CN109583588B - Short-term wind speed prediction method and system - Google Patents
Short-term wind speed prediction method and system Download PDFInfo
- Publication number
- CN109583588B CN109583588B CN201811539880.4A CN201811539880A CN109583588B CN 109583588 B CN109583588 B CN 109583588B CN 201811539880 A CN201811539880 A CN 201811539880A CN 109583588 B CN109583588 B CN 109583588B
- Authority
- CN
- China
- Prior art keywords
- wind speed
- data
- prediction
- historical
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a short-term wind speed prediction method and a short-term wind speed prediction system. The method comprises the following steps: acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data; optimizing the neural network model through a genetic algorithm; training the optimized neural network model through the historical wind speed data to obtain a prediction model; performing variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data; and predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data. The method or the system can effectively improve the reliability of wind speed prediction accuracy.
Description
Technical Field
The invention relates to the field of wind speed prediction, in particular to a short-term wind speed prediction method and a short-term wind speed prediction system.
Background
Since the 21 st century, energy and environmental problems are becoming more and more global, and the scarcity of fossil energy and irreversibility of environmental pollution have prompted countries in the world to make energy sustainable development strategies and to construct diversified energy systems. The development of green renewable energy sources is a key for solving the current energy problems, and is a necessary way for promoting the transformation of global energy sources.
Wind power generation is a new energy power generation technology with zero emission and low operation cost, and large-scale development and utilization are realized worldwide. According to data issued by the global wind energy council, the cumulative installed capacity of the global wind power market in 2017 exceeds 500GW, the newly added installed capacity exceeds 50GW, wherein the cumulative installed capacity 188GW accounts for 34.8% of the global cumulative installed capacity, the newly added installed capacity is 19.5GW which accounts for 37% of the global new installed capacity, and the global wind power market is continuously brought forward. In recent years, offshore wind power has a good development tendency, and by 2017, the installed capacity 4.331GW is added to offshore wind power worldwide, and the accumulated installed capacity reaches 18.814 GW. However, as the installed capacity is continuously increased, the problem of power grid consumption is increasingly highlighted. The power grid is low in consumption level, and the wind abandoning and electricity limiting are severe, so that immeasurable economic loss is brought to each wind power enterprise, the power generation utilization rate of new energy is greatly weakened, and the development of the new energy industry is limited. Accurate wind speed prediction is a key link for effectively improving the power grid absorption capacity.
Disclosure of Invention
The invention aims to provide a short-term wind speed prediction method and a short-term wind speed prediction system, which are used for effectively improving the reliability of wind speed prediction accuracy.
In order to achieve the purpose, the invention provides the following scheme:
a short-term wind speed prediction method, the method comprising:
acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data;
optimizing the neural network model through a genetic algorithm;
training the optimized neural network model through the historical wind speed data to obtain a prediction model;
performing variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data;
and predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data.
Optionally, the acquiring historical wind speed data specifically includes:
acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data;
and screening out the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range by a hierarchical clustering method, wherein the first initial historical wind speed data is first historical wind speed data.
Optionally, the optimizing the neural network model by using a genetic algorithm specifically includes:
optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold;
and determining a neural network model according to the optimal initial weight and the optimal threshold value to obtain an optimized neural network.
Optionally, the training of the optimized neural network model through the historical wind speed data to obtain a prediction model specifically includes:
carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
if yes, determining the optimized neural network model as a prediction model;
if not, adjusting the optimized neural network model to enable the error between the output data and the first nonlinear fluctuation wind speed characteristic data to be within the range of an error threshold value.
Optionally, the predicting the plurality of second nonlinear fluctuation wind speed characteristic data by the prediction model to obtain wind speed prediction data specifically includes:
predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data;
and linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
A short term wind speed prediction system, the system comprising:
the wind speed data acquisition module is used for acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data;
the optimization module is used for optimizing the neural network model through a genetic algorithm;
the training module is used for training the optimized neural network model through the historical wind speed data to obtain a prediction model;
the decomposition module is used for carrying out variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data;
and the prediction module is used for predicting the wind speed through the plurality of second nonlinear fluctuation wind speed characteristic data and the prediction model to obtain the predicted wind speed.
Optionally, the wind speed data acquiring module specifically includes:
the wind speed data acquisition unit is used for acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data;
and the screening unit is used for screening the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range through a hierarchical clustering method, and the first initial historical wind speed data is first historical wind speed data.
Optionally, the optimization module specifically includes:
the optimization unit is used for optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold;
and the first model determining unit is used for determining the neural network model according to the optimal initial weight and the optimal threshold value to obtain the optimized neural network model.
Optionally, the training module specifically includes:
the decomposition unit is used for carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
the input unit is used for taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
the judging unit is used for judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
a second model determining unit, configured to determine the optimized neural network model as a prediction model when an error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range;
and the adjusting unit is used for adjusting the optimized neural network model when the error between the output data and the first nonlinear fluctuating wind speed characteristic data is out of an error threshold range, so that the error between the output data and the first nonlinear fluctuating wind speed characteristic data is in the error threshold range.
Optionally, the prediction module specifically includes:
the prediction unit is used for predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data;
and the superposition unit is used for linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
Compared with the prior art, the invention has the following technical effects: according to the invention, a systematic clustering method is adopted to screen original data to obtain data with high correlation degree with a prediction day as a prediction modeling sample, and a variational modal decomposition method is combined to decompose deleted historical data, so that the complexity of the data is reduced. Meanwhile, the initial weight and the threshold of the BP neural network are optimized by utilizing a genetic algorithm, the possibility that the neural network falls into a local extreme point or even does not converge is reduced, meanwhile, the optimized neural network has a better fitting effect on nonlinear data, and the prediction performance of the neural network is improved. The method effectively weakens the intermittence and the fluctuation of the wind speed and improves the reliability of the wind speed prediction precision.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a short term wind speed prediction method according to an embodiment of the present invention;
FIG. 2 is a wind speed data distribution graph of a wind farm for 9 days according to an embodiment of the invention;
FIG. 3 is a schematic diagram of a clustering result of wind speed data according to an embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating a decomposition result of wind speed data according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating imf component prediction results according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a wind speed prediction curve for each model according to an embodiment of the present invention;
FIG. 7 is a schematic diagram of the prediction errors of four wind speed prediction models according to an embodiment of the present invention;
FIG. 8 is a graph of error distribution for a prediction model according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of a short-term wind speed prediction system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in FIG. 1, the short-term wind speed prediction method includes the following steps:
step 101: acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data. The acquiring of the historical wind speed data specifically comprises: acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data; and screening out the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range by a hierarchical clustering method, wherein the first initial historical wind speed data is first historical wind speed data.
Systematic Cluster analysis, also known as Hierarchical Cluster method (Hierarchical Cluster), is one of the commonly used Cluster analysis methods. The difference degree between the variables is a classification standard of the system clustering method and can be realized by calculating the distance between the variables. The common clustering analysis distance calculation method comprises the following steps: the method comprises a shortest distance method, a longest distance method, a class average method, a gravity center method, a middle distance method and a dispersion square sum method, wherein the dispersion square sum method is adopted to calculate the distance between variables.
The basic idea of the system clustering analysis is to regard n variables as one class, merge the closest classes according to the distance between the classes, calculate the distance between the new class and other classes, and select the class with the closest distance to merge, and reduce one class every time of merging, continuously calculate the process until all the variables are merged into one class. The basic steps of the system cluster analysis are as follows:
(1) calculating the distance between n variables { d }ijD is recorded;
(2) constructing n classes, wherein each class only comprises one variable;
(3) merging two types with the shortest distance into a new type;
(4) calculating the distance between the new class and the current classes, if the number of the classes is 1, turning to the step (5), and if not, returning to the step (3);
(5) drawing a clustering chart;
(6) and determining the number and the types of the classes.
Step 102: and optimizing the neural network model through a genetic algorithm. Optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold; and determining a neural network model according to the optimal initial weight and the optimal threshold value to obtain an optimized neural network.
The genetic algorithm simulates a coding string formed by parameters to be optimized into a biological evolution process, and generates the next generation through operations such as selection, crossing, mutation and the like, thereby continuously improving the fitness value of individuals in a population until a certain termination condition is met. The BP neural network is a multilayer feedforward artificial neural network, has strong self-learning and self-adaptive capabilities, and is particularly suitable for nonlinear data fitting. However, the initial weight and the initial threshold which are randomly generated easily cause the network to be trapped in a local extreme point or even not converged in the training process, so the initial weight and the threshold of the BP neural network are optimized by adopting a genetic algorithm. The process of optimizing the BP neural network by the genetic algorithm is given as follows:
(1) initializing GA algorithms and parameters of the neural network. The BP neural network structure is set to be 6-6-1, the training times are 100, the learning rate is 0.1, and the error target is 0.0001. The population scale in the genetic algorithm is set to be 10, the evolution times are set to be 60, the cross probability is set to be 0.3, and the variation probability is set to be 0.1.
(2) And carrying out real number coding to generate an initial population. And taking the network weight value and the threshold value which are cascaded in a certain sequence as a chromosome of the genetic algorithm. Within the weight and threshold, M such chromosomes are generated, constituting an initial population. The calculation formula of the code length L is as follows:
L＝i×j+j+j×k+k(1)
wherein i is the number of nodes of the input layer of the neural network, j is the number of nodes of the hidden layer of the neural network, and k is the number of nodes of the output layer of the neural network.
(3) And constructing a fitness function. The genetic algorithm searches by using the fitness value of each chromosome, and the higher the fitness value is, the higher the possibility of inheritance to the next generation is. The calculation formula of the fitness function f adopted by the invention is as follows:
wherein the content of the first and second substances,
to predict value, yiAre true values.
(4) And (4) selecting. I.e., selecting individuals for cross-mutation, let f (x)i) Is an individual xiThe probability of individual selection is:
wherein N is the number of the population.
(5) And (4) crossing. The method adopts a real number crossing method and specifically operates as follows
n is [0,1 ]]Random number between, mkjIs the j gene of the individual k, mljIs the j gene of the individual l.
(6) And (5) carrying out mutation. Selection of the j-th Gene m of an individual iijAnd performing mutation operation according to a formula.
Wherein m ismax，mminIs mijG is the number of iterations, r is the gene mijSpecific gravity in individual i.
Wherein r is2Is a random number, GmaxIs the maximum number of iterations.
(7) And (5) repeatedly carrying out the steps (3) to (5), selecting the optimal individual from all the individuals, and stopping evolution when the target function reaches a set value or reaches the maximum iteration number to obtain the initial weight and the threshold of the BP neural network.
Step 103: and training the optimized neural network model through the historical wind speed data to obtain a prediction model. The method specifically comprises the following steps:
carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
if yes, determining the optimized neural network model as a prediction model;
if not, adjusting the optimized neural network model to enable the error between the output data and the first nonlinear fluctuation wind speed characteristic data to be within the range of an error threshold value.
Step 104: and carrying out variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data.
The variational modal decomposition is a self-adaptive, quasi-orthogonal and completely non-recursive decomposition method, and is a variational problem solving process based on classical wiener filtering, Hilbert transform and frequency mixing. In the process of obtaining the decomposition variable, the frequency center and the bandwidth of each component are determined by iteratively searching the optimal solution of the variational model, and the estimation of the modes is converted into the solution of the variational problem, so that the sum of the estimation bandwidths of each mode is minimum, and the frequency domain subdivision of the signals and the effective separation of each component can be adaptively realized.
The variation modal decomposition method is completely different from the empirical modal decomposition method in principle. Empirical mode decomposition is only simple circular screening, and variational mode decomposition puts the decomposition of signals into a variational model, and realizes the decomposition of complex signals by searching the optimal solution of a constraint variational model. In the process, the center frequency and the bandwidth of each eigenmode function component are continuously and alternately updated in an iterative manner, and finally, the signal is adaptively decomposed according to the frequency band to obtain K narrow-band eigenmode function components set by the eigenmode function component. The specific steps of the variational modal decomposition are as follows:
construction variation problem (evaluation mode bandwidth)
(1) For each mode function uk(t) solving an analysis signal through Hilbert transform to obtain a single-edge spectrum;
wherein δ (t) is the dirac distribution; denotes convolution.
(2) The analytic signal of each mode is multiplied by the estimated central frequency
Move its spectrum to the baseband;
(3) solving the bandwidth of each modal function by Gaussian smoothing, wherein each modal function uk(t) the sum of which is the decomposed signal x (t), i.e.:
solving variational problems
(1) Introducing a secondary penalty factor alpha and a Lagrange multiplication operator lambda, and converting the constraint variable problem into an unconstrained variable problem, namely:
wherein u isk(t) is a mode function, ωkF (t) is the original input data.
(2) The above variational problem is solved by adopting a multiplication operator alternating direction method, and u is updated alternatelyk、ωkAnd λ seeks to extend the minimum point of the lagrange expression.
ukIteration of (2):
ωkiteration of (2):
iteration of λ:
(3) the specific process of VMD decomposition is as follows:
step one, initialization
λ1Setting the iteration number n to 1;
step two, the iteration times n are n + 1;
step three, for 1: k,
for all omega ≧ 0, update according to equation (11)
Updating according to equation (12)
Updating according to equation (13)
Step four, carrying out double lifting on all omega not less than 0
Step five, repeating the step two to the step four until an iteration constraint condition is met:
and finishing the whole cycle and outputting the result so as to obtain k narrow-band IMF components.
Step 105: and predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data. Predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data; and linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: according to the invention, a systematic clustering method is adopted to screen original data to obtain data with high correlation degree with a prediction day as a prediction modeling sample, and a variational modal decomposition method is combined to decompose deleted historical data, so that the complexity of the data is reduced. Meanwhile, the initial weight and the threshold of the BP neural network are optimized by utilizing a genetic algorithm, the possibility that the neural network falls into a local extreme point or even does not converge is reduced, meanwhile, the optimized neural network has a better fitting effect on nonlinear data, and the prediction performance of the neural network is improved. The method effectively weakens the intermittence and the fluctuation of the wind speed and improves the reliability of the wind speed prediction precision.
In order to verify that the method has good prediction performance on the wind speed data of the actual wind power plant, the wind speed of the Chinese wind power plant is adopted to carry out a data simulation experiment.
Data analysis
A wind power plant takes wind speed data every 10 minutes, 144 wind speed points are collected every day, FIG. 2 is a wind speed data distribution curve of the wind power plant for 9 days, the maximum wind speed is close to 18m/s, the minimum wind speed is only close to 0m/s, the wind speed fluctuation is large, and the wind speed fluctuation is maximum on the 7 th day.
Clustering analysis results
The wind speed sequence is a group of nonlinear time sequences, records the sequence and the size of data, and presents the dynamic characteristics of wind speed. The wind speed characteristics can be obviously distinguished along with the change of four seasons, and can also generate a certain rule due to the periodic change of sunrise and sunset. In order to fully utilize data information contained in historical wind speed data, the system clustering analysis method is used for classifying the data, Euclidean distance is used for measuring the similarity between two vectors, and a dispersion square sum method (Ward method) is used for carrying out parallel classification. Wind speed data of 9 days, the first 8 days are taken as training sets, the 9 th day is a prediction set, 1296 data are classified by using a clustering analysis method, the clustering result is shown in figure 3, and the abscissa represents the day of wind speed data collection. The clustering analysis result shows that the data with the highest correlation degree with the prediction set are the 4 th day and the 5 th day, then the 8 th day, and finally the 1 st day, the 2 nd day and the 3 rd day, so the complexity of the original wind speed data is effectively reduced by taking the data of the 1 st day, the 2 nd day, the 3 rd day, the 4 th day, the 5 th day and the 8 th day as input signals of the Variational Modal Decomposition (VMD).
VMD decomposition result of original wind speed sequence
The screened wind speed data was subjected to metamorphic mode decomposition with k being 5 and α being 500, and the decomposition results are shown in fig. 4. The black curve is wind speed data to be decomposed, the fluctuation range of the wind speed data is large, strong nonlinear characteristics are shown, imf 1-imf 5 are 5 components obtained by decomposing a wind speed signal, and the fluctuation of imf1 components to imf5 components is gradually enhanced, and the random fluctuation characteristics of an original wind speed sequence are shown.
Improved GA-BP wind speed prediction model modeling step based on variational modal decomposition
And (3) integrating the advantages of system clustering analysis, Variational Modal Decomposition (VMD), genetic algorithm and BP neural network to obtain a specific modeling process of the short-term wind speed prediction model, and naming the model as an HC-VMD-GA-BP model. The detailed modeling flow is as follows:
the method comprises the following steps: the raw data is classified based on a systematic clustering analysis. Performing system clustering analysis on the original data of 9 days, and screening high data y (t) related to the wind speed data of 9 days according to the clustering graph result;
step two: selecting parameters k and alpha, and performing Variational Modal Decomposition (VMD) on the screened data to obtain k empirical mode functions with nonlinear fluctuation characteristics: imf1, …, imfk;
step three: the GA algorithm optimizes the BP neural network. Optimizing the initial weight and the threshold of the BP neural network by using the global search capability of a genetic algorithm, and determining the optimal initial weight and the optimal initial threshold;
step four: selecting an optimized GA-BP neural network to predict k empirical mode functions respectively, and recording prediction results of k components: y is1(t),…,yk(t)；
Step five: and linearly superposing the component prediction results to obtain a wind speed prediction result:
step six: and carrying out error analysis on the wind speed prediction result. And comprehensively evaluating the prediction result by using the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE) and the Mean Absolute Percent Error (MAPE).
Wherein V (t) represents raw wind speed data,
indicating the predicted value of wind speed.
Wind speed prediction results and error analysis
1) Each component prediction result
The wind speed sequence is a typical nonlinear time sequence, and the inherent random fluctuation of the wind speed sequence brings a plurality of difficulties for improving the wind speed prediction accuracy. The wind speed sequence is decomposed by a VMD method, and the imf components are predicted by a GA-BP neural network to obtain the prediction result of each imf component, as shown in FIG. 5. In the figure, a solid curve represents an actual component sequence, a dashed curve represents a component prediction sequence, the fluctuation of imf1 and imf2 is small, the prediction curve is almost coincident with a real curve, the fluctuation of imf 3-imf 5 is gradually increased, but the trend of the prediction curve is basically consistent with that of the real sequence and is almost different in value. Therefore, the VMD decomposition deeply excavates the information contained in the original wind speed data, effectively lowers the fluctuation of the original wind speed data, and effectively improves the prediction performance of the GA-BP neural network.
2) Predicted result of wind speed
The wind speed prediction curves for each model are shown in FIG. 6. The abscissa is the prediction period and the ordinate is the wind speed value. Most of the prediction curves of the four models are below the real wind speed curve, which shows that most of the prediction results of the four models are lower than the real wind speed value. The wind speed prediction curves of the BP neural network model and the GA-BP model are consistent with the change trend of an original wind speed sequence, but the point curve is closer to the original wind speed sequence, and the difference between the dotted curve and the point curve is more obvious after the eighteenth prediction point, which fully shows that the prediction performance of the BP neural network is effectively optimized by the genetic algorithm. Meanwhile, compared with the prediction result of the GA-BP model, the HC-GA-BP model better describes the change trend of the original wind speed sequence, and the data processed by the system clustering analysis has important significance for improving the prediction precision. The HC-VMD-GA-BP model not only describes the dynamic changes of the original wind speed sequence more accurately, but also is closer to the original wind speed sequence in value. By combining the analysis, the prediction performance of the prediction model has sensitivity to the wind speed data, the wind speed sequence is a complex time sequence, the complexity of the wind speed data is low by adopting a system clustering analysis method, the implicit information of the historical wind speed data is mined by utilizing a VMD decomposition method, the strong nonlinear fitting capability and the generalization capability of the GA-BP neural network model are considered, and therefore, the HC-VMD-GA-BP model has good prediction performance and is suitable for prediction of real wind speed data, and the prediction precision is effectively improved.
3) Error analysis of wind speed prediction model
Prediction error of four wind speed prediction models such asAs shown in fig. 7, the error distributions are shown in fig. 8(a) - (d). Figure 7 middle belt
Curves and band + curves represent the prediction errors of the BP model and the GA-BP model, respectively, and the band curves are in [ -2m/s, 1.5m/s]Fluctuates between the two, and the band + curve is [ -1.5m/s, 1m/s]Although the fluctuation range of the band + curve is large, the large frequency error distribution interval of the BP model represented by the graph (a) of FIG. 8 is [ -1.5m/s, 0.5m/s]And the larger frequency error distribution interval of the GA-BP model represented in FIG. 8(b) is [ -1m/s, 0.5m/s]It can be seen that the error distribution of the GA-BP model is more concentrated but points with large individual prediction errors exist, but the overall prediction effect is better than that of the BP neural network. The o-curve in FIG. 7 represents the prediction error of HC-GA-BP model, which is [ -1.2m/s, 1m/s]The prediction error of HC-VMD-GA-BP model is [ -0.3m/s, 0.3m/s]The fluctuation, and the large frequency fluctuation interval in FIG. 8(c) is significantly larger than that in FIG. 8(d), the VMD decomposition improves the prediction performance of the HC-GA-BP model well. In conclusion, the HC-VMD-GA-BP model has obvious advantages in wind speed prediction and good prediction performance.
And comprehensively evaluating the wind speed prediction result of each model by using the mean absolute error (MSE), the Root Mean Square Error (RMSE) and the Mean Absolute Percent Error (MAPE). The specific prediction accuracy of each model is shown in table 1. The MAE, RMSE and MAPE of the BP neural network are 0.6545, 0.7895 and 11.87 percent respectively, compared with the error index of the BP neural network, the MAE, RMSE and MAPE of the GA-BP model are reduced by at least 38.29 percent, and the prediction performance of the GA-BP model is obviously far superior to that of the BP neural network. Three error indexes of the HC-GA-BP model are obviously smaller than those of the GA-BP model, and therefore, the prediction accuracy of the GA-BP can be effectively improved by utilizing the system clustering analysis to process the historical wind speed data. MAE, RMSE and MAPE of the HC-VMD-GA-BP model are respectively 0.1304, 0.1581 and 2.55 percent, and are reduced by at least 58.44 percent compared with the HC-GA-BP model, so that the random fluctuation of a wind speed sequence is effectively reduced by VMD decomposition on original wind speed data, and the short-term wind speed prediction precision is improved. Through the analysis, the three error indexes of the HC-VMD-GA-BP are better than those of the other three prediction models, and the HC-VMD-GA-BP model is proved to have good prediction performance again.
TABLE 1 error statistics for wind speed prediction
As shown in fig. 9, a short-term wind speed prediction system includes:
a wind speed data obtaining module 901, configured to obtain historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data.
The wind speed data obtaining module 901 specifically includes:
the wind speed data acquisition unit is used for acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data;
and the screening unit is used for screening the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range through a hierarchical clustering method, and the first initial historical wind speed data is first historical wind speed data.
And an optimizing module 902, configured to optimize the neural network model through a genetic algorithm.
The optimization module 902 specifically includes:
the optimization unit is used for optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold;
and the first model determining unit is used for determining the neural network model according to the optimal initial weight and the optimal threshold value to obtain the optimized neural network model.
And the training module 903 is used for training the optimized neural network model through the historical wind speed data to obtain a prediction model.
The training module 903 specifically includes:
the decomposition unit is used for carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
the input unit is used for taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
the judging unit is used for judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
a second model determining unit, configured to determine the optimized neural network model as a prediction model when an error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range;
and the adjusting unit is used for adjusting the optimized neural network model when the error between the output data and the first nonlinear fluctuating wind speed characteristic data is out of an error threshold range, so that the error between the output data and the first nonlinear fluctuating wind speed characteristic data is in the error threshold range.
And a decomposition module 904, configured to perform variational modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data.
And the predicting module 905 is configured to predict the wind speed through the plurality of second nonlinear fluctuation wind speed characteristic data and the prediction model to obtain a predicted wind speed.
The prediction module 905 specifically includes:
the prediction unit is used for predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data;
and the superposition unit is used for linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (4)
1. A method of short-term wind speed prediction, the method comprising:
acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data;
optimizing the neural network model through a genetic algorithm;
training the optimized neural network model through the historical wind speed data to obtain a prediction model;
performing variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data;
predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data;
the acquiring of the historical wind speed data specifically comprises:
acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data;
screening out the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range through a hierarchical clustering method, wherein the first initial historical wind speed data is first historical wind speed data;
training the optimized neural network model through the historical wind speed data to obtain a prediction model, and specifically comprising the following steps:
carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
if yes, determining the optimized neural network model as a prediction model;
if not, adjusting the optimized neural network model to enable the error between the output data and the first nonlinear fluctuation wind speed characteristic data to be within an error threshold range;
predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data and obtain predicted wind speed, and the method specifically comprises the following steps:
predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data;
and linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
2. The short-term wind speed prediction method according to claim 1, wherein the optimization of the neural network model by a genetic algorithm specifically comprises:
optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold;
and determining a neural network model according to the optimal initial weight and the optimal threshold value to obtain an optimized neural network.
3. A short term wind speed prediction system, characterized in that the system comprises:
the wind speed data acquisition module is used for acquiring historical wind speed data; the historical wind speed data comprises first historical wind speed data and second historical wind speed data, and the first historical wind speed data is data before the second historical wind speed data;
the optimization module is used for optimizing the neural network model through a genetic algorithm;
the training module is used for training the optimized neural network model through the historical wind speed data to obtain a prediction model;
the decomposition module is used for carrying out variation modal decomposition on the second historical wind speed data to obtain a plurality of second nonlinear fluctuation wind speed characteristic data;
the prediction module is used for predicting the second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain wind speed prediction data;
the wind speed data acquisition module specifically comprises:
the wind speed data acquisition unit is used for acquiring initial historical wind speed data; the initial historical wind speed data comprises first initial historical wind speed data and second historical wind speed data;
the screening unit is used for screening the first initial historical wind speed data with the correlation with the second historical wind speed data within a threshold value range through a hierarchical clustering method, and the first initial historical wind speed data is first historical wind speed data;
the training module specifically comprises:
the decomposition unit is used for carrying out variation modal decomposition on the first historical wind speed data to obtain a plurality of first nonlinear fluctuation wind speed characteristic data;
the input unit is used for taking a plurality of first nonlinear fluctuation wind speed characteristic data as the input of the optimized neural network model to obtain output data;
the judging unit is used for judging whether the error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range or not;
a second model determining unit, configured to determine the optimized neural network model as a prediction model when an error between the output data and the first nonlinear fluctuating wind speed characteristic data is within an error threshold range;
the adjusting unit is used for adjusting the optimized neural network model when the error between the output data and the first nonlinear fluctuating wind speed characteristic data is out of an error threshold range, so that the error between the output data and the first nonlinear fluctuating wind speed characteristic data is in the error threshold range;
the prediction module specifically comprises:
the prediction unit is used for predicting each second nonlinear fluctuation wind speed characteristic data through the prediction model to obtain a plurality of initial wind speed prediction data;
and the superposition unit is used for linearly superposing the initial wind speed prediction data to obtain wind speed prediction data.
4. The short term wind speed prediction system of claim 3, wherein the optimization module specifically comprises:
the optimization unit is used for optimizing the initial weight and the threshold of the neural network model through a genetic algorithm to obtain an optimal initial weight and an optimal threshold;
and the first model determining unit is used for determining the neural network model according to the optimal initial weight and the optimal threshold value to obtain the optimized neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811539880.4A CN109583588B (en) | 2018-12-17 | 2018-12-17 | Short-term wind speed prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811539880.4A CN109583588B (en) | 2018-12-17 | 2018-12-17 | Short-term wind speed prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109583588A CN109583588A (en) | 2019-04-05 |
| CN109583588B true CN109583588B (en) | 2021-09-10 |
Family
ID=65930263
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811539880.4A Active CN109583588B (en) | 2018-12-17 | 2018-12-17 | Short-term wind speed prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583588B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942170A (en) * | 2019-08-29 | 2020-03-31 | 华北电力大学（保定） | Short-term wind speed prediction method and system based on information processing |
| CN111553510A (en) * | 2020-04-08 | 2020-08-18 | 东华大学 | Short-term wind speed prediction method |
| CN113740931B (en) * | 2020-05-29 | 2023-12-22 | 金风科技股份有限公司 | Wind array detection method and device for wind generating set |
| CN113627674A (en) * | 2021-08-12 | 2021-11-09 | 中国华能集团清洁能源技术研究院有限公司 | Distributed photovoltaic power station output prediction method and device and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008305A (en) * | 2014-06-11 | 2014-08-27 | 国家电网公司 | Method for estimating distribution of wind resources capable of generating electricity for millions-kilowatt wind electricity base |
| CN104112167A (en) * | 2014-06-06 | 2014-10-22 | 国家电网公司 | Method for obtaining distribution of wind resources capable of power generation |
| CN107392397A (en) * | 2017-08-25 | 2017-11-24 | 广东工业大学 | A kind of short-term wind speed forecasting method, apparatus and system |
| CN108022025A (en) * | 2017-12-28 | 2018-05-11 | 华中科技大学 | A kind of wind speed interval Forecasting Methodology and system based on artificial neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480829A (en) * | 2017-08-25 | 2017-12-15 | 广东工业大学 | A kind of Short-term electricity price forecasting method, apparatus and system |
| CN108694477A (en) * | 2018-07-05 | 2018-10-23 | 广东工业大学 | A kind of Electricity price forecasting solution and relevant apparatus |
-
2018
- 2018-12-17 CN CN201811539880.4A patent/CN109583588B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104112167A (en) * | 2014-06-06 | 2014-10-22 | 国家电网公司 | Method for obtaining distribution of wind resources capable of power generation |
| CN104008305A (en) * | 2014-06-11 | 2014-08-27 | 国家电网公司 | Method for estimating distribution of wind resources capable of generating electricity for millions-kilowatt wind electricity base |
| CN107392397A (en) * | 2017-08-25 | 2017-11-24 | 广东工业大学 | A kind of short-term wind speed forecasting method, apparatus and system |
| CN108022025A (en) * | 2017-12-28 | 2018-05-11 | 华中科技大学 | A kind of wind speed interval Forecasting Methodology and system based on artificial neural network |
Non-Patent Citations (3)
| Title |
|---|
| 基于CEEMDAN-PE和QGA-BP的短期风速预测;赵辉;《电子技术应用》;20181206;说明书第0-4章 * |
| 基于相似样本的风速组合预测;谭沛然等;《太原理工大学学报》;20161130;第752-753页 * |
| 张超凡.铁路桥梁行车风速概率预测.《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》.2018, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109583588A (en) | 2019-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109214575B (en) | Ultrashort-term wind power prediction method based on small-wavelength short-term memory network | |
| CN109583588B (en) | Short-term wind speed prediction method and system | |
| Zhang et al. | Short-term wind speed prediction model based on GA-ANN improved by VMD | |
| CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
| CN110647900B (en) | Intelligent safety situation prediction method, device and system based on deep neural network | |
| CN106650767B (en) | Flood forecasting method based on cluster analysis and real-time correction | |
| CN112348271A (en) | Short-term photovoltaic power prediction method based on VMD-IPSO-GRU | |
| CN110309603B (en) | Short-term wind speed prediction method and system based on wind speed characteristics | |
| CN110580543A (en) | Power load prediction method and system based on deep belief network | |
| CN111784041B (en) | Wind power prediction method and system based on graph convolution neural network | |
| CN109886464B (en) | Low-information-loss short-term wind speed prediction method based on optimized singular value decomposition generated feature set | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN112966871A (en) | Traffic jam prediction method and system based on convolution long-short term memory neural network | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN113887787A (en) | Flood forecasting model parameter multi-objective optimization method based on long-time and short-time memory network and NSGA-II algorithm | |
| CN114169645A (en) | Short-term load prediction method for smart power grid | |
| CN111553510A (en) | Short-term wind speed prediction method | |
| CN104200096A (en) | Lightning arrester grading ring optimization method based on differential evolutionary algorithm and BP neural network | |
| CN112434891A (en) | Method for predicting solar irradiance time sequence based on WCNN-ALSTM | |
| CN111141879B (en) | Deep learning air quality monitoring method, device and equipment | |
| CN114912666A (en) | Short-time passenger flow volume prediction method based on CEEMDAN algorithm and attention mechanism | |
| CN114266416A (en) | Photovoltaic power generation power short-term prediction method and device based on similar days and storage medium | |
| CN116960978A (en) | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction | |
| CN117113054A (en) | Multi-element time sequence prediction method based on graph neural network and transducer | |
| CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |