CN114943368A - Sea surface wind speed prediction method based on Transformer - Google Patents
Sea surface wind speed prediction method based on Transformer Download PDFInfo
- Publication number
- CN114943368A CN114943368A CN202210445463.3A CN202210445463A CN114943368A CN 114943368 A CN114943368 A CN 114943368A CN 202210445463 A CN202210445463 A CN 202210445463A CN 114943368 A CN114943368 A CN 114943368A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- speed field
- module
- transformer
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a Transformer-based sea surface wind speed prediction method which comprises a feature extraction module, a Transformer module and a wind speed field reconstruction module, wherein when sea surface wind speed prediction is carried out, a complete wind speed field with the past 10 time steps is input, and the input wind speed field is coded into a feature vector of the wind speed field at each moment in a high-dimensional feature space through the feature extraction module. Inputting the feature vectors of the high-dimensional wind speed field at the past 10 moments into a Transformer module, wherein the Transformer module is used for establishing global long-term dependence among the feature vectors of the wind speed field at different moments and deducing the feature vectors of the wind speed field prediction result at the future 10 moments; and inputting the wind speed field characteristic vectors of 10 moments in the future into a wind speed field reconstruction module based on a convolutional neural network, and decoding the wind speed field characteristic vectors into a complete wind speed field by the wind speed field reconstruction module.
Description
Technical Field
The invention relates to a prediction algorithm of sea surface wind speed, in particular to a multi-step advanced wind speed prediction algorithm based on a Transformer.
Background
The prediction of sea surface wind speed plays a vital role in trade shipping and energy development. From the perspective of trade shipping, the safety and economic benefits of offshore operating personnel are damaged by extreme wind speed, and the normal operation of shipping can be effectively guaranteed by early warning of the extreme wind speed; from the perspective of energy development, the uncertainty of offshore wind resources has adverse effects on the stable operation of an offshore wind farm, and the accurate and timely wind speed prediction can effectively improve the safety and reliability of a wind power generation system.
Wind speed modeling and prediction methods can be divided into 3 categories: numerical mode methods, data-driven methods, and artificial intelligence prediction methods. The numerical mode method utilizes physics knowledge, oceanographic knowledge, atmospheric knowledge, fluid mechanics knowledge and the like to model the change process of the wind speed, and constructs a complex partial differential equation to solve the future wind speed; the data-driven method learns the change process of the wind speed from massive historical data by designing a learnable statistical model; the artificial intelligence prediction method utilizes the nonlinear fitting capability of the deep learning technology and the neural network to construct a learnable wind speed prediction network.
Disclosure of Invention
The invention provides a more accurate sea surface wind speed prediction method based on artificial intelligence; based on the problem of introducing a Transformer into wind speed prediction, a feature extraction module capable of excavating high-dimensional space representation of a wind speed field is designed firstly, and high-dimensional feature representation of the wind speed field with 10 past time steps is extracted; thirdly, capturing a long-term global relation represented by the high-dimensional characteristics by using a Transformer, realizing multi-step advanced prediction of the high-dimensional characteristics of the wind speed field at 10 moments in the future, and finally designing a wind speed field reconstruction module which restores the high-dimensional characteristics of the wind speed field into a complete wind speed field; the technical scheme is as follows:
a sea surface wind speed prediction method based on a Transformer comprises a feature extraction module, a Transformer module and a wind speed field reconstruction module, wherein when sea surface wind speed prediction is carried out, a complete wind speed field with the past 10 time steps is input, and the input wind speed field is coded into a feature vector of the wind speed field at each moment in a high-dimensional feature space through the feature extraction module; inputting the feature vectors of the high-dimensional wind speed field at the past 10 moments into a Transformer module, wherein the Transformer module is used for establishing global long-term dependence among the feature vectors of the wind speed field at different moments and deducing the feature vectors of the wind speed field prediction result at the future 10 moments; and inputting the wind speed field characteristic vectors of 10 moments in the future into a wind speed field reconstruction module based on a convolutional neural network, and decoding the wind speed field characteristic vectors into a complete wind speed field by the wind speed field reconstruction module.
Further, the feature extraction module and the wind field reconstruction module comprise the following contents:
the characteristic extraction module is used for extracting characteristics from the wind speed field, the wind speed field reconstruction module is used for reconstructing the wind speed field from the characteristics, and the wind speed field at a single moment is simultaneously used as input data and a label for training the characteristic extraction module and the wind speed field reconstruction module;
setting the wind speed field characteristic extraction module as h (-), the wind field reconstruction module as f (-), and the input wind speed field as I epsilon R H×W And H and W are respectively the length and the width of the wind speed field, and a high-dimensional feature x is extracted from the input wind speed field i through a feature extraction module and expressed as:
x=h(i)
the wind speed field reconstruction module takes the high-dimensional characteristics x as the input of the wind speed field reconstruction module and reconstructs the high-dimensional characteristics into a complete wind speed fieldThe formula is expressed as:
reconstructed wind velocity fieldThe smaller the error between the wind speed field and the input wind speed field i is, the more fully the model is trained, and the training uses a mean square error loss function to update the model parameters; by self-runningIn a supervised training mode, the wind speed field feature extraction module has the capability of extracting high-dimensional feature x from a wind speed field, and the wind speed field reconstruction module decodes the high-dimensional feature x into a complete wind field;
further, the Transformer module adopts the following steps:
step 1: feature extraction
Coding the wind speed field of each moment in the input data by using a feature extraction module to obtain a feature vector of the wind speed field of each moment in a high-dimensional feature space, wherein the feature vector is expressed as (X) 1 ,X 2 ,...,X t ) T ;
Step 2: position coding and time information generation
The Transformer encodes the position information of each input data, which is called position embedding; the information of the month, the year, the time and the like of the sequence data is coded, which is called as time embedding, and the input of the global time information is helpful for improving the capability of the model for capturing long-term dependence; adding position embedding information and time embedding information to input data after dimension conversion;
and step 3: encoding input data using Transformer encoder
And 4, step 4: generating a full 0 vector with the length of 10 as a placeholder;
and 5: splicing the placeholders and the intermediate result of the encoder to be used as the input of a Transformer decoder;
and 6: prediction of output data using a Transformer decoder: in the decoder, input data is subjected to multi-head probability sparse self-attention operation with a mask, then multi-head self-attention operation is carried out on the input data and an intermediate result output by an encoder, finally dimensionality of data output is adjusted through a full connection layer, and a prediction result Y is obtained (Y is equal to Y) t+1 ,Y t+2 ,...,Y t+10 ) T ;
And 7: coding the wind speed field at each moment by using a feature extraction module to obtain a feature vector of the wind speed field at each moment in a high-dimensional feature space, wherein the feature vector is expressed as Z-Z (Z-Z) 1 ,Z 2 ,...,Z t ) T ;
And 8: performing Euclidean distance calculation on Y and Z to obtain a characteristic error, and optimizing a Transformer model through back propagation;
and step 9: and (5) repeating the steps 1-8 until convergence, and obtaining a trained prediction model.
According to the transform-based sea surface wind speed prediction algorithm, an unsupervised self-coding structure is used for training a wind speed field coding and decoding module, effective and discriminative feature representation can be obtained by extracting images, and a real wind speed field is decoded through discriminative features; and learning the time sequence change represented by the wind field characteristics through a supervised Transformer model, and realizing the prediction of the future wind speed field. The user inputs the wind speed fields at the past ten moments, and the wind speed prediction algorithm can accurately predict the wind speed field changes at the future 10 moments. Quantitative and qualitative results obtained by experiments in the covered sea areas of the yellow sea and the east sea show the effectiveness of the wind speed prediction algorithm.
Drawings
FIG. 1 Transformer based sea surface wind speed prediction
FIG. 2Transformer Structure
FIG. 3 shows the results of the qualitative experiment, (a) and the quantitative experiment (b)
Detailed Description
The multi-step advanced wind speed prediction algorithm based on the Transformer comprises three modules: the system comprises a feature extraction module, a Transformer module and a wind speed field reconstruction module, and the whole system is shown in figure 1. The prediction algorithm provided by the invention comprises two stages of model training and application testing, wherein each module firstly determines model parameters in the training stage and then predicts the wind speed field in the application testing stage. The method comprises the following specific steps:
the first step is as follows: training feature extraction module and wind field reconstruction module
(1) Preparing input data and labels
The training feature extraction module is used for extracting features from the wind speed field, the wind speed field reconstruction module is used for reconstructing the wind speed field from the features, and the wind speed field at a single moment is simultaneously used as input data and a label for training the feature extraction module and the wind speed field reconstruction module.
(2) And (5) training a model.
Setting a wind speed field characteristic extraction module as h (-), a wind field reconstruction module as f (-), and an input wind speed field as I epsilon R H×W The method comprises the following steps of firstly processing X belongs to X through a characteristic extraction module, extracting a high-dimensional characteristic X from an input wind speed field i through the wind speed field characteristic extraction module, wherein H and W are respectively the length and the width of the wind speed field, and the formula is as follows:
x=h(i)
the wind speed field reconstruction module takes the high-dimensional characteristics x as the input of the wind speed field reconstruction module and reconstructs the high-dimensional characteristics into a complete wind speed fieldThe formula is expressed as:
self-supervised training mode, wind speed field to be reconstructedIs the same as the input wind speed field i, thus a reconstructed wind speed fieldThe smaller the error with the input wind speed field i, the more fully the model is trained. Therefore, the training at this stage uses the mean square error loss function to update the model parameters, which is formulated as
And an intermediate result x of the training process is the high-dimensional characteristic representation of the wind speed field. Through a self-supervision training mode, the wind speed field feature extraction module has the capability of extracting high-dimensional features x from the wind speed field, and the wind speed field reconstruction module can decode the key features x into a complete wind field.
The second step is that: training Transformer module
(1) Preparing input data and labels
The function of the algorithm is to predict the complete wind speed field 10 time steps in the future according to the complete wind speed field 10 time steps in the past. Therefore, the wind speed fields are sorted in time order, and 10 consecutive wind speed fields are taken as input data of the prediction model, and 10 consecutive wind speed fields after the input data are taken as tag data of the prediction model.
(2) Training of Transformer modules
Step 1: feature extraction
Coding the wind speed field at each moment in the input data by using a feature extraction module to obtain a feature vector of the wind speed field at each moment in a high-dimensional feature space, wherein the feature vector is expressed as (X) 1 ,X 2 ,...,X t ) T 。
Step 2: position coding and time information generation
For the time series prediction problem, the order relationship between data is important, in order to make the position relationship (i.e. order structure) not lost after the data is input into the model, the Transformer encodes (i.e. position embeds) the position information of each input data, and the specific operation of the position encoding PE is as follows:
in the formula, pos represents a position (sequence order), j represents a dimension, and d model Is the input dimension of the model. In addition, considering the importance of time information to the time series prediction problem, such as that summer in the east coast is influenced by the season weather, a lot of gusts of wind exist, typhoon is logged in sometimes, and the wind speed in winter is relatively stable, the Transformer encodes (i.e. embeds time) the information of month, year, time and the like of the sequence data, and the input of the global time information is helpful for improving the capability of the model for capturing long-term dependence. In the end of this process,the position embedding information and the time embedding information are added to the input data after the dimension is transformed.
And step 3: encoding input data using Transformer encoder
The encoder is internally formed by stacking a Multi-head probability sparse Self-attention (Multi-head ProbSparse Self-attention) module and a 'distillation' (Distilling) mechanism module.
Multiple heads allow parallel operation of the probabilistic sparseness self-attention mechanism, the purpose of the "distillation" operation being to compress the characteristic dimensions and extract the main information. After multiple operations of the multi-head probability sparse self-attention module and the distillation mechanism module, the encoder outputs an intermediate result.
And 4, step 4: placeholder generation
An all 0 vector of length 10 is generated as a placeholder.
And 5: the placeholders are concatenated with the encoder intermediate results as input to the transform decoder.
Step 6: prediction of output data using a Transformer decoder
In the decoder, input data is subjected to multi-head probability sparse self-attention operation with mask (Masked), multi-head self-attention operation is performed with an intermediate result output by the encoder, finally dimensionality of data output is adjusted through a full connection layer, and a prediction result Y is obtained (Y is equal to Y) t+1 ,Y t+2 ,...,Y t+10 ) T 。
And 7: and coding the wind speed field at each moment in the label by using a feature extraction module to obtain a feature vector of the wind speed field at each moment in a high-dimensional feature space, wherein the feature vector is expressed as Z-Z (Z-Z) 1 ,Z 2 ,...,Z t ) T 。
And 8: and performing Euclidean distance calculation on Y and Z to obtain a characteristic error, and optimizing a Transformer model through back propagation.
And step 9: and repeating the steps 1-8 until the model converges.
The third step: predictive model testing
The input of the algorithm is a complete wind speed field of the past 10 time steps, and the input wind speed field is coded into a feature vector of the wind speed field in a high-dimensional feature space at each moment through a feature extraction module; then, feature vectors of the high-dimensional wind speed field at the past 10 moments are input into a Transformer module, the Transformer module can establish global long-term dependence among the feature vectors of the wind speed field at different moments, and the feature vectors of the wind speed field prediction result at the future 10 moments are deduced according to the global long-term dependence; and finally, inputting the wind speed field characteristic vectors of 10 moments in the future into a wind speed field reconstruction module based on a convolutional neural network, and decoding the wind speed field characteristic vectors into a complete wind speed field by the wind speed field reconstruction module.
The following examples are given by way of illustration.
The encoder of the wind speed field encoding and decoding module is formed by stacking convolution with the size of 3 multiplied by 3, and a batch normalization and a Leaky ReLU activation function with the slope of 0.2 are added behind each convolution layer. The wind speed field outputs a characteristic vector of 2048 dimensions after passing through the encoder, the decoder consists of a reverse convolution module with the size of 3 multiplied by 3, and the scales of input and output pictures of the wind speed field coding and decoding module are the same. In the training phase, an Adam optimizer is adopted, and the initial learning rate is 3 multiplied by 10 -3 For each 30 rounds of training, the learning rate is set to 0.1 times of the original rate, and a total of 100 rounds of training are performed.
Training completion wind speed field codec fixed parameters are used in the training of the Transformer. In the training stage, a pre-training model is used to accelerate the convergence rate of the model, an SGD optimizer is adopted, the initial learning rate is 0.1, the learning rate is set to be 0.1 time of the original learning rate every 30 rounds of training, and 100 rounds of training are performed. At this point, the training phase of the transform network is completed.
For evaluating the performance of the algorithm, RMSE (root mean square error) and MAE (mean absolute error) are used as evaluation indexes. The model provided by the algorithm is experimentally verified on a CCMP sea surface wind speed data set in coastal areas of east China, the time range of data selection is 2018 to 2021 for 4 years, the distance between adjacent moments is 6 hours, and the data are calculated according to the following formula of 0.75: the data set is divided according to the proportion of 0.25, and in order to ensure time continuity, data in 2018 and 2020 are used as a training and verifying set, and data in 2021 are used as a testing set. The experimental results on the test set can be referred to fig. 3.
Claims (3)
1. A sea surface wind speed prediction method based on a Transformer comprises a feature extraction module, a Transformer module and a wind speed field reconstruction module, wherein when sea surface wind speed prediction is carried out, a complete wind speed field with the past 10 time steps is input, and the input wind speed field is coded into a feature vector of the wind speed field at each moment in a high-dimensional feature space through the feature extraction module; inputting the feature vectors of the high-dimensional wind speed field at the past 10 moments into a Transformer module, wherein the Transformer module is used for establishing global long-term dependence among the feature vectors of the wind speed field at different moments and deducing the feature vectors of the wind speed field prediction result at the future 10 moments; and inputting the wind speed field characteristic vectors of 10 moments in the future into a wind speed field reconstruction module based on a convolutional neural network, and decoding the wind speed field characteristic vectors into a complete wind speed field by the wind speed field reconstruction module.
2. The method for predicting the wind speed on the sea surface according to claim 1, wherein the feature extraction module and the wind field reconstruction module comprise the following contents:
the characteristic extraction module is used for extracting characteristics from the wind speed field, the wind speed field reconstruction module is used for reconstructing the wind speed field from the characteristics, and the wind speed field at a single moment is simultaneously used as input data and a label for training the characteristic extraction module and the wind speed field reconstruction module;
setting a wind speed field characteristic extraction module as h (-), a wind field reconstruction module as f (-), and an input wind speed field as I epsilon R H×W And H and W are respectively the length and the width of a wind speed field, and a high-dimensional feature x is extracted from an input wind speed field i through a feature extraction module and is expressed as follows:
x=h(i)
the wind speed field reconstruction module takes the high-dimensional characteristics x as the input of the wind speed field reconstruction module and reconstructs the high-dimensional characteristics into a complete wind speed fieldThe formula is expressed as:
reconstructed wind speed fieldThe smaller the error between the wind speed field and the input wind speed field i is, the more fully the model is trained, and the training uses a mean square error loss function to update the model parameters; through a self-supervision training mode, the wind speed field feature extraction module has the capability of extracting high-dimensional features x from the wind speed field, and the wind speed field reconstruction module decodes the high-dimensional features x into a complete wind field.
3. The method for predicting the wind speed on the sea surface according to claim 1, wherein the Transformer module adopts the following steps:
step 1: feature extraction
Coding the wind speed field of each moment in the input data by using a feature extraction module to obtain a feature vector of the wind speed field of each moment in a high-dimensional feature space, wherein the feature vector is expressed as (X) 1 ,X 2 ,...,X t ) T ;
And 2, step: position coding and time information generation
The Transformer encodes the position information of each input data, which is called position embedding; the information of the month, the year, the time and the like of the sequence data is coded, which is called as time embedding, and the input of the global time information is helpful for improving the capability of the model for capturing long-term dependence; adding position embedding information and time embedding information to input data after dimension conversion;
and 3, step 3: encoding input data using a Transformer encoder
And 4, step 4: generating a full 0 vector with the length of 10 as a placeholder;
and 5: splicing the placeholders and the intermediate result of the encoder to be used as the input of a Transformer decoder;
step 6: prediction of output data using a Transformer decoder: inside the decoder, the input data is first processed through the multi-head probability sparse self-attention operation with the mask and then processed withPerforming multi-head self-attention operation on the intermediate result output by the encoder, and finally adjusting the dimension of data output through the full-connection layer to obtain a prediction result Y (Y ═ Y) t+1 ,Y t+2 ,...,Y t+10 ) T ;
And 7: coding the wind speed field at each moment by using a feature extraction module to obtain a feature vector of the wind speed field at each moment in a high-dimensional feature space, wherein the feature vector is expressed as Z-Z (Z-Z) 1 ,Z 2 ,...,Z t ) T ;
And 8: performing Euclidean distance calculation on Y and Z to obtain a characteristic error, and optimizing a Transformer model through back propagation;
and step 9: and (5) repeating the steps 1-8 until convergence, and obtaining the trained prediction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210445463.3A CN114943368A (en) | 2022-04-26 | 2022-04-26 | Sea surface wind speed prediction method based on Transformer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210445463.3A CN114943368A (en) | 2022-04-26 | 2022-04-26 | Sea surface wind speed prediction method based on Transformer |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114943368A true CN114943368A (en) | 2022-08-26 |
Family
ID=82906823
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210445463.3A Pending CN114943368A (en) | 2022-04-26 | 2022-04-26 | Sea surface wind speed prediction method based on Transformer |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114943368A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545361A (en) * | 2022-12-02 | 2022-12-30 | 国网湖北省电力有限公司信息通信公司 | Method, system and medium for predicting climate environment of power grid transmission line |
-
2022
- 2022-04-26 CN CN202210445463.3A patent/CN114943368A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545361A (en) * | 2022-12-02 | 2022-12-30 | 国网湖北省电力有限公司信息通信公司 | Method, system and medium for predicting climate environment of power grid transmission line |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110209801B (en) | Text abstract automatic generation method based on self-attention network | |
Saeed et al. | Hybrid bidirectional LSTM model for short-term wind speed interval prediction | |
CN111784041B (en) | Wind power prediction method and system based on graph convolution neural network | |
CN116070799B (en) | Photovoltaic power generation amount prediction system and method based on attention and deep learning | |
CN115392595B (en) | Time-space short-term wind speed prediction method and system based on graph convolution neural network and Transformer | |
CN113988449A (en) | Wind power prediction method based on Transformer model | |
CN111242351A (en) | Tropical cyclone track prediction method based on self-encoder and GRU neural network | |
Li et al. | Deep spatio-temporal wind power forecasting | |
CN117096867A (en) | Short-term power load prediction method, device, system and storage medium | |
CN113868432A (en) | Automatic knowledge graph construction method and system for iron and steel manufacturing enterprises | |
CN117076931A (en) | Time sequence data prediction method and system based on conditional diffusion model | |
CN114943368A (en) | Sea surface wind speed prediction method based on Transformer | |
CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
CN116187508A (en) | Wind turbine generator system fault prediction method integrating knowledge graph and deep learning | |
CN114821569A (en) | Scene text recognition method and system based on attention mechanism | |
CN116596150A (en) | Event prediction method of transform Hoxwell process model based on multi-branch self-attention | |
CN115906857A (en) | Chinese medicine text named entity recognition method based on vocabulary enhancement | |
Tuyen et al. | A combination of novel hybrid deep learning model and quantile regression for short‐term deterministic and probabilistic PV maximum power forecasting | |
CN116757369B (en) | Attention mechanism-based carbon emission analysis method and system | |
CN112465929A (en) | Image generation method based on improved graph convolution network | |
CN116483036A (en) | Transformer-based self-encoder soft measurement modeling method | |
CN112735604B (en) | Novel coronavirus classification method based on deep learning algorithm | |
CN114896969A (en) | Method for extracting aspect words based on deep learning | |
CN112052685A (en) | End-to-end text entity relationship identification method based on two-dimensional time sequence network | |
CN116826727B (en) | Ultra-short-term wind power prediction method and prediction system based on time sequence representation and multistage attention |
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 |