CN118013457A - Wind speed prediction method and system based on multi-mode data - Google Patents
Wind speed prediction method and system based on multi-mode data Download PDFInfo
- Publication number
- CN118013457A CN118013457A CN202410154819.7A CN202410154819A CN118013457A CN 118013457 A CN118013457 A CN 118013457A CN 202410154819 A CN202410154819 A CN 202410154819A CN 118013457 A CN118013457 A CN 118013457A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- time step
- prediction result
- matrix
- comprehensive
- 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
- 238000000034 method Methods 0.000 title claims abstract description 49
- 239000011159 matrix material Substances 0.000 claims abstract description 112
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000012937 correction Methods 0.000 claims abstract description 14
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 230000009466 transformation Effects 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 69
- 238000000605 extraction Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 11
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000010845 search algorithm Methods 0.000 claims description 8
- 230000007613 environmental effect Effects 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 3
- 239000002131 composite material Substances 0.000 claims 1
- 230000015654 memory Effects 0.000 description 10
- 238000012360 testing method Methods 0.000 description 10
- 241000282461 Canis lupus Species 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 8
- 241000282421 Canidae Species 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 235000019580 granularity Nutrition 0.000 description 5
- 230000007774 longterm Effects 0.000 description 5
- 230000000306 recurrent effect Effects 0.000 description 5
- 238000005070 sampling Methods 0.000 description 5
- 238000010183 spectrum analysis Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000000638 solvent extraction Methods 0.000 description 3
- 230000001360 synchronised effect Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 241000282470 Canis latrans Species 0.000 description 1
- 241000283153 Cetacea Species 0.000 description 1
- 241000511338 Haliaeetus leucocephalus Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
- G06F18/256—Fusion techniques of classification results, e.g. of results related to same input data of results relating to different input data, e.g. multimodal recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- 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
- G06F18/2131—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on a transform domain processing, e.g. wavelet transform
-
- 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
- G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
- G06F18/21322—Rendering the within-class scatter matrix non-singular
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2123/00—Data types
- G06F2123/02—Data types in the time domain, e.g. time-series data
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a wind speed prediction method and a wind speed prediction system based on multi-mode data, comprising the following steps: acquiring a historical wind speed sequence and external environment characteristics; denoising the historical wind speed sequence by using an SSA algorithm to obtain a first intermediate wind speed sequence; performing wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, constructing a first wind speed matrix, and performing normalization processing on the first wind speed matrix to obtain a second wind speed matrix; dividing a second wind speed matrix according to a sliding window and slice combination mode, respectively extracting features of three matrix slices obtained by dividing, respectively inputting the extracted three features into a GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result, and overlapping by using a linear weight overlapping module to obtain an initial wind speed comprehensive prediction result; and correcting the initial wind speed comprehensive prediction result by using the trained error correction model to obtain a final wind speed prediction result.
Description
Technical Field
The invention belongs to the field of wind speed prediction, and particularly relates to a wind speed prediction method and system based on multi-mode data.
Background
With the increasing of greenhouse effect, the continuous development of fossil energy sources such as coal, petroleum and the like, the continuous exhaustion of traditional energy sources, and the search for alternatives to fossil energy sources is urgent. Wind power is used as a clean and renewable energy source, and has wide development prospect, but wind power grid connection can cause great impact to a power system due to volatility and instability of the wind power. Therefore, accurate prediction of wind speed is of great importance for planning and scheduling of wind power generation.
In order to better utilize wind energy, the great impact on a power grid caused by wind power grid connection is avoided, and short-term accurate prediction of wind speed is also important for running and planning of a wind power plant and stability and reliability of a power system.
Currently, short-term wind speed predictions mainly include physical methods, statistical methods, and machine learning methods. The physical method is to simulate the atmospheric kinematics process according to a numerical weather forecast model and combining with topographic data to predict the wind speed of a designated position. The method has large calculation amount and high requirement on calculation resources, but can provide better physical interpretation. Statistical methods such as ARMA, kalman filtering and other time sequence models can capture the statistical characteristics of time sequence data and predict. The method is simple in calculation, but has poor prediction effect on abnormal conditions. The machine learning method such as a neural network, a support vector machine and the like can automatically extract complex data characteristics, perform nonlinear modeling and realize end-to-end prediction. This method has high prediction accuracy but poor interpretability. The existing short-term wind speed prediction technology generally has the problems of low precision, poor prediction effect on abnormal conditions and the like. The wind power generation method is mainly characterized in that the randomness of wind power is strong, the wind power generation method is influenced by various meteorological factors, and the establishment of an accurate prediction model has certain difficulty. In order to improve the prediction precision, the current research is mainly focused on aspects of wind speed space-time correlation analysis, advanced data acquisition technology, combined prediction mode and the like, but each technology has certain defects and needs to be further improved.
Disclosure of Invention
In order to solve the problems in the background art, an aspect of the present invention provides a wind speed prediction method based on multi-modal data, including:
S1: acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
S2: denoising the historical wind speed sequence by using an SSA algorithm to obtain a first intermediate wind speed sequence;
S3: performing wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
s4: constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
S5: dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ T-T 2, T-1], a matrix slice with a time step length of [ T-T 3, T-1] and a matrix slice with a time step length of [ T-T 4, T-1], and respectively inputting the three divided matrix slices into a 2DCNN convolutional neural network to perform feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature, wherein T 1＞T2＞T3＞T4 is more than 1;
S6: respectively splicing the first intermediate feature, the second intermediate feature and the third intermediate feature with the external environment feature of the current time step t to obtain a first comprehensive feature, a second comprehensive feature and a third comprehensive feature;
S7: respectively inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into a GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
S8: inputting the first wind speed predicted result, the second wind speed predicted result and the third wind speed predicted result of the current time step t into a trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive predicted result of the current time step t; and correcting the initial wind speed comprehensive prediction result of the current time step t by using the trained error correction model to obtain the final wind speed prediction result of the current time step t.
Preferably, the external environmental features include: at least one of the environment temperature, the fan temperature, the environment humidity and the environment wind direction after normalization treatment.
Preferably, said constructing a first wind speed matrix from the N second intermediate wind speed sequences comprises:
Let X n, N e {1,2,..n } denote the nth second intermediate wind speed sequence, then the first wind speed matrix is expressed as:
Where T represents the transpose and X represents the first wind speed matrix.
Preferably, the comprehensive prediction result of the initial wind speed of the current time step t includes:
Wherein, The method comprises the steps of representing an initial wind speed comprehensive prediction result of a current time step t, W 1、W2 and W 3 representing weight parameters of a linear weight superposition module, Y t1 representing a first wind speed prediction result of the current time step t, Y t2 representing a second wind speed prediction result of the current time step t, and Y t3 representing a third wind speed prediction result of the current time step t.
Preferably, the training process of the linear weight superposition module includes:
Dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ t i-T2,ti -1], a matrix slice with a time step length of [ t i-T3,ti -1] and a matrix slice with a time step length of [ t i-T4,ti -1], [ t i∈[t-T1+T2, t-1];
generating a training sample set by taking a matrix slice with the time step length of [ t i-T2,ti -1], a matrix slice with the time step length of [ t i-T3,ti -1] and a matrix slice with the time step length of [ t i-T4,ti -1] as training samples and taking the real wind speed of the time step t i as a label of the training samples;
for each training sample, respectively splicing a matrix slice with the time step of [ t i-T2,ti -1], a matrix slice with the time step of [ t i-T3,ti -1] and a matrix slice with the time step of [ t i-T4,ti -1] with external environment features of the time step of t i to obtain a first training comprehensive feature, a second training comprehensive feature and a third training comprehensive feature;
Respectively inputting the first training comprehensive feature, the second training comprehensive feature and the third training comprehensive feature into a GRU prediction model to obtain a first training wind speed prediction result, a second training wind speed prediction result and a third training wind speed prediction result of a time step t i;
Inputting the first training wind speed prediction result, the second training wind speed prediction result and the third training wind speed prediction result of the time step t i into a linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the time step t i;
And taking the sum of the initial wind speed comprehensive prediction results of all training samples and the deviation value of the real wind speed as a fitness value function, and optimizing the weight parameters of the linear weight superposition module by using an intelligent search algorithm with the minimum fitness value as an optimization target to obtain the optimal weight parameters of the linear weight superposition module.
Preferably, the correcting the initial wind speed comprehensive prediction result by using the trained error correction model includes:
Calculating the deviation between the comprehensive predicted result of the initial wind speed of the time step t k∈[t-T5, t-1 and the real wind speed of the time step t k, and constructing a composition error sequence
Inputting an error sequence P t into an LSTM network model to predict the deviation between the comprehensive predicted result of the initial wind speed of the current time step t and the real wind speed of the current time step t;
And obtaining a final wind speed prediction result of the current time step t according to the initial wind speed comprehensive prediction result of the current time step t and the deviation addition between the initial wind speed comprehensive prediction result of the current time step t and the real wind speed of the current time step t.
In another aspect, the present invention provides a wind speed prediction system based on multi-modal data, where the system is applied to the wind speed prediction method based on multi-modal data, and the wind speed prediction method includes:
The data acquisition module is used for acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
the SSA noise reduction module is used for denoising the historical wind speed sequence to obtain a first intermediate wind speed sequence;
The wavelet transformation module is used for carrying out wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
the data processing module is used for constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
The matrix slicing module is used for dividing the second wind speed matrix according to a sliding window and slicing combination mode to obtain a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1], wherein T 1＞T2＞T3＞T4 is more than 1;
The feature extraction module is used for respectively inputting a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1] into the 2DCNN convolutional neural network for feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature;
the characteristic splicing module is used for respectively splicing the first intermediate characteristic, the second intermediate characteristic and the third intermediate characteristic with the external environment characteristic of the current time step t to obtain a first comprehensive characteristic, a second comprehensive characteristic and a third comprehensive characteristic;
the GRU prediction module is used for inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into the GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
The linear weight superposition module is used for inputting the first wind speed prediction result, the second wind speed prediction result and the third wind speed prediction result of the current time step t into the trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the current time step t;
The error correction module is used for correcting the comprehensive wind speed prediction result of the current time step t to obtain the final wind speed prediction result of the current time step t.
The invention has at least the following beneficial effects
Compared with the prior art, the method creatively introduces the implicit characteristic of multi-mode data, and depth information in the multi-mode data is further extracted from the original one-dimensional wind speed sequence. In addition, the fusion processing method with multiple time granularities is one of the advantages created by the invention, and the difference of the time granularities can greatly influence the prediction accuracy, so that in order to weaken the influence, the singular spectrum noise reduction method is improved for the different prediction time granularities from three different time granularities, the noise reduction processing is carried out on the original wind speed data by adopting an embedded time window, and meanwhile, the final different prediction results are linearly overlapped, so that the final fusion result is obtained. The error correction is another innovation point created by the invention, after the superimposed prediction result is obtained, the difference value between the superimposed prediction result and the true value is used as a data set to be input into the LSTM for training and prediction, so that the error value to be corrected is obtained, and the error value is superimposed with the prediction result in the last step, so that the final prediction result is obtained.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention;
FIG. 2 is a diagram of an embodiment of wavelet analysis according to the present invention;
FIG. 3 is a schematic diagram of an intermediate wind speed matrix process of the present invention;
FIG. 4 is a diagram illustrating the partitioning of a data set according to the present invention;
FIG. 5 is a sample and label slip schematic of the present invention;
Fig. 6 is a schematic diagram of a sample and tag sliding partitioning rule according to the present invention.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
Referring to fig. 1, the present invention provides a wind speed prediction method based on multi-mode data, including:
S1: acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
the external environmental features include: at least one of the environment temperature, the fan temperature, the environment humidity and the environment wind direction after normalization treatment.
S2: denoising the historical wind speed sequence by using an SSA (singular spectrum analysis (Singular Spectrum Analysis, SSA)) algorithm to obtain a first intermediate wind speed sequence;
Singular spectrum analysis (Singular Spectrum Analysis, SSA) is a method of processing nonlinear time series data that involves the decomposition and reconstruction of a trace matrix of the time series under investigation to extract different components (long-term trends, seasonal trends, noise, etc.) in the time series for analysis or denoising of the time series and for other tasks. The singular spectrum analysis mainly comprises four steps: embedding-decomposition-grouping-reconstruction.
S3: performing wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
The wavelet transform (Wavelet Transform, WT) is derived from fourier analysis, which is advantageous to improve the accuracy of wind speed predictions. Is a tool for analyzing multi-resolution signals of time and frequency windows by stretching and translating. The method can effectively extract the characteristic information from the signals, solves the problem that many Fourier transforms cannot solve, and improves the accuracy of wind speed forecast. The original data is subjected to wavelet analysis processing to obtain N subsequences, and the five subsequences are converted into a matrix so as to facilitate subsequent prediction processing.
S4: constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
S5: dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ T-T 2, T-1], a matrix slice with a time step length of [ T-T 3, T-1] and a matrix slice with a time step length of [ T-T 4, T-1], and respectively inputting the three divided matrix slices into a 2DCNN convolutional neural network to perform feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature, wherein T 1＞T2＞T3＞T4 is more than 1;
2DCNN is an abbreviation for Two-dimensional convolutional neural network (Two-Dimensional Convolutional Neural Network), which is a convolutional neural network used to process image data. The basic structure of 2DCNN includes a convolutional layer, a pooling layer, and a fully-connected layer. The convolution layer performs a convolution operation using the convolution kernel input to extract spatial information and features. The pooling layer downsamples the output of the convolution layer, and reduces the size of the feature map, thereby reducing the number of parameters and the amount of computation. The full connection layer is then used for the final classification or regression task.
In order to extract more hidden features from the original wind speed sequence, the method is different from the traditional mode of predicting by combining 1DCNN with other neural networks, and the 2DCNN is adopted to extract the earlier-stage features of the wind speed subsequence matrix.
S6: respectively splicing the first intermediate feature, the second intermediate feature and the third intermediate feature with the external environment feature of the current time step t to obtain a first comprehensive feature, a second comprehensive feature and a third comprehensive feature;
S7: respectively inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into a GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
GRU (Gated Recurrent Unit, gated loop unit) is a variant of a Recurrent Neural Network (RNN) that aims to solve the problem of gradient disappearance and the difficulties of long-term dependence modeling that exist with traditional RNNs. GRU was proposed by Cho et al in 2014, which references the concept of LSTM (long short term memory network) and attempts to simplify the structure of LSTM.
A recurrent neural network is a neural network for processing sequence data that has recurrent connections allowing information to be transferred between different time steps. However, the conventional RNN has problems of gradient extinction and gradient explosion, which results in difficulty in handling long sequences and capturing long-term dependencies. To solve the gradient vanishing problem, hochreiter and Schmidhuber proposed LSTM in 1997. LSTM introduces gating mechanisms, including input gates, forget gates, and output gates, to control the flow of information and preserve long-term dependencies. Although LSTM is very effective, its complex structure may not be sufficiently lightweight. GRU is proposed as a lightweight alternative to LSTM. It reduces the number of gates in the LSTM, including only update gates and reset gates, thereby reducing the complexity of the model. The structure of the GRU is simpler, but still effectively captures long-term dependencies.
The main components of the GRU include:
Update Gate (Update Gate): it is controlled whether the hidden state of the previous time step should be preserved and how to blend with the current input.
Reset gate (RESET GATE): it is decided how the concealment status of the previous time step should affect the candidate concealment status of the current time step.
Candidate hidden state (CANDIDATE HIDDEN STATE): candidate hidden states are calculated based on the current input and the hidden state of the previous time step.
Final hidden state: the hidden state of the previous time step of the updated gate control and the candidate hidden state are combined.
One of the advantages of a GRU is that it generally requires fewer parameters in training and inference and is therefore easier to train and deploy. It is excellent in many Natural Language Processing (NLP) and sequence modeling tasks, and becomes one of the recurrent neural network structures commonly used in deep learning.
S8: inputting the first wind speed predicted result, the second wind speed predicted result and the third wind speed predicted result of the current time step t into a trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive predicted result of the current time step t; and correcting the initial wind speed comprehensive prediction result of the current time step t by using the trained error correction model to obtain the final wind speed prediction result of the current time step t.
Preferably, said constructing a first wind speed matrix from the N second intermediate wind speed sequences comprises:
Let X n, N e {1,2,..n } denote the nth second intermediate wind speed sequence, then the first wind speed matrix is expressed as:
Where T represents the transpose and X represents the first wind speed matrix.
Preferably, the comprehensive prediction result of the initial wind speed of the current time step t includes:
Wherein, The method comprises the steps of representing an initial wind speed comprehensive prediction result of a current time step t, W 1、W2 and W 3 representing weight parameters of a linear weight superposition module, Y t1 representing a first wind speed prediction result of the current time step t, Y t2 representing a second wind speed prediction result of the current time step t, and Y t3 representing a third wind speed prediction result of the current time step t.
Preferably, the training process of the linear weight superposition module includes:
Dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ t i-T2,ti -1], a matrix slice with a time step length of [ t i-T3,ti -1] and a matrix slice with a time step length of [ t i-T4,ti -1], [ t i∈[t-T1+T2, t-1];
generating a training sample set by taking a matrix slice with the time step length of [ t i-T2,ti -1], a matrix slice with the time step length of [ t i-T3,ti -1] and a matrix slice with the time step length of [ t i-T4,ti -1] as training samples and taking the real wind speed of the time step t i as a label of the training samples;
for each training sample, respectively splicing a matrix slice with the time step of [ t i-T2,ti -1], a matrix slice with the time step of [ t i-T3,ti -1] and a matrix slice with the time step of [ t i-T4,ti -1] with external environment features of the time step of t i to obtain a first training comprehensive feature, a second training comprehensive feature and a third training comprehensive feature;
Respectively inputting the first training comprehensive feature, the second training comprehensive feature and the third training comprehensive feature into a GRU prediction model to obtain a first training wind speed prediction result, a second training wind speed prediction result and a third training wind speed prediction result of a time step t i;
Inputting the first training wind speed prediction result, the second training wind speed prediction result and the third training wind speed prediction result of the time step t i into a linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the time step t i;
And taking the sum of the initial wind speed comprehensive prediction results of all training samples and the deviation value of the real wind speed as a fitness value function, and optimizing the weight parameters of the linear weight superposition module by using an intelligent search algorithm with the minimum fitness value as an optimization target to obtain the optimal weight parameters of the linear weight superposition module.
The historic wind speed sequence and external environmental characteristics etc. in this embodiment are derived from the SCADA (SupervisoryControlandDataAcquisition) system provided in the wind turbine, the SCADA system being a computerized system for monitoring and controlling industrial processes. The data in this embodiment is 184 days of historical data stored in a fan SCADA system of a certain wind field, and 26496 groups of data in total include historical wind speed data, environmental temperature, fan temperature, environmental temperature and environmental wind direction, and the sampling interval of the collected data is 15 minutes. It should be noted that, in the following practical application, the minimum data time length T1 required for prediction is only greater than the set maximum time step, i.e. the preset 24 hours.
In this embodiment, one of the innovation points is that SSA noise reduction analysis is performed on different embedding windows for different data partitioning and prediction step sizes, that is, specific embedding windows are set to be 24, 48 and 96 respectively for three different time steps of 6 hours, 12 hours and 24 hours, and subsequent decomposition, grouping and reconstruction modes are consistent. Different embedded windows are set for different data partitions and training step sizes, hidden features under different conditions are extracted, and a first intermediate wind speed sequence is obtained and is used for subsequent further processing.
The intermediate sequences of three different implicit features obtained after SSA noise reduction processing are processed by wavelet analysis according to different frequencies to obtain N subsequences, in the embodiment, 5 subsequences are taken as an example. The wavelet analysis processing results in the examples are shown in fig. 2, and it is clear from fig. 2 that the original sequence obtains sub-sequences of 5 different frequencies after wavelet analysis.
The five subsequences, namely the second intermediate wind speed sequence, are transposed and combined to form a first wind speed matrix to form a wind speed matrix with the shape of (26496,5), wherein the number of rows is the number of wind speed data, and the number of columns is the number of subsequences. The second wind speed matrix is further normalized to be included between [ -1,1] so as to improve the training speed and the prediction accuracy of the subsequent model. The process is shown in fig. 3.
The second wind speed matrix and other characteristic data obtained through the steps need to be subjected to data division and characteristic extraction to obtain the related information implied therein. In order to facilitate the prediction and processing of the subsequent model, all data in this embodiment needs to be divided into a training set, a validation set, and a test set, and the number of the data is 19872, 3312, 3312, that is, the division ratio is 0.75,0.125,0.125. The specific division is shown in fig. 4. The divided training set, verification set and test set are used for training, verifying and testing the subsequent model respectively, wherein the test set can be considered as the practical application test of the invention.
Due to the time series prediction requirement, in actual prediction, regression prediction is often performed according to data of a time step before a prediction time. The different poles of the time steps affect the extraction of the model to the input data characteristics to a great extent, so that the accuracy of the prediction result of the model is further affected, and therefore, the division of the selected time steps is very important. The input data and label data of the wind speed prediction model such as the second wind speed matrix need to be further divided in a sliding manner and processed by Reshape according to the input data rule of the model, and the processing rule and mode are shown in fig. 5.
The sliding division rule of the sample and the label can be generally understood from the schematic diagram 5, taking fig. 5 as an example, the sliding window of the input sample slides down sequentially according to a certain time step for sampling, and each time slides one sample, so as to form continuous sample data, while for the label data, the first sliding sampling is located at the later time point of the sample sampling data, and only one sampling time data is extracted, which is very clearly shown in the figure. Fig. 6 is a brief overview of the sliding division principle of samples and labels. The samples associated with the subsequent different time steps, that is, [ti-T2,ti-1],[ti-T3,ti-1],[ti-T4,ti-1] different time steps obtained by each division, correspond to the p value of T i -1 and the q value of T i in fig. 6, and finally form training, verifying and testing samples and label data of different time steps, and in this embodiment, a person skilled in the art can select different values of T 1～T5 according to actual prediction effects.
Taking 24, 28, 96 in the embodiment, namely corresponding time steps of 6 hours, 12 hours and 24 hours respectively as an example, the final wind speed characteristic matrix data set is formed to be (19848, 24,5), (3288, 24,5), (3288, 24,5), the external characteristic data set is formed to be (19848, 24, 1), (3288, 24, 1), and the wind speed label data set is formed to be (19848,1), (3288,1), (3288,1). Taking 6 hours of time step as an example, the results are a training set, a verification set and a test set in sequence from left to right, and aiming at the three-dimensional data shape format, the first dimension is the sample length in the set, the second dimension is the time step, and the third dimension is the number of 'features'; for two-dimensional data, the first dimension is the sample length in the set and the second dimension is the "feature quantity". It should be noted that, because the set length of the training set, the validation set and the test set after the division needs to be subtracted by the corresponding time step, the data that needs to be imported by the system in practical application must be larger than the maximum time step.
And inputting a second wind speed matrix obtained by processing and dividing based on different time steps into a 2DCNN for feature extraction, further performing Reshape operation on the extracted feature map after flattening layer processing to obtain first, second and third intermediate features, and splicing the first, second and third intermediate features with external environment features of corresponding time steps to obtain first, second and third comprehensive features taking different time steps as distinguishing points.
The comprehensive characteristics are input into the GRU model for training and prediction, and in the same way, in order to solve the problem that the prediction result time is not synchronous due to the fact that the prediction result lengths under three time granularities are different caused by different time steps, redundant parts of other prediction results need to be omitted by taking the prediction time period of the maximum time step as a reference, and time synchronization of the prediction results is achieved so that the follow-up prediction results can be overlapped linearly. After time synchronization is completed, the COA is adopted to optimize the superposition weight, the error between the superposed predicted result and the true value is used as a fitness function, and the optimal weight and the initial wind speed predicted result are obtained after iterative optimization.
After the initial prediction result is obtained, taking the error value of the initial prediction result and the original wind speed in the corresponding time period as a new data set, after the new error data set is obtained, dividing the training set, the verification set and the test set according to the method, inputting the training set, the verification set and the test set into an LSTM network for training and predicting to obtain the error prediction result, and superposing the error value and the initial error result in the previous step to obtain the final wind speed prediction result.
The intelligent search algorithm adopted in the embodiment is suburban wolf algorithm, suburban wolf optimization algorithm (coyote optimization algorithm, COA) is a heuristic optimization algorithm based on suburban wolf population behaviors in nature, and the algorithm is proposed by Pierezan equal to 2018. The method simulates the behavior of suburban wolves in searching food in group cooperation, and is used for solving the optimization problem. Because the algorithm has unique search structure, a new mechanism is provided for the exploration and exploitation balance process, and the algorithm has higher convergence rate under the condition of ensuring population diversity, so the algorithm has certain local and global search capability.
The core of the COA is the social structure and survival adaptability of suburban wolf population, the algorithm is mainly divided into population initialization and random grouping, suburban wolves growing in each group, birth and death of suburban wolves in the simulated genetics, suburban wolves with the best adaptability screening, and the optimization steps of the COA algorithm comprise:
Step 1: initializing COA parameters, including: the number of suburban wolves, the number of suburban wolves contained in each suburban wolves, the space dimension for solving the optimization problem and the maximum iteration number; the COA parameters are initialized to: suburban wolf population number is 5, each group comprises individual number is 10, and problem dimension is 3 (corresponding to 3 weight parameters of the linear weight superposition module).
Step2: calculating the fitness value of each suburban wolf by using a fitness value function, searching an individual optimal position and a global optimal position, updating the optimal fitness value, and updating the status of the suburban wolf; the specific updating method refers to the traditional suburban wolf algorithm mode, and is not specifically described in detail herein;
Step 3: judging whether the maximum iteration times are reached, if the maximum iteration times are reached, transmitting the obtained optimal parameters to a linear weight superposition module, and if the optimal parameters do not reach the requirements, carrying out iteration optimization.
Given fitness value functions of smart search algorithms and parameters to be updated, it is clear to a person skilled in the art how to update target parameters with smart search algorithms, according to the characteristics of different smart search algorithms, common smart search algorithms include: suburban wolf algorithm, particle swarm algorithm, whale algorithm, bald eagle algorithm or ant colony algorithm, etc.
Preferably, the correcting the initial wind speed comprehensive prediction result by using the trained error correction model includes:
Calculating the deviation between the comprehensive predicted result of the initial wind speed of the time step t k∈[t-T5, t-1 and the real wind speed of the time step t k, and constructing a composition error sequence
Inputting an error sequence P t into an LSTM network model to predict the deviation between the comprehensive predicted result of the initial wind speed of the current time step t and the real wind speed of the current time step t;
According to the initial wind speed comprehensive prediction result of the current time step t and the deviation addition between the initial wind speed comprehensive prediction result of the current time step t and the real wind speed of the current time step t, the final wind speed prediction result of the current time step t is obtained, the training process of the error correction model can be based on the actual application process of the error correction model, and the parameters of the error correction model can be updated by adopting a common cross entropy loss function.
In another aspect, the present invention provides a wind speed prediction system based on multi-modal data, where the system is applied to the wind speed prediction method based on multi-modal data, and the wind speed prediction method includes:
The data acquisition module is used for acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
the SSA noise reduction module is used for denoising the historical wind speed sequence to obtain a first intermediate wind speed sequence;
The wavelet transformation module is used for carrying out wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
the data processing module is used for constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
The matrix slicing module is used for dividing the second wind speed matrix according to a sliding window and slicing combination mode to obtain a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1], wherein T 1＞T2＞T3＞T4 is more than 1;
The feature extraction module is used for respectively inputting a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1] into the 2DCNN convolutional neural network for feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature;
the characteristic splicing module is used for respectively splicing the first intermediate characteristic, the second intermediate characteristic and the third intermediate characteristic with the external environment characteristic of the current time step t to obtain a first comprehensive characteristic, a second comprehensive characteristic and a third comprehensive characteristic;
the GRU prediction module is used for inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into the GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
The linear weight superposition module is used for inputting the first wind speed prediction result, the second wind speed prediction result and the third wind speed prediction result of the current time step t into the trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the current time step t;
The error correction module is used for correcting the comprehensive wind speed prediction result of the current time step t to obtain the final wind speed prediction result of the current time step t.
Those skilled in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, where the program may be stored in a non-volatile computer readable storage medium, and where the program, when executed, may include processes in the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.
Claims (7)
1. A method of predicting wind speed based on multimodal data, comprising:
S1: acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
S2: denoising the historical wind speed sequence by using an SSA algorithm to obtain a first intermediate wind speed sequence;
S3: performing wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
s4: constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
S5: dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ T-T 2, T-1], a matrix slice with a time step length of [ T-T 3, T-1] and a matrix slice with a time step length of [ T-T 4, T-1], and respectively inputting the three divided matrix slices into a 2DCNN convolutional neural network to perform feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature, wherein T 1>T2>T3>T4 >1;
S6: respectively splicing the first intermediate feature, the second intermediate feature and the third intermediate feature with the external environment feature of the current time step t to obtain a first comprehensive feature, a second comprehensive feature and a third comprehensive feature;
S7: respectively inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into a GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
S8: inputting the first wind speed predicted result, the second wind speed predicted result and the third wind speed predicted result of the current time step t into a trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive predicted result of the current time step t; and correcting the initial wind speed comprehensive prediction result of the current time step t by using the trained error correction model to obtain the final wind speed prediction result of the current time step t.
2. A method of wind speed prediction based on multimodal data as claimed in claim 1, wherein the external environmental features include: at least one of the environment temperature, the fan temperature, the environment humidity and the environment wind direction after normalization treatment.
3. A method of predicting wind speed based on multimodal data as claimed in claim 1, wherein constructing a first wind speed matrix from the N second intermediate wind speed sequences comprises:
Let X n, N ε {1,2, … N } denote the nth second intermediate wind speed sequence, then the first wind speed matrix is expressed as:
Where T represents the transpose and X represents the first wind speed matrix.
4. The method for predicting wind speed based on multi-modal data according to claim 1, wherein the initial wind speed comprehensive prediction result of the current time step t comprises:
Wherein, The method comprises the steps of representing an initial wind speed comprehensive prediction result of a current time step t, W 1、W2 and W 3 representing weight parameters of a linear weight superposition module, Y t1 representing a first wind speed prediction result of the current time step t, Y t2 representing a second wind speed prediction result of the current time step t, and Y t3 representing a third wind speed prediction result of the current time step t.
5. The method of claim 4, wherein the training process of the linear weight superposition module comprises:
Dividing a second wind speed matrix according to a sliding window and slice combination mode to obtain a matrix slice with a time step length of [ t i-T2,ti -1], a matrix slice with a time step length of [ t i-T3,ti -1] and a matrix slice with a time step length of [ t i-T4,ti -1], [ t i∈[t-T1+T2, t-1];
Generating a training sample set by taking a matrix slice with the time step length of [ t i-T2,ti -1], a matrix slice with the time step length of [ t i-T3,ti -1] and a matrix slice with the time step length of [ t i-T4,ti -1] as training samples and taking the real wind speed of the time step t i as a label of the training samples;
For each training sample, respectively splicing a matrix slice with the time step of [ t i-T2,ti -1], a matrix slice with the time step of [ t i-T3,ti -1] and a matrix slice with the time step of [ t i-T4,ti -1] with external environment features of the time step of t i to obtain a first training comprehensive feature, a second training comprehensive feature and a third training comprehensive feature;
Respectively inputting the first training comprehensive feature, the second training comprehensive feature and the third training comprehensive feature into a GRU prediction model to obtain a first training wind speed prediction result, a second training wind speed prediction result and a third training wind speed prediction result of a time step t i;
Inputting the first training wind speed prediction result, the second training wind speed prediction result and the third training wind speed prediction result of the time step t i into a linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the time step t i;
And taking the sum of the initial wind speed comprehensive prediction results of all training samples and the deviation value of the real wind speed as a fitness value function, and optimizing the weight parameters of the linear weight superposition module by using an intelligent search algorithm with the minimum fitness value as an optimization target to obtain the optimal weight parameters of the linear weight superposition module.
6. The method of claim 5, wherein correcting the initial wind speed composite prediction result using the trained error correction model comprises:
Calculating the deviation between the comprehensive predicted result of the initial wind speed of the time step t k∈[t-T5, t-1 and the real wind speed of the time step t k, and constructing a composition error sequence 1＜T5＜T1；
Inputting an error sequence P t into an LSTM network model to predict the deviation between the comprehensive predicted result of the initial wind speed of the current time step t and the real wind speed of the current time step t;
And obtaining a final wind speed prediction result of the current time step t according to the initial wind speed comprehensive prediction result of the current time step t and the deviation addition between the initial wind speed comprehensive prediction result of the current time step t and the real wind speed of the current time step t.
7. A multi-modal data based wind speed prediction system, the system being applied to a multi-modal data based wind speed prediction method as claimed in any one of claims 1 to 6, comprising:
The data acquisition module is used for acquiring a historical wind speed sequence of the time step [ T-T 1, T-1] and external environment characteristics of the time step [ T-T 1, T ];
the SSA noise reduction module is used for denoising the historical wind speed sequence to obtain a first intermediate wind speed sequence;
The wavelet transformation module is used for carrying out wavelet transformation on the first intermediate wind speed sequence to obtain N second intermediate wind speed sequences, wherein N represents the level of wavelet decomposition;
the data processing module is used for constructing a first wind speed matrix according to the N second intermediate wind speed sequences, and carrying out normalization processing on the first wind speed matrix to obtain a second wind speed matrix;
The matrix slicing module is used for dividing the second wind speed matrix according to a sliding window and slicing combination mode to obtain a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1], wherein T 1＞T2＞T3＞T4 is more than 1;
The feature extraction module is used for respectively inputting a matrix slice with the time step length of [ T-T 2, T-1], a matrix slice with the time step length of [ T-T 3, T-1] and a matrix slice with the time step length of [ T-T 4, T-1] into the 2DCNN convolutional neural network for feature extraction to obtain a first intermediate feature, a second intermediate feature and a third intermediate feature;
the characteristic splicing module is used for respectively splicing the first intermediate characteristic, the second intermediate characteristic and the third intermediate characteristic with the external environment characteristic of the current time step t to obtain a first comprehensive characteristic, a second comprehensive characteristic and a third comprehensive characteristic;
the GRU prediction module is used for inputting the first comprehensive feature, the second comprehensive feature and the third comprehensive feature into the GRU prediction model to obtain a first wind speed prediction result, a second wind speed prediction result and a third wind speed prediction result of the current time step t;
The linear weight superposition module is used for inputting the first wind speed prediction result, the second wind speed prediction result and the third wind speed prediction result of the current time step t into the trained linear weight superposition module for superposition to obtain an initial wind speed comprehensive prediction result of the current time step t;
The error correction module is used for correcting the comprehensive wind speed prediction result of the current time step t to obtain the final wind speed prediction result of the current time step t.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410154819.7A CN118013457A (en) | 2024-02-04 | 2024-02-04 | Wind speed prediction method and system based on multi-mode data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410154819.7A CN118013457A (en) | 2024-02-04 | 2024-02-04 | Wind speed prediction method and system based on multi-mode data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118013457A true CN118013457A (en) | 2024-05-10 |
Family
ID=90949468
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410154819.7A Pending CN118013457A (en) | 2024-02-04 | 2024-02-04 | Wind speed prediction method and system based on multi-mode data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118013457A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118211494A (en) * | 2024-05-21 | 2024-06-18 | 哈尔滨工业大学（威海） | Wind speed prediction hybrid model construction method and system based on correlation matrix |
| CN118245733A (en) * | 2024-05-21 | 2024-06-25 | 深圳市北电仪表有限公司 | Data preprocessing method based on operation error monitoring model and intelligent ammeter |
| CN118673464A (en) * | 2024-08-22 | 2024-09-20 | 南京信息工程大学 | Wind speed prediction method, device and storage medium |
-
2024
- 2024-02-04 CN CN202410154819.7A patent/CN118013457A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118211494A (en) * | 2024-05-21 | 2024-06-18 | 哈尔滨工业大学（威海） | Wind speed prediction hybrid model construction method and system based on correlation matrix |
| CN118245733A (en) * | 2024-05-21 | 2024-06-25 | 深圳市北电仪表有限公司 | Data preprocessing method based on operation error monitoring model and intelligent ammeter |
| CN118673464A (en) * | 2024-08-22 | 2024-09-20 | 南京信息工程大学 | Wind speed prediction method, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shamshirband et al. | A survey of deep learning techniques: application in wind and solar energy resources | |
| Alencar et al. | Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil | |
| CN118013457A (en) | Wind speed prediction method and system based on multi-mode data | |
| CN109829587A (en) | Zonule grade ultra-short term and method for visualizing based on depth LSTM network | |
| CN109948845A (en) | A kind of distribution network load shot and long term Memory Neural Networks prediction technique | |
| CN114548591B (en) | Sequential data prediction method and system based on mixed deep learning model and Stacking | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| Liu et al. | Heating load forecasting for combined heat and power plants via strand-based LSTM | |
| Cardoso et al. | Improve irrigation timing decision for agriculture using real time data and machine learning | |
| CN106709588A (en) | Prediction model construction method and equipment and real-time prediction method and equipment | |
| CN113344243B (en) | Wind speed prediction method and system for optimizing ELM (ELM) based on improved Harris eagle algorithm | |
| Saffari et al. | Deep convolutional graph rough variational auto-encoder for short-term photovoltaic power forecasting | |
| CN117688846A (en) | Reinforced learning prediction method and system for building energy consumption and storage medium | |
| CN110738363B (en) | Photovoltaic power generation power prediction method | |
| CN117421566A (en) | Photovoltaic power generation power prediction method based on IMRFO-StemNN | |
| Tripathy et al. | Weather forecasting using ANN and PSO | |
| Setiawan et al. | Indoor Climate Prediction Using Attention-Based Sequence-to-Sequence Neural Network | |
| Bian et al. | A novel study on Power Consumption of an HVAC system using CatBoost and AdaBoost algorithms combined with the Metaheuristic Algorithms | |
| Ravi et al. | Design of Deep Learning Model for Predicting Rainfall | |
| CN116613732A (en) | Multi-element load prediction method and system based on SHAP value selection strategy | |
| Alankar et al. | Predictive analytics for weather forecasting using back propagation and resilient back propagation neural networks | |
| CN115600498A (en) | Wind speed forecast correction method based on artificial neural network | |
| CN113361476B (en) | Zhang Heng one-number pre-earthquake abnormal signal identification method based on artificial intelligence technology | |
| CN107679478A (en) | The extracting method and system of transmission line of electricity space load state | |
| Zhou et al. | A short-term wind speed prediction method utilizing rolling decomposition and time-series extension to avoid information leakage |
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 |