A wind farm cluster power prediction method
By establishing a wind farm power prediction model and training a neural network using historical data, the problem of wind energy being greatly affected by changes in wind field conditions was solved, and stable collection and utilization of wind energy was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEBEI UNIV OF TECH
- Filing Date
- 2023-09-15
- Publication Date
- 2026-05-05
AI Technical Summary
Wind energy generation is greatly affected by changes in wind field conditions, and existing technologies make it difficult to accurately predict the short-term output power of wind fields, which limits wind energy collection and utilization.
By acquiring historical three-dimensional matrices, training an initial neural network, and establishing a power prediction model, the power prediction of wind farm clusters is performed using parameter-related data and power-related data, predicting the short-term changes in the total output power of the wind farm clusters.
It enables accurate prediction of short-term output power of wind farm clusters, provides dynamic decision-making guidance for wind farm control, and ensures stable wind energy output.
Smart Images
Figure CN117150242B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power generation technology, and more specifically to a method for predicting the power of a wind farm cluster. Background Technology
[0002] With fossil fuel reserves dwindling, cutting-edge technologies such as photovoltaic power generation, tidal power generation, and wind power generation are flourishing. Among these, wind energy, as a usable clean resource, has received widespread attention from researchers both domestically and internationally in recent years. However, wind energy generation is significantly affected by changes in wind field conditions. If the wind field control system cannot make dynamic decisions based on wind field conditions in a timely manner, the collection and utilization of wind energy will inevitably be limited.
[0003] Wind field conditions can be characterized by parameters such as wind speed, temperature, and wind direction. However, these wind field parameters are objectively affected by multiple factors such as time and season, and their uncertainty distribution and mutual coupling are complex. Accurately measuring the uncertainty and correlation of these parameters is beneficial to analyzing their impact on the short-term output power of the wind field, and ultimately to predicting the short-term output power of the wind field, thereby effectively guiding wind field control decisions. Summary of the Invention
[0004] In view of the above-mentioned defects or deficiencies in the prior art, the present invention aims to provide a method for predicting wind farm group power, comprising:
[0005] Obtain a first sample set, which includes a historical three-dimensional matrix of the start time of different target historical periods and the actual total output power of the wind field cluster at the end time of the target historical period; the historical three-dimensional matrix is based on the historical periods prior to the target historical period. a The power-related data and parameter-related data are obtained from the end time of each historical period; wherein the power-related data is obtained by processing the actual total output power of the wind farm group during the historical period, and the parameter-related data is obtained by processing the parameters of multiple wind farm groups.
[0006] Using the historical 3D matrix as input and the actual total output power of the wind field cluster as output, the initial neural network is trained to obtain a power prediction model; the target 3D matrix at the start of the period to be predicted is obtained, the target 3D matrix being based on the period prior to the period to be predicted. a The power-related data and parameter-related data at the end of each historical time period are processed to obtain the target three-dimensional matrix. The target three-dimensional matrix is then input into the power prediction model to obtain the predicted total power of the wind field group at the end of the time period to be predicted.
[0007] According to the technical solution provided by the present invention, obtaining the historical three-dimensional matrix at the start time of the target historical period includes the following steps:
[0008] Obtain a stationary time matrix, which is composed of historical data prior to the target time period. a The data consists of stable data on the actual total output power of the wind farm group at the cut-off time of each historical period.
[0009] Obtain the joint probability matrix of parameters. The wind field has multiple parameters that affect its output power. The joint probability matrix of parameters is derived from the historical data of the target period prior to this period. a The joint probability or conditional joint probability of each parameter at the cutoff time of each of the historical periods is composed of the joint probability or conditional joint probability between each of the two wind fields;
[0010] Obtain the parameter cross feature matrix, which is composed of parameters from the previous historical time period of the target. a The parameters at the cutoff time of each of the historical periods are composed of the cross-features between the base wind field and each of the remaining wind fields; the remaining wind fields are the other wind fields in the wind field group besides the base wind field.
[0011] The stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters are assembled according to time sequence into the historical three-dimensional matrix at the start time of the target historical period.
[0012] According to the technical solution provided by the present invention, obtaining a stationary time matrix includes the following steps:
[0013] Obtain the original time series, which includes the period prior to the target historical period. a The original data of the actual total output power of the wind farm group at the cutoff time of the aforementioned historical period;
[0014] The original time series is subjected to stationarity detection. The original data at the cutoff time of the historical period that fails the stationarity detection is detrended to obtain the stationary data at the cutoff time of the historical period.
[0015] According to the technical solution provided by the present invention, the wind farm group includes multiple wind farms, and the original data of the actual total output power of the wind farm group is the sum of the actual output power of each wind farm at the end of the historical period; the following formula (1) is used to perform detrending processing on the original data that fails the stationarity test:
[0016] Formula (1)
[0017] in, t This indicates the cutoff time of the historical period that failed the stationarity test. Indicates the first t Stable data at the cutoff point of each historical period Indicates the firstj The raw data of the actual output power of each wind farm. Indicates the first j Polynomial fitting data of the actual output power of each wind farm m This represents the total number of wind farms in the wind farm cluster.
[0018] According to the technical solution provided by the present invention, the joint probability or conditional joint probability of each parameter between every two wind fields is calculated using the following formula (2):
[0019] Formula (2)
[0020] in, t Indicates the end time for different historical periods. Indicates the first j The first wind farm i The values of the parameters, n The number of wind field parameters is used to simplify the formula. As an alternative F i express m dimensional joint distribution function, Indicates parameters Marginal distribution function; f i express m Joint probability density function Indicates parameters The marginal probability density function; C i For Copula functions, c i is the Copula density function.
[0021] According to the technical solution provided by the present invention, formula (2) is simplified by formula (3):
[0022] Formula (3)
[0023] in, Indicates the first i The parameters are in the 1st, 2nd, and 3rd positions. , m –1 wind field condition m The conditional marginal probability density function of a wind field. Indicates the first i The parameters are in the Copula density function for the first and second wind fields. Indicates the first i The conditional Copula density function of the parameters under the second wind field condition for the first and third wind fields. Indicates the first i The parameter in the first... k Under the first wind field condition, the first j The conditional marginal probability distribution function of a wind field;
[0024] The first step is calculated using the Copula function and its density function according to the above formula. i The parameter in the first... j The wind farm and the first k The joint distribution function and joint probability density function of the i-th wind field, where the Copula function value is the i-th i The parameter in the first... j The wind farm and the first k The joint probability or conditional joint probability between wind fields.
[0025] According to the technical solution provided by the present invention, the cross-characteristics of each parameter between the base wind field and each of the other wind fields are calculated using the following formula (4):
[0026] Formula (4)
[0027] in, b Indicates the basic wind field. j Indicates the remaining wind fields, i.e., none of them are... b wind field, t Indicates the end time for different historical periods. Indicates the basic wind field b The i The parameters are in t The value at the end of the historical period. Indicates the remaining wind fields j The i The parameters are in t The value at the end of the historical period. Indicates in t The first time at the end of the historical period i These parameters are in the basic wind field. b Compared with other wind fields j The cross-features between them.
[0028] According to the technical solution provided by the present invention, assembling the stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters into the historical three-dimensional matrix at the start time of the target historical period according to time sequence includes the following steps:
[0029] The stationary data of the actual total output power of the wind field group at the end time of the same historical period, the joint probability or conditional joint probability of each parameter between each two wind fields, and the cross features of each parameter between the base wind field and each of the other wind fields are sequentially assembled and rearranged to obtain the historical two-dimensional matrix at the end time of the historical period.
[0030] The historical two-dimensional matrices are reassembled according to the time sequence of the end times of the historical periods to obtain the historical three-dimensional matrix.
[0031] According to the technical solution provided by the present invention, obtaining the actual total output power of the wind farm group at the cutoff time of the target historical period includes the following steps:
[0032] Obtain the actual output power of each wind field at the cutoff time of the target historical period;
[0033] The actual output power of the wind farm group is obtained by summing the actual individual output power of each wind farm and normalizing the result.
[0034] According to the technical solution provided by the present invention, the parameters include at least temperature, wind speed, and wind direction, and each of the parameters is time-varying.
[0035] In summary, this invention proposes a method for predicting wind farm cluster power. This method includes the following steps: obtaining a first sample set, which includes historical three-dimensional matrices at the start time of different target historical periods and the actual total output power of the wind farm cluster at the end time of the target historical period; training an initial neural network using the historical three-dimensional matrix as input and the actual total output power of the wind farm cluster as output to obtain a power prediction model; obtaining the target three-dimensional matrix at the start time of the period to be predicted, inputting the target three-dimensional matrix into the power prediction model to obtain the predicted total output power of the wind farm cluster at the end time of the period to be predicted.
[0036] This method can train a power prediction model using past data. The model can predict the power change in the current period based on the influence of parameters on the short-term output power of the wind farm group. It can be based on the power-related data and parameter-related data at the end time of the previous period (i.e., the start time of the current period). If the predicted total output power of the wind farm group at the end time of a certain period is unstable compared with the previous data, the predicted value can provide control guidance. The wind farm control terminal can make dynamic decisions in a timely manner to ensure the stability of the actual total output power of the wind farm group, thereby ensuring the stable output of wind energy. Attached Figure Description
[0037] Figure 1 A flowchart illustrating the steps of the wind farm group power prediction method provided by this invention;
[0038] Figure 2A schematic diagram of the structure of the historical three-dimensional matrix provided by this invention.
[0039] The text labels in the image represent:
[0040] 1. Historical two-dimensional matrix. Detailed Implementation
[0041] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0043] As mentioned in the background section, to address the problems in the existing technology, this invention proposes a method for predicting wind farm cluster power. Please refer to [link / reference]. Figure 1 As shown, it includes:
[0044] S1. Obtain a first sample set, which includes a historical three-dimensional matrix of the start time of different target historical periods and the actual total output power of the wind field cluster at the end time of the target historical period; the historical three-dimensional matrix is based on the historical periods before the target historical period. a The power-related data and parameter-related data are obtained from the end time of each historical period; wherein the power-related data is obtained by processing the actual total output power of the wind farm group during the historical period, and the parameter-related data is obtained by processing the parameters of multiple wind farm groups.
[0045] S2. Using the historical three-dimensional matrix as input and the actual total output power of the wind farm group as output, train the initial neural network to obtain a power prediction model; obtain the target three-dimensional matrix at the start of the period to be predicted, the target three-dimensional matrix being based on the period prior to the period to be predicted. a The power-related data and parameter-related data at the end of each historical time period are processed to obtain the target three-dimensional matrix. The target three-dimensional matrix is then input into the power prediction model to obtain the predicted total power of the wind field group at the end of the time period to be predicted.
[0046] The method for obtaining the first sample set specifically includes: the historical data includes power-related data and parameter-related data at the end times of several historical time periods; after selecting one of these historical time periods as the target historical time period, the data is processed by analyzing its preceding data. aThe power-related data and parameter-related data at the end time of each historical period are processed to obtain the historical three-dimensional matrix at the start time of the target historical period. The historical three-dimensional matrix at the start time of the target historical period is also the historical three-dimensional matrix at the end time of the previous historical period adjacent to the target historical period. By selecting a target historical period, a corresponding historical three-dimensional matrix and a corresponding actual total output power can be obtained. This serves as a sample. Repeatedly selecting different target historical periods will yield multiple sets of samples, which together form the first sample set.
[0047] Optionally, a wind farm group includes three wind farms. The wind speed, wind direction, and temperature parameters of the wind farms, as well as the annual changes in the actual total output power of the wind farm group, are collected at a frequency of 15 minutes per time. The total number of data records is 35,136, so the duration of each historical period is 15 minutes.
[0048] The method for obtaining the historical three-dimensional matrix at the beginning of a given historical period is as follows: the historical three-dimensional matrix at the beginning of the historical period is obtained by recording the wind speed, wind direction, and temperature parameters of the wind field corresponding to each of the previous historical periods, as well as the actual total output power of the wind field group. Then, the actual total output power of the wind field group at the end of the historical period is obtained by recording the actual total output power of the wind field group corresponding to the historical period itself.
[0049] The period to be predicted is a period that has not occurred in the short term. The duration of the period to be predicted is the same as the duration of each of the historical periods. The start time of the period to be predicted is the end time of the historical period adjacent to it. The method for obtaining the target three-dimensional matrix is the same as the method for obtaining the historical three-dimensional matrix.
[0050] Specifically, the first sample set is divided into two parts: 80% of the samples in the first sample set are used as the training set, and 20% of the samples are used as the validation set. A CNN-LSTM neural network is then built, which is the power prediction model. Specifically, the filter size is set to 3×3, and the number of filters is 50; the optimization function is the Adam function; the initial learning rate is 0.01, the learning rate strategy is piecewise, and the regularization factor is 1×10⁻⁶. -6 The gradient threshold was set to 1; the maximum number of training rounds was 10; the sequence length was 96; and the MiniBatchSize was 32. The relative error between the power prediction model and the measured power of the wind farm group was 6%, the RMSE was 4.6MW, and the training time was 18 minutes.
[0051] Taking a historical period and a predicted period of 15 minutes as an example, the target three-dimensional matrix at the start of the predicted period can be input into the power prediction model to predict the total output power of the wind farm group at the end of the predicted period (a future time that has not yet occurred). Obviously, using this method, the total output power of the wind farm at the end of a certain period that has not yet occurred can be predicted from the start of the period (the predicted value of the wind farm group power for that period can be calculated without waiting for 15 minutes).
[0052] This method can train a power prediction model using past data. The model can predict the power change in the current period based on the influence of parameters on the short-term output power of the wind farm group, by inputting the parameter values at the end time of the previous period (i.e. the start time of the current period). If the predicted total output power of the wind farm group at the end time of a certain period is unstable compared with the previous data, then the predicted value can provide control guidance. Dynamic decisions can be made in a timely manner through the wind farm control terminal to ensure the stable output of wind energy.
[0053] In a preferred embodiment, obtaining the historical three-dimensional matrix at the start time of the target historical period includes the following steps:
[0054] Obtain a stationary time matrix, which is composed of historical data prior to the target time period. a The data consists of stable data on the actual total output power of the wind farm group at the cut-off time of each historical period.
[0055] Obtain the joint probability matrix of parameters. The wind field has multiple parameters that affect its output power. The joint probability matrix of parameters is derived from the historical data of the target period prior to this period. a The joint probability or conditional joint probability of each parameter at the cutoff time of each of the historical periods is composed of the joint probability or conditional joint probability between each of the two wind fields;
[0056] Obtain the parameter cross feature matrix, which is composed of parameters from the previous historical time period of the target. a The parameters at the cutoff time of each of the historical periods are composed of the cross-features between the base wind field and each of the remaining wind fields; the remaining wind fields are the other wind fields in the wind field group besides the base wind field.
[0057] The stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters are assembled according to time sequence into the historical three-dimensional matrix at the start time of the target historical period.
[0058] In a preferred embodiment, the parameters include at least temperature, wind speed, and wind direction, and each of the parameters is time-varying.
[0059] In a preferred embodiment, assembling the stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters into the historical three-dimensional matrix at the start time of the target historical period according to time sequence includes the following steps:
[0060] The stationary data of the actual total output power of the wind field group at the end of the same historical period, the joint probability or conditional joint probability of each parameter between each two wind fields, and the cross features of each parameter between the base wind field and each of the other wind fields are sequentially assembled and rearranged to obtain the historical two-dimensional matrix 1 at the end of the historical period.
[0061] The historical two-dimensional matrices 1 are reassembled according to the time sequence of the end times of each historical period to obtain the historical three-dimensional matrix.
[0062] Specifically, the stationary time matrix W, the parameter joint probability matrix C, and the parameter cross-feature matrix Fc are as shown in formula (5). The stationary data of the actual total output power of the wind field group at the cutoff time of the same historical period, the joint probability or conditional joint probability of each parameter between each two wind fields, and the cross-feature of each parameter between the base wind field and each of the other wind fields are sequentially assembled to obtain the initial matrix I. Each row of the initial matrix I is rearranged to obtain the historical two-dimensional matrix I as shown in formula (6). Indicates, such as Figure 2 As shown, the historical two-dimensional matrix 1 is then assembled into the historical three-dimensional matrix according to the time sequence, and the size of the historical three-dimensional matrix is determined by the matrix components. m and n They are collectively identified as a square matrix or an approximate square matrix.
[0063] The historical three-dimensional matrix can accurately measure the uncertainty and correlation of factors such as wind speed, temperature, and wind direction in wind field clusters.
[0064] Formula (5)
[0065] Formula (6)
[0066] For the stationary time matrix W, with the cutoff times of different historical periods as columns, each row represents the stationary data of the actual total output power of the wind farm group at the cutoff time of each historical period;
[0067] For the joint probability matrix C of the parameters, with the cutoff times of different historical periods as columns, each row represents the joint probability or conditional joint probability of each parameter between every two wind fields at the cutoff time of each historical period. For any element in matrix C, Indicates the first tThe first historical period at the end time i The parameter in the first... m The wind farm and the first n The joint probability or conditional joint probability between wind fields (for example, for the first and second wind fields, the joint probability of the first and second wind fields is required; for the first and third wind fields, the conditional joint probability of the first and third wind fields under the condition of the second wind field is required; for the first and fourth wind fields, the conditional joint probability of the first and fourth wind fields under the condition of the second and third wind fields is required).
[0068] Specifically, the base wind field is pre-selected based on wind field operation data, geographical conditions, and operating conditions. For the parameter cross-feature matrix Fc, with different historical time periods as columns, each row represents the cross-feature composition of each parameter between the base wind field and other wind fields at each historical time period's cutoff time. For any element in matrix Fc... Indicates the first t The first historical period at the end time i These parameters are in the basic wind field. b and any one not for b wind field j The cross-features between them.
[0069] In a preferred embodiment, obtaining the stationary time matrix includes the following steps:
[0070] Obtain the original time series, which includes the period prior to the target historical period. a The original data of the actual total output power of the wind farm group at the cutoff time of the aforementioned historical period;
[0071] The original time series is subjected to stationarity detection. The original data at the end of the historical period that fails the stationarity detection is detrended to obtain the stationary data of the wind field cluster at the end of the historical period.
[0072] In a preferred embodiment, the wind farm group includes multiple wind farms, and the original data of the actual total output power of the wind farm group is the sum of the actual output power of each wind farm at the end of the historical period; the original data that failed the stationarity test are detrended using the following formula (1):
[0073] Formula (1)
[0074] in, t This indicates the cutoff time of the historical period that failed the stationarity test. Indicates the first tStable data at the cutoff point of each historical period Indicates the first j The raw data of the actual output power of each wind farm. Indicates the first j Polynomial fitting data of the actual output power of each wind farm m This represents the total number of wind farms in the wind farm cluster.
[0075] Specifically, the target historical period is obtained from historical data. a The original data of the actual total output power of the wind field group at the cutoff time of each of the historical periods are obtained. The original data are subjected to ADF stationarity detection. The original data that passes the stationarity detection are retained as stationary data in the corresponding historical period. The original data that fails the detection are corrected by formula (1) to obtain the stationary data in the historical period. Finally, the stationary time matrix is obtained.
[0076] In a preferred embodiment, the joint probability or conditional joint probability of each parameter between every two wind fields is calculated using the following formula (2):
[0077] Formula (2)
[0078] in, t Indicates the end time for different historical periods. Indicates the first j The first wind farm i The values of the parameters, n The number of wind field parameters is used to simplify the formula. As an alternative F i express m dimensional joint distribution function, Indicates parameters Marginal distribution function; f i express m Joint probability density function Indicates parameters The marginal probability density function; C i For Copula functions, c i is the Copula density function.
[0079] Specifically, the Copula function can accurately fit any uncertainty distribution when there are sufficient samples, and measure linear / nonlinear correlation. Each column in matrix C requires a Copula function with different parameters.
[0080] In a preferred embodiment, formula (2) is simplified by formula (3):
[0081] Formula (3)
[0082] in, Indicates the first i The parameters are in the 1st, 2nd, and 3rd positions. , m –1 wind field condition m The conditional marginal probability density function of a wind field. Indicates the first i The parameters are in the Copula density function for the first and second wind fields. Indicates the first i The conditional Copula density function of the parameters under the second wind field condition for the first and third wind fields. Indicates the first i The parameter in the first... k Under the first wind field condition, the first j The conditional marginal probability distribution function of a wind field;
[0083] The first step is calculated using the Copula function and its density function according to the above formula. i The parameter in the first... j The wind farm and the first k The joint distribution function and joint probability density function of the i-th wind field, where the Copula function value is the i-th i The parameter in the first... j The wind farm and the first k The joint probability or conditional joint probability between wind fields.
[0084] In a preferred embodiment, the cross-characteristics of each parameter between the base wind field and the remaining wind fields are calculated using the following formula (4):
[0085] Formula (4)
[0086] in, b Indicates the basic wind field. j This indicates the remaining wind fields, that is, any one that is not... b wind field, t Indicates the end time for different historical periods. Indicates the basic wind field b The i The parameters are in t The value at the end of the historical period. Indicates the remaining wind fields j The i The parameters are in t The value at the end of the historical period. Indicates in t The first time at the end of the historical period i These parameters are in the basic wind field. b Compared with other wind fields j The cross-features between them.
[0087] In a preferred embodiment, obtaining the actual total output power of the wind farm group at the end of the target historical period includes the following steps:
[0088] Obtain the actual output power of each wind field at the cutoff time of the target historical period;
[0089] The actual output power of the wind farm group is obtained by summing the actual individual output power of each wind farm and normalizing the result.
[0090] Specifically, a wind farm group includes three wind farms. At the end of the target historical period, each wind farm has its own actual output power. After summing the actual output power of the three wind farms, considering that the output power may vary greatly due to various factors, a normalization process is performed to normalize the total actual output power of the wind farm group to the range of 0-1.
[0091] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.
Claims
1. A method for predicting the power of a wind farm cluster, characterized in that, Includes the following steps: Obtain a first sample set, which includes a historical three-dimensional matrix of the start time of different target historical periods and the actual total output power of the wind field group at the end time of the target historical period; The historical three-dimensional matrix is based on the period prior to the target historical time. a The power-related data and parameter-related data at the end of each historical period were obtained; The power-related data is obtained by processing the actual total output power of the wind farm group during the historical period, and the parameter-related data is obtained by processing the parameters of multiple wind farm groups. Using the historical 3D matrix as input and the actual total output power of the wind field cluster as output, the initial neural network is trained to obtain a power prediction model; the target 3D matrix at the start of the period to be predicted is obtained, the target 3D matrix being based on the period prior to the period to be predicted. a The power-related data and parameter-related data at the end of each historical period were processed to obtain the data. The target three-dimensional matrix is input into the power prediction model to obtain the predicted total output power of the wind field group at the end of the predicted period. Obtaining the historical three-dimensional matrix at the start time of the target historical period includes the following steps: Obtain a stationary time matrix, which is composed of historical data prior to the target time period. a The data consists of stable data on the actual total output power of the wind farm group at the cut-off time of each historical period. Obtain the joint probability matrix of parameters. The wind field has multiple parameters that affect its output power. The joint probability matrix of parameters is derived from the historical data of the target period prior to this period. a The joint probability or conditional joint probability of each parameter at the cutoff time of each of the historical periods is composed of the joint probability or conditional joint probability between each of the two wind fields; Obtain the parameter cross feature matrix, which is composed of parameters from the previous historical time period of the target. a The parameters at the cutoff time of each of the historical periods are composed of the cross-features between the base wind field and each of the remaining wind fields; the remaining wind fields are the other wind fields in the wind field group besides the base wind field. The stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters are assembled according to time sequence into the historical three-dimensional matrix at the start time of the target historical period.
2. The wind farm group power prediction method according to claim 1, characterized in that, Obtaining a stationary time matrix involves the following steps: Obtain the original time series, which includes the period prior to the target historical period. a The original data of the actual total output power of the wind farm group at the cutoff time of the aforementioned historical period; The original time series is subjected to stationarity detection. The original data at the cutoff time of the historical period that fails the stationarity detection is detrended to obtain the stationary data at the cutoff time of the historical period.
3. The wind farm group power prediction method according to claim 2, characterized in that, The wind farm group includes multiple wind farms, and the original data of the actual total output power of the wind farm group is the sum of the actual output power of each wind farm at the end of the historical period. The following formula (1) is used to detrend the original data that fails the stationarity test: Official (1) in, t This indicates the cutoff time of the historical period that failed the stationarity test. Indicates the first t Stable data at the cutoff point of each historical period Indicates the first j The raw data of the actual output power of each wind farm. Indicates the first j Polynomial fitting data of the actual output power of each wind farm m This represents the total number of wind farms in the wind farm group.
4. The wind farm group power prediction method according to claim 3, characterized in that, The joint probability or conditional joint probability of each parameter between every two wind fields is calculated using the following formula (2): Official (2) in, t Indicates the end time for different historical periods. Indicates the first j The first wind farm i The values of the parameters, n The number of wind field parameters is used to simplify the formula. As an alternative F i express m dimensional joint distribution function, Indicates parameters Marginal distribution function; f i express m Joint probability density function Indicates parameters The marginal probability density function; C i For Copula functions, c i is the Copula density function.
5. The wind farm group power prediction method according to claim 4, characterized in that, Formula (2) is simplified by formula (3): Official (3) in, Indicates the first i The parameters are in the 1st, 2nd, and 3rd positions. , m –1 wind field condition m The conditional marginal probability density function of a wind field. Indicates the first i The parameters are in the Copula density function for the first and second wind fields. Indicates the first i The conditional Copula density function of the parameters under the second wind field condition for the first and third wind fields. Indicates the first i The parameter in the first... k Under the first wind field condition, the first j The conditional marginal probability distribution function of a wind field; The first step is calculated using the Copula function and its density function according to the above formula. i The parameter in the first... j The wind farm and the first k The joint distribution function and joint probability density function of the i-th wind field, where the Copula function value is the i-th i The parameter in the first... j The wind farm and the first k The joint probability or conditional joint probability between wind fields.
6. The wind farm group power prediction method according to claim 1, characterized in that, The cross-characteristics of each parameter between the base wind field and the remaining wind fields are calculated using the following formula (4): Official (4) in, b Indicates the basic wind field. j Indicates the remaining wind fields, i.e., none of them are... b wind field, t Indicates the end time for different historical periods. Indicates the basic wind field b The i The parameters are in t The value at the end of the historical period. Indicates the remaining wind fields j The i The parameters are in t The value at the end of the historical period. Indicates in t The first time at the end of the historical period i These parameters are in the basic wind field. b Compared with other wind fields j The intersecting features between them.
7. The wind farm group power prediction method according to claim 1, characterized in that, Assemble the stationary time matrix, the joint probability matrix of parameters, and the cross-feature matrix of parameters into the historical three-dimensional matrix at the start time of the target historical period according to the time sequence, including the following steps: The stationary data of the actual total output power of the wind field group at the end of the same historical period, the joint probability or conditional joint probability of each parameter between each two wind fields, and the cross-feature of each parameter between the base wind field and each other wind field are sequentially assembled and rearranged to obtain the historical two-dimensional matrix (1) at the end of the historical period. The historical two-dimensional matrices (1) are reassembled according to the time sequence of the end time of each historical period to obtain the historical three-dimensional matrix.
8. The wind farm group power prediction method according to claim 1, characterized in that, To obtain the actual total output power of the wind farm group at the cutoff time of the target historical period, the following steps are included: Obtain the actual output power of each wind field at the cutoff time of the target historical period; The actual output power of the wind farm group is obtained by summing the actual individual output power of each wind farm and normalizing the result.
9. The wind farm group power prediction method according to claim 1, characterized in that, The parameters include at least temperature, wind speed, and wind direction, and each of the parameters is time-varying.
Citation Information
Patent Citations
Wind power plant cluster short-term power prediction method based on space-time diagram convolutional neural network
CN112529282A