(h,t) is the device spatial distribution density function of the hth node in the sub-region in the tth time monitoring interval, which is generated by kernel density estimation, t∈(1,2,…,W), and W is the total number of time monitoring intervals;

[0168] The device failure probability is calculated by the following formula:

[0169]

[0170] wherein, Pb is the device failure probability, β0, β1 and β2 are calculation parameters. In this embodiment, β0, β1 and β2 are set to 0.2, 0.4 and 0.4 respectively.

[0171] In one embodiment of the present application, according to the transient characteristics of the node and the power quality, operating efficiency and environmental adaptability of the sub-region when a transient event occurs, an integrated multi-source data fusion algorithm and deep learning evaluation model are constructed to evaluate the state of the node, including the following steps:

[0172] Obtain the transient characteristic data in the historical output data of the node when the pause event occurs and in the preset time period before the pause event occurs, and calculate the corresponding power quality, operating efficiency and environmental adaptability when the transient event occurs;

[0173] According to whether the node sends a fault within a preset time period before and after the occurrence of the suspension event, labels of potential fault data or normal data are added to corresponding transient characteristic data and power quality, operation efficiency and environmental adaptability at the time of the transient event;

[0174] An integrated multi-source data fusion algorithm and a deep learning evaluation model are constructed, weights of power quality, operation efficiency and environmental adaptability at the time of the transient event are dynamically calculated based on an entropy weight method, and weighted fusion is performed;

[0175] The transient characteristic data and the weighted fused data are input into an LSTM-Attention model for training, and the model outputs node data as potential fault data or normal data;

[0176] After the power quality, operation efficiency and environmental adaptability data of the corresponding sub-region of the latest monitored node at the time of the transient event are weighted fused, the transient characteristic data are input into the trained LSTM-Attention model, and the data are identified as potential fault data or normal data.

[0177] The above examples are only used to illustrate the technical method of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A system for online monitoring and status assessment of photovoltaic grid-connected power generation under transient disturbances, characterized in that, include: Data acquisition layer: Photovoltaic grid-connected nodes are set up in the photovoltaic grid-connected power generation area, and high-precision synchronous measurement sensors are deployed at the photovoltaic grid-connected nodes to monitor the electrical and environmental data of the grid-connected points in real time; Edge computing layer: The photovoltaic grid-connected power generation area is equivalent to a multi-feeder-multi-bus topology, and the photovoltaic grid-connected nodes are marked at the bus of the topology. Based on the data collected from the acquisition layer, the electrical distance between the nodes is quantified. Using the node impedance matrix as input, a spectral clustering algorithm is used to divide the initial region. According to the electromagnetic compatibility standard, the voltage sag events of each node in the initial region are detected, and the transient characteristics of each node are recorded. Based on the transient characteristics and power properties of the nodes in the initial region, the initial region is divided into sub-regions; For the divided sub-regions, a regional analysis model is constructed to evaluate the power quality, operating efficiency, and environmental adaptability of the sub-regions during transient events. Cloud-based analytics layer: Based on the transient characteristics of nodes and the power quality, operating efficiency, and environmental adaptability of sub-regions during transient events, an integrated multi-source data fusion algorithm and deep learning evaluation model are constructed to evaluate node status. The initial region is divided into sub-regions based on the transient characteristics and power properties of the nodes in the initial region, including the following steps: Based on the transient energy distribution of the regional nodes and the correlation of transient characteristics among the nodes in the initial region, the nodes in the initial region are divided into sub-branches for the second time. Historical power output data of each sub-branch is collected, and the DTW dynamic time warping distance is calculated. Using the DTW distance as input, hierarchical clustering is used to merge branches with similar power characteristics into sub-regions. Based on the transient energy distribution of the regional nodes and the correlation of transient characteristics among the nodes in the initial region, the internal nodes of the initial region are further divided into sub-branches, including the following steps: Wavelet decomposition is performed on the transient signal of the node, and the maximum number of decomposition layers is set. In each layer, the energy value of the layer in each frequency band is calculated, and the energy value of each frequency band is used as the element to construct the node energy feature vector. The vector correlation is calculated by taking the ratio of the product of the covariance and the standard deviation of the energy eigenvectors of two nodes. Where ρij is the vector correlation between nodes, Cov(Eni,Enj) is the covariance of the energy feature vector Eni of node i and the energy feature vector Enj of node j, and σEni and σEnj are the standard deviations between the energy feature vectors Eni and Enj of nodes. By calculating the cosine similarity of the energy feature vectors of two nodes, and then weighting the vector correlation with the cosine similarity, the correlation between nodes can be obtained. Each node is treated as an independent cluster using a clustering algorithm; Calculate the correlation between all clusters based on the correlation between nodes, merge the two clusters with the lowest correlation, and repeat this step until all nodes are merged into k clusters. Each cluster corresponds to a sub-branch, and the nodes within the cluster are members of that sub-branch, forming a sub-branch of secondary partitioning.

2. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient disturbances according to claim 1, characterized in that, The data acquisition layer includes: Electrical quantity acquisition unit: Real-time monitoring of three-phase voltage / current at the grid connection point, DC side voltage / current, and inverter output power; Environmental sensing unit: monitors the light intensity and temperature at the grid connection point; Transient event triggering unit: Sets trigger threshold through hardware comparator to detect transient characteristics in real time, including voltage sag / surge amplitude, frequency offset, and harmonic distortion.

3. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient interference as described in claim 1, characterized in that, The photovoltaic grid-connected power generation area is equivalent to a multi-feeder-multi-bus topology, and the photovoltaic grid-connected nodes are marked at the busbars of the topology. Based on the data acquisition layer, the electrical distance between nodes is quantified, including the following steps: By equating the lines connecting the photovoltaic array to the grid connection point to feeders, and the photovoltaic branch junction points or grid connection points to buses, the photovoltaic grid-connected power generation area is equating to a multi-feeder-multi-bus topology. Photovoltaic grid-connected nodes include: Photovoltaic injection node: located on the AC side of the bus and inverter; Grid connection node: located at the connection point between the busbar and the upstream power grid; Load node: located at the connection point of the busbar electrical structure; Mark the photovoltaic injection node, grid connection node, and grid connection node at the bus of the topology, and number the nodes according to their type; Based on the data acquired at the acquisition layer, the electrical distance between nodes is quantized using a node impedance matrix: ； Where Dij is the electrical distance between node i and node j, Zii is the self-impedance of node i, Zjj is the self-impedance of node j, and Zij is the mutual impedance between nodes i and j.

4. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient interference as described in claim 1, characterized in that, Using the nodal impedance matrix as input, the initial region is divided using a spectral clustering algorithm, including the following steps: The node admittance matrix Ybus is formed, and then the matrix inversion is used to obtain Zbus=Ybus-1. The node impedance matrix Zbus is established through the complex impedance between nodes. Calculate the similarity between nodes using the Gaussian kernel function: ； Where Wij is the similarity between nodes i and j, |Zij| is the mutual impedance between nodes i and j, and σ is the similarity attenuation coefficient; Construct a similarity matrix W based on the similarity Wij between nodes; Calculate the Laplacian matrix L=DW, where D is the degree matrix, and the diagonal elements of the degree matrix are equal to the degree of the corresponding node; Calculate the eigenvectors corresponding to the first k smallest eigenvalues ​​of L, and form the eigenvector matrix U; For each node of the eigenvector matrix U, the eigenvector corresponding to the eigenvector is divided into k clusters using the K-means clustering algorithm, thus completing the initial region division.

5. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient disturbances according to claim 1, characterized in that, According to electromagnetic compatibility standards, voltage sag events at each node in the initial region are detected, and the transient characteristics of each node are recorded, including the following steps: According to electromagnetic compatibility standards, voltage dips at each node in the initial region are detected. Extracting transient energy distribution using the WPT wavelet packet transform method: ； Where dj,k is the coefficient of the kth sub-band of the jth node, Ej is the concentration of the transient energy of the jth node in the frequency domain, k∈(1,2,…,N), and N is the total number of sub-bands; The transient energy concentration of each node in the frequency domain is statistically analyzed and recorded in the transient characteristics of each node.

6. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient disturbances according to claim 1, characterized in that, Collect historical power output data for each sub-branch, calculate the DTW dynamic time warping distance, and use the DTW distance as input to merge branches with similar power characteristics into sub-regions using hierarchical clustering, including the following steps: By collecting historical power output data of each photovoltaic branch, the transient characteristics of each photovoltaic branch are obtained based on the historical power output data; Transient features are mapped to the [-1,1] interval using the Min-Max normalization algorithm; Correlation analysis of power fluctuations is performed using time series X=(x1,x2,…,xn) and Y=(y1,y2,…,ym) of the transient characteristics between nodes to calculate the dynamic time-warped distance of the DTW. ； Where DTW(X,Y) is the DTW distance between nodes, X,Y are the power time series of the two branches, d(xi,yj) is the Euclidean distance between nodes, n is the total number of time series of transient features of node i, and m is the total number of time series of transient features of node j. Calculate the DTW distance between each pair of all branches to form a symmetric distance matrix; Each branch is initialized as an independent cluster using hierarchical clustering. Perform iterative merging, find the two clusters Cp and Cq with the smallest current distance, merge Cp and Cq, update the distance matrix, and calculate the distance of the new cluster using the average link strategy; Termination conditions can be set including: preset number of clusters and maximum inter-cluster distance threshold; Set electrical constraint verification: Perform electrical distance check, the electrical distance between nodes within the merged sub-region is <0.3pu.

7. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient disturbances according to claim 1, characterized in that, For the divided sub-regions, a regional analysis model is constructed to evaluate the power quality, operating efficiency, and environmental adaptability of the sub-regions during transient events, including the following steps: By constructing a regional analysis model to evaluate the power quality of sub-regions: By analyzing harmonic distortion and combining sag depth, duration, and occurrence frequency, the power quality factor of the sub-region during transient events is calculated: ； Where TE is the power quality coefficient of the sub-region, Uh is the effective value of the harmonic voltage of the h-th node in the sub-region, and U0 is the effective value of the fundamental voltage of the h-th node in the sub-region; h∈(1,2,…,H), H is the total number of nodes in the sub-region, wh is the dynamic weight; Vch is the residual voltage during the voltage sag at the h-th node in the sub-region, V0h is the rated voltage of the h-th node in the sub-region; th is the mean of the duration of the voltage sag at the h-th node in the sub-region, T is the statistical period; λh is the correction coefficient for the frequency of voltage sag at the h-th node in the sub-region, which is fitted by a Poisson distribution; a1 and a2 are the corresponding coefficients; By constructing a regional analysis model to evaluate the operational efficiency of sub-regions: The operating efficiency of the sub-regional power grid is calculated by constructing a BCC model, and the obtained efficiency value is used as the operating efficiency coefficient of the sub-region when a transient event occurs. ； ； ； ； Where θ is the efficiency value (0≤θ≤1). The efficiency value of the input for the c-th item of all nodes in the sub-region. For the r-th output of all nodes in the sub-region, For non-Archimedean infinitesimals; xch is the c-th input of the h-th node in the subregion; yrh is the r-th output of the h-th node in the subregion; sc - and sr + Let c be a slack variable, representing input redundancy and output insufficiency, where c∈(1,2,…,g), r∈(1,2,…,s), g is the total number of input items for all nodes in the subregion, and s is the total number of output items for all nodes in the subregion. By constructing a regional analysis model to assess the environmental adaptability of sub-regions: By analyzing the impact of extreme weather on the power grid, the environmental adaptability coefficient of the sub-region during transient events is calculated: ； Where Re is the environmental adaptability coefficient of the sub-region, Pb(h,t) is the equipment failure probability of the h-th node in the t-th time monitoring interval of the sub-region, f(h,t) is the equipment spatial distribution density function of the h-th node in the t-th time monitoring interval of the sub-region, which is generated by kernel density estimation, t∈(1,2,…,W), and W is the total number of time monitoring intervals; The probability of equipment failure is calculated using the following formula: ； Where Pb is the probability of equipment failure, and β0, β1 and β2 are calculation parameters.

8. The online monitoring and status assessment system for photovoltaic grid-connected power generation under transient disturbances according to claim 1, characterized in that, Based on the transient characteristics of nodes and the power quality, operating efficiency, and environmental adaptability of sub-regions during transient events, an integrated multi-source data fusion algorithm and deep learning evaluation model are constructed to evaluate node status, including the following steps: Acquire transient characteristic data from the node's historical output data when a pause event occurs, as well as during a preset time period before the pause event, and calculate the corresponding power quality, operating efficiency, and environmental adaptability at the time of the transient event. Based on whether a node experiences a fault within a preset time period before and after a pause event, tags for potential fault data or normal data are added to the corresponding transient characteristic data, as well as the power quality, operating efficiency, and environmental adaptability at the time of the transient event. An integrated multi-source data fusion algorithm and deep learning evaluation model are constructed. Based on the entropy weight method, the weights of power quality, operating efficiency and environmental adaptability during transient events are dynamically calculated and weighted. Transient feature data and weighted fused data are input into the LSTM-Attention model for training. The model outputs node data as potential fault data or normal data. The power quality, operating efficiency, and environmental adaptability data of the corresponding sub-regions of the latest monitored nodes at the time of transient events are weighted and fused together, and then input into the trained LSTM-Attention model along with the transient feature data to identify whether the data is potential fault data or normal data.

Images