CN-117150242-B - Wind field group power prediction method

CN117150242BCN 117150242 BCN117150242 BCN 117150242BCN-117150242-B

Abstract

The invention provides a wind field group power prediction method, which belongs to the technical field of wind field power generation and comprises the steps of obtaining a first sample set, wherein the first sample set comprises a historical three-dimensional matrix of starting moments of different target historical time periods and actual output total power of a wind field group at a cut-off moment, taking the historical three-dimensional matrix as input, taking the actual output total power of the wind field group as output, training to obtain a power prediction model, obtaining a target three-dimensional matrix at the starting moment of a period to be predicted, inputting the target three-dimensional matrix into the power prediction model, and obtaining the wind field group prediction output total power at the cut-off moment of the period to be predicted. According to the method, a power prediction model is trained through previous data, the model can predict the power change of the current period based on the influence of parameters on the output power of the wind field group in a short period, the parameter value of the deadline of the current period is input, the prediction value can provide control guidance, and dynamic decisions are timely made through a wind field control end to ensure stable output of wind energy.

Inventors

OUYANG HENG
WANG HAOYANG
HAN XU
DUAN SHUYONG

Assignees

河北工业大学

Dates

Publication Date: 20260505
Application Date: 20230915

Claims (9)

1. The wind field group power prediction method is characterized by comprising the following steps of: Acquiring a first sample set, wherein the first sample set comprises a historical three-dimensional matrix of starting moments of different target historical time periods and actual output total power of wind field groups at the ending moments of the target historical time periods, the historical three-dimensional matrix is obtained based on power related data and parameter related data of a historical time period ending moments before the target historical time periods, the power related data is obtained by processing the actual output total power of the wind field groups of the historical time periods, and the parameter related data is obtained by processing a plurality of wind field group parameters; The method comprises the steps of taking a historical three-dimensional matrix as input, taking actual output total power of a wind field group as output, training an initial neural network to obtain a power prediction model, obtaining a target three-dimensional matrix of a starting moment of a period to be predicted, and processing the target three-dimensional matrix based on power-related data and parameter-related data of a historical period cut-off moment before the period to be predicted; acquiring the historical three-dimensional matrix of the target historical period starting time, comprising the following steps: Acquiring a stable time matrix, wherein the stable time matrix consists of stable data of the total power actually output by the wind field group at a time of the cut-off of the historical time periods a before the target historical time period; Acquiring a parameter joint probability matrix, wherein the wind field is provided with a plurality of parameters affecting the output power of the wind field, and the parameter joint probability matrix consists of joint probability or conditional joint probability of each parameter between every two wind fields at a historical period deadline time before the target historical period; Acquiring a parameter cross characteristic matrix, wherein the parameter cross characteristic matrix consists of cross characteristics of each parameter between a basic wind field and each other wind field at a historical period cut-off time before the target historical period; And assembling the stable time matrix, the parameter joint probability matrix and the parameter cross characteristic matrix into the historical three-dimensional matrix of the starting moment of the target historical period according to time sequence.
2. The wind farm group power prediction method according to claim 1, wherein the obtaining of the stationary time matrix comprises the steps of: Acquiring an original time sequence, wherein the original time sequence comprises original data of the total power actually output by the wind field group at a time of the expiration of a historical time periods before the target historical time period; And carrying out stationarity detection on the original time sequence, and carrying out trending treatment on the original data under the historical period deadline which does not pass through the stationarity detection, so as to obtain the stationary data under the historical period deadline.
3. The method for predicting the power of a wind farm group according to claim 2, wherein the wind farm group comprises a plurality of wind farms, the raw data of the total power actually output by the wind farm group is the sum of the actual output powers of the wind farms at the time when the historical period is cut off, and the raw data which does not pass the stationarity detection is subjected to a trending process by adopting the following formula (1): Formula (1) Wherein t represents the historical period expiration time that fails the stationarity detection, Represents stationary data at the expiration time of the t-th history period, Raw data representing the actual output power of the jth wind farm, Polynomial fit data representing the actual output power of the jth wind farm, m being the total number of wind farms for the wind farm group.
4. A wind farm group power prediction method according to claim 3, wherein the joint probability or conditional joint probability between each two wind farms for each of the parameters is calculated using the following formula (2): Formula (2) Wherein t represents the expiration time of different history periods, The value of the ith parameter representing the jth wind field, n representing the number of wind field parameters, is expressed by a simplified formula, and is described below Instead, F i represents an m-dimensional joint distribution function, Representing parameters F i represents an m-dimensional joint probability density function, Representing parameters C i is a Copula function, and C i is a Copula density function.
5. The wind farm group power prediction method according to claim 4, wherein equation (2) is simplified by equation (3): Formula (3) Wherein, the Indicating that the i-th parameter is at the level of 1, 2, A conditional edge probability density function of an mth wind field under m-1 wind field conditions, Representing Copula density functions of the ith parameter at the 1 st wind farm and the 2 nd wind farm, Representing the condition Copula density function of the ith parameter in the 1 st wind field and the 3 rd wind field under the 2 nd wind field condition, Representing a conditional edge probability distribution function of the jth wind field of the ith parameter under the kth wind field condition; and calculating the joint distribution function and the joint probability density function of the ith parameter in the jth wind field and the kth wind field according to the formula through the Copula function and the density function thereof, wherein the Copula function value is the joint probability or the conditional joint probability of the ith parameter between the jth wind field and the kth wind field.
6. The wind farm group power prediction method according to claim 1, wherein the intersection characteristics of each of the parameters between the base wind farm and the respective remaining wind farms are calculated using the following formula (4): formula (4) Wherein b represents a basic wind field, j represents the rest of wind fields, namely, any wind field which is not b, t represents the expiration time of different history periods, The value of the ith parameter representing the basic wind park b at the expiration of the t history period, The value of the i-th parameter representing the remaining wind park j at the expiration of the t history period, Representing the crossing characteristics of the ith parameter at the expiration of the tschet period between the base wind park b and the remaining wind parks j.
7. The wind farm group power prediction method according to claim 1, wherein assembling the stationary time matrix, the parameter joint probability matrix, and the parameter cross feature matrix into the historical three-dimensional matrix of the target historical period start time according to a time sequence comprises the steps of: Sequentially assembling and rearranging stable data of actual total output power of the wind field group at the same historical period cut-off time, joint probability or conditional joint probability of each parameter between every two wind fields, and cross characteristics of each parameter between a basic wind field and each other wind field to obtain a historical two-dimensional matrix (1) at the historical period cut-off time; and reassembling each history two-dimensional matrix (1) according to the time sequence of each history period deadline to obtain the history three-dimensional matrix.
8. The wind farm group power prediction method according to claim 1, wherein obtaining the wind farm group actual output total power at the target historical period cutoff time comprises the steps of: acquiring actual output single power of each wind field at the target historical period cut-off moment; And summing the actual output single power of each wind field and carrying out normalization processing to obtain the actual output total power of the wind field group.
9. The method of claim 1, wherein the parameters include at least temperature, wind speed, wind direction, each of the parameters having a time-varying nature.

Description

Wind field group power prediction method Technical Field The invention relates to the technical field of wind farm power generation, in particular to a wind farm group power prediction method. Background As fossil energy reserves are gradually depleted, the results of leading-edge fields such as photovoltaic power generation technology, tidal power generation technology, wind power generation technology and the like bloom throughout. Among them, wind energy is a clean resource which can be utilized, and has recently received a great deal of attention from researchers at home and abroad. However, the generation of wind energy is greatly influenced by the change of the wind field state, and if the wind field control end cannot make a dynamic decision in time according to the wind field state, the collection and the utilization of wind energy are limited. The wind field state can be represented by parameters such as wind speed, temperature, wind direction and the like, the wind field parameters are objectively influenced by multiple factors such as time, seasons and the like, the uncertainty distribution and the mutual coupling form are complex, the uncertainty and the correlation of the parameters are accurately measured, the influence rule of the uncertainty and the correlation on the output power in the short period of the wind field is favorably analyzed, and finally the prediction of the short period of the output power of the wind field is realized, so that the wind field control decision is effectively guided. Disclosure of Invention In view of the foregoing drawbacks or shortcomings of the prior art, the present invention is directed to a wind farm group power prediction method, comprising: Acquiring a first sample set, wherein the first sample set comprises a historical three-dimensional matrix of starting moments of different target historical time periods and actual output total power of wind field groups at the ending moments of the target historical time periods, the historical three-dimensional matrix is obtained based on power related data and parameter related data of a historical time period ending moments before the target historical time periods, the power related data is obtained by processing the actual output total power of the wind field groups of the historical time periods, and the parameter related data is obtained by processing a plurality of wind field group parameters; The method comprises the steps of taking the historical three-dimensional matrix as input, taking the actual total power output by a wind field group as output, training an initial neural network to obtain a power prediction model, obtaining a target three-dimensional matrix of the starting moment of a period to be predicted, processing the target three-dimensional matrix based on power related data and parameter related data of a historical period cut-off moment before the period to be predicted, and inputting the target three-dimensional matrix into the power prediction model to obtain the wind field group prediction total power output of the period to be predicted at the cut-off moment. According to the technical scheme provided by the invention, the historical three-dimensional matrix of the target historical period starting time is obtained, and the method comprises the following steps: Acquiring a stable time matrix, wherein the stable time matrix consists of stable data of the total power actually output by the wind field group at a time of the cut-off of the historical time periods a before the target historical time period; Acquiring a parameter joint probability matrix, wherein the wind field is provided with a plurality of parameters affecting the output power of the wind field, and the parameter joint probability matrix consists of joint probability or conditional joint probability of each parameter between every two wind fields at a historical period deadline time before the target historical period; Acquiring a parameter cross characteristic matrix, wherein the parameter cross characteristic matrix consists of cross characteristics of each parameter between a basic wind field and each other wind field at a historical period cut-off time before the target historical period; And assembling the stable time matrix, the parameter joint probability matrix and the parameter cross characteristic matrix into the historical three-dimensional matrix of the starting moment of the target historical period according to time sequence. According to the technical scheme provided by the invention, the stable time matrix is obtained, and the method comprises the following steps: Acquiring an original time sequence, wherein the original time sequence comprises original data of the total power actually output by the wind field group at a time of the expiration of a historical time periods before the target historical time period; And carrying out stationarity detection on the original time sequence, and carrying out trendin