CN115796361A - Wind speed interval prediction method and device for ground stage of overhead line engineering - Google Patents
Wind speed interval prediction method and device for ground stage of overhead line engineering Download PDFInfo
- Publication number
- CN115796361A CN115796361A CN202211509211.9A CN202211509211A CN115796361A CN 115796361 A CN115796361 A CN 115796361A CN 202211509211 A CN202211509211 A CN 202211509211A CN 115796361 A CN115796361 A CN 115796361A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- data
- sequence
- value
- power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method and a device for predicting a wind speed interval of a researched stage of an overhead line project, belongs to the technical field of data preprocessing and recognition, and solves the problem of low accuracy of current wind speed determination. The method comprises the following steps: collecting historical wind speed data and preprocessing the historical wind speed data, wherein the preprocessing comprises data correction, continuity inspection and processing of abnormal values and missing values; decomposing the wind speed into a plurality of subsequences by combining variational modal decomposition, reconstructing the subsequences into a stable trend, ordered fluctuation and an unordered concussion sequence according to sample entropy, and identifying and extracting the maximum value, the minimum value and the average value of the unordered concussion sequence based on fuzzy information granulation; and respectively constructing prediction models of stable trend, ordered fluctuation, maximum value, minimum value and average value sequence by adopting a particle swarm optimization neural network, and superposing prediction results of the prediction models to determine a wind speed interval of a ground stage of the overhead line engineering. The improved particle swarm optimization GRU is adopted to effectively improve the interval coverage rate and the prediction accuracy.
Description
Technical Field
The invention relates to the technical field of data preprocessing and identification, in particular to a method and a device for predicting a wind speed interval of an exploratory stage of an overhead line engineering.
Background
Because the long-term weather stations are sparsely distributed, the observation fields are positioned in suburban boundary areas or suburbs, the wind measuring environment generally degrades along with the acceleration of the urbanization process, and the wind speed spatial distribution rule of different areas such as open fields in the open fields, mountain windward slopes, coastal strong wind areas and the like which pass through line engineering cannot be objectively reflected. The change of the wind speed directly has direct influence on the construction cost, the design wind speed is a key parameter for designing the overhead transmission line, and the value of the design wind speed directly relates to the economy, the safety and the applicability of line engineering. At present, the designed wind speed of the overhead transmission line is mainly based on the wind speed data of a long-term meteorological station near a line corridor, and the accuracy needs to be improved. Therefore, the wind speed is accurately predicted, the wind speed is scientifically estimated in the research stage, certain support and reference can be provided for equipment model selection, scheme formulation and the like, and the method has important significance for accurate manufacturing cost, and improvement of economic benefit and market competitiveness.
For a long time, domestic and foreign scholars have conducted a great deal of research and study on wind speed prediction methods, and have obtained a certain prediction result using a point prediction method, however, the point prediction method is difficult to describe wind speed randomness, and reliability of prediction cannot be ensured.
Disclosure of Invention
In view of the foregoing analysis, embodiments of the present invention provide a method and an apparatus for predicting a wind speed interval in a research-able stage of an overhead line project, so as to solve a problem that a model constructed by an existing method cannot embody scientific estimation of wind speed parameters in different research-able stages of the line project due to an actual problem that the accuracy of determining a current wind speed is not high.
On one hand, the embodiment of the invention provides a wind speed interval prediction method for a researched stage of an overhead line project, which comprises the following steps: collecting historical wind speed data, and preprocessing the historical wind speed data, wherein the preprocessing comprises data correction, continuity check, and abnormal value and missing value processing; decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered concussion sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered concussion sequence based on fuzzy information granulation; and respectively constructing prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting a particle swarm optimization neural network, and superposing prediction results of the prediction models to determine a wind speed interval of a researched stage of the overhead line engineering.
The beneficial effects of the above technical scheme are as follows: the wind speed is decomposed into a plurality of subsequences by combining a variational modal decomposition method, the subsequences are reconstructed according to the sample entropy, the subsequences comprise three parts of a stable trend, ordered fluctuation and disordered oscillation so as to improve the prediction accuracy, and then the maximum value, the minimum value and the average value of the disordered oscillation part are identified and extracted based on a fuzzy information granulation theory. By adopting the improved particle swarm optimization GRU network structure, the coverage rate of the regions is effectively improved, the average error is reduced, and the reliability of the predicted regions is higher.
Based on further improvement of the method, the historical wind speed data comprises the historical coastal observation data of the national meteorological station and the maximum 10-minute average wind speed observation data of the past year, the 10-minute average wind speed observation data of the regional automatic meteorological station, the coordinates of the poles and towers of the established line engineering with different voltage levels, the designed wind speed, the wind disaster accident and the basic geographic information data.
Based on the further improvement of the method, the historical wind speed data comprises discontinuous wind speed data and continuous wind speed data, wherein the preprocessing of the historical wind speed data comprises the following steps: performing significance test on a wind speed sequence before interruption and a wind speed sequence after interruption in the interrupted wind speed data to verify the continuity of the data and correcting the interrupted wind speed data by adopting a power exponent formula; performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm; and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparison analysis and correlation analysis of different meteorological elements.
Based on the further improvement of the method, the significance test of the wind speed sequence before and after interruption in the interrupted wind speed data to verify the continuity of the data and the correction of the interrupted wind speed data by adopting a power exponent formula comprises the following steps: the pre-discontinuity wind speed sequence is x 1 ,x 2 ,…,x n1 The sequence of wind speeds after interruption is l 1 ,l 2 ,…,l n2 The average value of the wind speed sequence before interruption and the wind speed sequence after interruption is the average value of all wind speed data
The sum of the squared deviations of the total wind speed data from the mean is:
judging the wind speed sequence before interruption and the wind speed sequence after interruption by the following formula, and performing significance test to verify the continuity of the wind speed data after interruption:
correcting abnormal data in the discontinuous wind speed data by adopting the following power exponent formula:
wherein V is the wind speed at standard height, Z is the anemometer height recorded in the historical edge leather of the meteorological station, and V 0 Observing the wind speed by an anemoscope, wherein alpha is the ground roughness coefficient of an open and flat area, and the value is 0.15;
the wind speed is then converted using the following linear regression model:
V 10 ＝aV 2 +b；
wherein, V 10 Is a 10 minute sequence of maximum values of mean wind speed, V 2 Is a 2-minute average wind speed maximum value sequence, selects annual self-recording 10-minute average maximum wind speed values from the stations with wind speed self-recording, and converts the values to a standard height to obtain V 10 A sequence; selecting average maximum wind speed value of 2 minutes from the same year of corresponding station and converting to standard height to obtain V 2 Sequencing; the constants a, b are found according to the least squares method.
Based on the further improvement of the method, the abnormal value processing method is adopted to carry out subsection cleaning on the abnormal data in the continuous wind speed data, and the reconstruction of the cleaned abnormal data through the four-point interpolation subdivision algorithm comprises the following steps: dividing the historical wind speed data into s intervals, and dividing power data of the ith wind speed-power interval in the s intervals intoSample is P i ＝{p i,1 ,p i,2 ,…,p i,m Where i =1,2, \ 8230;, s, p i,1 ≤p i,2 ≤…≤p i,m And m is the number of power points in each wind speed-power interval; calculating the average value of power in the ith wind speed-power interval by the following formula:
wherein, P i,j The power value of each power point in the ith wind speed-power interval is obtained; calculating a standard deviation of the power point data within the ith wind speed-power interval by the following formula:
calculating the absolute value of the deviation of each power point data in the ith wind speed-power interval by the following formula:
the absolute value delta of the deviation of the power point data when in the ith wind speed-power interval i,j At maximum, the power point data is an abnormal value to be eliminated; the value of τ is calculated by the following equation:
wherein t is a t distribution value of the power sample data; α is the significance level, the value of which affects the adequacy of the power data, when δ i,j ≥τS i When the power point data is an outlier, when delta i,j ≤τS i The power point data is a normal point.
Based on the further improvement of the method, the method is used for cleaning through a four-point interpolation subdivision algorithmReconstructing the washed abnormal data comprises the following steps: arranging the cleaned abnormal data from small to large and dividing the abnormal data into four equal parts, taking the data at the positions of 3 dividing points as quartiles, and calculating a second quantile M by the following formula i ：
M i For power data samples P i' The median, the first quantile and the third quantile of the data respectively represent P i' The position of the middle before and after 25% power data point is divided; when M is an even number, passing M i Will P i' Dividing into two sequences P of equal length i_1 ＝{p i,1 ,p i,2 ,…,p i,(m-1)/2 And P i_2 ＝{p i,(m+1)/2 ,p i,(m+3)/2 ,…,p i,m }，Q 1,i Is P i_1 Median of (2), Q 3,i Is P i_2 When m =4k +3 (k =0,1. -), Q is calculated by the following formula 1,i And Q 3,i ：
When m =4k +1 (k =0,1.), Q is calculated by the following formula 1,i And Q 3,i ：
P is calculated by the following formula i' A quarter-bit pitch I i ：
I i ＝Q 3,i -Q 1,i ；
Upper limit W calculated by the following formula u,i Lower limit W d,i To exclude outliers in the data:
W u,i ＝Q 3,i +1.5I i ；
W d,i ＝Q 1,i -1.5I i ；
after the error is identified, calculating power points to be interpolated by utilizing the adjacent 4 power values corresponding to the wind speed through a four-point interpolation subdivision algorithm:
wherein, P i-1 ，P i ，P i+1 ，P i+2 Respectively corresponding power values of 4 wind speeds closest to the power point to be interpolated; omega is a tensor parameter, when omega is epsilon [0,0.125]Then, an interpolation point is obtained.
Based on further improvement of the method, a variation modal decomposition method is combined, the wind speed is decomposed into a plurality of subsequences, the subsequences are reconstructed into a stable trend sequence, an ordered fluctuation sequence and an unordered concussion sequence according to sample entropy, and then based on fuzzy information granulation, the identification and extraction of the maximum value, the minimum value and the average value of the unordered concussion sequence comprise the following steps: for each mode function u k (t) computing the corresponding analysis signal by means of a hilbert transform to obtain its single-sided spectrum:
wherein δ (t) is a dirac function; k is the number of modes to be decomposed; for said each mode function u k (t) by aliasing and mode function u k (t) center frequency ω k Index term e of -jωkt Modulating the spectrum of each mode to the corresponding fundamental band:
and estimating the bandwidth of each mode signal by using a Gaussian smoothing method of the mediation signal to solve the variation problem under the constraint condition, wherein the objective function is as follows:
decomposing an original wind speed sequence into k subsequences through a variation mode, and obtaining a stable trend sequence, an ordered fluctuation sequence and an unordered oscillation sequence based on a sample entropy value; processing the disordered oscillation sequence by adopting a fuzzy information granulation method to obtain the minimum value, the average value and the maximum value of each window; establishing fuzzy information particles for each window by adopting the following membership functions:
wherein x is a wind speed sequence, and a, m and b are the minimum value, the average value and the maximum value of each fuzzy information particle of the fuzzy granulation window.
Based on the further improvement of the method, the method for respectively constructing the prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting the particle swarm optimization neural network comprises the following steps: respectively establishing a gated cycle unit GRU prediction model of a particle swarm optimization PSO for the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence, and performing joint optimization on an embedding dimension m, a delay time tau and a hidden layer node number L through the PSO; the GRU prediction model is used for processing a time sequence, selectively forgetting or reserving input historical information in a gating mode to predict future wind speed data, and comprises a training part and a prediction part, wherein the prediction part comprises an input layer, a hidden layer and an output layer, the GRU prediction model is trained by adopting a gradient descent algorithm according to sample loss, so that a sample loss function value is minimum, and the weight of each network parameter is updated:
z t ＝σ(W xz x t +W hz h t-1 +b z )；
r t ＝σ(W rz x t +W hr h t-1 +b r )；
y t ＝σ(W ho h t +b y )；
wherein, the phase point of the phase space reconstruction matrix is used as the network input x, the sequence x = { x (1 + M τ), x (2 + M τ),. } is used as the network output y, h is the hidden layer, σ is the sigmoid activation function, z is the sigmoid activation function t ,r t Update and reset gates, W, respectively, of the circulation network xz Is the weight between the input layer and the refresh gate at time t; w hz Is the weight between the hidden layer at time t-1 and the update gate at time t, W xr Is the weight between the input layer and the reset gate at time t, W hr Is the weight between the hidden layer at time t-1 and the reset gate at time t, W ho Is the weight between the hidden layer and the output layer at time t and theta is the Hadamard product of the matrix.
On the other hand, an embodiment of the present invention provides a wind speed interval prediction apparatus for a researched stage of an overhead line engineering, including: the data acquisition module is used for collecting historical wind speed data; the preprocessing module is used for preprocessing the historical wind speed data, wherein the preprocessing comprises data correction, continuity inspection and abnormal value and missing value processing; the decomposition and reconstruction module is used for decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered oscillation sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered oscillation sequence based on fuzzy information granulation; and the prediction module is used for respectively constructing prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting a particle swarm optimization neural network, superposing prediction results of the prediction models and determining a wind speed interval of the ground stage of the overhead line engineering.
Based on further improvement of the device, the historical wind speed data comprises discontinuous wind speed data and continuous wind speed data, wherein the preprocessing module is used for performing significance test on a wind speed sequence before the discontinuity and a wind speed sequence after the discontinuity in the discontinuous wind speed data to verify the continuity of the data and correcting the discontinuous wind speed data by adopting a power exponent formula; performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm; and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparison analysis and correlation analysis of different meteorological elements.
Compared with the prior art, the invention can realize at least one of the following beneficial effects:
1. firstly, the data is checked for observation data consistency and data record reliability. And analyzing the wind speed data which is obviously larger or smaller, and verifying whether the data record is reliable or not through weather system process analysis, inter-station comparison analysis and related analysis of different meteorological elements. And corresponding methods are respectively provided for continuous and discontinuous abnormal values, so that the accuracy of data is improved. And (3) adopting a sequence-dividing significance test mode aiming at discontinuous data, verifying the continuity of the data by performing significance test on sequence data before and after the discontinuity, and correcting the data by adopting a power exponent formula. And (3) performing segmented cleaning on abnormal data by adopting an abnormal value identification method aiming at the continuous data, and reconstructing the cleaned abnormal data through a four-point interpolation subdivision algorithm.
2. Aiming at the aspect of risk data series, a variational modal decomposition method is combined to decompose the wind speed into a plurality of subsequences, the subsequences are reconstructed according to the sample entropy, the subsequences comprise three parts of a stable trend, ordered fluctuation and unordered oscillation so as to improve the prediction accuracy, and then the maximum value, the minimum value and the average value of the unordered oscillation part are identified and extracted based on a fuzzy information granulation theory; five prediction sub-modules are formed and predicted, the prediction accuracy of the interval prediction model which is directly established for the wind speed sequence is low, and the prediction accuracy can be obviously improved by using a proper data preprocessing method.
3. The improved particle swarm optimization GRU network structure is adopted, so that the coverage rate of the regions is effectively improved, the average error is reduced, and the reliability of the predicted regions is higher; compared with a shallow neural network, the prediction model established based on the GRU can dynamically track the change of the wind speed signal during single-step and multi-step interval prediction, and has stronger learning capacity and higher prediction precision.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flow chart of a method for predicting wind speed intervals during a research phase of an overhead line engineering according to an embodiment of the present invention;
FIG. 2 is a block diagram of a variational modal decomposition and reconstruction in accordance with an embodiment of the present invention;
FIG. 3 is a graph of a time series of raw wind speeds according to an embodiment of the present invention;
FIG. 4 is a predictive block diagram of a gated loop unit GRU according to an embodiment of the invention;
FIGS. 5A, 5B, 5C, 5D, and 5E are graphs of a steady trend sequence, an ordered wave sequence, a maximum sequence, a mean sequence, and a minimum sequence, respectively, according to an embodiment of the present invention;
fig. 6 is a flowchart of a weight initialization process of a gated cyclic unit GRU according to an embodiment of the present invention; and
fig. 7 is a block diagram of a wind speed interval prediction apparatus at a research stage of an overhead line engineering according to an embodiment of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
Referring to fig. 1, an embodiment of the present invention discloses a method for predicting a wind speed interval of a research stage of an overhead line engineering, including: in step S102, collecting historical wind speed data, and preprocessing the historical wind speed data, wherein the preprocessing includes data correction, continuity check, and processing of abnormal values and missing values; in step S104, a variational modal decomposition method is combined to decompose the wind speed into a plurality of subsequences, the subsequences are reconstructed into a stable trend sequence, an ordered fluctuation sequence and a disordered concussion sequence according to the sample entropy, and then the maximum value, the minimum value and the average value of the disordered concussion sequence are identified and extracted based on fuzzy information granulation; and in step S106, respectively constructing a prediction model of a stable trend sequence, an ordered fluctuation sequence, a maximum value sequence, a minimum value sequence and an average value sequence by adopting a particle swarm optimized neural network, and superposing the prediction results of the prediction models to determine the wind speed interval of the ground-possible stage of the overhead line engineering.
Compared with the prior art, in the wind speed interval prediction method in the ground stage of the overhead line engineering, a variational modal decomposition method is combined to decompose the wind speed into a plurality of subsequences, the subsequences are reconstructed according to the sample entropy, the subsequences comprise three parts of a stable trend, ordered fluctuation and disordered oscillation so as to improve the prediction accuracy, and then the maximum value, the minimum value and the average value of the disordered oscillation part are identified and extracted based on a fuzzy information granulation theory. By adopting the improved particle swarm optimization GRU network structure, the interval coverage rate is effectively improved, the average error is reduced, and the reliability of the predicted interval is higher.
Hereinafter, each step of the wind speed interval prediction method of the exploitable stage of the overhead line engineering according to the embodiment of the invention will be described in detail with reference to fig. 1.
In step S102, historical wind speed data is collected, and the historical wind speed data is preprocessed, wherein the preprocessing includes data modification, continuity check, and processing of abnormal values and missing values. Specifically, the historical wind speed data comprises historical edge leather of a national weather station and maximum 10-minute average wind speed observation data of the national weather station, 10-minute average wind speed observation data of a regional automatic weather station, coordinates of existing line engineering towers with different voltage levels, design wind speed, wind disaster accidents and basic geographic information data.
The historical wind speed data comprises intermittent wind speed data and continuous wind speed data, wherein the step of preprocessing the historical wind speed data comprises the following steps: the method comprises the following steps of (1) carrying out significance test on a wind speed sequence before interruption and a wind speed sequence after interruption in interrupted wind speed data to verify the continuity of the data and correcting the interrupted wind speed data by adopting a power exponent formula; performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm; and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparative analysis and related analysis of different meteorological elements.
Specifically, the step of performing significance test on the wind speed sequence before interruption and the wind speed sequence after interruption in the interrupted wind speed data to verify the continuity of the data and correcting the interrupted wind speed data by adopting a power exponent formula comprises the following steps: the wind speed sequence before interruption is x 1 ,x 2 ,…,x n1 The sequence of wind speeds after interruption is l 1 ,l 2 ,…,l n2 The average value of the wind speed sequence before interruption and the wind speed sequence after interruption is the average value of all wind speed data
The sum of the squared deviations of all wind speed data from the mean is:
judging the wind speed sequence before interruption and the wind speed sequence after interruption by the following formula, and performing significance test to verify the continuity of the interrupted wind speed data:
and correcting abnormal data in the discontinuous wind speed data by adopting the following power exponent formula:
wherein V is the standard altitude wind speed, Z is the anemometer altitude recorded in the historical welfare of the meteorological station, V 0 Observing the wind speed by an anemoscope, wherein alpha is the ground roughness coefficient of an open and flat area, and the value is 0.15;
the wind speed is then converted using the following linear regression model:
V 10 ＝aV 2 +b；
wherein, V 10 Is a 10 minute sequence of maximum values of mean wind speed, V 2 Is a 2-minute average wind speed maximum value sequence, selects annual self-recording 10-minute average maximum wind speed values from the stations with wind speed self-recording, and converts the values to a standard height to obtain V 10 Sequencing; selecting average maximum wind speed value of 2 minutes at the same year of corresponding station and converting to standard height to obtain V 2 A sequence; the constants a, b are found according to the least squares method.
Specifically, the step of performing segmentation cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and the step of reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm comprises the following steps:
dividing historical wind speed data into s intervals, wherein a power data sample of the ith wind speed-power interval in the s intervals is P i ＝{p i,1 ,p i,2 ,…,p i,m Where i =1,2, \8230;, s, p i,1 ≤p i,2 ≤…≤p i,m And m is the number of power points in each wind speed-power interval;
calculating the average power value in the ith wind speed-power interval by the following formula:
wherein, P i,j The power value of each power point in the ith wind speed-power interval is obtained;
calculating the standard deviation of the power point data in the ith wind speed-power interval by the following formula:
calculating the absolute value of the deviation of each power point data in the ith wind speed-power interval by the following formula:
absolute value of deviation delta of power point data when in the ith wind speed-power interval i,j At maximum, the power point data is an abnormal value to be eliminated;
the value of τ is calculated by the following equation:
wherein t is a t distribution value of the power sample data; α is a significance level, the value of which affects the adequacy of the power data whenδ i,j ≥τS i When the power point data is an outlier, when delta i,j ≤τS i The power point data is a normal point.
Specifically, the reconstructing the cleaned abnormal data through a four-point interpolation subdivision algorithm comprises the following steps: arranging the cleaned abnormal data from small to large and dividing the abnormal data into four equal parts, taking the data at the positions of 3 dividing points as quartiles, and calculating a second quartile M by the following formula i ：
M i For power data samples P i' The median, the first quantile and the third quantile of the data respectively represent P i' The position values of the front and rear 25% power data points are divided;
when M is an even number, passing M i Will P i' Dividing into two sequences P of equal length i_1 ＝{p i,1 ,p i,2 ,…,p i,(m-1)/2 And P i_2 ＝{p i,(m+1)/2 ,p i,(m+3)/2 ,…,p i,m }，Q 1,i Is P i_1 Median of (2), Q 3,i Is P i_2 When m =4k +3 (k =0,1. -), Q is calculated by the following formula 1,i And Q 3,i ：
When m =4k +1 (k =0,1,.), Q is calculated by the following equation 1,i And Q 3,i ：
P is calculated by the following formula i' A quarter-bit pitch I i ：
I i ＝Q 3,i -Q 1,i ；
Upper limit W calculated by the following formula u,i Lower limit W d,i To exclude outliers in the data:
W u,i ＝Q 3,i +1.5I i ；
W d,i ＝Q 1,i -1.5I i ；
after the error is identified, calculating power points to be interpolated by using the adjacent 4 power values corresponding to the wind speed through a four-point interpolation subdivision algorithm:
wherein, P i-1 ，P i ，P i+1 ，P i+2 Respectively corresponding power values of 4 wind speeds closest to the power point to be interpolated; omega is a tensor parameter, when omega is epsilon [0,0.125]Then, an interpolation point is obtained.
In step S104, the wind speed is decomposed into a plurality of sub-sequences by combining a variational modal decomposition method, the plurality of sub-sequences are reconstructed into a stable trend sequence, an ordered fluctuation sequence and a disordered concussion sequence according to the sample entropy, and then the maximum value, the minimum value and the average value of the disordered concussion sequence are identified and extracted based on fuzzy information granulation.
Specifically, decomposing the wind speed into a plurality of subsequences by combining a variation modal decomposition method, reconstructing the plurality of subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered concussion sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered concussion sequence based on fuzzy information granulation comprises the following steps:
for each mode function u k (t) computing the corresponding analysis signal by hilbert transform to obtain its single-sided spectrum:
wherein δ (t) is a dirac function; k is the number of modes to be decomposed;
for each mode function u k (t) by aliasing and mode function u k (t) center frequency ω corresponding to k Index term e of -jωkt Modulating the frequency spectrum of each mode to a corresponding basic frequency band;
and (3) estimating the bandwidth of each mode signal by using a Gaussian smoothing method for mediating signals to solve the variation problem under the constraint condition, wherein the objective function is as follows:
decomposing an original wind speed sequence into k subsequences through a variational mode, and obtaining a stable trend sequence, an ordered fluctuation sequence and an unordered oscillation sequence based on a sample entropy value;
processing the disordered oscillation sequence by adopting a fuzzy information granulation method to obtain the minimum value, the average value and the maximum value of each window;
fuzzy information particles are established for each window by adopting the following membership functions:
wherein x is a wind speed sequence, and a, m and b are the minimum value, the average value and the maximum value of each fuzzy information particle of the fuzzy granulation window.
In step S106, a neural network optimized by using particle swarm is used to respectively construct prediction models of a steady trend sequence, an ordered fluctuation sequence, a maximum value sequence, a minimum value sequence and an average value sequence, and the prediction results of the prediction models are superimposed to determine a wind speed interval of the exploitable stage of the overhead line engineering.
Specifically, the method for respectively constructing the prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting the particle swarm optimization neural network comprises the following steps:
respectively establishing a gated cycle unit GRU prediction model of the particle swarm optimization PSO for the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence, and performing combined optimization on the embedding dimension m, the delay time tau and the hidden layer node number L through the PSO;
the GRU prediction model is used for processing a time sequence, selectively forgetting or reserving input historical information in a gating mode to predict future wind speed data, comprises a training part and a prediction part, comprises an input layer, a hidden layer and an output layer, is trained by a gradient descent algorithm according to sample loss, minimizes a sample loss function value, and updates the weight of each network parameter:
z t ＝σ(W xz x t +W hz h t-1 +b z )；
r t ＝σ(W rz x t +W hr h t-1 +b r )；
y t ＝σ(W ho h t +b y )；
wherein, the phase point of the phase space reconstruction matrix is used as the network input x, the sequence x = { x (1 + M τ), x (2 + M τ) } is used as the network output y, h is the hidden layer, σ is the sigmoid activation function, z is the hidden layer, and x (M + M τ) } is used as the network output y t ,r t Update and reset gates, W, respectively, of the circulation network xz Is the weight between the input layer and the refresh gate at time t; w hz Is the weight between the hidden layer at time t-1 and the update gate at time t, W xr Is the weight between the input layer and the reset gate at time t, W hr Is t-1 time hiddenWeight between hidden layer and reset gate at time t, W ho Is the weight between the hidden layer and the output layer at time t and theta is the Hadamard product of the matrix.
Referring to fig. 7, an embodiment of the present invention discloses a wind speed interval prediction apparatus for a ground stage of an overhead line engineering, including: a data acquisition module 702 for collecting historical wind speed data; the preprocessing module 704 is used for preprocessing historical wind speed data, wherein the preprocessing comprises data correction, continuity check and processing of abnormal values and missing values; the decomposition and reconstruction module 706 is used for decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the plurality of subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered oscillation sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered oscillation sequence based on fuzzy information granulation; and the prediction module 708 is used for respectively constructing prediction models of a stable trend sequence, an ordered fluctuation sequence, a maximum value sequence, a minimum value sequence and an average value sequence by adopting a particle swarm optimization neural network, superposing the prediction results of the prediction models and determining the wind speed interval of the researched stage of the overhead line engineering.
The historical wind speed data comprises intermittent wind speed data and continuous wind speed data, wherein the preprocessing module 704 is used for performing significance test on a wind speed sequence before interruption and a wind speed sequence after interruption in the intermittent wind speed data to verify the continuity of the data and correcting the intermittent wind speed data by adopting a power exponent formula; performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm; and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparison analysis and correlation analysis of different meteorological elements.
Hereinafter, a wind speed interval prediction method of a exploratory stage of an overhead line engineering according to an embodiment of the present invention will be described in detail by way of specific examples with reference to fig. 2 to 6.
The method for predicting the wind speed interval in the exploitable stage of the overhead line engineering comprises the following steps: collecting historical data including average wind speed data for national weather stations and regional weather stations; preprocessing the data, including data correction and continuity test, and processing abnormal values and missing values; decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the subsequences according to the sample entropy, wherein the subsequences comprise three parts of a stable trend, ordered fluctuation and disordered oscillation, and identifying and extracting the maximum value, the minimum value and the average value of the disordered oscillation part based on a fuzzy information granulation theory; and finally, respectively constructing prediction models of the 5 modules by using a particle swarm optimized neural network model, and superposing the prediction results of the submodules to scientifically determine the wind speed interval at the researched stage.
The beneficial effects of the above technical scheme are as follows: by collecting data and combining data characteristics, a main data processing mode is provided from continuous and discontinuous dimensions, the reliability of the data is finally improved, and the reasonability of a wind speed determination result is ensured. And (3) adopting a sequence-dividing significance test mode aiming at discontinuous data, verifying the continuity of the data by performing significance test on sequence data before and after the discontinuity, and correcting the data by adopting a power exponent formula. And (3) performing segmented cleaning on abnormal data by adopting an abnormal value identification method aiming at the continuous data, and reconstructing the cleaned abnormal data through a four-point interpolation subdivision algorithm.
And (3) adopting a sequence-dividing significance test mode aiming at discontinuous data, verifying the continuity of the data by performing significance test on sequence data before and after the discontinuity, and correcting the data by adopting a power exponent formula.
And (3) performing segmented cleaning on abnormal data by adopting an abnormal value identification method aiming at the continuous data, and reconstructing the cleaned abnormal data through a four-point interpolation subdivision algorithm.
The data is first checked for observation consistency and data record reliability. And analyzing the wind speed data which is obviously larger or smaller, and verifying whether the data record is reliable or not through weather system process analysis, inter-station comparison analysis and related analysis of different meteorological elements.
Outlier processing can be used to find isolated points in an independent variable. The wind speed is divided into s intervals according to a certain size, and abnormal power data points in each interval are identified and removed in a segmented mode by an abnormal value method.
The wind speed sequence before the break year (excluding the year) is subsequence 1, followed by subsequence 2. Let subsequence 1 be x 1 ,x 2 ,...,x n1 Subsequence 2 is l 1 ,l 2 ,...,l n2 . The average values of the subsequences 1 and 2 are respectively the average value of the total data
The sum of the squared deviations of all data from the mean is:
and (3) judging whether the difference between groups is significant, and performing significance test (verifying whether the data is large or small) by using an F test method:
and correcting the abnormal data, including wind speed altitude correction and time conversion.
The wind speed altitude correction adopts a power exponent formula:
wherein V is the wind speed at standard altitude, Z is the anemoscope height recorded in the historical edge leather of the meteorological station, V0 is the anemoscope observation wind speed, and alpha is the roughness coefficient of the ground in the open flat area, and the value is 0.15.
The time conversion of the wind speed adopts a linear regression model:
V 10 ＝aV 2 +b；
V 10 is a 10 minute average wind speed maximum sequence. V 2 Is a 2 minute average wind speed maximum sequence. Respectively picking out the average maximum wind speed value of annual self-recording for 10 minutes from the station with wind speed self-recording, and converting the average maximum wind speed value into the standard height to obtain V 10 And (4) sequencing.
Selecting the average maximum wind speed value with the timing of 2 minutes from the same year of the corresponding station, and converting the average maximum wind speed value to the standard height to obtain V 2 And (4) sequencing.
The constants a, b are found according to the least squares method.
Abnormal value detection is performed for the continuous data as follows.
Dividing the wind speed into s intervals according to a certain size, and recording a power data sample of the ith wind speed-power interval as P i ＝{p i,1 ,p i,2 ,..p i,m Where i =1,2 i,1 ≤P i,2 ≤...≤P i,m And m is the number of power points in each wind speed-power interval.
Firstly, calculating the mean value of power data in an interval as follows:
in the formula: p i,j Are the power values in the ith interval.
The standard deviation of the power data in the ith interval is as follows:
recording the absolute value of the deviation of each power sample data in the interval
If in the ith interval, the absolute value delta of the deviation of the observed power data i,j At maximum, then the point may be an outlier to be culled. When the power deviation value is maximum, the power value is maximum or minimum in the interval, and the possibility that the point is the power abnormal value is higher.
The tau value in Thompson tau method is calculated as follows
In the formula: t is a t distribution value of the power sample data; α is the significance level, the value of which affects the adequacy of the power data.
When delta i,j ≥τS i When the detected power value is an abnormal point; otherwise if delta i,j ≤τS i Then the detected power value is the normal point. When the power value is detected as an abnormal value, the power value is eliminated, and the average value and the standard deviation in the interval are recalculated. The value of tau will also change with the number of power data in the interval until no new power abnormal value is found.
After the first identification is finished by the method, the number of the remaining power points in each wind speed interval is m ', the number of the power points m' in the interval is still recorded as m for simplifying the naming, and the power data sample of the ith wind speed-power interval after the first identification is P i '＝{p i',1 ,p i',2 ,..p i',m In which P is i',1 ≤P i',2 ≤...≤P i',m 。
The quartile is a quantile in statistics, namely all data are arranged from small to large and divided into four equal parts, and the data at the positions of 3 dividing points are the quartile. Second fraction M i Represents a sample P i' The median of the data is calculated as follows.
The first and third quantiles represent P i' The position of the middle 25% data point before and after segmentation. The total data points m in the interval are different, and the calculation formula is slightly different.
When M is an even number, M i Will P i' Divided into two sequences P of equal length i_1 ＝{p i,1 ,p i,2 ,…,p i,(m-1)/2 And P i_2 ＝{p i,(m+1)/2 ,p i,(m+3)/2 ,…,p i,m }。Q 1,i Is P i_1 The median of (3). Q 3,i Is P i_2 The median of (3).
When m =4k +3 (k =0, 1.), the calculation formula is
When m =4k +1 (k =0,1.), the calculation formula is
To obtain P i' A quarter-bit pitch I i Is composed of
I i ＝Q 3,i -Q 1,i ；
In the quartile method, an upper limit W is used u,i Lower limit W d,i To eliminate abnormal values in the data, the calculation formula is
W u,i ＝Q 3,i +1.5I i ；
W d,i ＝Q 1,i -1.5I i ；
After the error is identified, a new interpolation point is calculated by using 4 adjacent points by using a four-point interpolation subdivision algorithm, and the same operation rule is used in each calculation. The method has theoretical feasibility by searching 4 wind speed points close to the wind speed to be interpolated and utilizing a four-point interpolation algorithm to interpolate missing data.
Wherein P is 2i+1 Is a power point to be interpolated; p i-1 ，P i ，P i+1 ，P i+2 Respectively corresponding power values of 4 wind speeds closest to the power point to be interpolated; omega is a tensor parameter, and when omega is epsilon [0]In the process, a satisfactory interpolation point can be obtained, and the middle point omega =1/16 in the interval is taken.
Referring to fig. 2, the wind speed is decomposed into a plurality of subsequences by combining a variational modal decomposition method, the subsequences are reconstructed according to the sample entropy, and the subsequences comprise three parts of a stable trend, ordered fluctuation and disordered oscillation, then the maximum value, the minimum value and the average value of the disordered oscillation part are identified and extracted based on a fuzzy information granulation theory, prediction is more accurate, and the prediction accuracy can be improved by further decomposing without oscillation.
The variational modal decomposition is a processing mode of self-adaptive and non-recursive signal decomposition, the stability of a decomposed sequence is good, the singularity characteristic of a signal can be well reflected, and the noise robustness is good. In the process of obtaining each component, the algorithm searches for the optimal solution of the constraint variational model by converting the decomposition process of the signal into a variational frame, finally realizes the self-adaptive decomposition of the signal and gets rid of the mode of circularly screening and stripping the processed signal by empirical mode decomposition.
The steps for constructing the variation problem are as follows:
for each mode function u k (t) computing the corresponding analysis signal by means of a hilbert transform to obtain its single-sided spectrum.
Wherein δ (t) is a dirac function; k is the number of modes to be decomposed.
For each mode function u k (t) by aliasing and mode function u k (t) center frequency ω k Index term e of -jωkt Each dieThe spectrum of the states is modulated to the corresponding fundamental frequency band.
The bandwidth of each mode signal is estimated by using a Gaussian smoothing method for adjusting the signals, the variation problem under the constraint condition is solved, and the objective function is as follows:
the wind speed sequence is decomposed into K subsequences through a variational mode, and then three sequences are obtained on the basis of a sample entropy method, wherein the three sequences respectively have a stable trend, ordered fluctuation and disordered oscillation.
And processing the disordered oscillation sequence by adopting a fuzzy information granulation method to obtain the minimum value, the average value and the maximum value of each window.
An information particle is a collection of elements that are held together by hard-to-distinguish, or similar, or close or functional features. Information particles are ubiquitous around us as the representation form of information, and are a basic concept of understanding the world by human beings. When people know the world, people often put a part of similar things together as a whole to study the properties or characteristics of the things, and in fact, the way of handling things is information granulation. And the "whole" of the study is called an information particle.
In the information granulation, the non-fuzzy granulation mode (c-granulation) plays an important role in a plurality of process technologies, but in the reasoning and concept formation of almost all people, the granules are fuzzy (f-granulation), and the non-fuzzy granulation does not reflect the fact. Fuzzy information granulation is inspired by human granulation information methods and inferred accordingly.
Firstly, an original wind speed sequence x (refer to fig. 3) is divided into subsequences with equal length according to time as operation windows, a fuzzy information particle P is established for each window by adopting a membership function, and a fuzzy set is formedInstead of all the valid characteristic information of the original sequence, i.e. a fuzzy concept G that can describe x,
it is therefore essentially the process of determining the membership function of G, such that P = a (x).
Triangular fuzzy particles are selected for research, and the membership function of the triangular fuzzy particles is as follows:
in the formula: x is a wind speed sequence; a. m, b-minimum, average, maximum per particle of the fuzzy granulation window.
Based on the modules, the invention constructs a GRU prediction model for improving particle swarm optimization. And respectively establishing a GRU prediction model for particle swarm optimization for the 5 modules, and performing joint optimization on the embedding dimension m, the delay time tau and the number L of nodes of the hidden layer through PSO.
The input is equivalent to that the wind speed can be divided into three parts, namely a more stable sequence (refer to fig. 5A), a sequence with ordered fluctuation (refer to fig. 5B), and a sequence with disordered oscillation, wherein the disordered oscillation comprises three values of a maximum value (refer to fig. 5C), an average value (refer to fig. 5D) and a minimum value (refer to fig. 5E), the five sequences are respectively used as prediction input, models are respectively constructed, and finally, the five sequences are reconstructed to form a final wind speed prediction result.
The Gated Recurrent Unit (GRU) is a network model for processing time sequences, compared with the traditional BP algorithm, the GRU has stronger learning capacity, and can selectively forget or retain input historical information in a gated mode, so that the prediction of future data is realized. The network structure is divided into 2 parts of training and prediction, the prediction part comprises 3 parts of an input layer, a hidden layer, an output layer and the like, a GRU network is trained by adopting a gradient descent algorithm according to sample loss, so that a sample loss function value is minimum, and the weight of each network parameter is updated.
z t ＝σ(W xz x t +W hz h t-1 +b z )；
r t ＝σ(W rz x t +W hr h t-1 +b r )；
y t ＝σ(W ho h t +b y )；
The prediction block diagram of the GRU network is shown in FIG. 4, the network structure is divided into 2 parts of training and prediction, the prediction part comprises 3 parts of an input layer, a hidden layer, an output layer and the like, the GRU network is trained by adopting a gradient descent algorithm according to sample loss, so that the value of a sample loss function is minimum, and the weight of each network parameter is updated; the lower part is an expanded view of a hidden layer of a GRU network, wherein the phase points of a phase space reconstruction matrix are taken as an input part x of the network, a sequence x = { x (1 + M τ), x (2 + M τ) } is taken as y of an output part, h is the hidden layer, σ is a sigmoid activation function, z is an expansion table of the hidden layer of the GRU network, x = { x (1 + M τ), x (2 + M τ) } is taken as the hidden layer, z is taken as the hidden layer, b is taken as the activation function of the sigmoid, and z is taken as the input part x of the network t ,r t Respectively an update gate and a reset gate of the circulation network. W is a group of xz Is the weight between the input layer and the update gate at time t. W hz Is the weight between the hidden layer at time t-1 and the update gate at time t, W xr Is the weight between the input layer and the reset gate at time t. W hr Is the weight between the hidden layer at time t-1 and the reset gate at time t, W ho Is the weight between the hidden layer and the output layer at time t. And Θ is the Hadamard product of the matrix. b z 、b r 、b y 、b h Respectively, the respective offset.
Referring to FIG. 4, when state x is input t After entering, the signals are respectively output to two gates. In the status line, x t And a weight W sh Multiplying, resetting gate r t Hidden state h from the previous moment t-1 Multiplication by vector elementsThen vector-summed to the input vector of the hidden layer
The learning speed of the GRU is fast, the training error is small, but initialization weights between hidden layers and between the hidden layers and output layers are randomly selected, and the GRU lacks pertinence to a subsequent training process. The improved particle swarm optimization algorithm has strong global optimization capability, and the model prediction capability can be further improved by introducing the improved particle swarm optimization algorithm into the network training process of the GRU. Referring to fig. 6, based on the improved particle swarm optimization, the GRU weight initialization process is as follows:
for a given training set [ x ] i ,y i ]Initializing N p A parameter vector t of dimension E (E = Kn +1, where K is the number of hidden layers of GRU and n is the number of input neurons) q,g (q＝1，2，…，N p ) Any dimension has a value range of [ -1,1]And g is the number of iterations.
The initial weight matrix and the offset vector of the GRU form particle swarm individuals, and for the individual t in each swarm q,g And calculating the output results of each gate control unit, the candidate state and the network.
And calculating a GRU prediction result corresponding to each particle, taking the root mean square error as a fitness function of the IPSO algorithm, and optimizing the particle corresponding to the GRU with the minimum root mean square error.
Updating population optima g best And individual optimum p best 。
And calculating the concentration degree.
When the degree of aggregation reaches a threshold, the velocity and position of the particles are updated according to the new dynamic topology and particle search behavior proposed herein.
And finishing the initialization process of the GRU model after the condition of the number of times of the particle swarm termination iteration is reached. After the initialization process of the GRU is completed, the algorithm is continuously used for learning the network parameters. And respectively constructing and training a GRU model for each module, and finally integrating the prediction results of each component through a new GRU network.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
While the invention has been described with reference to specific preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the following claims.
Claims (10)
1. A wind speed interval prediction method for a ground stage of an overhead line project is characterized by comprising the following steps:
collecting historical wind speed data, and preprocessing the historical wind speed data, wherein the preprocessing comprises data correction, continuity check, and abnormal value and missing value processing;
decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered concussion sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered concussion sequence based on fuzzy information granulation; and
and respectively constructing prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting a particle swarm optimization neural network, and superposing the prediction results of the prediction models to determine the wind speed interval of the exploitable stage of the overhead line engineering.
2. The method for predicting wind speed intervals in the exploitable stage of overhead line engineering according to claim 1, wherein the historical wind speed data comprises the historical edgewise and historical 10-minute average wind speed observation data of a national meteorological station, the 10-minute average wind speed observation data of a regional automatic meteorological station, coordinates of poles and towers of the constructed line engineering with different voltage levels, designed wind speed, wind disaster accidents and basic geographic information data.
3. The method for forecasting the wind speed interval in the exploratory stage of the overhead line engineering according to claim 1, wherein the historical wind speed data comprises intermittent wind speed data and continuous wind speed data, and wherein the preprocessing the historical wind speed data comprises:
performing significance test on a wind speed sequence before interruption and a wind speed sequence after interruption in the interrupted wind speed data to verify the continuity of the data and correcting the interrupted wind speed data by adopting a power exponent formula;
performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm;
and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparative analysis and correlation analysis of different meteorological elements.
4. The method for predicting the wind speed interval in the exploitable stage of the overhead line engineering according to claim 3, wherein the step of performing significance check on the wind speed sequence before and after interruption in the interrupted wind speed data to verify the continuity of the data and correcting the interrupted wind speed data by adopting a power exponent formula comprises the following steps:
the wind speed sequence before interruption is x 1 ,x 2 ,…,x n1 The sequence of wind speeds after interruption is l 1 ,l 2 ,…,l n2 The average value of the wind speed sequence before interruption and the wind speed sequence after interruption is the average value of all wind speed data
The sum of the squared deviations of the total wind speed data from the mean is:
judging the wind speed sequence before interruption and the wind speed sequence after interruption by the following formula, and carrying out significance test to verify the continuity of the interrupted wind speed data:
correcting abnormal data in the discontinuous wind speed data by adopting the following power exponent formula:
wherein V is the standard altitude wind speed, Z is the anemometer altitude recorded in the historical welfare of the meteorological station, V 0 Observing the wind speed by an anemoscope, wherein alpha is the ground roughness coefficient of an open and flat area, and the value is 0.15;
the wind speed is then converted using the following linear regression model:
V 10 ＝aV 2 +b；
wherein, V 10 Is a 10 minute sequence of maximum values of mean wind speed, V 2 Is a 2-minute average wind speed maximum value sequence, selects annual self-recording 10-minute average maximum wind speed values from the stations with wind speed self-recording, and converts the values to a standard height to obtain V 10 A sequence; selecting average maximum wind speed value of 2 minutes at the same year of corresponding station and converting to standard height to obtain V 2 A sequence; the constants a, b are found according to the least squares method.
5. The method for predicting the wind speed interval in the exploitable stage of the overhead line engineering according to claim 3, wherein the step of cleaning abnormal data in the continuous wind speed data in a segmented manner by adopting an abnormal value processing method, and the step of reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm comprises the following steps:
dividing the historical wind speed data into s intervals, wherein the power data sample of the ith wind speed-power interval in the s intervals is P i ＝{p i,1 ,p i,2 ,…,p i,m Where i =1,2, \ 8230;, s, p i,1 ≤p i,2 ≤…≤p i,m And m is the number of power points in each wind speed-power interval;
calculating the average power value in the ith wind speed-power interval by the following formula:
wherein, P i,j The power value of each power point in the ith wind speed-power interval is obtained;
calculating a standard deviation of the power point data within the ith wind speed-power interval by the following formula:
calculating the absolute value of the deviation of each power point data in the ith wind speed-power interval by the following formula:
the absolute value delta of the deviation of the power point data when in the ith wind speed-power interval i,j At maximum, the power point data is an abnormal value to be eliminated;
the value of τ is calculated by the following equation:
wherein t is a t distribution value of the power sample data; alpha is the significance level, the value of which affects the adequacy of the power data, when delta i,j ≥τS i When the power point data is an abnormal point, when delta i,j ≤τS i The power point data is a normal point.
6. The method for predicting the wind speed interval in the exploitable stage of the overhead line engineering according to claim 5, wherein the reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm comprises:
arranging the cleaned abnormal data from small to large and dividing the abnormal data into four equal parts, taking the data at the positions of 3 dividing points as quartiles, and calculating a second quartile M by the following formula i ：
M i For power data samples P i' The median, the first quantile and the third quantile of the data respectively represent P i' The position of the middle before and after 25% power data point is divided;
when M is an even number, passing M i Will P i' Dividing into two sequences P of equal length i_1 ＝{p i,1 ,p i,2 ,…,p i,(m-1)/2 And P i_2 ＝{p i,(m+1)/2 ,p i,(m+3)/2 ,…,p i,m }，Q 1,i Is P i_1 Median of (2), Q 3,i Is P i_2 When m =4k +3 (k =0,1. -), Q is calculated by the following formula 1,i And Q 3,i ：
When m =4k +1 (k =0,1,.), Q is calculated by the following equation 1,i And Q 3,i ：
P is calculated by the following formula i' A quarter-bit pitch I i ：
I i ＝Q 3,i -Q 1,i ；
Upper limit W calculated by the following formula u,i Lower limit W d,i To exclude outliers in the data:
W u,i ＝Q 3,i +1.5I i ；
W d,i ＝Q 1,i -1.5I i ；
after the error is identified, calculating power points to be interpolated by using the adjacent 4 power values corresponding to the wind speed through a four-point interpolation subdivision algorithm:
wherein, P i-1 ，P i ，P i+1 ，P i+2 The power values are respectively corresponding to 4 wind speeds closest to the power point to be interpolated; omega is a tensor parameter, when omega is epsilon [0,0.125]Then, an interpolation point is obtained.
7. The method for predicting the wind speed interval in the exploratory stage of the overhead line engineering according to claim 3, wherein the step of decomposing the wind speed into a plurality of subsequences by combining a variation modal decomposition method, reconstructing the plurality of subsequences into a stable trend sequence, an ordered fluctuation sequence and an unordered concussion sequence according to sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the unordered concussion sequence based on fuzzy information granulation comprises the steps of:
for each mode function u k (t) calculating the corresponding analysis signal by means of a Hilbert transformTo obtain its single-sided spectrum:
wherein δ (t) is a dirac function; k is the number of modes to be decomposed;
for said each mode function u k (t) by aliasing and mode function u k (t) center frequency ω corresponding to k Index term e of -jωkt Modulating the spectrum of each mode to a corresponding fundamental frequency band;
and (3) estimating the bandwidth of each mode signal by using a Gaussian smoothing method for mediating signals to solve the variation problem under the constraint condition, wherein the objective function is as follows:
decomposing an original wind speed sequence into k subsequences through a variation mode, and obtaining a stable trend sequence, an ordered fluctuation sequence and an unordered oscillation sequence based on a sample entropy value;
processing the disordered oscillation sequence by adopting a fuzzy information granulation method to obtain the minimum value, the average value and the maximum value of each window;
establishing fuzzy information particles for each window by adopting the following membership functions:
wherein x is a wind speed sequence, and a, m and b are the minimum value, the average value and the maximum value of each fuzzy information particle of the fuzzy granulation window.
8. The method for predicting wind speed intervals in the exploitable stage of the overhead line engineering according to claim 3, wherein the step of respectively constructing the prediction models of the steady trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by using a particle swarm optimization neural network comprises the following steps of:
respectively establishing a gated cycle unit GRU prediction model of a particle swarm optimization PSO for the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence, and performing joint optimization on an embedding dimension m, a delay time tau and a hidden layer node number L through the PSO;
the GRU prediction model is used for processing a time sequence, selectively forgetting or reserving input historical information in a gating mode to predict future wind speed data, comprises a training part and a prediction part, comprises an input layer, a hidden layer and an output layer, is trained by a gradient descent algorithm according to sample loss, enables a sample loss function value to be minimum, and updates the weight of each network parameter:
z t ＝σ(W xz x t +W hz h t-1 +b z )；
r t ＝σ(W rz x t +W hr h t-1 +b r )；
y t ＝σ(W ho h t +b y )；
wherein, the phase point of the phase space reconstruction matrix is used as the network input x, the sequence x = { x (1 + M τ), x (2 + M τ) } is used as the network output y, h is the hidden layer, σ is the hidden layersigmoid activation function, z t ,r t Update and reset gates, W, respectively, of the circulation network xz Is the weight between the input layer and the refresh gate at time t; w is a group of hz Is the weight between the hidden layer at time t-1 and the update gate at time t, W xr Is the weight between the input layer and the reset gate at time t, W hr Is the weight between the hidden layer at time t-1 and the reset gate at time t, W ho Is the weight between the hidden layer and the output layer at time t and theta is the Hadamard product of the matrix.
9. An apparatus for predicting a wind speed interval in a exploitable stage of an overhead line project, comprising:
the data acquisition module is used for collecting historical wind speed data;
the preprocessing module is used for preprocessing the historical wind speed data, wherein the preprocessing comprises data correction, continuity inspection and processing of abnormal values and missing values;
the decomposition and reconstruction module is used for decomposing the wind speed into a plurality of subsequences by combining a variational modal decomposition method, reconstructing the subsequences into a stable trend sequence, an ordered fluctuation sequence and a disordered oscillation sequence according to the sample entropy, and then identifying and extracting the maximum value, the minimum value and the average value of the disordered oscillation sequence based on fuzzy information granulation; and
and the prediction module is used for respectively constructing prediction models of the stable trend sequence, the ordered fluctuation sequence, the maximum value sequence, the minimum value sequence and the average value sequence by adopting a particle swarm optimization neural network, superposing the prediction results of the prediction models and determining the wind speed interval of the exploitable stage of the overhead line engineering.
10. The device for predicting the wind speed interval in the exploitable stage of the overhead line engineering according to claim 9, wherein the historical wind speed data comprises discontinuous wind speed data and continuous wind speed data, and the preprocessing module is used for performing significance check on a wind speed sequence before the discontinuity and a wind speed sequence after the discontinuity in the discontinuous wind speed data to verify the continuity of the data and correcting the discontinuous wind speed data by adopting a power exponent formula; performing segmented cleaning on abnormal data in the continuous wind speed data by adopting an abnormal value processing method, and reconstructing the cleaned abnormal data by a four-point interpolation subdivision algorithm; and analyzing the historical wind speed data which is obviously larger or smaller, and determining whether the data record is reliable or not through weather system process analysis, inter-station comparative analysis and correlation analysis of different meteorological elements.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211509211.9A CN115796361A (en) | 2022-11-29 | 2022-11-29 | Wind speed interval prediction method and device for ground stage of overhead line engineering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211509211.9A CN115796361A (en) | 2022-11-29 | 2022-11-29 | Wind speed interval prediction method and device for ground stage of overhead line engineering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115796361A true CN115796361A (en) | 2023-03-14 |
Family
ID=85442896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211509211.9A Pending CN115796361A (en) | 2022-11-29 | 2022-11-29 | Wind speed interval prediction method and device for ground stage of overhead line engineering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115796361A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117725540A (en) * | 2024-02-07 | 2024-03-19 | 宇恒数智(北京)科技有限公司 | Method, system, equipment and medium for calculating dynamic baseband |
-
2022
- 2022-11-29 CN CN202211509211.9A patent/CN115796361A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117725540A (en) * | 2024-02-07 | 2024-03-19 | 宇恒数智(北京)科技有限公司 | Method, system, equipment and medium for calculating dynamic baseband |
| CN117725540B (en) * | 2024-02-07 | 2024-05-07 | 宇恒数智(北京)科技有限公司 | Method, system, equipment and medium for calculating dynamic baseband |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kankanala et al. | AdaBoost $^{+} $: An ensemble learning approach for estimating weather-related outages in distribution systems | |
| CN113392961B (en) | Method for extracting mesoscale eddy track stable sequence and predicting cyclic neural network | |
| CN114297036B (en) | Data processing method, device, electronic equipment and readable storage medium | |
| Wu et al. | A hybrid support vector regression approach for rainfall forecasting using particle swarm optimization and projection pursuit technology | |
| Paoli et al. | Solar radiation forecasting using ad-hoc time series preprocessing and neural networks | |
| CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
| CN110020712B (en) | Optimized particle swarm BP network prediction method and system based on clustering | |
| CN106526710A (en) | Haze prediction method and device | |
| CN111695290A (en) | Short-term runoff intelligent forecasting hybrid model method suitable for variable environment | |
| CN114492922A (en) | Medium-and-long-term power generation capacity prediction method | |
| Zhang et al. | A hybrid forecasting system with complexity identification and improved optimization for short-term wind speed prediction | |
| CN114372631A (en) | Data-lacking area runoff prediction method based on small sample learning and LSTM | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN115796361A (en) | Wind speed interval prediction method and device for ground stage of overhead line engineering | |
| CN114970946A (en) | PM2.5 pollution concentration long-term space prediction method based on deep learning model and empirical mode decomposition coupling | |
| CN117116045A (en) | Traffic flow prediction method and device based on space-time sequence deep learning | |
| Georgoulas et al. | Examining nominal and ordinal classifiers for forecasting wind speed | |
| CN116486611A (en) | Urban road vehicle speed prediction method | |
| CN116822742A (en) | Power load prediction method based on dynamic decomposition-reconstruction integrated processing | |
| CN114741952A (en) | Short-term load prediction method based on long-term and short-term memory network | |
| Putranto et al. | Deep Learning Approach for Heavy Rainfall Prediction Using Himawari-8 And RDCA Data | |
| CN110188967A (en) | Electric load probabilistic forecasting method based on chaos crowd algorithm and Bayesian network | |
| Huang et al. | Ultra-short-term prediction of microgrid source load power considering weather characteristics and multivariate correlation | |
| CN114169585B (en) | Wind power generation power prediction method based on Markov chain and combined model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |