CN-122023060-A - Electric power data acquisition method and system for breeze power generation system

CN122023060ACN 122023060 ACN122023060 ACN 122023060ACN-122023060-A

Abstract

The invention belongs to the technical field of data acquisition, and particularly relates to a power data acquisition method and system for a breeze power generation system, which are used for solving the technical problems that the geometrical space of the existing power data processing method is not properly selected and the physical priori constraint of an operation mode is lacking. The acquisition method comprises the following steps of S1, obtaining time sequence power data in the running process of the breeze power generation system, wherein the time sequence power data form high-dimensional power data, S2, constructing a high-dimensional probability map of the high-dimensional power data, S3, performing dimension reduction embedding on the high-dimensional probability map in a normal negative curvature space, and S4, outputting a low-dimensional embedding point layout and an optimal normal negative curvature value obtained by the optimization algorithm, and taking the low-dimensional embedding point layout and the optimal normal negative curvature value as a structural processing result of the high-dimensional power data. The invention can display the hierarchical structure in the data and is suitable for the relation of the power generation system among different operation modes.

Inventors

Pang Senxiang
TANG HUIHONG
Pang Gunhong
CHEN YANPING
LU WEIPENG
NIU WEIDONG
TAN RUNHUA
ZHANG JIANJUN
GAO JIABIN
WU ZICHENG
LAI JIANHONG
LIU XINLONG

Assignees

广州佰宏新能源科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (10)

1. The electric power data acquisition method for the breeze power generation system is characterized by comprising the following steps of: s1, acquiring time sequence power data in the running process of a breeze power generation system, wherein the time sequence power data form high-dimensional power data; S2, constructing a high-dimensional probability map of high-dimensional power data, wherein the construction process comprises the steps of defining a local Riemann metric based on local wind speed fluctuation rate and pitch angle adjustment frequency of each high-dimensional power data point, and calculating geodesic distances among the high-dimensional power data points to obtain initial connection weights; S3, performing dimension reduction embedding on the high-dimensional probability map in a normally negative curvature space, namely projecting the high-dimensional power data onto a theoretical power curve based on a Betts limit theory, initializing one dimension of low-dimensional embedding representation by using a projection position of the high-dimensional power data, constructing a normally negative curvature value which takes the structural difference between the minimum high-dimensional and low-dimensional probability maps as an objective function, iterating through an optimization algorithm, and optimizing the layout of low-dimensional embedding points and the normally negative curvature value of the normally negative curvature space; and S4, outputting the low-dimensional embedded point layout and the optimal constant negative curvature value obtained by the optimization algorithm, and taking the low-dimensional embedded point layout and the optimal constant negative curvature value as a structural processing result of the high-dimensional power data.
2. The method for collecting power data for a breeze power generation system according to claim 1, wherein in S1, when time-series power data during operation of the breeze power generation system is obtained, wind speed, active power, pitch angle and rotor rotation speed of a generator set are synchronously collected to form time-series power data at least comprising four dimensions.
3. The method for collecting power data for a breeze power generation system according to claim 1, wherein the process of defining a local Riemann metric based on the local wind speed fluctuation rate and the pitch angle adjustment frequency of each high-dimensional power data point and calculating the geodesic distance between the high-dimensional power data points to obtain the initial connection weight comprises: for each high-dimensional power data point Calculating the standard deviation of wind speed in a window formed by 10 sampling points before and after as the local wind speed fluctuation rate And calculating the rate of change of pitch angle in the window as the pitch angle adjustment frequency ; Local wind speed volatility for all high-dimensional power data points And pitch angle adjustment frequency Respectively carrying out normalization processing to obtain normalized fluctuation rate And normalizing the frequency ; Obtaining a metric scaling factor from a formula Wherein And For preset weight, the measurement scale factor Scale for defining local Riemann metrics and constructing Riemann metric tensors Wherein Is a unit matrix; Calculating geodesic distance between high-dimensional power data points by adopting fast travelling algorithm And measure the ground wire distance Is a negative exponential function value of (2) As an initial connection weight.
4. A method for collecting power data for a breeze power generation system according to claim 3, wherein the process of adjusting the initial connection weight by using a preset state transition probability matrix comprises: distributing a determined operation mode label to each high-dimensional power data point, wherein the operation mode is one of starting, grid-connected power generation, fault and stopping; for any two high-dimensional power data points, searching a probability value of transferring the operation mode label of the first high-dimensional power data point to the operation mode label of the second high-dimensional power data point from the state transfer probability matrix; Multiplying the probability value with the initial connection weight between the two high-dimensional power data points to obtain the adjusted connection weight.
5. The method for collecting power data for a breeze power generation system according to claim 1, wherein the process of projecting the high-dimensional power data onto a theoretical power curve based on the betz limit theory and initializing one dimension of the low-dimensional embedded representation using the projection position thereof comprises: generating a standard wind speed-power theoretical curve according to rated parameters of the generator set; for each high-dimensional power data point, extracting the values of wind speed and active power dimension to form a two-dimensional point, and searching a theoretical point closest to Euclidean distance of the two-dimensional point on the wind speed-power theoretical curve; And taking the abscissa value of the theoretical point as an initial set value of a first-dimension coordinate of the high-dimension power data point in the low-dimension embedded representation, wherein the abscissa value is the theoretical wind speed.
6. The method for collecting power data for breeze power generation system according to claim 5, wherein the process of constructing an objective function targeting to minimize structural differences between high-dimensional and low-dimensional probability maps, iterating through an optimization algorithm, and optimizing the layout of low-dimensional embedded points and the constant negative curvature value of the space comprises: performing symmetry and normalization processing on the adjusted connection weights to obtain high-dimensional connection weight probability distribution ; Objective function Probability distribution for high-dimensional connection weights Probability distribution induced by hyperbolic distance between embedded points in low dimension KL divergence between: ; Wherein the method comprises the steps of For a low-dimensional embedded point set, Is a constant negative curvature of the space and, For embedding points in low dimensions And Probability values between; adopting an Adam optimizer with a learning rate of 0.01 to carry out gradient descent on the objective function, synchronously updating the coordinates of the low-dimensional embedded points and the constant negative curvature value of the space, and iterating until the variation of the objective function value is continuously iterated 100 times to be smaller than And then converges.
7. A power data acquisition system for a breeze power generation system, comprising: The acquisition module is used for acquiring time sequence power data in the operation process of the breeze power generation system, and the time sequence power data form high-dimensional power data; The construction module is used for constructing a high-dimensional probability map of high-dimensional power data, and the construction process comprises the steps of defining a local Riemann metric based on local wind speed fluctuation rate and pitch angle adjustment frequency of each high-dimensional power data point, calculating geodesic distance between the high-dimensional power data points to obtain initial connection weight, and adjusting the initial connection weight by using a preset state transition probability matrix, wherein the state transition probability matrix is used for representing the possibility of conversion of a generator among starting, grid-connected power generation, fault and shutdown operation modes; The optimization module is used for performing dimension reduction embedding on the high-dimensional probability map in a normally negative curvature space, namely projecting the high-dimensional power data onto a theoretical power curve based on a Betts limit theory, initializing one dimension of low-dimensional embedding representation by using the projection position of the high-dimensional power data, constructing a normally negative curvature value which takes the structural difference between the minimum high-dimensional and low-dimensional probability maps as an objective function, iterating through an optimization algorithm, and optimizing the layout of low-dimensional embedding points and the normally negative curvature space; and the output module is used for outputting the low-dimensional embedded point layout and the optimal constant negative curvature value obtained by the optimization algorithm as a structural processing result of the high-dimensional power data.
8. The system according to claim 7, wherein the acquisition module acquires the wind speed, active power, pitch angle and rotor rotation speed of the generator set synchronously to form time series power data including at least four dimensions when acquiring time series power data during operation of the breeze power generation system.
9. The system for collecting power data for a breeze power generation system according to claim 7, wherein the process of defining the local Riemann metric based on the local wind speed fluctuation rate and the pitch angle adjustment frequency of each high-dimensional power data point and calculating the geodesic distance between the high-dimensional power data points to obtain the initial connection weight comprises: for each high-dimensional power data point Calculating the standard deviation of wind speed in a window formed by 10 sampling points before and after as the local wind speed fluctuation rate And calculating the rate of change of pitch angle in the window as the pitch angle adjustment frequency ; Local wind speed volatility for all high-dimensional power data points And pitch angle adjustment frequency Respectively carrying out normalization processing to obtain normalized fluctuation rate And normalizing the frequency ; Obtaining a metric scaling factor from a formula Wherein And For preset weight, the measurement scale factor Scale for defining local Riemann metrics and constructing Riemann metric tensors Wherein Is a unit matrix; Calculating geodesic distance between high-dimensional power data points by adopting fast travelling algorithm And measure the ground wire distance Is a negative exponential function value of (2) As an initial connection weight.
10. The system for collecting power data for breeze power generation system according to claim 9, wherein the process of adjusting the initial connection weight by using the preset state transition probability matrix in the construction module comprises: distributing a determined operation mode label to each high-dimensional power data point, wherein the operation mode is one of starting, grid-connected power generation, fault and stopping; for any two high-dimensional power data points, searching a probability value of transferring the operation mode label of the first high-dimensional power data point to the operation mode label of the second high-dimensional power data point from the state transfer probability matrix; Multiplying the probability value with the initial connection weight between the two high-dimensional power data points to obtain the adjusted connection weight.

Description

Electric power data acquisition method and system for breeze power generation system Technical Field The invention belongs to the technical field of data acquisition, and particularly relates to a method and a system for acquiring electric power data for a breeze power generation system. Background Breeze power generation is an important component of distributed energy, and is increasingly widely applied to remote areas and urban building scenes. During the running process of the breeze power generation system, the sensor can acquire a large amount of time sequence power data including wind speed, wind direction, generator rotating speed, output power, pitch angle, grid frequency, voltage and current. These data have the characteristics of high dimensionality, strong nonlinearity, tight coupling and noise inclusion, forming a complex high-dimensional data stream. The traditional electric power data processing method is difficult to capture deep nonlinear association and internal geometric structures among data samples, and the deep learning-based method can be used for processing nonlinear problems, but is often used as a 'black box' model, the processing result lacks visual physical interpretation, and the system operation mode conversion rule contained in the data cannot be fully utilized. To reveal the manifold structure behind high-dimensional power data, some advanced techniques began to introduce manifold learning or graph embedding methods that attempted to preserve the neighborhood structure in low-dimensional space by constructing a neighbor relation graph between data points, thereby achieving nonlinear dimension reduction and visualization. However, the existing methods have some defects that the existing methods default to embed data into a flat Euclidean space, which has inherent limitations for expressing operation modes with natural hierarchical structures or complex topological relations, and meanwhile, the existing methods fail to guide and correct the construction and embedding processes of a graph structure by taking the objectively existing transition probability between different operation modes as priori knowledge, so that the dimension reduction result cannot reflect the real evolution logic of the system. Disclosure of Invention The invention provides a power data acquisition method and system for a breeze power generation system, which are used for solving the technical problems that the geometrical space of the existing power data processing method is not properly selected and the physical priori constraint of an operation mode is lacking. In a first aspect, the present invention provides a method for collecting power data for a breeze power generation system, including the steps of: s1, acquiring time sequence power data in the running process of a breeze power generation system, wherein the time sequence power data form high-dimensional power data; S2, constructing a high-dimensional probability map of high-dimensional power data, wherein the construction process comprises the steps of defining a local Riemann metric based on local wind speed fluctuation rate and pitch angle adjustment frequency of each high-dimensional power data point, and calculating geodesic distances among the high-dimensional power data points to obtain initial connection weights; S3, performing dimension reduction embedding on the high-dimensional probability map in a normally negative curvature space, namely projecting the high-dimensional power data onto a theoretical power curve based on a Betts limit theory, initializing one dimension of low-dimensional embedding representation by using a projection position of the high-dimensional power data, constructing a normally negative curvature value which takes the structural difference between the minimum high-dimensional and low-dimensional probability maps as an objective function, iterating through an optimization algorithm, and optimizing the layout of low-dimensional embedding points and the normally negative curvature value of the normally negative curvature space; and S4, outputting the low-dimensional embedded point layout and the optimal constant negative curvature value obtained by the optimization algorithm, and taking the low-dimensional embedded point layout and the optimal constant negative curvature value as a structural processing result of the high-dimensional power data. Further, in S1, when time-series power data in the running process of the breeze power generation system is obtained, wind speed, active power, pitch angle and rotor rotation speed of the generator set are synchronously collected, so as to form time-series power data at least including four dimensions. Further, a process of defining a local Riemann metric based on local wind speed fluctuation and pitch angle adjustment frequency for each high-dimensional power data point and calculating geodesic distances between the high-dimensional power data points to obtain initial conne