CN116960978A - Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction - Google Patents
Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction Download PDFInfo
- Publication number
- CN116960978A CN116960978A CN202310949782.2A CN202310949782A CN116960978A CN 116960978 A CN116960978 A CN 116960978A CN 202310949782 A CN202310949782 A CN 202310949782A CN 116960978 A CN116960978 A CN 116960978A
- Authority
- CN
- China
- Prior art keywords
- power
- decomposition
- wind speed
- offshore wind
- lstm
- 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
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 238000013135 deep learning Methods 0.000 claims abstract description 5
- 230000008878 coupling Effects 0.000 claims abstract description 3
- 238000010168 coupling process Methods 0.000 claims abstract description 3
- 238000005859 coupling reaction Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 12
- 241000287127 Passeridae Species 0.000 claims description 10
- 230000002431 foraging effect Effects 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000015654 memory Effects 0.000 claims description 4
- 230000006399 behavior Effects 0.000 claims description 3
- 230000000087 stabilizing effect Effects 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims description 2
- 230000003044 adaptive effect Effects 0.000 claims 1
- 238000010845 search algorithm Methods 0.000 claims 1
- 238000013528 artificial neural network Methods 0.000 abstract description 6
- 230000006641 stabilisation Effects 0.000 abstract 1
- 238000011105 stabilization Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000029087 digestion Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Power Engineering (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The application discloses a wind speed-power combination decomposition reconstruction-based offshore wind power prediction method, which comprises the following steps: performing stabilization treatment on the offshore wind power sequence based on a combined decomposition method; reconstructing and dimension-reducing the decomposed data according to the complexity and similarity characteristics of the components; decomposing the power according to the coupling correlation between the offshore wind speed and the power, and establishing a predictor model by combining with LSTM; according to the network structure characteristics of deep learning, super parameters are optimized by using a Bayesian algorithm, a BO-LSTM model is constructed, and the prediction results of all the BO-LSTM sub-models are overlapped to obtain a final prediction value. The method can solve the problems of strong data fluctuation, incomplete decomposition and large data scale caused by decomposition in the offshore wind power prediction, enhance the capability of a neural network for exploring the correlation between wind speed data and power data, and improve the offshore wind power prediction precision.
Description
Technical Field
The application belongs to the field of offshore wind power prediction, and particularly relates to an offshore wind power prediction method based on wind speed-power combination decomposition reconstruction and Bayesian optimization.
Background
Under the age background of greatly developing renewable clean energy sources and striving to realize the aims of carbon peak and carbon neutralization, the offshore wind power industry in China rapidly develops by virtue of the advantages of abundant resources, high power generation efficiency, close to a load center and the like. However, due to randomness and fluctuation of the offshore wind power output, large-scale offshore wind power grid connection brings serious challenges to safe operation of coastal provinces of China. The method can accurately predict the output of the offshore wind power, can flexibly allocate the spare capacity of the system, ensure the safety and stability of the power system and provide important basis for improving the offshore wind power digestion capability. Along with the rapid development of machine learning, a statistical model and a machine learning prediction method are widely applied to new energy power prediction, such as a neural network, a support vector machine, deep learning and other methods, so that characteristic relations between input and output can be mined, and the method has higher calculation speed and higher precision. Although the machine learning algorithm can effectively improve the precision of the new energy generated power prediction, the traditional machine learning prediction model is difficult to capture the characteristic information of the offshore wind power data due to the randomness and fluctuation of the offshore wind power output, so that the power prediction precision is poor and the ideal effect is difficult to achieve.
Therefore, a new solution is needed to solve this problem.
Disclosure of Invention
The application aims to: in order to overcome the defect of poor power prediction accuracy in the prior art, the application provides an offshore wind power prediction method based on wind speed-power combination decomposition reconstruction and Bayesian optimization, which can improve the problems of strong data volatility, incomplete decomposition and large data scale caused by decomposition in offshore wind power prediction, enhance the capability of a neural network for exploring the correlation between wind speed data and power data, and improve the offshore wind power prediction accuracy.
The technical scheme is as follows: the application provides a wind speed-power combination decomposition reconstruction-based offshore wind power prediction method, which comprises the following steps of:
s1: stabilizing wind speed data based on a combined decomposition method;
s2: reconstructing and dimension-reducing the data decomposed in the step S1 based on the fuzzy entropy FE according to the complexity and similarity characteristics of the components;
s3: according to the coupling correlation between the offshore wind speed and the power, decomposing the historical power, establishing a long-term memory network LSTM predictor model, and selecting wind speed reconstruction data and power data with a certain time scale as inputs of the predictor model;
the time scale of the input data is determined by analyzing the pearson correlation coefficient of the machine output at time t and the wind speed and power history data at the previous n times, i.e. the history data at the previous n times is used for prediction.
S4: according to the network structure characteristics of deep learning, a Bayesian algorithm BO is used for optimizing LSTM super parameters to construct a BO-LSTM model, and the prediction results of all the BO-LSTM sub-models are overlapped to obtain a final prediction value.
Further, the method for decomposing the wind speed data in step S1 includes:
the primary decomposition of the wind speed sequence using ICEEMDAN includes: defining wind speed sequence as v and k-order modal component of EMD decomposition as E k (. Cndot.) adding noise to the wind velocity sequence, calculating the firstResidual R of sub-decomposition es1 And a first order eigenmode component D 1 。
D 1 ＝v-R es1 (1)
Where ε is the signal-to-noise ratio of white noise, w (i) For the added i-th group of gaussian white noise, M is the number of groups to which white noise is added. Repeating the steps to continue adding white noise, and calculating R of the 2 nd decomposition es2 and D2 . The kth decomposition residual and the k-th order modal component continue to be calculated until decomposition cannot be continued.
Aiming at the strong fluctuation of the once decomposed high-frequency component IMF1, adopting a VMD method to decompose the IMF1 into the following components: defining an eigenmode function IMF of limited bandwidth k The constraint variation problem is constructed as shown in a formula (4), and the variation problem is solved to obtain modal components.
IMF k (t)＝A k (t)cos(Φ(t)) (3)
Wherein: k is the number of components after decomposition, A k (t) and φ (t) are amplitude and phase, ω k For the center frequency of each modal component,the gradient and convolution operation symbols are respectively, delta (t) is a dirac function, and X (t) is a sequence to be decomposed.
Solving the formula (4), and introducing a penalty factor alpha to convert the constraint variation problem into an unconstrained variation problem so as to reduce Gaussian noise interference. The expression is as follows:
wherein: alpha is a penalty factor and lambda is a Lagrangian operator.
The SSA algorithm is used for optimizing VMD decomposition parameters to solve the problem that the VMD cannot adaptively select the component number k and the punishment coefficient alpha, avoids the influence of subjective factors of manually set parameters, improves the decomposition quality, and specifically comprises the following steps:
a1: initializing the SSA population quantity, and setting the initial [ K, alpha ] and the maximum iteration number.
A2: updating the position of a finder in the sparrow population, wherein the finder is a member with the best fitness in the population and is responsible for foraging, and the position updating formula is as follows:
in the formula ,represents the location of the i-th finder in the d-dimension, iter max Represents the maximum iteration number, R 2 For the early warning value, the value range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the S represents a safety value, the value range is 0.5,1]The method comprises the steps of carrying out a first treatment on the surface of the Q is a random number obeying normal distribution; l is a1×d matrix with all elements being 1.
A3: the joiner observes the behavior of the discoverer at any time and adjusts his own position accordingly. The location update formula is:
in the formula ,xworst Representing the current worst-case location of the current,representing the current global optimal position; a is a matrix with elements of 1 or-1 and A + ＝A T (A T ) -1 。
A4: a certain number of sparrows are responsible for warning in each iteration population, and foraging is abandoned after danger is found and the sparrows go to a safe area to continue foraging. The location update formula is:
in the formula ,xbest Representing the current global optimal position; beta is a random number obeying normal distribution; f (f) i Is the current fitness; f (f) g For optimal fitness in the population, I is [0,1]Random values within the range, μ being a constant.
A5: repeating the iteration until the optimal [ K, alpha ] value is obtained.
Further, in the step S2, the method for reconstructing and dimension-reducing the decomposed data based on FE includes: the FE values of the components are analyzed, grouping is carried out according to the complexity and similarity characteristics of each component, the components of the same component are reconstructed into one component, and the specific steps are as follows:
b1: let the sub-sequence obtained by combining and decomposing the offshore wind speed data be x 1 ,x 2 ,…,x n Phase space reconstruction is carried out on the subsequence to obtain X i ；
Wherein N is x i Is a length of (2); m represents the embedding dimension, typically taken as 2;representing the mean of m adjacent sequences;
b2: defining fuzzy membership functionsCalculate each sequence X i Is a fuzzy entropy of (2);
in the formula ,represents the maximum value of the difference between the corresponding endpoints of the two sequences X (i, X (j)), r is the noise margin, and the standard deviation of 0.15 times of the processed data is taken;
b3: FE of a subsequence is a dimensionless number with a value of [0,1], and the larger the value is, the higher the complexity of the sequence is, and the larger the probability of generating a new mode is. Each sub-sequence is divided into 3 groups according to its fuzzy entropy value, each group of sub-sequences being reconstructed into one component.
The method for performing SSA-VMD decomposition modeling on the power in the step S3 comprises the following steps: and (2) decomposing the once decomposed high-frequency component IMF1 again by using the SSA-VMD in the step (S1), decomposing the historical power sequence by using the SSA-VMD, and establishing a predictor model for each power component. Meanwhile, the pearson correlation coefficient between the offshore wind power output and the historical wind speed and the historical power at the previous n moments is analyzed, and the time scale of the input of the model is determined.
Further, in the step S3, the method for establishing the LSTM predictor model for the power component includes:
constructing an LSTM network, wherein each LSTM layer consists of a plurality of LSTM units, each LSTM unit is provided with three gates, including a forgetting gate, an input gate and an output gate, and the forgetting gate determines the current time c t Preserving the last time cell state c t-1 The number of input gates determines the cell state X t Current time c of (2) t The output gate controls the current output value h output to the LSTM t Cell state c of (2) t The specific formula is expressed as follows:
f t ＝σ(W f ·(h t-1 ,x t )+b f ) (13)
i t ＝σ(W i ·(h t-1 ,x t )+b i ) (14)
o t ＝σ(W o ·(h t-1 ,x t )+b o ) (15)
c′ t ＝tanh(W c ·(h t-1 ,x t )+b c ) (16)
c t ＝f t ×c t-1 +i t ×c′ t (17)
h t ＝o t ×tanh(c t ) (18)
wherein ,ft 、i t 、o t Information of a forget gate, an input gate and an output gate respectively; w (W) f 、W i 、W o Is a weight matrix; x is x t and ht-1 The input at the moment t and the hidden layer state at the moment t-1 are respectively; b f 、b i 、 b o is the offset; sigma is a sigmoid function, c' t Inputting information of a memory module for the time t; w (W) c Is a weight matrix; tanh is a hyperbolic tangent function.
Further, in the step S4, the method for optimizing the super parameter of the LSTM network by using the bayesian algorithm includes:
by evaluating 2 groups of super parameters, a probability model updated sequentially is established to obtain the prior probability of the optimization problem, and then the optimal solution meeting the objective function f (z) is searchedThe objective function formula is as follows:
wherein The optimal value of the range Z is optimized for the super parameter.
According to the scheme, the whole flow of the method can be summarized as follows: firstly, the marine wind power data is combined and decomposed by using a signal decomposition method, so that the influence of randomness of an original signal and the problem of incomplete decomposition on prediction accuracy is improved. And then reconstructing the modal components by using fuzzy entropy analysis, and reducing the dimension of the input data of the prediction model. And finally, establishing a predictive sub-model for each power component, inputting wind speed reconstruction data and historical power data into a long-period memory neural network subjected to Bayesian optimization, and superposing the predictive results of the sub-models to obtain a final power predictive value.
The technical points involved in the present application are comprehensively described below:
combined decomposition of offshore wind speed data: aiming at the non-stationarity of offshore wind power data and the influence of decomposed high-frequency modal components on power prediction accuracy, ICEEMDAN and VMD are adopted to carry out combined decomposition on offshore wind speed data, the data is processed stably, and SSA algorithm is used to optimize VMD so as to solve the problem that VMD cannot select decomposition parameters in a self-adaptive mode.
FE-based data reconstruction: the number of components of the offshore wind power original data after twice decomposition is too large, so that the dimension of the input data is too large, and the direct processing is very complex. The application analyzes the complexity of each sub-component by the FE method, reconstructs sub-sequence groups with correlation FE characteristics, reduces the data dimension and improves the calculation speed.
SSA-VMD decomposition modeling of historical power: in the prediction modeling stage, the offshore wind power data has strong time sequence correlation, and the non-stationarity of the power data also seriously influences the prediction precision. Therefore, the SSA-VMD is used for decomposing the historical power sequence of the offshore wind power, a predictor model is built for each power component, and interference of fluctuation of the historical power sequence on prediction accuracy is reduced. According to the application, a wind speed reconstruction signal and a historical power signal on a certain time scale are selected as the input of a prediction model, so that the learning capacity of LSTM on the correlation characteristic and the time sequence characteristic between wind speed data and historical power data is improved.
BO-LSTM prediction model: the super-parameter setting of the neural network has important influence on the performance of the model, and as the super-parameter selection is very difficult, the BO algorithm with stronger generalization capability and faster convergence speed is selected to optimize the super-parameter of the LSTM network.
The beneficial effects are that: compared with the prior art, the offshore wind power prediction method based on wind speed-power combination decomposition reconstruction is provided, the influence of fluctuation of historical data on power prediction is effectively reduced, the problems of incomplete decomposition, modal aliasing and large data scale after repeated decomposition in the signal decomposition process are solved, and meanwhile, the performance of an LSTM network is enhanced, so that a neural network can better discover relevant characteristic information between an offshore wind speed sequence and a historical power sequence, and the offshore wind power prediction precision is effectively improved.
Drawings
FIG. 1 is a frame diagram of an offshore wind power prediction model;
FIG. 2 is a diagram of an LSTM network architecture;
FIG. 3 is an exploded view of the wind speed sequence ICEEMDAN;
FIG. 4 is a schematic representation of an iterative evolutionary display of the SSA algorithm;
FIG. 5 is a graph showing SSA-VMD results for a wind speed high frequency component IMF 1;
FIG. 6 is a graph showing SSA-VMD results for a historical power sequence;
FIG. 7 is a diagram showing the results of component FE analysis;
FIG. 8 is a diagram showing the results of component reconstruction;
FIG. 9 is a graph showing the four seasons of prediction RMSE errors for different models;
FIG. 10 is a diagram showing the four seasons prediction result of the decomposition model.
Detailed Description
The present application is further illustrated in the accompanying drawings and detailed description which are to be understood as being merely illustrative of the application and not limiting of its scope, and various modifications of the application, which are equivalent to those skilled in the art upon reading the application, will fall within the scope of the application as defined in the appended claims.
The application provides a wind speed-power combination decomposition reconstruction-based offshore wind power prediction method, which comprises the following steps with reference to FIG. 1:
s1: stabilizing wind speed data based on a combined decomposition method;
the original signal is decomposed to reduce the influence of volatility and randomness on the prediction accuracy, and the method is an effective method for improving the prediction accuracy of the new energy generated power. In the step, ICEEMDANs are adopted to decompose the wind speed sequence of the offshore wind power for the first time, specifically, the k-order modal component defining the wind speed sequence as v and EMD decomposition is E k (. Cndot.) adding noise to the wind velocity sequence, calculating the first decomposed residual R es1 And a first order eigenmode component D 1 。
D 1 ＝v-R es1 (1)
Where ε is the signal-to-noise ratio of white noise, w (i) For the added i-th group of gaussian white noise, M is the number of groups to which white noise is added. Repeating the steps to continue adding white noise, and calculating R of the 2 nd decomposition es2 and D2 . The kth decomposition residual and the k-th order modal component continue to be calculated until decomposition cannot be continued.
The VMD method is adopted to decompose the once decomposed high-frequency modal component IMF1 again, and the method specifically comprises the following steps: defining an eigenmode function IMF of limited bandwidth k The constraint variation problem is constructed as shown in a formula (4), and the variation problem is solved to obtain modal components.
IMF k (t)＝A k (t)cos(Φ(t)) (3)
Wherein: k is the number of components after decomposition, A k (t) and φ (t) are amplitude and phase, ω k For the center frequency of each modal component,* The gradient and convolution operation symbols are respectively, delta (t) is a dirac function, and X (t) is a sequence to be decomposed.
Solving the formula (4), and introducing a penalty factor alpha to convert the constraint variation problem into an unconstrained variation problem so as to reduce Gaussian noise interference. The expression is as follows:
wherein: alpha is a penalty factor and lambda is a Lagrangian operator.
The SSA algorithm is used for optimizing VMD decomposition parameters to solve the problem that the VMD cannot adaptively select the component number k and the punishment coefficient alpha, avoids the influence of subjective factors of manually set parameters, improves the decomposition quality, and specifically comprises the following steps:
(1) Initializing the SSA population quantity, and setting the initial [ K, alpha ] and the maximum iteration number.
(2) Updating the position of a finder in the sparrow population, wherein the finder is a member with the best fitness in the population and is responsible for foraging, and the position updating formula is as follows:
in the formula ,represents the location of the i-th finder in the d-dimension, iter max Represents the maximum iteration number, R 2 For the early warning value, the value range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the S represents a safety value, the value range is 0.5,1]The method comprises the steps of carrying out a first treatment on the surface of the Q is a random number obeying normal distribution; l is a1×d matrix with all elements being 1.
(3) The joiner observes the behavior of the discoverer at any time and adjusts his own position accordingly. The location update formula is:
in the formula ,xworst Representing the current worst-case location of the current,representing the current global optimal position; a is a matrix with elements of 1 or-1 and A + ＝A T (A T ) -1 。
(4) A certain number of sparrows are responsible for warning in each iteration population, and foraging is abandoned after danger is found and the sparrows go to a safe area to continue foraging. The location update formula is:
in the formula ,xbest Representing the current global optimal position; beta is a random number obeying normal distribution; f (f) i Is the current fitness; f (f) g For optimal fitness in the population, I is [0,1]Random values within the range, μ being a constant.
(5) Repeating the iteration until the optimal [ K, alpha ] value is obtained.
S2: and reconstructing and dimension-reducing the decomposed data based on FE according to the complexity and similarity characteristics of the components: the number of components of the offshore wind power original data after twice decomposition is excessive, so that the dimension of the input data is overlarge. The fuzzy entropy of each component is analyzed, grouping is carried out according to the complexity and similarity characteristics of each component, and the same component is reconstructed into one component, which comprises the following steps:
(1) Let the sub-sequence obtained by combining and decomposing the offshore wind speed data be x 1 ,x 2 ,…,x n Phase space reconstruction is carried out on the subsequence to obtain X i 。
Wherein N is x i Is a length of (2); m represents the embedding dimension, typically taken as 2;representing the mean of m adjacent sequences.
(2) Defining fuzzy membership functionsCalculate each sequence X i Is a fuzzy entropy of (c).
in the formula ,represents the maximum value of the difference value of the corresponding endpoints between the two sequences X (i) and X (j); r is the noise margin, taking a standard deviation of 0.15 times the processed data.
(3) FE of a subsequence is a dimensionless number with a value of [0,1], and the larger the value is, the higher the complexity of the sequence is, and the larger the probability of generating a new mode is. Each sub-sequence is divided into 3 groups according to its fuzzy entropy value, each group of sub-sequences being reconstructed into one component.
S3: and carrying out SSA-VMD decomposition on the power, establishing an LSTM prediction sub-model, and selecting wind speed reconstruction data and power data with a certain time scale as sub-model input. And (2) decomposing the once decomposed high-frequency component IMF1 again by using the SSA-VMD in the step (S1), decomposing the historical power sequence by using the SSA-VMD, and establishing a predictor model for each power component. Meanwhile, the pearson correlation coefficient between the offshore wind power output and the historical wind speed and the historical power at the previous n moments is analyzed, and the time scale of the input of the model is determined.
In this step, the formula for establishing the LSTM predictor model for each power component is:
f t ＝σ(W f ·(h t-1 ,x t )+b f ) (13)
i t ＝σ(W i ·(h t-1 ,x t )+b i ) (14)
o t ＝σ(W o ·(h t-1 ,x t )+b o ) (15)
c t '＝tanh(W c ·(h t-1 ,x t )+b c ) (16)
c t ＝f t ×c t-1 +i t ×c t ' (17)
h t ＝o t ×tanh(c t ) (18)
wherein ft 、i t 、o t Information of a forget gate, an input gate and an output gate respectively; w (W) f 、W i 、W o Is a weight matrix; x is x t and ht-1 The input at time t and the hidden layer state at time t-1 are respectively, b f 、b i 、b o For offset, σ is a sigmoid function, c' t Inputting information of the memory module for the time t, W c For the weight matrix, tanh is a hyperbolic tangent function.
S4: optimizing LSTM super parameters by using a Bayesian algorithm to form a BO-LSTM model, and superposing the predicted results of all the BO-LSTM sub-models to obtain a final predicted value: by evaluating 2 groups of super parameters, a probability model updated sequentially is established to obtain the prior probability of the optimization problem, and then the optimal solution meeting the objective function f (z) is searchedThe objective function formula is as follows:
wherein The optimal value of the range Z is optimized for the super parameter.
According to the offshore wind power prediction method based on the data decomposition reconstruction, simulation analysis is performed in the embodiment, and the method specifically comprises the following steps:
taking the actual measurement data of a marine wind farm with a installed capacity of 100MW in the United states as an example for 2018, the data sampling interval is 10min, and 144 recording points are recorded every day. Seasonal weather changes in offshore wind farms have a large impact on offshore wind power predictions, and in order to verify the model's practicability in different seasons, this example selects data of four typical months, 1 month, 4 months, 7 months and 10 months, to represent four seasons, winter, spring, summer and autumn, respectively. The front 30d or 29d in each month data is a training set, the back 1d is a verification set, and one-step, two-step and three-step prediction is respectively carried out, namely, the offshore wind power after 10min, 20min and 30min is respectively predicted. To eliminate the effect of the deep learning randomness, the average of 30 predictions is taken.
1. Combined decomposition of offshore wind speed data
Taking 1 month winter data as an example, as shown in fig. 2, after the original wind speed sequence is decomposed by the icemdan, a total of 12 IMF components and one residual component (only IMF1-IMF5 are shown) are obtained, and it can be seen that IMF1 after the decomposition by the icemdan still has strong volatility.
The process of optimizing the decomposition parameters of the VMD by the SSA algorithm is as shown in FIG. 3, and the optimal penalty coefficient is obtained as 205, and the number of components is 4. The results of the re-decomposition of IMF1 by SSA-VMD are shown in FIG. 4. After the combined decomposition, a total of 16 components were obtained.
2. FE-based wind speed component reconstruction
And (3) calculating FE values of 16 wind speed sequences obtained after secondary decomposition (the number of 4 iMF sub-components obtained by decomposing IMF1 is 13-16, and the result is shown in figure 5. According to the calculation result, reconstructing the sequence with the FE value larger than 0.6 into a component R1, wherein the component entropy is maximum, the fluctuation is most complex and the random information content is maximum, reconstructing the sequence with the FE value lower than 0.2 into R2, wherein the component entropy is minimum, the fluctuation is smaller, the trend of the basic wind speed change is represented, and the sequence with the FE value between 0.2 and 0.6 is reconstructed into R3, and the component complexity and the information content are between R1 and R2. After wind speed data is reconstructed by FE, the data scale is reduced from 16 time sequences to 3.
3. And carrying out SSA-VMD decomposition on the power, establishing an LSTM prediction sub-model, and selecting wind speed reconstruction data and power data with a certain time scale as sub-model input.
In the decomposition modeling of the historical power, the optimization result of the SSA algorithm is that the optimal penalty coefficient is 303, the number of components is 4, and the decomposition result is shown in fig. 6.
The time lag characteristic is provided between the wind speed and the power of the offshore wind power, and the offshore wind power output at the moment t is closely related to the historical wind speed and the historical power at the moment n. The time sequence Pearson correlation of the offshore wind power data is shown in table 1, and the historical data of the first 12 times with the power correlation of more than 0.85 (ρ is more than or equal to 0.85) at the t time is selected as the input data of the prediction model.
TABLE 1 time-series Pearson coefficient of offshore wind data
As shown in FIG. 7, the LSTM network structure used in this example includes 3 control gates per LSTM unit, and the forget gate determines the current time c t Preserving the last time cell state c t-1 The number of input gates determines the cell state X t Current time c of (2) t The output gate controls the current output value h output to the LSTM t Cell state c of (2) t Is a number of (3).
4. Optimizing LSTM super parameters by using a Bayesian algorithm to form a BO-LSTM model: the setting of the optimizing range and the optimizing result of the Bayesian algorithm parameters are shown in table 2, a Bayesian optimizing process is shown in fig. 8, the model is continuously updated through the observed points, the next point with the minimum error is continuously searched according to the updated model, and the LSTM super-parameter with the minimum prediction error is obtained after multiple iterations.
TABLE 2 parameter optimization results
In order to clearly and intuitively evaluate the performance of model prediction, the Root Mean Square Error (RMSE), mean Absolute Error (MAE) and Mean Average Percent Error (MAPE) are selected for evaluation in this embodiment, and the formula is shown below.
The application was compared with classical models, and the following control groups were also set up:
model 1: and (3) carrying out ICEEMDAN and SSA-VMD combined decomposition on the wind speed sequence, and inputting the decomposed components into an LSTM model for prediction after reconstructing.
Model 2: and (3) decomposing and modeling the wind power sequence by using an SSA-VMD method, and respectively predicting and finally superposing to obtain a result.
Model 3: and (3) carrying out ICEEMDAN decomposition and SSA-VMD combined decomposition on the wind speed sequence, reconstructing by applying an FE method, simultaneously carrying out SSA-VMD decomposition modeling on the historical power, and respectively predicting and superposing to obtain a result.
Model 4: based on model 3, bayesian optimization of the super parameters of LSTM, the proposed method of the present application, is used.
The result pairs of four-season power prediction and multi-step prediction of the method and other methods according to the application are shown in tables 3 and 4 and figures 9 and 10:
table 3 comparison of prediction error estimates for different models
Table 4 Multi-step prediction error index (1 month)
Therefore, the method has higher precision in offshore wind power prediction and multi-step prediction in four seasons, can effectively reduce the influence of historical data fluctuation on prediction precision, and has more obvious improvement effect as the data fluctuation is stronger.
Claims (10)
1. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction is characterized by comprising the following steps of:
s1: stabilizing wind speed data based on a combined decomposition method;
s2: reconstructing and dimension-reducing the data decomposed in the step S1 based on the fuzzy entropy FE according to the complexity and similarity characteristics of the components;
s3: according to the coupling correlation between the offshore wind speed and the power, decomposing the historical power, establishing a long-term memory network LSTM predictor model, and selecting wind speed reconstruction data and power data with a certain time scale as inputs of the predictor model;
s4: according to the network structure characteristics of deep learning, a Bayesian algorithm BO is used for optimizing LSTM super parameters to construct a BO-LSTM model, and the prediction results of all the BO-LSTM sub-models are overlapped to obtain a final prediction value.
2. The method for predicting the wind power at sea based on wind speed-power combination decomposition reconstruction of claim 1, wherein the method for combining and decomposing wind speed data in step S1 is as follows: and carrying out combined decomposition on the offshore wind speed data by adopting a fully-assembled empirical mode decomposition ICEEMDAN and a variable mode decomposition VMD for improving the self-adaptive noise, and solving the VMD optimal decomposition parameters by using a sparrow search algorithm SSA.
3. The method for predicting the power of the offshore wind power based on the wind speed-power combination decomposition reconstruction according to claim 2, wherein the method for decomposing the icemdan by adopting the complete set of empirical modes for improving the adaptive noise in the step S1 is as follows:
defining a wind speed sequence v, and carrying out ICEEMDAN decomposition on the wind speed sequence, wherein an eigenmode function D of the ICEEMDAN decomposition is carried out 1 And residual component R es1 The calculation formula is as follows:
D 1 ＝v-R es1 (1)
wherein epsilon is the signal-to-noise ratio of white noise; w (w) (i) White gaussian noise for the i th group added; m is the number of groups to which white noise is added.
4. The method for predicting the offshore wind power based on wind speed-power combination decomposition reconstruction according to claim 2, wherein the method for decomposing the VMD in the variable mode in the step S1 is as follows:
performing VMD secondary decomposition on the decomposed high-frequency component, constructing a variation problem according to the following formula, and solving the variation problem to obtain an intrinsic mode function IMF k
IMF k (t)＝A k (t)cos(Φ(t)) (3)
Wherein: k is the number of components after decomposition, A k (t) and φ (t) are amplitude and phase, ω k For the center frequency of each modal component,* The method comprises the steps of respectively obtaining gradient and convolution operation symbols, wherein delta (t) is a dirac function, and X (t) is a sequence to be decomposed;
solving a formula (4), and introducing a penalty factor alpha to convert a constraint variation problem into a non-constraint variation problem so as to reduce Gaussian noise interference, wherein the formula is as follows:
wherein: alpha is a penalty factor and lambda is a Lagrangian operator.
5. The method for predicting the power of the offshore wind power based on the wind speed-power combination decomposition reconstruction of claim 4, wherein in the step S1, in the secondary decomposition of the wind speed data, the VMD decomposition parameters are optimized by using SSA algorithm to solve the problem that the VMD cannot adaptively select the component number k and the penalty coefficient α, specifically comprising the following steps:
a1: initializing SSA population quantity, and setting initial [ K, alpha ] and maximum iteration times;
a2: updating the position of a finder in the sparrow population, wherein the finder is a member with the best fitness in the population and is responsible for foraging, and the position updating formula is as follows:
in the formula ,represents the location of the i-th finder in the d-dimension, iter max Represents the maximum iteration number, R 2 For the early warning value, the value range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the S represents a safety value; q is a random number obeying normal distribution; l is a1 x d matrix with all elements being 1;
a3: the joiner observes the behavior of the discoverer at any time and adjusts the position of the joiner, and the position updating formula is as follows:
in the formula ,xworst Representing the current worst-case location of the current,representing the current global optimal position; a is a matrix with elements of 1 or-1 and A + ＝A T (A T ) -1 ；
A4: sparrows are responsible for warning in each iteration population, foraging is abandoned after danger is found, the sparrows go to a safe area to continue foraging, and a position updating formula is as follows:
in the formula ,xbest Representing the current global optimal position; beta is a random number obeying normal distribution; f (f) i Is the current fitness; f (f) g For optimal fitness in the population, I is [0,1]Random values in the range, μ being a constant;
a5: repeating the iteration until the optimal [ K, alpha ] value is obtained.
6. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction of claim 1, wherein the method comprises the following steps: the method for reconstructing and dimension-reducing the decomposed data based on FE in the step S2 comprises the following steps:
b1: let the sub-sequence obtained by combining and decomposing the offshore wind speed data be x 1 ,x 2 ,…,x n Phase space reconstruction is carried out on the subsequence to obtain X i ；
Wherein N is x i Is a length of (2); m represents an embedding dimension;representing the mean of m adjacent sequences;
b2: defining fuzzy membership functionsCalculate each sequence X i Is a fuzzy entropy of (2);
in the formula ,represents the maximum value of the difference value of the corresponding endpoints between the two sequences X (i) and X (j); r is the noise margin;
b3: FE of the subsequences is a dimensionless number with a value of [0,1], the subsequences are divided into 3 groups according to fuzzy entropy values of the subsequences, and each group of subsequences is reconstructed into a component.
7. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction of claim 1, wherein the method comprises the following steps: the method for decomposing and modeling the historical power sequence in the step S3 comprises the following steps: performing SSA-VMD decomposition on the historical power sequence, establishing an LSTM predictor model for each power component after the decomposition, simultaneously analyzing the pearson correlation coefficient between the offshore wind power output and the historical wind speed and the historical power at the previous n moments, and determining the input time scale of the model.
8. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction of claim 7, wherein the method comprises the following steps: the method for establishing the LSTM predictor model for the power component in the step S3 comprises the following steps:
constructing an LSTM network, wherein each LSTM layer consists of a plurality of LSTM units, each LSTM unit is provided with three gates, including a forgetting gate, an input gate and an output gate, and the forgetting gate determines the current time c t Preserving the last time cell state c t-1 The number of input gates determines the cell state X t Current time c of (2) t The output gate controls the current output value h output to the LSTM t Cell state c of (2) t The specific expression is as follows:
f t ＝σ(W f ·(h t-1 ,x t )+b f ) (13)
i t ＝σ(W i ·(h t-1 ,x t )+b i ) (14)
o t ＝σ(W o ·(h t-1 ,x t )+b o ) (15)
c′ t ＝tanh(W c ·(h t-1 ,x t )+b c ) (16)
c t ＝f t ×c t-1 +i t ×c′ t (17)
h t ＝o t ×tanh(c t ) (18)
wherein ,ft 、i t 、o t Information of a forget gate, an input gate and an output gate respectively; w (W) f 、W i 、W o Is a weight matrix; x is x t and ht-1 The input at the moment t and the hidden layer state at the moment t-1 are respectively; b f 、b i 、b o Is the offset; sigma is a sigmoid function, c' t Inputting information of a memory module for the time t; w (W) c Is a weight matrix; tanh is a hyperbolic tangent function.
9. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction of claim 1, wherein the method comprises the following steps: the method for building the BO-LSTM model by adopting the Bayesian algorithm in the step S4 comprises the following steps: by evaluating 2 groups of super parameters, a probability model updated sequentially is established to obtain the prior probability of the optimization problem, and then the optimal solution meeting the objective function f (z) is searched
10. The offshore wind power prediction method based on wind speed-power combination decomposition reconstruction of claim 9, wherein the method comprises the following steps: the optimal solutionThe objective function formula of (2) is as follows:
wherein ,the optimal value of the range Z is optimized for the super parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310949782.2A CN116960978A (en) | 2023-07-31 | 2023-07-31 | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310949782.2A CN116960978A (en) | 2023-07-31 | 2023-07-31 | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116960978A true CN116960978A (en) | 2023-10-27 |
Family
ID=88461705
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310949782.2A Pending CN116960978A (en) | 2023-07-31 | 2023-07-31 | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116960978A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117252448A (en) * | 2023-11-20 | 2023-12-19 | 华东交通大学 | Wind power prediction method based on space-time feature extraction and secondary decomposition polymerization |
| CN117892113A (en) * | 2024-03-13 | 2024-04-16 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
-
2023
- 2023-07-31 CN CN202310949782.2A patent/CN116960978A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117252448A (en) * | 2023-11-20 | 2023-12-19 | 华东交通大学 | Wind power prediction method based on space-time feature extraction and secondary decomposition polymerization |
| CN117252448B (en) * | 2023-11-20 | 2024-02-20 | 华东交通大学 | Wind power prediction method based on space-time feature extraction and secondary decomposition polymerization |
| CN117892113A (en) * | 2024-03-13 | 2024-04-16 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
| CN117892113B (en) * | 2024-03-13 | 2024-06-11 | 广东工业大学 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shamshirband et al. | A survey of deep learning techniques: application in wind and solar energy resources | |
| Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
| CN110309603B (en) | Short-term wind speed prediction method and system based on wind speed characteristics | |
| CN116960978A (en) | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction | |
| US20210334658A1 (en) | Method for performing clustering on power system operation modes based on sparse autoencoder | |
| Yin et al. | A novel asexual-reproduction evolutionary neural network for wind power prediction based on generative adversarial networks | |
| CN110264012B (en) | Renewable energy power combination prediction method and system based on empirical mode decomposition | |
| CN109583588B (en) | Short-term wind speed prediction method and system | |
| CN112434891A (en) | Method for predicting solar irradiance time sequence based on WCNN-ALSTM | |
| CN116316591A (en) | Short-term photovoltaic power prediction method and system based on hybrid bidirectional gating cycle | |
| CN116562908A (en) | Electric price prediction method based on double-layer VMD decomposition and SSA-LSTM | |
| CN116644970A (en) | Photovoltaic power prediction method based on VMD decomposition and lamination deep learning | |
| CN112949610A (en) | Improved Elman neural network prediction method based on noise reduction algorithm | |
| CN114036850A (en) | Runoff prediction method based on VECGM | |
| CN111222689A (en) | LSTM load prediction method, medium, and electronic device based on multi-scale temporal features | |
| CN114692947A (en) | VMD-Lasso-CNN-LSTM-based user electricity consumption prediction method | |
| CN115511657A (en) | Wind power output and photovoltaic output evaluation method based on combined prediction model | |
| CN110782013A (en) | Thermal power plant NO based on generalized mutual entropy self-encoder xEmission amount prediction method | |
| CN113836823A (en) | Load combination prediction method based on load decomposition and optimized bidirectional long-short term memory network | |
| CN117520809A (en) | Transformer fault diagnosis method based on EEMD-KPCA-CNN-BiLSTM | |
| CN115936062A (en) | Photovoltaic power day-ahead interval prediction method based on scene generation and deep learning | |
| CN116681154A (en) | Photovoltaic power calculation method based on EMD-AO-DELM | |
| CN114386666A (en) | Wind power plant short-term wind speed prediction method based on space-time correlation | |
| CN115498627A (en) | Combined load prediction method integrating sequence decomposition and feature selection | |
| CN115796327A (en) | Wind power interval prediction method based on VMD (vertical vector decomposition) and IWOA-F-GRU (empirical mode decomposition) -based models |
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 |