CN117892113A - Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism - Google Patents
Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism Download PDFInfo
- Publication number
- CN117892113A CN117892113A CN202410283502.3A CN202410283502A CN117892113A CN 117892113 A CN117892113 A CN 117892113A CN 202410283502 A CN202410283502 A CN 202410283502A CN 117892113 A CN117892113 A CN 117892113A
- Authority
- CN
- China
- Prior art keywords
- dimension
- power
- wind power
- matrix
- reduction
- 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.)
- Granted
Links
- 230000007246 mechanism Effects 0.000 title claims abstract description 42
- 230000009977 dual effect Effects 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 37
- 239000011159 matrix material Substances 0.000 claims abstract description 93
- 230000009467 reduction Effects 0.000 claims abstract description 63
- 238000005457 optimization Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims description 22
- 230000008859 change Effects 0.000 claims description 20
- 239000002245 particle Substances 0.000 claims description 18
- 230000003044 adaptive effect Effects 0.000 claims description 11
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000002238 attenuated effect Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012935 Averaging Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000005065 mining Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 abstract description 5
- 238000003062 neural network model Methods 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
-
- 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
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Computing Systems (AREA)
- Human Resources & Organizations (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Power Engineering (AREA)
- Tourism & Hospitality (AREA)
- Water Supply & Treatment (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of wind power prediction, in particular to a wind power prediction method of a self-adaptive VMD and a dual dimension-reduction attention mechanism, wherein the dual dimension-reduction attention mechanism can effectively reduce dimension of high-dimension wind power data, can realize dimension reduction and feature extraction of the high-dimension wind power data, reduces data redundancy and is beneficial to improving the prediction precision of wind power; the self-adaptive VMD algorithm can decompose the power signal into components which are easier to train, so that the training efficiency is improved; and the super-parameter matrix of the network model is optimized by adopting a crisscross optimization algorithm, so that the problem of local optimum possibly existing in the network model built by the neural network can be solved. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism not only has higher training efficiency, but also can reduce dimension of high-dimension wind power data and solve the problem of local optimization of a neural network model, thereby improving the prediction precision of wind power.
Description
Technical Field
The invention relates to the technical field of wind power prediction, in particular to a wind power prediction method of a self-adaptive VMD and a dual dimension-reduction attention mechanism.
Background
Wind energy is used as a new energy source, and with the continuous development of wind power generation technology, the grid connection of large-capacity wind power brings challenges to the safe and stable operation of a large power grid. Therefore, the accurate prediction of the wind power generation output power can schedule the wind power grid-connected capacity in advance, reduce the fluctuation of the wind power grid-connected power, and has important significance for safe and stable operation of a power system. The wind power sequence is a time sequence with strong randomness and volatility, the wind power is mainly influenced by meteorological factors such as wind speed, wind direction and air pressure, and the wind power prediction is usually performed by accurately predicting the wind power by using a deep learning network so as to obtain the coupling relation between the power and different wind power characteristics. Wind power characteristics of different dimensions have different coupling relations with wind power output power, and the prior art only aims at relations between low-dimensional meteorological characteristics recorded by a wind power plant and power. However, with the development of wind power plants, the meteorological feature dimension recorded in the wind power plants is continuously increased, and the higher feature dimension can cause feature redundancy.
Disclosure of Invention
The invention aims to overcome the defects that fitting and prediction precision are poor between high-dimensional wind power characteristics and power cannot be performed in the prior art, and provides a wind power prediction method with a self-adaptive VMD and a dual dimension-reduction attention mechanism, which is used for fitting between the high-dimensional wind power characteristics and the power and can improve the prediction precision.
In order to solve the technical problems, the invention adopts the following technical scheme:
the wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism comprises the following steps:
S10, acquiring a power sequence of a target wind power plant, a wind speed sequence, a temperature sequence and a barometric pressure sequence with different heights, and preprocessing;
S20, forming a characteristic matrix X by the preprocessed power sequence, the wind speed sequences, the temperature sequences and the air pressure sequences with different heights;
S30, constructing a dual dimension-reducing attention mechanism and a gating circulation unit network model comprising an active dimension-reducing layer and a passive dimension-reducing layer;
S40, inputting the feature matrix X in the step S20 into the active dimension reduction layer to perform feature screening dimension reduction, and mining out features with the largest influence on power by using an average influence value MIV dimension reduction method;
S50, decomposing the power sequence by using a crisscross optimization algorithm and taking the weighted envelope change rate as an adaptability function of the variation modal decomposition VMD, and adding the decomposed power sequence into a feature matrix after dimension reduction;
S60, conveying the feature matrix containing the decomposed power sequences after dimension reduction to the passive dimension reduction layer in the step S30, fully excavating the relations between different features and power again, and giving weights to obtain a feature matrix/> given the weights;
S70, conveying the feature matrix given with the weight to a gating circulation unit network, and fully excavating an implicit relation existing in the feature matrix/> given with the weight by the gating circulation unit network;
S80, optimizing a super-parameter matrix theta in a gate control circulation unit network by adopting a longitudinal and transverse cross optimization algorithm to complete the training of the prediction model;
S90, predicting a power time sequence of the future moment of the target wind power plant by using the trained prediction model.
According to the wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism, the dimension reduction of the high-dimension wind power data can be effectively carried out by the dual dimension-reduction attention mechanism, the dimension reduction and feature extraction of the high-dimension wind power data can be realized, the data redundancy is reduced, and the prediction precision of the wind power is improved; the self-adaptive VMD algorithm can decompose the power signal into components which are easier to train, so that the training efficiency is improved; and the super-parameter matrix of the network model is optimized by adopting a crisscross optimization algorithm, so that the problem of local optimum possibly existing in the network model built by the neural network can be solved. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism not only has higher training efficiency, but also can reduce dimension of high-dimension wind power data and solve the problem of local optimization of a neural network model, thereby improving the prediction precision of wind power.
Preferably, in step S10, the power sequence and the wind speed sequence, the temperature sequence and the air pressure sequence with different heights are normalized by min-max, so as to obtain a processed power sequence P, a processed wind speed sequence WS, a processed temperature sequence Tem and a processed air pressure sequence Pre.
Preferably, in step S20, x= [ D 1,D2,...,Dm ], where D m represents a matrix of features at the m-th height t-1 to t-n, and the feature matrix X is expressed as:
where P t-n is the power at times t-1 to t-n, 、/> and/> are the wind speed, temperature and air pressure at times t-1 to t-n, respectively, for the mth altitude.
Preferably, in step S30, the dual dimension reduction attention mechanism and gating loop unit network model further includes an input layer and an output layer, where the number of neurons of the input layer and the output layer is equal to the number of features.
Preferably, step S40 comprises the steps of:
S41, increasing the characteristic amplitude of each dimension by A% to obtain X up, and obtaining m characteristic enhancement matrixes F up:
Wherein denotes the enhanced ith column feature,/> denotes the ith column feature,/> denotes the W column feature; a% is 10% -50%;
S42, respectively reducing the characteristic amplitude of each dimension by A% to obtain X reduce, and obtaining m characteristic attenuation matrixes F reduce:
S43, respectively sending the F up and the F reduce into a gating circulation unit for training and prediction, wherein the length of a predicted sample is N, respectively obtaining a predicted result Y up with enhanced characteristics and a predicted result Y reduce with attenuated characteristics, and subtracting the predicted result Y up with enhanced characteristics from the predicted result Y reduce with attenuated characteristics to obtain an influence value of the characteristics i on power prediction output:
Wherein is a sequence of length N;
Averaging yields the average impact of the ith feature on the predicted power output/> :
And S44, reserving the characteristics with the influence value IV higher than the average influence value , and eliminating the characteristics with the influence value IV lower than the average influence value/> to obtain a feature matrix/> after dimension reduction.
Preferably, in step S50, the process of calculating the envelope change rate includes the steps of:
S51, carrying out Hilbert transformation on a continuous sample sequence power sequence x (t) to obtain :
S52, defining envelope of x (t) as , wherein the expression is shown in the following formula:
S53, calculating the envelope change rate ECR as follows:
where T is the sample length of x (T), phi is the rate of change interval, and the mean () function acts to calculate the average;
S54, regarding N sub-components generated by the decomposition of the variation mode of x (t), a weighted summation mode is adopted to obtain WECR:
Where denotes the envelope/> ,/> denotes the envelope of x (t)/> denotes the envelope rate of change of/> and WECR denotes the sum of the weighted envelope rates of change of all the following power sub-sequences;
And S55, optimizing the decomposition parameters by using a crisscross optimization algorithm, and finally adding the decomposed power sequences into the feature matrix after dimension reduction.
Preferably, step S60 comprises the steps of:
S61, pearson correlation coefficient/> between the j-th dimension feature and power in the feature matrix after dimension reduction of the decomposed power sequence is as follows:
Wherein represents the ith value of the mean value of the jth dimension characteristic after the active dimension reduction layer is reconstructed,/> represents the ith value of the power,/> and/> represent the mean value of the jth dimension characteristic and the mean value of the power after the dimension reduction layer is reconstructed respectively;
s62, combining the characteristics after dimension reduction into a power correlation matrix R:
S63, solving an average of the power correlation matrix R:
Wherein represents the feature dimension after dimension reduction;
S64, transforming R by taking as an evaluation index to obtain a transformation matrix num_0_w:
Wherein, part is called a non-key feature, part/() is called a key feature, w is a non-key feature weight, and the value is (0, 1);
S65, performing matrix point multiplication on the initial weight of the attention layer and the input matrix by using num_0_w as the initial weight of the attention layer to obtain a feature matrix after the weight is given.
Preferably, in step S70, a two-layer gated loop unit network is built with the weighted feature matrix as input, and the activation function is tanh:
Wherein 、/>、/>、/>、/>、/> is a weight parameter matrix,/> 、/>、/> is a bias parameter matrix,/> is a matrix multiplication,/> is a Sigmod function,/> is a reset gate,/> is an update gate,/> is a candidate state of an hidden layer at the current time,/> is a current hidden state,/> is a hidden state at the previous time, and/> is an input state at the current time.
Preferably, in step S80, the hyper-parameter matrix θ is formed by the weight parameters and bias parameters of the output layer of the gated loop unit network, and is represented by the following formula:
。
Preferably, in step S80, the specific step of performing optimization of the hyper-parameter matrix θ by the crossbar optimization algorithm includes:
s81, taking the improvement of prediction precision as a target, adopting a root mean square error minimum as an fitness function in a crisscross optimization algorithm, wherein the following formula is adopted:
Wherein is a fitness function, wherein/> is the number of samples, and/> and/> represent a true value and a predicted value, respectively;
S82, randomly selecting nth-dimension interdigitation by parent particles and/> in the parameters;
Wherein is the nth dimension child generated by transverse intersection of/> and/> , r1, r2 and c1, c2 are random numbers of (0, 1) and (-1, 1), respectively,/> is the nth dimension of particle/> , and/> is the nth dimension of particle/> ;
s83, randomly selecting a v-th dimension and a k-th dimension to be intersected by each other by using a parent particle theta (q) in the parameters;
wherein is a v-th dimension child generated by longitudinally crossing/> and/> , r is a random number of (0, 1), v-th dimension of/> is particle/> , and k-th dimension of/> is particle/> ;
s84, repeating the steps S62 and S63 according to the set iteration times, and stopping iteration when the preset iteration times are reached.
Compared with the prior art, the invention has the beneficial effects that:
according to the wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism, the dual dimension-reduction attention mechanism can effectively reduce dimension of high-dimension wind power data, and is beneficial to improving the prediction precision of wind power; the self-adaptive VMD algorithm can decompose the power signal into components which are easier to train, so that the training efficiency is improved; and the super-parameter matrix of the network model is optimized by adopting a crisscross optimization algorithm, so that the problem of local optimum possibly existing in the network model built by the neural network can be solved. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism not only has higher training efficiency, but also can reduce dimension of high-dimension wind power data and solve the problem of local optimization of a neural network model, thereby improving the prediction precision of wind power.
Drawings
FIG. 1 is a schematic diagram of a wind power prediction method for an adaptive VMD and a dual dimension-reduction attention mechanism;
fig. 2 is a diagram showing the effect of prediction including a predicted value and a real value in the second embodiment.
Detailed Description
The invention is further described below in connection with the following detailed description.
Example 1
The embodiment is a first embodiment of a wind power prediction method of a self-adaptive VMD and a dual dimension reduction attention mechanism, firstly, meteorological features with different heights are required to be obtained, then, a feature matrix is formed after preliminary processing, then, dimension reduction is carried out on the features by adopting an active dimension reduction algorithm to obtain a feature matrix after dimension reduction, then, the feature matrix after dimension reduction is decomposed by adopting the self-adaptive VMD algorithm by taking a weighted envelope change rate as an index, the decomposed feature matrix is input into a prediction model, the dimension reduction is carried out on high-dimension features again by adopting a passive dimension reduction attention mechanism in the prediction model, meanwhile, hidden relations existing in the feature matrix are mined, then, training of the prediction model is completed by adopting a criss-cross optimization algorithm to optimize a super-parameter matrix in a gating circulation unit network, and finally, wind power is predicted by using the prediction model. Specifically, the method comprises the following steps:
S10, acquiring a power sequence of a target wind power plant, a wind speed sequence, a temperature sequence and a barometric pressure sequence with different heights, and preprocessing;
S20, forming a characteristic matrix X by the preprocessed power sequence, the wind speed sequences, the temperature sequences and the air pressure sequences with different heights;
S30, constructing a dual dimension-reducing attention mechanism and a gating circulation unit network model comprising an active dimension-reducing layer and a passive dimension-reducing layer;
S40, inputting the feature matrix X in the step S20 into the active dimension reduction layer to perform feature screening dimension reduction, and mining out features with the largest influence on power by using an average influence value MIV dimension reduction method;
S50, decomposing the power sequence by using a crisscross optimization algorithm and taking the weighted envelope change rate as an adaptability function of the variation modal decomposition VMD, and adding the decomposed power sequence into a feature matrix after dimension reduction;
S60, conveying the feature matrix containing the decomposed power sequences after dimension reduction to the passive dimension reduction layer in the step S30, fully excavating the relations between different features and power again, and giving weights to obtain a feature matrix/> given the weights;
S70, conveying the feature matrix given with the weight to a gating circulation unit network, and fully excavating an implicit relation existing in the feature matrix/> given with the weight by the gating circulation unit network;
S80, optimizing a super-parameter matrix theta in a gate control circulation unit network by adopting a longitudinal and transverse cross optimization algorithm to complete the training of the prediction model;
S90, predicting a power time sequence of the future moment of the target wind power plant by using the trained prediction model.
In step S10, the power sequence, the wind speed sequence, the temperature sequence and the air pressure sequence with different heights are subjected to min-max normalization processing to obtain a processed power sequence P, a processed wind speed sequence WS, a processed temperature sequence Tem and a processed air pressure sequence Pre. Of course, the invention can also use other common data preprocessing modes to preprocess the power sequence, the wind speed sequence, the temperature sequence and the air pressure sequence with different heights.
In step S20, x= [ D 1,D2,...,Dm ], where D m represents a matrix of features at the m-th height t-1 to t-n, and the feature matrix X is expressed as:
Where P t-n is the power at times t-1 to t-n, 、/> and/> are the wind speed, temperature and air pressure at times t-1 to t-n, respectively, for the mth altitude.
In step S30, the dual dimension-reduction attention mechanism and gating cycle unit network model further includes an input layer and an output layer, where the number of neurons in the input layer and the number of neurons in the output layer are equal to the number of features.
Step S40 includes the steps of:
S41, increasing the characteristic amplitude of each dimension by A% to obtain X up, and obtaining m characteristic enhancement matrixes F up:
Wherein denotes the enhanced ith column feature,/> denotes the ith column feature,/> denotes the W column feature; a% is 10% -50%, preferably 10%;
S42, respectively reducing the characteristic amplitude of each dimension by A% to obtain X reduce, and obtaining m characteristic attenuation matrixes F reduce:
wherein denotes the attenuated ith column feature;
S43, respectively sending the F up and the F reduce into a gating circulation unit for training and prediction, wherein the length of a predicted sample is N, respectively obtaining a predicted result Y up with enhanced characteristics and a predicted result Y reduce with attenuated characteristics, and subtracting the predicted result Y up with enhanced characteristics from the predicted result Y reduce with attenuated characteristics to obtain an influence value of the characteristics i on power prediction output:
wherein is a sequence of length N;
averaging yields the average impact of the ith feature on the predicted power output/> :
S44, averaging the obtained W-1 MIVs to obtain parameters of different characteristics on the importance of the predicted power output:
And taking as a characteristic active dimension reduction index, eliminating the corresponding characteristic component X i of which MIV (i) is less than/> , so that the original wind power data is reconstructed into new wind power data after dimension reduction, and the characteristic quantity of a characteristic matrix/> after dimension reduction is/> .
In step S50, the process of calculating the envelope change rate includes the steps of:
s51, carrying out Hilbert transformation on a continuous sample sequence power sequence x (t) to obtain :
s52, defining envelope of x (t) as , wherein the expression is shown in the following formula:
S53, calculating the envelope change rate ECR as follows:
Wherein, T is the sample length of x (T), phi is the change rate interval, the mean () function is to calculate the average value, generally taking phi=1, namely calculating the change rate of two adjacent sample values, when the sample length is very long, the value of phi can be properly increased to reduce the calculated amount and the memory occupation;
S54, regarding N sub-components generated by the decomposition of the variation mode of x (t), a weighted summation mode is adopted to obtain WECR:
Where denotes the envelope/> ,/> denotes the envelope of x (t)/> denotes the envelope rate of change of/> and WECR denotes the sum of the weighted envelope rates of change of all the following power sub-sequences;
And S55, optimizing the decomposition parameters by using a crisscross optimization algorithm, and finally adding the decomposed power sequences into the feature matrix after dimension reduction.
Step S60 includes the steps of:
S61, pearson correlation coefficient/> between the j-th dimension feature and power in the feature matrix after dimension reduction of the decomposed power sequence is as follows:
Wherein represents the ith value of the mean value of the jth dimension characteristic after the active dimension reduction layer is reconstructed,/> represents the ith value of the power,/> and/> represent the mean value of the jth dimension characteristic and the mean value of the power after the dimension reduction layer is reconstructed respectively;
s62, combining the characteristics after dimension reduction into a power correlation matrix R:
S63, solving an average of the power correlation matrix R:
Wherein represents the feature dimension after dimension reduction;
S64, transforming R by taking as an evaluation index to obtain a transformation matrix num_0_w:
Wherein the part is called a non-key feature, the part/() is called a key feature, w is a non-key feature weight, the value is (0, 1), the larger the value of w is, the more non-key feature information is reserved, and when the feature dimension is high, the purpose of passive dimension reduction is achieved by generally taking w=0.001, and meanwhile, the inactivation of neurons caused by w=0 is avoided;
S65, performing matrix point multiplication on the initial weight of the attention layer and the input matrix by using num_0_w as the initial weight of the attention layer to obtain a feature matrix after the weight is given.
In step S70, a gating unit network is built by taking the feature matrix with the given weight as input, and the gating unit network continuously learns, trains and iterates the weights of different features in the feature matrix/> in a gradient descending manner through a plurality of gating neurons in the gating unit network, so as to finally obtain the weight relation of different features in the feature matrix/> , and realize the best fitting of power. The activation function is the last output layer in the gated loop cell network, and the activation function is generally changed to be nonlinear for output. In this embodiment, two layers of gating circulation unit networks are built, the number of neurons is 4 and 8 respectively, and the activation function is tanh:
Wherein 、/>、/>、/>、/>、/> is a weight parameter matrix,/> 、/>、/> is a bias parameter matrix,/> is a matrix multiplication,/> is a Sigmod function,/> is a reset gate,/> is an update gate,/> is a candidate state of an hidden layer at the current time,/> is a current hidden state,/> is a hidden state at the previous time, and/> is an input state at the current time.
In step S80, iterative optimization is performed again on the neuron weights in the gated loop unit network, so as to obtain a final determined model. Specifically, the hyper-parameter matrix θ is composed of weight parameters and bias parameters of the output layer of the gated loop unit network, and the following formula is shown:
。
in step S80, the specific steps of optimizing the hyper-parameter matrix θ by the crisscross optimization algorithm include:
s81, taking the improvement of prediction precision as a target, adopting a root mean square error minimum as an fitness function in a crisscross optimization algorithm, wherein the following formula is adopted:
Wherein is a fitness function, wherein/> is the number of samples, and/> and/> represent a true value and a predicted value, respectively;
S82, randomly selecting nth-dimension interdigitation by parent particles and/> in the parameters;
wherein is the nth dimension child generated by transverse intersection of/> and/> , r1, r2 and c1, c2 are random numbers of (0, 1) and (-1, 1), respectively,/> is the nth dimension of particle/> , and/> is the nth dimension of particle/> ;
s83, randomly selecting a v-th dimension and a k-th dimension to be intersected by each other by using a parent particle theta (q) in the parameters;
Wherein is a v-th dimension child generated by longitudinally crossing/> and/> , r is a random number of (0, 1), v-th dimension of/> is particle/> , and k-th dimension of/> is particle/> ;
s84, repeating the steps S62 and S63 according to the set iteration times, and stopping iteration when the preset iteration times are reached.
Through the steps, the dual dimension reduction attention mechanism can effectively reduce dimension of the high-dimension wind power data, can realize dimension reduction and feature extraction of the high-dimension wind power data, reduces data redundancy, and is beneficial to improving the prediction precision of wind power; the self-adaptive VMD algorithm can decompose the power signal into components which are easier to train, so that the training efficiency is improved; and the super-parameter matrix of the network model is optimized by adopting a crisscross optimization algorithm, so that the problem of local optimum possibly existing in the network model built by the neural network can be solved. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism not only has higher training efficiency, but also can reduce dimension of high-dimension wind power data and solve the problem of local optimization of a neural network model, thereby improving the prediction precision of wind power.
Example two
The embodiment is an application embodiment of a wind power prediction method of an adaptive VMD and a dual dimension-reduction attention mechanism in the first embodiment, in step S1, obtaining measured data of a wind farm, wherein the total data is 54 dimensions, and the method comprises the following steps: wind power output power, wind speed at 0m, wind direction, temperature, humidity and air pressure, wind speed at 30m, 50m, 70m, 100m, 120m, 150m, 170m, 200m, 220m, 250m, 300m, 350m, and characteristics are 54-dimensional in total;
In step S4, the activation function of the spatial attention mechanism is a linear rectification function ReLU, and the number of neurons in the input layer and the output layer is 4;
In the step S6, the population number of the crisscross cross optimization algorithm is set to be 25, the transverse cross rate is 1, the longitudinal cross rate is 0.6, and the iteration times are 200 times;
The above data were substituted into the short-term wind power prediction method of example 1 to obtain the wind power prediction effect shown in fig. 2. From the prediction effect in fig. 2, it can be seen that the prediction sequence obtained by the short-term wind power prediction method is basically consistent with the actual measurement sequence, so that the accuracy of short-term wind power prediction can be effectively improved.
Example III
The embodiment is an embodiment of a prediction system for implementing the wind power prediction method of the adaptive VMD and the dual dimension reduction attention mechanism described in any one of the first embodiment or the second embodiment, and includes an acquisition module and a processing module that are communicatively connected;
The acquisition module is used for acquiring power, wind speed, wind direction and temperature data of a target wind power plant and adjacent wind power plants, performing preliminary processing, and outputting the processed wind power plant power, wind speed, wind direction and temperature data to the processing module to form a feature matrix X;
The processing module is constructed with a prediction model comprising a spatial attention mechanism and a gating circulation unit network;
after the feature matrix X is input into the prediction model, spatial relationships between different wind power plants and target wind power plants are mined by a spatial attention mechanism in the prediction model, and corresponding weights are respectively given to the corresponding wind power plants according to the mined spatial relationships; then, the feature matrix given with the weight is transmitted to a gating circulation unit network, and the gating circulation unit network digs the hidden relation existing in the feature matrix ; then optimizing a super-parameter matrix theta in a gating circulation unit network by the prediction model by adopting a crisscross optimization algorithm to finish the training of the prediction model; and finally, predicting the power of the target wind power plant by using the trained prediction model, and obtaining a power time sequence of the corresponding wind power plant.
The acquisition module and the processing module in the system can be integrated in a circuit board or can be arranged separately.
According to the prediction system, the dual dimension reduction attention mechanism can effectively reduce dimension of the high-dimension wind power data, can realize dimension reduction and feature extraction of the high-dimension wind power data, reduces data redundancy, and is beneficial to improving the prediction precision of wind power; the self-adaptive VMD algorithm can decompose the power signal into components which are easier to train, so that the training efficiency is improved; and the super-parameter matrix of the network model is optimized by adopting a crisscross optimization algorithm, so that the problem of local optimum possibly existing in the network model built by the neural network can be solved. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism not only has higher training efficiency, but also can reduce dimension of high-dimension wind power data and solve the problem of local optimization of a neural network model, thereby improving the prediction precision of wind power.
In the specific content of the above embodiment, any combination of the technical features may be performed without contradiction, and for brevity of description, all possible combinations of the technical features are not described, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
It is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.
Claims (10)
1. A wind power prediction method of a self-adaptive VMD and a dual dimension-reduction attention mechanism is characterized by comprising the following steps:
S10, acquiring a power sequence of a target wind power plant, a wind speed sequence, a temperature sequence and a barometric pressure sequence with different heights, and preprocessing;
S20, forming a characteristic matrix X by the preprocessed power sequence, the wind speed sequences, the temperature sequences and the air pressure sequences with different heights;
S30, constructing a dual dimension-reducing attention mechanism and a gating circulation unit network model comprising an active dimension-reducing layer and a passive dimension-reducing layer;
S40, inputting the feature matrix X in the step S20 into the active dimension reduction layer to perform feature screening dimension reduction, and mining out features with the largest influence on power by using an average influence value MIV dimension reduction method;
S50, decomposing the power sequence by using a crisscross optimization algorithm and taking the weighted envelope change rate as an adaptability function of the variation modal decomposition VMD, and adding the decomposed power sequence into a feature matrix after dimension reduction;
s60, conveying the feature matrix containing the decomposed power sequences after dimension reduction to the passive dimension reduction layer in the step S30, fully excavating the relations between different features and power again, and giving weights to obtain a feature matrix given the weights;
S70, conveying the feature matrix given with the weight to a gating circulation unit network, and fully excavating an implicit relation existing in the feature matrix/> given with the weight by the gating circulation unit network;
S80, optimizing a super-parameter matrix theta in a gate control circulation unit network by adopting a longitudinal and transverse cross optimization algorithm to complete the training of the prediction model;
S90, predicting a power time sequence of the future moment of the target wind power plant by using the trained prediction model.
2. The wind power prediction method of the self-adaptive VMD and the dual dimension-reduction attention mechanism according to claim 1, wherein in step S10, a power sequence and wind speed sequences, temperature sequences and air pressure sequences with different heights are subjected to min-max normalization processing to obtain a processed power sequence P, a processed wind speed sequence WS, a processed temperature sequence Tem and an processed air pressure sequence Pre.
3. The method for predicting wind power by using an adaptive VMD and a dual dimension-reduction attention mechanism according to claim 2, wherein in step S20, x= [ D 1,D2,...,Dm ], wherein D m represents a matrix formed by features at the m-th height t-1 to t-n, and the feature matrix X is represented as:
Where P t-n is the power at times t-1 to t-n, 、/> and/> are the wind speed, temperature and air pressure at times t-1 to t-n, respectively, for the mth altitude.
4. The method for predicting wind power of an adaptive VMD and a dual dimension reduction attention mechanism according to claim 1, wherein in step S30, the dual dimension reduction attention mechanism and the gated loop unit network model further comprises an input layer and an output layer, and the number of neurons of the input layer and the output layer is equal to the number of features.
5. The method for predicting wind power by adaptive VMD and dual dimension-reduction attention mechanism of claim 1, wherein step S40 comprises the steps of:
s41, increasing the characteristic amplitude of each dimension by A% to obtain X up, and obtaining m characteristic enhancement matrixes :
wherein denotes the enhanced ith column feature,/> denotes the ith column feature,/> denotes the W column feature; a% is 10% -50%;
S42, respectively reducing the characteristic amplitude of each dimension by A% to obtain X reduce, and obtaining m characteristic attenuation matrixes :
Wherein denotes the attenuated ith column feature;
S43, respectively sending the F up and the F reduce into a gating circulation unit for training and prediction, wherein the length of a predicted sample is N, respectively obtaining a predicted result Y up with enhanced characteristics and a predicted result Y reduce with attenuated characteristics, and subtracting the predicted result Y up with enhanced characteristics from the predicted result Y reduce with attenuated characteristics to obtain an influence value of the characteristics i on power prediction output:
wherein is a sequence of length N;
Averaging yields the average impact of the ith feature on the predicted power output/> :
And S44, reserving the characteristics of the influence value IV higher than the average influence value , and eliminating the characteristics of the influence value IV lower than the average influence value to obtain a feature matrix/> after dimension reduction.
6. The wind power prediction method of the adaptive VMD and the dual dimension reduction attention mechanism according to claim 5, wherein in step S50, the calculation process of the envelope change rate includes the steps of:
S51, carrying out Hilbert transformation on a continuous sample sequence power sequence x (t) to obtain :
S52, defining envelope of x (t) as , wherein the expression is shown in the following formula:
S53, calculating the envelope change rate ECR as follows:
where T is the sample length of x (T), phi is the rate of change interval, and the mean () function acts to calculate the average;
S54, regarding N sub-components generated by the decomposition of the variation mode of x (t), a weighted summation mode is adopted to obtain WECR:
Where denotes the envelope/> ,/> denotes the envelope of x (t)/> denotes the envelope rate of change of/> and WECR denotes the sum of the weighted envelope rates of change of all the following power sub-sequences;
and S55, optimizing the decomposition parameters by using a crisscross optimization algorithm, and finally adding the decomposed power sequences into the feature matrix after dimension reduction.
7. The method for predicting wind power by adaptive VMD and dual dimension-reduction attention mechanism of claim 6, wherein step S60 comprises the steps of:
S61, pearson correlation coefficient/> between the j-th dimension feature and power in the feature matrix after dimension reduction of the decomposed power sequence is as follows:
wherein represents the ith value of the mean value of the jth dimension characteristic after the active dimension reduction layer is reconstructed,/> represents the ith value of the power,/> and/> represent the mean value of the jth dimension characteristic and the mean value of the power after the dimension reduction layer is reconstructed respectively;
s62, combining the characteristics after dimension reduction into a power correlation matrix R:
s63, solving an average of the power correlation matrix R:
Wherein represents the feature dimension after dimension reduction;
s64, transforming R by taking as an evaluation index to obtain a transformation matrix num_0_w:
Wherein, part is called a non-key feature, part/() is called a key feature, w is a non-key feature weight, and the value is (0, 1);
S65, performing matrix point multiplication on the initial weight of the attention layer and the input matrix by using num_0_w as the initial weight of the attention layer to obtain a feature matrix after the weight is given.
8. The wind power prediction method of the adaptive VMD and the dual dimension reduction attention mechanism according to claim 7, wherein in step S70, a two-layer gating cyclic unit network is built with a weighted feature matrix as input, and an activation function is tanh:
Wherein 、/>、/>、/>、/>、/> is a weight parameter matrix,/> 、/>、/> is a bias parameter matrix,/> is a matrix multiplication,/> is a Sigmod function,/> is a reset gate,/> is an update gate,/> is a candidate state of an hidden layer at the current time,/> is a current hidden state,/> is a hidden state at the previous time, and/> is an input state at the current time.
9. The method for predicting wind power by using an adaptive VMD and a dual dimensionality reduction and attention mechanism of claim 8, wherein in step S80, the hyper-parameter matrix θ is composed of weight parameters and bias parameters of a network output layer of a gating cycle unit, and the following formula is shown:
。
10. the method for predicting wind power by using an adaptive VMD and a dual dimension-reduction attention mechanism according to claim 9, wherein in step S80, the specific step of optimizing the hyper-parameter matrix θ by using the crisscross optimization algorithm includes:
s81, taking the improvement of prediction precision as a target, adopting a root mean square error minimum as an fitness function in a crisscross optimization algorithm, wherein the following formula is adopted:
wherein is a fitness function, wherein/> is the number of samples, and/> and/> represent a true value and a predicted value, respectively;
S82, randomly selecting nth-dimension interdigitation by parent particles and/> in the parameters;
Wherein is the nth dimension child generated by transverse intersection of/> and/> , r1, r2 and c1, c2 are random numbers of (0, 1) and (-1, 1), respectively,/> is the nth dimension of particle/> , and/> is the nth dimension of particle/> ;
s83, randomly selecting a v-th dimension and a k-th dimension to be intersected by each other by using a parent particle theta (q) in the parameters;
Wherein is a v-th dimension child generated by longitudinally crossing/> and/> , r is a random number of (0, 1), is a v-th dimension of particle/> , and/> is a k-th dimension of particle/> ;
s84, repeating the steps S62 and S63 according to the set iteration times, and stopping iteration when the preset iteration times are reached.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410283502.3A CN117892113B (en) | 2024-03-13 | 2024-03-13 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410283502.3A CN117892113B (en) | 2024-03-13 | 2024-03-13 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117892113A true CN117892113A (en) | 2024-04-16 |
| CN117892113B CN117892113B (en) | 2024-06-11 |
Family
ID=90645066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410283502.3A Active CN117892113B (en) | 2024-03-13 | 2024-03-13 | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117892113B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113193556A (en) * | 2021-05-12 | 2021-07-30 | 燕山大学 | Short-term wind power prediction method based on probability prediction model |
| CN113642767A (en) * | 2021-07-09 | 2021-11-12 | 武汉科技大学 | Multi-dimensional feature combination prediction method based on MI-VMD-DA-EDLSTM-VEC |
| CN114169251A (en) * | 2021-12-27 | 2022-03-11 | 广东工业大学 | Ultra-short-term wind power prediction method |
| CN114429078A (en) * | 2021-12-22 | 2022-05-03 | 广东工业大学 | Short-term wind power prediction method and system |
| CN115630726A (en) * | 2022-09-01 | 2023-01-20 | 华能江苏综合能源服务有限公司 | Roof photovoltaic power prediction method based on VMD-BILSTM neural network fusion attention mechanism |
| CN115907120A (en) * | 2022-11-11 | 2023-04-04 | 盐城工学院 | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method |
| WO2023123593A1 (en) * | 2021-12-30 | 2023-07-06 | 浙大城市学院 | Variational mode decomposition and residual network-based aviation bearing fault diagnosis method |
| CN116960978A (en) * | 2023-07-31 | 2023-10-27 | 河海大学 | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction |
-
2024
- 2024-03-13 CN CN202410283502.3A patent/CN117892113B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113193556A (en) * | 2021-05-12 | 2021-07-30 | 燕山大学 | Short-term wind power prediction method based on probability prediction model |
| CN113642767A (en) * | 2021-07-09 | 2021-11-12 | 武汉科技大学 | Multi-dimensional feature combination prediction method based on MI-VMD-DA-EDLSTM-VEC |
| CN114429078A (en) * | 2021-12-22 | 2022-05-03 | 广东工业大学 | Short-term wind power prediction method and system |
| CN114169251A (en) * | 2021-12-27 | 2022-03-11 | 广东工业大学 | Ultra-short-term wind power prediction method |
| WO2023123593A1 (en) * | 2021-12-30 | 2023-07-06 | 浙大城市学院 | Variational mode decomposition and residual network-based aviation bearing fault diagnosis method |
| CN115630726A (en) * | 2022-09-01 | 2023-01-20 | 华能江苏综合能源服务有限公司 | Roof photovoltaic power prediction method based on VMD-BILSTM neural network fusion attention mechanism |
| CN115907120A (en) * | 2022-11-11 | 2023-04-04 | 盐城工学院 | Attention mechanism-based VMD-CNN-LSTM short-term wind power prediction method |
| CN116960978A (en) * | 2023-07-31 | 2023-10-27 | 河海大学 | Offshore wind power prediction method based on wind speed-power combination decomposition reconstruction |
Non-Patent Citations (2)
| Title |
|---|
| 殷豪;董朕;孟安波;: "基于VMD-SE-IPSO-BNN的超短期风电功率预测", 电测与仪表, no. 02, 25 January 2018 (2018-01-25) * |
| 邹仕富;李嘉周;毛启均;宋明军;: "数据降维和特征分析的GOA-LSSVM短期负荷预测", 实验室研究与探索, no. 11, 15 November 2019 (2019-11-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117892113B (en) | 2024-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tian | Modes decomposition forecasting approach for ultra-short-term wind speed | |
| CN111091233B (en) | Short-term wind power prediction modeling method for wind power plant | |
| CN114429078B (en) | Short-term wind power prediction method and system | |
| CN111401755B (en) | Multi-new-energy output scene generation method, device and system based on Markov chain | |
| CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
| CN111506868B (en) | Ultra-short-term wind speed prediction method based on HHT weight optimization | |
| CN112232561A (en) | Power load probability prediction method based on constrained parallel LSTM quantile regression | |
| CN115115125A (en) | Photovoltaic power interval probability prediction method based on deep learning fusion model | |
| CN114298140A (en) | Wind power short-term power prediction correction method considering unit classification | |
| CN116341613A (en) | Ultra-short-term photovoltaic power prediction method based on Informar encoder and LSTM | |
| CN116169670A (en) | Short-term non-resident load prediction method and system based on improved neural network | |
| CN116702937A (en) | Photovoltaic output day-ahead prediction method based on K-means mean value clustering and BP neural network optimization | |
| CN115456312A (en) | Short-term power load prediction method and system based on octyl geometric modal decomposition | |
| CN115409369A (en) | Comprehensive energy system reliability evaluation method based on mechanism and data hybrid driving | |
| CN115713144A (en) | Short-term wind speed multi-step prediction method based on combined CGRU model | |
| CN110458341A (en) | A kind of ultra-short term wind-powered electricity generation prediction technique and system considering meteorological features | |
| CN110738363A (en) | photovoltaic power generation power prediction model and construction method and application thereof | |
| CN112418504B (en) | Wind speed prediction method based on mixed variable selection optimization deep belief network | |
| CN117892113B (en) | Wind power prediction method of self-adaptive VMD and dual dimension-reduction attention mechanism | |
| CN115936062A (en) | Photovoltaic power day-ahead interval prediction method based on scene generation and deep learning | |
| CN116911418A (en) | Photovoltaic power generation power ultra-short-term prediction method based on wavelet transformation and BP neural network optimization | |
| CN115659201A (en) | Gas concentration detection method and monitoring system for Internet of things | |
| CN112564181B (en) | Online probability safety monitoring method considering uncertainty of new energy output | |
| CN114897260A (en) | Short-term wind speed prediction model modeling method and prediction method based on LSTM neural network | |
| CN114971022A (en) | Wind power prediction method based on D-value-LSTM neural network model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |