A photovoltaic power station power fluctuation feature extraction method and device for distinguishing micro-meteorological disturbance from equipment failure

Abstract

The application discloses a photovoltaic power station power fluctuation feature extraction method and device for distinguishing micro-meteorological disturbance from equipment failure, relates to the technical field of feature extraction, and comprises the following steps: acquiring original power data of a photovoltaic power station and processing the original power data to obtain segmented marked power data; performing wavelet analysis and spectrum feature enhancement on the segmented marked power data to obtain enhanced power data and wavelet energy features; performing power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features, simultaneously performing phase space reconstruction and dynamic feature analysis to extract phase sequence features and extreme value network features; and performing causal relationship analysis and pattern recognition according to the wavelet energy features, the elasticity parameters, the entropy gradient features, the phase sequence features and the extreme value network features to obtain a judgment result for distinguishing micro-meteorological disturbance from equipment failure. The application improves the accuracy of distinguishing micro-meteorological disturbance from equipment failure, guarantees stable operation of the photovoltaic power station, and improves photovoltaic energy utilization efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power feature extraction, in particular to a photovoltaic power plant power fluctuation feature extraction method and device for distinguishing micro-meteorological disturbances and equipment failures. BACKGROUND

[0002] With the increasing demand for clean energy worldwide, photovoltaic power generation has been rising in the energy sector due to its clean and renewable characteristics. However, the power output of photovoltaic power plants is easily affected by micro-meteorological disturbances (such as cloud cover, wind speed changes) and equipment failures (such as component aging, inverter abnormalities), showing significant volatility and instability. This unstable power output poses great challenges to grid scheduling, operation, and power system stability, for example, sudden micro-meteorological disturbances can cause power surges, affecting the balance between supply and demand of the power grid. Therefore, accurately distinguishing whether the power fluctuation of the photovoltaic power plant is caused by micro-meteorological disturbances or equipment failures is crucial for ensuring efficient and stable operation of photovoltaic power plants and improving the reliability of the power system.

[0003] Traditional methods for distinguishing the causes of photovoltaic power plant power fluctuations have many limitations. Some methods based on a single physical model can explain power changes to some extent, but they are not adaptable to complex and variable micro-meteorological and equipment failure scenarios, making it difficult to accurately distinguish between them. For example, relying solely on the theoretical relationship between light intensity and power cannot effectively identify power fluctuations caused by internal equipment failures. Some statistical models, such as simple time series analysis, lack sufficient data feature mining and cannot fully capture the multidimensional characteristics of power fluctuations, making it easy to misjudge when faced with similar power fluctuation curves caused by micro-meteorological disturbances and equipment failures. Existing methods mostly lack deep processing and multi-feature fusion analysis of power data, making it difficult to extract features with high recognition from complex power fluctuations, resulting in low accuracy in distinguishing between them. SUMMARY

[0004] To address the problem of lack of deep processing and multi-feature fusion analysis of power data in existing technology, which makes it difficult to extract features with high recognition from complex power fluctuations, resulting in low accuracy in distinguishing between them, the present application provides a photovoltaic power plant power fluctuation feature extraction method and device for distinguishing micro-meteorological disturbances and equipment failures, which can significantly improve the accuracy of distinguishing between micro-meteorological disturbances and equipment failures, effectively ensure the stable operation of photovoltaic power plants, and improve the efficiency of photovoltaic energy utilization. The specific technical solutions are as follows:

[0005] In a first aspect, the present application provides a photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbances and equipment failures, comprising:

[0006] The original power data of the photovoltaic power station are acquired, and are subjected to denoising and standardization processing, and the processed data are segmented and labeled to obtain segmented and labeled power data;

[0007] The segmented and labeled power data are subjected to wavelet analysis and spectrum feature enhancement to obtain enhanced power data and wavelet energy features;

[0008] Based on the enhanced power data, power recovery elasticity analysis is performed to extract elasticity parameters and entropy gradient features, and phase space reconstruction and dynamic feature analysis are performed to extract phase sequence features and extreme value network features;

[0009] According to the wavelet energy features, elasticity parameters, entropy gradient features, phase sequence features and extreme value network features, causal relationship analysis and pattern recognition are performed to obtain a determination result for distinguishing micro-meteorological disturbances from equipment failures.

[0010] Preferably, the wavelet analysis and spectrum feature enhancement of the segmented and labeled power data comprise:

[0011] The segmented and labeled power data are subjected to wavelet decomposition by using a preselected optimal wavelet basis function and an adaptive threshold method based on local fluctuation characteristics to obtain wavelet coefficients;

[0012] The wavelet coefficients are subjected to nonlinear enhancement processing and reconstruction to obtain enhanced power data;

[0013] Multi-scale wavelet energy distribution is calculated based on the enhanced power data to obtain wavelet energy features.

[0014] Preferably, the power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features comprises:

[0015] The enhanced power data are subjected to linear detrending processing to separate fluctuation components to obtain a detrended fluctuation sequence, a state-dependent spring-damping system model is established based on the detrended fluctuation sequence, and elasticity parameters are extracted;

[0016] The enhanced power data are subjected to sliding window information entropy calculation and entropy gradient analysis to extract entropy gradient features.

[0017] Preferably, the sliding window information entropy calculation and entropy gradient analysis of the enhanced power data to extract entropy gradient features comprise:

[0018] An information entropy sequence in a sliding time window is calculated;

[0019] An entropy gradient sequence is calculated based on the information entropy sequence;

[0020] An entropy gradient fluctuation response index is extracted based on the entropy gradient sequence, and the entropy gradient fluctuation response index is used as a key characteristic parameter for distinguishing between disturbances and faults.

[0021] Preferably, the enhanced power data is subjected to phase space reconstruction and dynamic characteristic analysis to extract phase sequence characteristics and extreme value network characteristics based on predetermined optimal time delay parameters and embedding dimensions.

[0022] The enhanced power data is subjected to phase space reconstruction to obtain a phase space trajectory based on predetermined optimal time delay parameters and embedding dimensions.

[0023] The phase space trajectory is subjected to phase sequence analysis to extract phase sequence characteristics including an angular velocity variation coefficient, a phase jump index, and a phase angle root mean square error.

[0024] A complex network is constructed based on multi-scale extreme value points of the enhanced power data to extract extreme value network characteristics including a network average degree and a clustering coefficient ratio.

[0025] Preferably, the wavelet energy characteristics, the elasticity parameters, the entropy gradient characteristics, the phase sequence characteristics, and the extreme value network characteristics are subjected to causal relationship analysis and pattern recognition to obtain a determination result for distinguishing between micro-meteorological disturbances and equipment faults.

[0026] A comprehensive feature vector is constructed based on the wavelet energy characteristics, the elasticity parameters, the entropy gradient characteristics, the phase sequence characteristics, and the extreme value network characteristics.

[0027] The comprehensive feature vector is subjected to feature optimization, and a L1 regularization method is used for feature selection to obtain an optimized feature set.

[0028] Based on the optimized feature set, a bidirectional Granger causality is constructed in combination with auxiliary measurement parameters including meteorological parameters and equipment parameters to extract causal flow characteristics.

[0029] The optimized feature set and the causal flow characteristics are fused, and a determination result for distinguishing between micro-meteorological disturbances and equipment faults is obtained through cosine similarity matching with a historical feature template library.

[0030] In a second aspect, the present application also provides a photovoltaic power station power fluctuation feature extraction device for distinguishing between micro-meteorological disturbances and equipment faults, which applies the aforementioned method and comprises:

[0031] A data acquisition and processing module is configured to acquire original power data of a photovoltaic power station, perform denoising and standardization processing, segment and label the processed data, and obtain segmented and labeled power data.

[0032] A data enhancement module is configured to perform wavelet analysis and frequency spectrum feature enhancement on the segmented and labeled power data to obtain enhanced power data and wavelet energy characteristics.

[0033] a feature extraction module configured to perform power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features, and simultaneously perform phase space reconstruction and dynamic feature analysis to extract phase sequence features and extreme value network features;

[0034] a feature recognition module configured to perform causality analysis and pattern recognition based on the wavelet energy features, elasticity parameters, entropy gradient features, phase sequence features and extreme value network features to obtain a determination result distinguishing between micro-meteorological disturbances and equipment failures.

[0035] Compared with the prior art, the present application has the following advantages:

[0036] The photovoltaic power station power fluctuation feature extraction method distinguishing between micro-meteorological disturbances and equipment failures of the present application can effectively enhance the differences between micro-meteorological disturbances and equipment failures in the frequency spectrum by denoising and standardizing the original power data, combining segmentation and category marking, wavelet analysis and spectrum feature enhancement. Equipment failures are often accompanied by high-frequency energy mutations. By extracting wavelet energy features, a key frequency dimension basis is provided for distinguishing between the two. The elasticity parameters and entropy gradient features extracted by power recovery elasticity analysis, and the phase sequence features and extreme value network features obtained by phase space reconstruction and dynamic feature analysis are integrated with the wavelet energy features to construct a comprehensive feature vector that fully reflects the nature of power fluctuations and improves the accuracy of the distinction. Based on the optimized feature set, bidirectional Granger causality is constructed to extract causality flow features, which are combined with feature template matching to associate the causality between power data and auxiliary measurement parameters, and accurately identify the source of power fluctuations. Compared with traditional methods, the present application can significantly improve the accuracy of distinguishing between micro-meteorological disturbances and equipment failures, effectively ensure the stable operation of photovoltaic power stations, and improve the utilization efficiency of photovoltaic energy. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the specific embodiments or the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0038] Figure 1 The flowchart of the photovoltaic power station power fluctuation feature extraction method distinguishing between micro-meteorological disturbances and equipment failures of the present application.

[0039] Figure 2 The flowchart of an embodiment of the photovoltaic power station power fluctuation feature extraction method distinguishing between micro-meteorological disturbances and equipment failures of the present application.

[0040] Figure 3A flow chart of another embodiment of the photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbance from equipment failure.

[0041] Figure 4 A flow chart of another embodiment of the photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbance from equipment failure.

[0042] Figure 5 A schematic diagram of the photovoltaic power plant power fluctuation feature extraction system for distinguishing micro-meteorological disturbance from equipment failure. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0044] It should be understood that, when used in the specification, the terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or sets thereof.

[0045] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and the appended claims of the present application, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0046] It should be further understood that the term "and / or" used in the specification of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0047] The following embodiments refer to Figures 1 to 5 .

[0048] The embodiments of the present application provide a photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbance from equipment failure, comprising:

[0049] Step S1, acquiring original power data of a photovoltaic power plant, and performing denoising and standardization processing, segmenting and classifying the processed data, to obtain segmented and labeled power data;

[0050] The original power data is acquired from a photovoltaic power station monitoring system, and a sampling frequency of 1 Hz is set to cover the complete power fluctuation and recovery process. The original power data is subjected to median filtering and statistical outlier detection to remove short-term noise and outliers, thereby obtaining denoised power data.

[0051] First, adaptive window median filtering is applied, and the window size is dynamically adjusted according to the data fluctuation amplitude. Then, the 3σ rule is used to detect and process the remaining abnormal points, and the points exceeding the 3σ range are replaced by local smoothing.

[0052] The denoised power data is subjected to standardization processing, and in this embodiment, the system rated power is used as the reference to obtain normalized power data.

[0053] The normalized power data is subjected to event segmentation based on the power change rate and fluctuation duration, and the complete power fluctuation and recovery process is identified to obtain segmented and marked power data.

[0054] The power change rate is calculated as follows:

[0055] ΔP(t)=[P normalized (t+1)-P normalized (t)] / Δt

[0056] In the formula, P normalized is the normalized power data, and Δt is the time interval between adjacent power data collection points.

[0057] When |ΔP(t)|>ΔP threshold (the threshold value of the power change rate) and the duration is greater than T min (the minimum duration of fluctuation start), it is marked as a fluctuation start point; when |ΔP(t)|<ΔP stable (the change rate threshold value of power stability) and the duration is greater than T stable (the minimum duration of fluctuation end), it is marked as a fluctuation end point.

[0058] Step S2, wavelet analysis and spectral feature enhancement are performed on the segmented and marked power data to obtain enhanced power data and wavelet energy features.

[0059] Specifically, the wavelet analysis and spectral feature enhancement of the segmented and marked power data include:

[0060] S21, using a preselected optimal wavelet basis function, the segmented and marked power data is subjected to wavelet decomposition using an adaptive threshold method based on local fluctuation characteristics to obtain wavelet coefficients.

[0061] According to the characteristics of photovoltaic power fluctuation, wavelet base function adaptability evaluation is performed on the segmented marked power data, and the optimal wavelet base function is selected. The fitting effects of various wavelet base functions (such as db4, sym8, coif4, etc.) on the power data are evaluated; the information entropy criterion of each wavelet base function is calculated; and the wavelet base function with the minimum information entropy is selected as the optimal wavelet base function.

[0062] In this embodiment, the segmented marked power data is subjected to wavelet decomposition by using the improved adaptive threshold method, and a set of wavelet coefficients is obtained.

[0063] The segmented marked power data is subjected to multi-scale decomposition by using the optimal wavelet base function; and the adaptive threshold value of the jth scale is calculated.

[0064] λ j =σ j ·

[0065] Where σ j is the noise standard deviation of the jth scale, N is the data length, V j is the local variation coefficient, and α is the adaptive coefficient.

[0066] The improved adaptive threshold method effectively removes the noise component while retaining the fluctuation characteristics.

[0067] S22, the wavelet coefficients are subjected to nonlinear enhancement processing and reconstruction, and enhanced power data are obtained.

[0068] The nonlinear enhancement function is applied to the wavelet coefficients:

[0069] W enhanced (j,k)=sign(W coeffs (j,k))·|W coeffs (j,k)|^γ j

[0070] Where W coeffs is the set of wavelet coefficients; sign is a sign function for extracting the wavelet coefficients; and γ j is a scale-dependent nonlinear coefficient.

[0071] Different enhancement strategies are adopted for different frequency band characteristics; the wavelet coefficients after enhancement are subjected to inverse transformation, and enhanced power data are reconstructed.

[0072] S23, the multi-scale wavelet energy distribution is calculated based on the enhanced power data, and the wavelet energy features are obtained.

[0073] The wavelet energy of each scale j is calculated:

[0074] E j =Σ[|W enhanced(j, k) | |2]

[0075] Constructing wavelet energy feature vector:

[0076] E wavelet = [E1, E2,..., En] j ]

[0077] Calculating energy ratio feature:

[0078] ER j = E j / ∑[E j ]

[0079] Based on the above calculation results, the normalized energy distribution is obtained.

[0080] Step S3, based on the enhanced power data, power recovery elasticity analysis is carried out to extract elasticity parameters and entropy gradient features, and phase space reconstruction and dynamic feature analysis are carried out to extract phase sequence features and extreme value network features;

[0081] Specifically, the power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features comprises:

[0082] S31, linear detrend processing is carried out on the enhanced power data, the fluctuation component is separated to obtain a detrended fluctuation sequence, a state-dependent spring-damping system model is established based on the detrended fluctuation sequence, and elasticity parameters are extracted.

[0083] In this embodiment, the trend component is adaptively extracted by using empirical mode decomposition method; the detrended fluctuation sequence is calculated:

[0084] F fluctuation = P enhanced (t) - Trend(t)

[0085] In the formula, P enhanced (t) is the enhanced power data; Trend(t) is the trend component, which is adaptively extracted by empirical mode decomposition method in this embodiment, and represents the trend part of the power data.

[0086] Based on the calculated detrended fluctuation sequence, a power recovery elasticity function is constructed to obtain a set of elasticity parameters.

[0087] Establishing elasticity response model:

[0088] M·d 2 P / dt 2 + C(P)·dP / dt + K(P)·P = F(t)

[0089] Wherein, M is the system inertia parameter, C(P) is the state-dependent damping coefficient, K(P) is the state-dependent stiffness coefficient; P is the power; F(t) is the external force function, the external excitation acting on the system.

[0090] By solving the differential equation in reverse, the elastic parameter set P is extracted resilience ={M,C(P),K(P)}. The micro-meteorological disturbance characteristics include low inertia (M is small), variable damping (C changes greatly); the equipment failure characteristics include high inertia (M is large), nonlinear stiffness (K is nonlinear).

[0091] S32, sliding window information entropy calculation and entropy gradient analysis are performed on the enhanced power data, and entropy gradient features are extracted.

[0092] Calculate the information entropy sequence in the sliding time window:

[0093]

[0094] Wherein, p(x,t) is the probability distribution estimation of power value x at t time;

[0095] Calculate the entropy gradient sequence:

[0096]

[0097] Extract the entropy gradient fluctuation response index:

[0098]

[0099] Based on the entropy gradient sequence, the entropy gradient fluctuation response index is extracted, which is used as a key feature parameter to distinguish disturbances and failures.

[0100] Specifically, the phase space reconstruction and dynamic feature analysis are performed based on the enhanced power data to extract phase sequence features and extreme value network features, including:

[0101] S33, based on the predetermined optimal time delay parameter and embedding dimension, the enhanced power data is reconstructed to obtain a phase space trajectory;

[0102] The optimal time delay parameter is calculated by mutual information method for enhanced power data, and the optimal delay is obtained.

[0103] Calculate the mutual information value of different time delay τ:

[0104]

[0105] In the formula, is The marginal probability density function; is The marginal probability density function; For and of .

[0106] Select the first local minimum of mutual information as the optimal delay τ opt .

[0107] Based on the enhanced power data and the optimal delay, the optimal embedding dimension is determined by the false nearest neighbor method, and the optimal embedding dimension is obtained. And calculate the false nearest neighbor ratio FNN(m) under different embedding dimension m; When FNN(m)<ε threshold (false nearest neighbor ratio threshold), determine m as the optimal embedding dimension;

[0108] Using enhanced power data, optimal delay and optimal embedding dimension, reconstruct the phase space trajectory to obtain the phase space trajectory. Construct the phase space trajectory:

[0109] Y ps (i)=[P enhanced (i),P enhanced (i+τ opt ),...,P enhanced (i+(m opt -1)·τ opt )]

[0110] S34, the phase sequence analysis is carried out on the phase space trajectory, and the phase sequence characteristics including the coefficient of variation of angular velocity, the phase jump index and the root mean square error of phase angle are extracted;

[0111] Calculate the instantaneous phase angle:

[0112] θ(t)=arctan(Y ps (t)2 / Y ps (t)1)

[0113] Calculate the angular velocity:

[0114] ω(t)=dθ(t) / dt

[0115] Calculate the coefficient of variation of angular velocity:

[0116]

[0117] In the formula, is the standard deviation of angular velocity; is the mean value of angular velocity.

[0118] Calculate the phase jump index:

[0119]

[0120] Where, is the step function, is the angular velocity jump threshold value;

[0121] Micro-meteorological disturbance characteristics: smooth change in phase angle velocity (V ω small), few jumps (Jump i index small);

[0122] Equipment failure characteristics: significant phase jump (Jump i index large), high variability (V ω large).

[0123] S35, constructing a complex network based on the multi-scale extreme points of the enhanced power data, and extracting extreme network features including network average degree and clustering coefficient ratio.

[0124] Extracting multi-scale extreme points from enhanced power data as network nodes; constructing a weighted directed network based on the time distance and amplitude relationship between extreme points; calculating network asymmetry index:

[0125] NAI=|Out degree -In degree | / (Out degree +In degree )

[0126] Where, Out degree , In degree represent out-degree and in-degree, out-degree refers to the number of edges from a node to other nodes, representing the number of connections emitted by the node as an information source. In-degree refers to the number of edges pointing to the node, representing the number of connections received by the node.

[0127] Calculate the clustering coefficient ratio CCR=C positive / C negative ; Where C positive and C negative are the clustering coefficients of the positive and negative change subnetworks, respectively.

[0128] Micro-meteorological disturbance characteristics: balanced network structure (low NAI), clustering coefficient ratio close to 1 (CCR≈1);

[0129] Equipment failure characteristics: network structure asymmetry (high NAI), clustering coefficient ratio deviates from 1 (CCR≠1).

[0130] Step S4, according to the wavelet energy features, elastic parameters, entropy gradient features, phase sequence features and extreme network features, causal relationship analysis and pattern recognition are performed to obtain the determination result of distinguishing micro-meteorological disturbances and equipment failures, specifically including:

[0131] S41. Construct a comprehensive feature vector based on the wavelet energy characteristics, elastic parameters, entropy gradient characteristics, phase sequence characteristics, and extreme value network characteristics.

[0132] Integrating the wavelet energy eigenvectors E from the preceding steps wavelet Elastic parameter set P resilience Entropy gradient feature set H gradient Phase sequence feature set F phase and the feature set G of the extreme value network features Construct a comprehensive feature vector F combined .

[0133] Constructing feature vectors:

[0134] F combined =[E wavelet ,P resilience H gradient ,F phase G features ]

[0135] S42. Perform feature optimization on the comprehensive feature vector, and use L1 regularization to select features to obtain an optimized feature set;

[0136] Calculate the correlation matrix between features:

[0137] Corr(i,j)=Cov(F i ,F j ) / (σ i× σ j )

[0138] Feature subsets are selected based on the maximum relevance minimum redundancy (mRMR) principle; an optimized feature set F is constructed. opt It contains multi-dimensional, low-redundancy, and highly discriminative features;

[0139] S43. Based on the optimized feature set, a bidirectional Granger causal relationship is constructed by combining auxiliary measurement parameters, and causal flow features are extracted. The auxiliary measurement parameters include meteorological parameters and equipment parameters.

[0140] Constructing an optimized feature set F opt The bidirectional Granger causality relationship between the power data and auxiliary measurement parameters (such as temperature, irradiance, etc.) yields the causal flow feature set C. features .

[0141] Calculate the causal flow matrix:

[0142] C flow (i,j)=[GC(i→j)-GC(j→i)] / [GC(i→j)+GC(j→i)]

[0143] where GC(i→j) represents the Granger causality strength of parameter i on parameter j;

[0144] Extracting causality flow topology features:

[0145] CFT =∑[sign(C flow (i, P)) · |C flow (i, P) |]

[0146] Micro-meteorological disturbance feature: environment parameter dominant causality flow (CFT < 0);

[0147] Equipment failure feature: power dominant causality flow (CFT > 0).

[0148] S44, fusing the optimized feature set and the causality flow feature, obtaining the determination result of distinguishing micro-meteorological disturbance and equipment failure by cosine similarity matching with the historical feature template library.

[0149] Corresponding feature templates are established for different types of micro-meteorological disturbances (such as cloud cover, wind speed change, etc.), and corresponding feature templates are established for different types of equipment failures (such as inverter failure, component hot spot, etc.); a distance measurement method is constructed for evaluating the similarity between the feature vector and the template;

[0150] Calculate the matching degree of the optimized feature set F opt of the current sample with each type of template in the feature template library T_lib, to obtain a matching degree score vector S match .

[0151] Calculate the weighted Mahalanobis distance of the optimized feature set F opt with each template in the feature template library T_lib; based on the distance, calculate the matching degree score vector S match =[s1, s2,..., s k ]; where s k k represents the matching degree score of the kth type of template.

[0152] Based on the matching degree score vector, determine the type and confidence of the power fluctuation, and output the disturbance / failure determination result.

[0153] Select the type corresponding to the maximum matching degree score as the recognition result; calculate the confidence index of the recognition result; output the final diagnosis result R final ={Type: "type", Confidence: "confidence value"}; provide feature contribution degree analysis to explain the basis for decision-making.

[0154] The photovoltaic power plant power fluctuation feature extraction method distinguishing micro-meteorological disturbance and equipment failure of the application can effectively strengthen the difference of micro-meteorological disturbance and equipment failure in the frequency spectrum by denoising, standardizing, combining segmentation and category marking, wavelet analysis and spectrum feature enhancement of original power data, and the equipment failure is often accompanied by high-frequency energy mutation, and the wavelet energy feature is extracted to provide a key spectrum dimension basis for distinguishing the two; the elasticity parameters and entropy gradient features extracted by power recovery elasticity analysis, and the phase sequence features and extreme value network features obtained by phase space reconstruction and dynamic feature analysis, the comprehensive feature vector is constructed by integrating these features and wavelet energy features, which comprehensively reflects the nature of power fluctuation and improves the accuracy of distinction; based on the optimized feature set, bidirectional Granger causality is constructed, causal flow features are extracted, and the causal relationship between power data and auxiliary measurement parameters is associated by combining feature template matching, so that the source of power fluctuation is accurately identified.

[0155] In another embodiment of the application, the adaptive wavelet analysis and spectrum feature enhancement include:

[0156] The segmented and marked power data is evaluated by a multi-dimensional wavelet basis function, and the optimal wavelet basis function is selected by a comprehensive scoring function; the segmented and marked power data is adaptively thresholded and decomposed by the optimal wavelet basis function to obtain a wavelet coefficient set; the wavelet coefficient set is applied to nonlinear enhancement processing and reconstruction to obtain enhanced power data; the multi-scale wavelet energy distribution of the enhanced power data is calculated to obtain a wavelet energy feature vector, which is used for subsequent feature integration and disturbance type identification.

[0157] In this embodiment, the optimal wavelet basis is selected by multi-dimensional wavelet basis function evaluation, and the adaptive threshold decomposition and nonlinear enhancement processing can more accurately capture the frequency spectrum features of photovoltaic power fluctuation, improve the signal reconstruction accuracy, reduce the loss of key features, enhance the frequency spectrum differences of different types of fluctuations (micro-meteorological disturbance and equipment failure), and the multi-scale wavelet energy distribution feature can effectively distinguish the frequency characteristics of the disturbance (such as the wideband fluctuation of micro-meteorological disturbance and the specific frequency band mutation of equipment failure), which provides a key spectrum basis for type identification.

[0158] In another embodiment of the application, the power recovery elasticity analysis includes:

[0159] Adaptive detrending is performed on the enhanced power data to separate the trend fluctuation sequence and the trend component; based on the trend fluctuation sequence and the trend component, a full-state correlation elastic system model is constructed to extract an elastic parameter function set; the elastic parameter function set and the trend fluctuation sequence are used to construct a power state transition network to obtain a state transition feature set; multi-time scale entropy gradient analysis is performed on the enhanced power data to extract an entropy gradient feature set; the elastic parameter function set, the state transition feature set, and the entropy gradient feature set together construct power recovery elasticity features, which are used for subsequent feature integration and disturbance type identification.

[0160] The adaptive detrending process can accurately separate the trend component and the fluctuation component, avoiding feature distortion caused by fixed detrending methods; the full-state correlation elastic system model (inertia, damping, and stiffness parameters) can quantify the recovery "elasticity" differences of different fluctuation types (such as the rapid elastic recovery of micro-meteorological disturbances and the rigid decay of equipment failures); the state transition network and the multi-time scale entropy gradient analysis supplement the feature dimension from the perspectives of dynamic evolution and complexity, improving the adaptability to complex fluctuation scenarios and reducing the risk of misjudgment of single features.

[0161] This example takes the power data of a 100kW photovoltaic power station as the research object, and provides a power fluctuation feature extraction method for distinguishing between micro-meteorological disturbances and equipment failures. The complete process is divided into the following five steps:

[0162] (1) Photovoltaic power station power data processing

[0163] Obtain 1Hz sampling frequency, 2 hours of original power data , containing a cloud cover event and an inverter failure event, data range 0 to 95.3kW, where there is an abnormal value of 98.7kW.

[0164] Adaptive window median filtering (initial window 5, dynamic adjustment range 3-9) is used for denoising, combined with the 3σ rule to identify abnormal values (replaced with the local average value 95.1kW) to obtain denoised data ;

[0165] Normalize to the rated power 100kW:

[0166]

[0167] The normalized data range is 0 to 0.953;

[0168] Based on the power change rate threshold (0.02), minimum duration (10s), and other parameters, two fluctuation events are identified:

[0169] Event 1 (cloud cover, to );

[0170] Event 2 (inverter failure, to );

[0171] Get labeled power data .

[0172] (II) Adaptive wavelet analysis and spectrum feature enhancement

[0173] Evaluate db4, sym8, coif3, bior3.9, dmey 5 wavelet bases, calculate comprehensive score (weight is 0.4, 0.3, 0.3) through information retention (RI), energy concentration (EC), feature saliency (FS), and finally select bior3.9 with the highest score as the optimal wavelet base.

[0174] 5-scale wavelet decomposition is performed on the labeled data, and adaptive threshold is used:

[0175]

[0176] Among them, Soft threshold processing is performed to enhance the features;

[0177] Calculate the wavelet energy of each scale and the normalized distribution to get the wavelet energy feature vector:

[0178]

[0179] The wavelet energy feature vector contains the energy proportion of each scale and the scale energy ratio.

[0180] (III) Power recovery resilience analysis

[0181] Through the EMD method enhanced by change point detection, 7 significant change points are identified and segmented, and the 4th IMF component is extracted as the trend component to get the detrended fluctuation sequence.

[0182] Establish a full-state related elastic system model, divide the power range into 4 intervals, solve the polynomial parameters of inertia (M), damping (C), and stiffness (K), and extract the inertia step index (ISI=0.732), the damping change rate (DCR=0.351), and the stiffness nonlinear index (KNI=0.104);

[0183] Construct a state transition network to get the state residence time distribution (STD), jump distance distribution (JDD), and transition asymmetry index (TAI=0.153 and 0.478 respectively) of event 1 and event 2;

[0184] The multi-time scale entropy gradient analysis was performed, and the entropy gradient fluctuation response index (EGRI=2.048) and entropy scale complexity (ESC=0.922) of event 1 were calculated. The EGRI and ESC of event 2 were 2.987 and 0.702, respectively.

[0185] (Four) Phase space reconstruction and dynamic characteristic analysis

[0186] Optimal time delay based on mutual information method (mutual information appears the first local minimum value at Optimal embedding dimension based on false nearest neighbor method (false nearest neighbor ratio <0.05 when

[0187] Reconstructing the phase space trajectory, the phase sequence characteristics were calculated:

[0188] The angular velocity variation coefficient (V_ω=0.271) and phase jump index (Jump_index=12) of event 1;

[0189] The V_ω and Jump_index of event 2 were 0.844 and 41, respectively;

[0190] The network asymmetry index (NAI=0.011) and clustering coefficient ratio (CCR=1.060) of event 1 were obtained by constructing the extreme value network. The NAI and CCR of event 2 were 0.108 and 1.279, respectively.

[0191] (Five) Causal relationship analysis and pattern recognition

[0192] Integrating wavelet energy, elastic parameter, state transition, entropy gradient, phase sequence, and extreme value network characteristics, a comprehensive feature vector was formed, and the optimal feature subset was selected by the maximum correlation minimum redundancy (mRMR) principle.

[0193] The bidirectional Granger causality between power data and irradiance, temperature, and wind speed was constructed, and the causal flow topology characteristics (CFT=-1.694 for event 1 and CFT=0.743 for event 2) were calculated, reflecting the differences in the dominant relationship between environmental parameters and power.

[0194] Based on the feature template library (containing 50 micro-meteorological disturbance and 50 equipment failure templates), the matching degree was calculated by weighted Mahalanobis distance. It includes:

[0195] Event 1 (cloud cover) has a higher matching degree with micro-meteorological disturbance templates (reliability 0.692), and is determined as "micro-meteorological disturbance";

[0196] Event 2 (inverter failure) has a higher matching degree with equipment failure templates (reliability 0.726), and is determined as "equipment failure".​

[0197] The embodiment shows the effectiveness of the photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbance and equipment failure through specific numerical calculation and processing process, realizes effective distinction of two types of power fluctuation events of the photovoltaic power plant, and verifies the feasibility of the method.

[0198] In the embodiment, the adaptive threshold wavelet analysis based on local fluctuation characteristics can adopt differentiated processing strategies for different fluctuation intensity regions; the full-state related elastic system model models the system inertia parameter as a state-related function, which is more in line with the physical characteristics of the photovoltaic system. The multi-time scale entropy gradient analysis can capture the dynamic change characteristics of the system order from the information entropy angle. The extreme scale network and its topological characteristics describe the power fluctuation characteristics from the perspective of complex network. The bidirectional causal flow topological characteristics provide evidence of the causal relationship between environmental parameters and power fluctuations.

[0199] In a second aspect, the embodiment of the present application also provides a photovoltaic power plant power fluctuation feature extraction device for distinguishing micro-meteorological disturbance and equipment failure, which applies the method described above, and comprises:

[0200] A data acquisition processing module is configured to acquire original power data of the photovoltaic power plant, and perform denoising and standardization processing on the original power data, segment and label the processed data, and obtain segmented and labeled power data.

[0201] A data enhancement module is configured to perform wavelet analysis and spectrum feature enhancement on the segmented and labeled power data, and obtain enhanced power data and wavelet energy features.

[0202] A feature extraction module is configured to perform power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features, and simultaneously perform phase space reconstruction and dynamic feature analysis to extract phase sequence features and extreme value network features.

[0203] A feature recognition module is configured to perform causal relationship analysis and pattern recognition based on the wavelet energy features, the elasticity parameters, the entropy gradient features, the phase sequence features and the extreme value network features, and obtain a judgment result for distinguishing micro-meteorological disturbance and equipment failure.

[0204] The functions of each unit in the embodiment are the same as those of the photovoltaic power plant power fluctuation feature extraction method for distinguishing micro-meteorological disturbance and equipment failure, and the technical effects are the same, and thus will not be repeated here.

[0205] Those skilled in the art can appreciate that the units of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0206] In the embodiments provided by the present application, it should be understood that the division of units is only a logical functional division, and there can be another division manner in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0207] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0208] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0209] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the description of the present application.

Claims

1. A method for extracting power fluctuation characteristics of a photovoltaic power plant to distinguish between micro-meteorological disturbances and equipment failures, characterized in that, include: The raw power data of the photovoltaic power station is acquired and denoised and standardized. The processed data is then segmented and categorized to obtain segmented and categorized power data. Wavelet analysis and spectral feature enhancement are performed on the segmented marked power data to obtain enhanced power data and wavelet energy features; Based on the enhanced power data, power recovery elasticity analysis is performed to extract elastic parameters and entropy gradient features. Simultaneously, phase space reconstruction and dynamic feature analysis are performed to extract phase sequence features and extreme value network features. Specifically, this includes: reconstructing the phase space of the enhanced power data based on predetermined optimal time delay parameters and embedding dimensions to obtain a phase space trajectory; performing phase sequence analysis on the phase space trajectory to extract phase sequence features including angular velocity variation coefficient, phase jump index, and root mean square error of phase angle; and constructing a complex network based on the multi-scale extreme points of the enhanced power data to extract extreme value network features including network average degree and clustering coefficient ratio. Based on the wavelet energy features, elastic parameters, entropy gradient features, phase sequence features, and extreme value network features, causal relationship analysis and pattern recognition are performed to obtain a judgment result that distinguishes between micro-meteorological disturbances and equipment failures. Specifically, this includes: constructing a comprehensive feature vector based on the wavelet energy features, elastic parameters, entropy gradient features, phase sequence features, and extreme value network features; optimizing the comprehensive feature vector by using L1 regularization to select features and obtain an optimized feature set; constructing a bidirectional Granger causal relationship based on the optimized feature set and auxiliary measurement parameters, including meteorological parameters and equipment parameters, and extracting causal flow features; fusing the optimized feature set and causal flow features, and obtaining a judgment result that distinguishes between micro-meteorological disturbances and equipment failures by cosine similarity matching with a historical feature template library.

2. The method for extracting power fluctuation characteristics of a photovoltaic power station to distinguish between micro-meteorological disturbances and equipment failures according to claim 1, characterized in that, The wavelet analysis and spectral feature enhancement of the segmented marked power data include: Using the pre-selected optimal wavelet basis function, an adaptive thresholding method based on local wave characteristics is employed to perform wavelet decomposition on the segmented labeled power data to obtain wavelet coefficients. The wavelet coefficients are subjected to nonlinear enhancement processing and reconstructed to obtain enhanced power data; Based on the enhanced power data, the multi-scale wavelet energy distribution is calculated to obtain the wavelet energy characteristics.

3. The method for extracting power fluctuation characteristics of a photovoltaic power station to distinguish between micro-meteorological disturbances and equipment failures according to claim 1, characterized in that, The step of performing power recovery elasticity analysis based on the enhanced power data to extract elasticity parameters and entropy gradient features includes: The enhanced power data is linearly detrended to separate the fluctuation components and obtain a detrended fluctuation sequence. Based on the detrended fluctuation sequence, a state-dependent spring-damped system model is established and the elastic parameters are extracted. The enhanced power data is subjected to sliding window information entropy calculation and entropy gradient analysis to extract entropy gradient features.

4. The method for extracting power fluctuation characteristics of a photovoltaic power station to distinguish between micro-meteorological disturbances and equipment failures according to claim 3, characterized in that, The step of performing sliding window information entropy calculation and entropy gradient analysis on the enhanced power data, and extracting entropy gradient features, includes: By calculating the information entropy sequence within the sliding time window; Calculate the entropy gradient sequence based on the information entropy sequence; The entropy gradient fluctuation response index is extracted based on the entropy gradient sequence. The entropy gradient fluctuation response index is used as a key feature parameter to distinguish between disturbances and faults.

5. A photovoltaic power plant power fluctuation feature extraction device for distinguishing between micro-meteorological disturbances and equipment failures, characterized in that, The method described by any one of claims 1 to 4 includes: The data acquisition and processing module is used to acquire the raw power data of the photovoltaic power station, and perform noise reduction and standardization processing. The processed data is segmented and categorized to obtain segmented and categorized power data. The data augmentation module is used to perform wavelet analysis and spectral feature enhancement on the segmented marked power data to obtain enhanced power data and wavelet energy features; The feature extraction module is used to perform power recovery elasticity analysis based on the enhanced power data to extract elastic parameters and entropy gradient features, and simultaneously perform phase space reconstruction and dynamic feature analysis to extract phase sequence features and extreme value network features. Specifically, it includes: reconstructing the phase space of the enhanced power data based on predetermined optimal time delay parameters and embedding dimensions to obtain a phase space trajectory; performing phase sequence analysis on the phase space trajectory to extract phase sequence features including angular velocity variation coefficient, phase jump index, and root mean square error of phase angle; and constructing a complex network based on the multi-scale extreme points of the enhanced power data to extract extreme value network features including network average degree and clustering coefficient ratio. The feature recognition module is used to perform causal relationship analysis and pattern recognition based on the wavelet energy features, elastic parameters, entropy gradient features, phase sequence features, and extreme value network features to obtain a judgment result that distinguishes between micro-meteorological disturbances and equipment failures. Specifically, it includes: constructing a comprehensive feature vector based on the wavelet energy features, elastic parameters, entropy gradient features, phase sequence features, and extreme value network features; optimizing the comprehensive feature vector by using L1 regularization to select features and obtain an optimized feature set; constructing a bidirectional Granger causal relationship based on the optimized feature set and auxiliary measurement parameters, including meteorological parameters and equipment parameters, and extracting causal flow features; fusing the optimized feature set and causal flow features, and obtaining a judgment result that distinguishes between micro-meteorological disturbances and equipment failures by matching with the cosine similarity of the feature template library.

Images