CN115099296A - Sea wave height prediction method based on deep learning algorithm - Google Patents

Sea wave height prediction method based on deep learning algorithm Download PDF

Info

Publication number
CN115099296A
CN115099296A CN202210398198.8A CN202210398198A CN115099296A CN 115099296 A CN115099296 A CN 115099296A CN 202210398198 A CN202210398198 A CN 202210398198A CN 115099296 A CN115099296 A CN 115099296A
Authority
CN
China
Prior art keywords
sea wave
sea
data
data information
wave height
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
Application number
CN202210398198.8A
Other languages
Chinese (zh)
Inventor
梅春晓
谭建鑫
卢盛欣
孟雷
于力强
张清清
张国峰
侯元柏
吴伟强
李练兵
高国强
韩旭
李永建
李佳琪
陈程
贾超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hebei Construction Investment Offshore Wind Power Co ltd
Hebei University of Technology
Original Assignee
Hebei Construction Investment Offshore Wind Power Co ltd
Hebei University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hebei Construction Investment Offshore Wind Power Co ltd, Hebei University of Technology filed Critical Hebei Construction Investment Offshore Wind Power Co ltd
Priority to CN202210398198.8A priority Critical patent/CN115099296A/en
Publication of CN115099296A publication Critical patent/CN115099296A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • 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

Abstract

The invention relates to a sea wave height prediction method based on a deep learning algorithm. The method comprises the steps of obtaining sea wave data information, analyzing and preprocessing the sea wave data information, analyzing the correlation between the sea wave data information and the height of the sea wave, performing characteristic extraction on the sea wave data information by adopting a Pearson, GRA and PCA combined method, and realizing optimization on model input characteristic quantity. And in the second stage, the processed data is used as the input of the model, the network structure is determined, and a network model of a bidirectional gating circulation unit based on a thought evolution algorithm is established. And in the third stage, setting MEA parameters to obtain optimal weight and threshold values, and training an MEA-BIGRU network model. And finally, establishing a BP neural network model and a BIGRU network model at the same time for comparison and verification, thereby realizing accurate prediction of the wave height of the sea waves.

Description

Sea wave height prediction method based on deep learning algorithm
Technical Field
The invention belongs to the technical field of sea wave height prediction, relates to a sea wave height prediction method based on a thought evolution algorithm and deep learning, and particularly relates to a multivariable sea wave height prediction method based on a thought evolution algorithm and a bidirectional gating cycle unit (BiGRU).
Background
Nowadays, the offshore wind power industry is actively developing worldwide, and offshore wind power generation has gradually become an important development direction for research and development of renewable energy sources. Naturally, offshore wind power business in China is required to closely follow world trend, and comprehensive development of offshore wind power resources is enhanced to improve resource utilization rate. Since the operation and maintenance of offshore wind power is affected by weather and sea conditions, significant costs including maritime, marine, long-term outages, etc. are incurred in addition to the costs of replacing or maintaining the components themselves. Therefore, the real-time refined sea wave prediction system has important significance for safe operation and maintenance of the offshore wind power plant, improvement of operation and maintenance efficiency and reduction of cost.
With the wide application of machine learning, the application prospect of predicting the marine environment by using a machine learning algorithm is great. The Support Vector Machine (SVM) is a very perfect classification and regression model in machine learning, and has excellent generalization capability. In documents James S C, Zhang Y, O' Donncha F.A mac bone left learning frame to for wave conditions [ J ] Coastal Engineering,2018,137:1-10, sea wave elements are predicted by machine learning, and effective wave height and characteristic period of sea waves are predicted by using a Support Vector Machine (SVM) and a multi-layer sensing Machine (MLP), respectively, so that application of machine learning in a sea wave prediction direction is realized. In recent years, deep learning algorithms gradually show the advantages of the deep learning algorithms in the aspect of wave height prediction. The output information of the long-short term memory network (LSTM) is controlled by an input gate, a forgetting gate and an output gate together, and a gating circulation unit (GRU) is a modified LSTM which integrates the forgetting gate and the input gate into a new gate, reduces network parameters, improves the efficiency of model training and is not easy to overfit. However, the gating cycle unit usually ignores text information in the wave sequence, and cannot effectively capture the time sequence rule in the wave sequence, and a single gating cycle unit cannot quickly and accurately track the change characteristics of the sequence.
The document Shuntao Fan, Nianhao Xiao, Sheng Dong.A novel model to predict significant wave height on long-term memory network [ J ]. Ocean Engineering,2020,205:10-18 proposes a long short-term memory (LSTM) network, which can predict significant wave height quickly and with higher precision. The LSTM network makes 1 hour and 6 hour predictions at 10 stations of different environmental conditions. The LSTM prediction results were obtained using the past 4h wind speed and the past 1h wave height and wind direction as input parameters.
The literature, Rupeng, Santa Claus, Zhouyou, Zhonghua, Wang Zhenhua, Zhengzong, wave height prediction based on variational modal decomposition and attention machine system [ J ] ocean mapping, 2021,41(02):34-39 ] the sea height prediction based on Variational Modal Decomposition (VMD) improves the traditional long short-term memory (LSTM) neural network algorithm on the basis of introducing the attention machine system (AM), and provides a wave height prediction algorithm based on a hybrid model. The algorithm predicts the time series of the sea wave heights through 3 main steps of preprocessing, predicting and reconstructing.
However, the correlation and weight problem of the wave height influence factors are not considered in the algorithm, so that the method extracts the characteristics of the wave data information, analyzes the correlation to determine the weight of the wave data information, and predicts the wave height through the BiGRU.
Disclosure of Invention
The invention aims to overcome the defects that the existing wave height prediction method has large data demand and processing capacity and poor adaptability and cannot fully mine the relation between wave height data, provides a wave height prediction method which has small data demand and processing capacity and good adaptability and can fully mine the relation between the wave height data, and provides a wave height prediction method based on a thought evolution algorithm and deep learning. The method considers the technical field of time sequence prediction, and adopts a sea wave height prediction method based on a thought evolutionary algorithm (MEA) and a bidirectional gating cycle unit (BiGRU) to analyze and preprocess sea wave data information, the processed data is used as the input of a model, the sea wave data information is extracted in a characteristic way by adopting a method of combining Pearson, GRA and PCA, a network structure is determined, a network model of the bidirectional gating cycle unit (MEA-BiGRU) based on the thought evolutionary algorithm is established, MEA parameters are set to obtain the optimal weight and threshold, the MEA-BiGRU network model is trained, and the accurate prediction of the sea wave height is realized;
in order to achieve the purpose, the invention provides the following technical scheme:
a sea wave height prediction method based on a deep learning algorithm is characterized by comprising the steps of obtaining sea wave data information, and analyzing and preprocessing the sea wave data information; the processed data are used as the input of a model, a network structure is determined, a network model of a bidirectional gating circulation unit based on a thought evolution algorithm is established, MEA parameters are set, and an MEA-BiGRU network model is trained; and establishing a BP neural network model and a BiGRU network model for comparison and verification, and realizing accurate prediction of the wave height of the sea wave.
Compared with the prior art, the invention has the beneficial effects that:
1. aiming at the nonlinear relation between the sea wave data information and the sea wave height, the invention provides a Pearson, GRA and PCA combined method for extracting the characteristics of the sea wave data information, so that the optimization on the model input characteristic quantity is realized, the characteristics of the sea wave height are better shown, the information of the sea wave parameters is better covered, the loss of the sea wave parameters is reduced, and the sea wave height is better predicted.
2. In the sea field, the sea wave height prediction, the thinking evolution algorithm and the deep learning are combined to expand the sea wave prediction method, and the weight and the deviation of the model are better optimized based on the thinking evolution algorithm, so that the model can better predict the sea wave height, the error of the sea wave height prediction is reduced, and the prediction precision is improved.
Drawings
FIG. 1 is a general block diagram of the planning method of the present invention;
FIG. 2 is a flow chart of sea wave data information feature quantity optimization;
FIG. 3 is a parameter graph of wave data information;
FIG. 4 is a heat map of sea wave data information correlation;
FIG. 5 is a schematic diagram of a BiGRU structure;
FIG. 6 is a MEA-BiGRU model flow diagram;
fig. 7 is a diagram for predicting the height of ocean waves.
Detailed Description
The technical solutions of the present invention are described in detail below with reference to the accompanying drawings and specific examples, but the scope of the present invention is not limited thereto.
In this embodiment, a structural diagram of the sea wave height prediction is shown in fig. 1, sea wave data information is obtained, data analysis and preprocessing are performed on 9 features of date, time, air temperature, wind direction, wind speed, air pressure, sea temperature, the wave height at the last moment and the wave period at the last moment, correlation between the sea wave data information and the sea wave height is analyzed, Pearson, GRA and PCA correlation analysis methods are respectively adopted to extract the features of the sea wave height, the processed data is used as input of a model to determine a network structure, a network model of a bidirectional gating cycle unit (MEA-BiGRU) based on a thought evolutionary algorithm is established, MEA parameters are set to obtain optimal weight and threshold values, and the MEA-grbiu network model is trained to realize accurate prediction of the sea wave height.
Step one, wave data information in the example is derived from Chinese station observation data of a national ocean science data center (NMDC), and mainly comprises 10248 groups of data of measurement date, time (0-24 hours), air temperature (DEG C), wind direction (DEG C), wind speed (m/s), air pressure (hpa), sea temperature (DEG C), wind wave period(s), wind wave height (m) and the like from 2019 to 2020 8 months in the small Changshan ocean station (122.7 DEG E, 39.2 DEG N). And analyzing and preprocessing the data of the 9 characteristics of the date, the time, the air temperature, the wind direction, the wind speed, the air pressure, the sea temperature, the wave height of the wind and the wave period of the last moment. ()
The sea wave data information is preprocessed, as shown in fig. 2: screening abnormal data by using a Mahalanobis average distance method, replacing abnormal values by using an average method, inputting the preprocessed sea wave data information into python, obtaining a characteristic parameter curve diagram as shown in FIG. 2, and observing the change trend of each characteristic parameter; the wave data information is analyzed, and the linear correlation between each characteristic and the wave height of the waves is measured through a Pearson correlation coefficient (Pearson), wherein the Pearson correlation coefficient changes between [ -1, 1], and the calculation formula is as follows:
Figure BDA0003598350160000031
the example can obtain a value of Pearson correlation based on python, and as shown in the following table 1, it can be found that the linear correlation between the wave data information and the wave height is not good, wherein the correlation between the wind speed is the highest, the correlation between the sea temperature is the next, the correlation between other wave data is obviously poor, and the linear correlation is weak, so that the example selects data with positive correlation for further analysis.
TABLE 1Pearson correlation values
Figure BDA0003598350160000034
And measuring the trend correlation between each characteristic and the wave height of the sea through gray correlation analysis (GRA), wherein the gray correlation coefficient is changed between [0 and 1], and the calculation formula is as follows:
Figure BDA0003598350160000032
in the embodiment, based on python, a GRA correlation value can be obtained, a GRA correlation heat map of wave data information is obtained as shown in fig. 3, and the correlation between the wave data information and the height of waves is obtained, as shown in table 2 below, it can be found that the correlation between the whole wave data information and the height of waves is higher, wherein the correlation between the wind speed is highest, and then the height of waves at the last moment is higher.
TABLE 2GRA correlation values
Figure BDA0003598350160000033
Figure BDA0003598350160000041
Obtaining correlation coefficients of each feature and the wave heights Pearson and GRA, wherein the wave is a strong nonlinear process, and the linear relation between the features is weak, so that the features with positive correlation of Pearson are selected for analysis, and the wind speed, date and time, the wave height at the last moment and the wave period at the last moment are selected; the GRA is selected to be analyzed with the characteristics having a large correlation, and because the correlation values of the GRA are large, the wind speed, the date, the time, the height of the wind wave at the last moment and the period of the wind wave at the last moment are also selected.
The correlation degrees of the two are compared, firstly, the wind speed is selected as an input characteristic, then, fusion processing is carried out (the model operation speed is made to be faster) based on a Principal Component Analysis (PCA), and the date, the time (hour), the height of the storm wave at the last moment and the period of the storm wave at the last moment are subjected to dimensionality reduction processing to generate two sea wave data fusion parameters which are named as time data and historical data respectively, so that the characteristics of more data are kept, and the data saturation and low generalization capability are avoided; carrying out dimensionality reduction processing on the data based on PCA to obtain sea wave fusion data parameters, wherein the calculation process is as follows:
firstly, performing decentralization on selected sea wave characteristics, namely subtracting respective average value from each bit of characteristics; secondly, calculating a covariance matrix C, and solving an eigenvalue lambda of the covariance matrix C and a corresponding eigenvector u; and finally projecting the original features onto the selected feature vectors to obtain the features of the new dimensionality after dimensionality reduction, wherein the calculation formula is as follows:
Figure BDA0003598350160000042
Cu＝λu (15)
Figure BDA0003598350160000043
obtaining the analyzed sea wave data information, carrying out normalization processing on the data, and normalizing the range of all data values to be between [0 and 1] by a normalization method, wherein the calculation formula is as follows:
Figure BDA0003598350160000044
the wind speed, time data and historical data are divided into a training set and a testing set, partial data at the bottom of 7 months in 2019 and data from 8 months in 2019 to 7 months in 2020 are collated to be used as the data set, wherein 80% of the data set is used as the training set, and 20% of the data set is used as the testing set. And training the model by taking the training set as the input features of the model.
Step two, establishing a BiGRU network model, as shown in fig. 5, determining a model structure and a topology structure, wherein the BiGRU is a bidirectional gating cycle unit and is composed of an input layer, a forward hidden layer, a backward hidden layer and an output layer. The input layer contains input data, and the data is simultaneously transmitted to the forward hidden layer and the backward hidden layer at each moment, namely the data simultaneously flows to two GRU networks with opposite directions, and the output sequence of the output layer is jointly determined by the two GRUs. One forward low-level GRU reads input data information from front to back, and one backward high-level GRU refines the data information from back to front again; let x be t For the input vector of the time instant, the calculation process of the GRU network is expressed as follows:
z t ＝σ(W z x t +U z h t-1 +b z ) (17)
r t ＝σ(W r x t +U r h t-1 +b r ) (18)
Figure BDA0003598350160000051
Figure BDA0003598350160000052
in the formula: z is a radical of t And r t Representing an update gate and a reset gate;
Figure BDA0003598350160000053
representing waiting timeSelecting a hidden layer state; h is t-1 And h t Respectively representing hidden layer states at t-1 and t; w and U are weights; b is an offset; σ represents Sigmoid function;
the mathematical expression of the BiGRU network structure is as follows:
Figure BDA0003598350160000054
Figure BDA0003598350160000055
Figure BDA0003598350160000056
in the formula:
Figure BDA0003598350160000057
and
Figure BDA0003598350160000058
the states of the forward hidden layer and the backward hidden layer at the time t are respectively;
Figure BDA0003598350160000059
and
Figure BDA00035983501600000510
weights of forward and backward hidden layer states at time t are respectively; b t Is the bias of the buried layer state at time t.
Step three, setting MEA parameters, setting parameter population size (popsize), number of winning sub-populations (bestsize), number of temporary sub-populations (tempsize) and sub-population Size (SG), wherein the calculation formula is as follows:
SG＝popsize/(bestsize+tempsize) (24)
setting the population size to be 180, the number of the winner sub-populations to be 3, the number of the temporary sub-populations to be 3, the size of the obtained sub-populations to be 30, and optimizing a BiGRU network model based on MEA; firstly, coding a weight value and a threshold value of a network connection network, adopting a real number coding method, and generating an initial population by adopting a reciprocal of a sample mean square error as a scoring function; the individual scores are ranked, a plurality of winning individuals and temporary individuals are generated according to the ranking, new individuals are generated around the individuals by taking the individuals as the center, and a set formed by the new individuals is called a winning sub-population and a temporary sub-population.
Firstly, executing convergence operation, and then judging whether each sub-population is mature or not by using a population maturation discriminant function ismate (). If the child population is mature, the convergence operation is finished, if the child population is immature, the child population is generated by a new center, and then the convergence operation is carried out until the child population is mature. And searching out the individual with the highest score in each sub-static, and taking the score of the individual as the score of the sub-population.
If the score of the temporary sub-population is higher than that of the sub-population of the winning sub-population, carrying out a dissimilarity operation, replacing the individuals in the winning sub-population by the individuals in the temporary sub-population, releasing the individuals originally in the winning sub-population, and searching the individuals in the global scope to form a new temporary sub-population so as to ensure that the number of the temporary sub-population is unchanged.
And judging whether the result reaches an ending condition or not, and if so, entering the next step. If the iteration times do not reach the set value of the network, the next step is also carried out, and if the iteration times are not full, the convergence operation is returned; according to the encoding rule, the weight and the threshold value assigned to the BiGRU model by the optimal individualization are decoded, the BiGRU network is trained, and the overall process of MEA-BiGRU is shown in fig. 6.
Step four, in order to explain the superiority of the MEA-BiGRU algorithm model, a BP neural network model and a BiGRU network model are established for comparison verification, mean square error, mean absolute error and maximum relative error are used as estimation indexes of prediction performance of each model, in the embodiment, comparison is carried out through the BP neural network model and the BiGRU network model, and the prediction result pair of the method and other models for predicting the sea wave height is shown in FIG. 7 and is shown in Table 3 together with other model prediction result evaluation indexes.
Table 3 compares the predicted results with other model evaluation indexes
Model (model) E MAPE E MAE E RMSE
MEA-BiGRU 0.1132 0.3866 0.4735
BiGRU 0.1562 0.5433 0.6024
BP 0.2187 0.6278 0.7043
As can be seen from the table 3, the prediction result error by adopting the method is lower, the MAPE error is stabilized below 12.0%, and the MEA-BiGRU algorithm can be seen from the table, so that the three evaluation indexes are respectively improved by 10.55%, 0.2412 and 0.2308; compared with a BiGRU algorithm, the three evaluation indexes are respectively improved by 4.3%, 0.1567 and 0.1289; compared with other algorithms, the MEA-BiGRU model has the advantage that the accuracy of the prediction result is obviously improved.
The prediction method based on the optimization of the sea wave data information characteristic quantity and the optimization of the BiGRU network based on the MEA improves the accuracy of sea wave height prediction to a certain extent.
The above description is only an application scenario of the present invention, and the present invention shall be covered by the scope of the present invention when the present invention is applied to the equal change made in the claims of the present invention or the prediction of the wave height of other waves.
The invention is not the best known technology.

Claims (6)

1. A sea wave height prediction method based on a deep learning algorithm is characterized by comprising the steps of obtaining sea wave data information, and analyzing and preprocessing the sea wave data information; taking the processed data as the input of a model, determining a network structure, establishing a network model of a bidirectional gating cycle unit based on a thought evolution algorithm, setting MEA parameters, and training an MEA-BiGRU network model; and establishing a BP neural network model and a BiGRU network model for comparison and verification, and realizing accurate prediction of the wave height of the sea wave.
2. A wave height prediction method based on a deep learning algorithm as claimed in claim 1, characterized in that the wave data information is date, time, air temperature, wind direction, wind speed, air pressure, sea temperature, wave height at the last moment and wave period at the last moment.
3. A sea wave height prediction method based on a deep learning algorithm as claimed in claim 1, wherein the analyzing and preprocessing of the sea wave data information refers to screening abnormal data by a Mahalanobis average distance method, replacing abnormal values by an average value method, analyzing correlation between the sea wave data information and the sea wave height, performing characteristic extraction on the sea wave data information by a method combining Pearson, GRA and PCA, and performing normalization processing on the extracted data.
4. A sea wave height prediction method based on a deep learning algorithm as claimed in claim 3, wherein the method of combining Pearson, GRA and PCA is that Pearson and GRA correlation analysis is respectively performed on sea wave data information to obtain correlation coefficients of the sea wave data information and the sea wave height, and PCA dimension reduction processing is performed on the sea wave data information based on the relation of the correlation coefficients to obtain characteristic data.
5. A sea wave height prediction method based on a deep learning algorithm as claimed in claim 4, characterized in that the characteristic data are wind speed, time data and historical data, wherein the time data and the historical data are fusion data processed by PCA dimension reduction, the time data are fusion parameters of date and time, and the historical data are fusion parameters of the height of the wind wave at the last moment and the period of the wind wave at the last moment.
6. A sea wave height prediction method based on a deep learning algorithm as claimed in claim 1, wherein the network model of the bidirectional gating cyclic unit of the thought evolution algorithm is based on the thought evolution algorithm to optimize the bidirectional gating cyclic unit and obtain optimal weights and thresholds of the BiGRU.
CN202210398198.8A 2022-04-15 2022-04-15 Sea wave height prediction method based on deep learning algorithm Pending CN115099296A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210398198.8A CN115099296A (en) 2022-04-15 2022-04-15 Sea wave height prediction method based on deep learning algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210398198.8A CN115099296A (en) 2022-04-15 2022-04-15 Sea wave height prediction method based on deep learning algorithm

Publications (1)

Publication Number Publication Date
CN115099296A true CN115099296A (en) 2022-09-23

Family

ID=83287497

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210398198.8A Pending CN115099296A (en) 2022-04-15 2022-04-15 Sea wave height prediction method based on deep learning algorithm

Country Status (1)

Country Link
CN (1) CN115099296A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116523125A (en) * 2023-04-13 2023-08-01 宁波市气象台 Wave height forecasting method based on sea surface wind speed forecasting
CN116992781A (en) * 2023-09-27 2023-11-03 浙江大学海南研究院 Multi-step multi-element storm forecasting method based on deep learning
CN117031310A (en) * 2023-09-05 2023-11-10 南京工业大学 Method for predicting residual service life of power battery of electric automobile
CN117314330A (en) * 2023-09-01 2023-12-29 湖南工商大学 Intelligent manufacturing system based on digital twinning

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116523125A (en) * 2023-04-13 2023-08-01 宁波市气象台 Wave height forecasting method based on sea surface wind speed forecasting
CN116523125B (en) * 2023-04-13 2023-10-20 宁波市气象台 Wave height forecasting method based on sea surface wind speed forecasting
CN117314330A (en) * 2023-09-01 2023-12-29 湖南工商大学 Intelligent manufacturing system based on digital twinning
CN117031310A (en) * 2023-09-05 2023-11-10 南京工业大学 Method for predicting residual service life of power battery of electric automobile
CN116992781A (en) * 2023-09-27 2023-11-03 浙江大学海南研究院 Multi-step multi-element storm forecasting method based on deep learning
CN116992781B (en) * 2023-09-27 2023-12-26 浙江大学海南研究院 Multi-step multi-element storm forecasting method based on deep learning

Similar Documents

Publication Publication Date Title
CN108022001B (en) Short-term load probability density prediction method based on PCA (principal component analysis) and quantile regression forest
CN115099296A (en) Sea wave height prediction method based on deep learning algorithm
CN111815033A (en) Offshore wind power prediction method based on RCNN and meteorological time sequence characteristics
CN108549962B (en) Wind power prediction method based on historical segmented sequence search and time sequence sparsification
CN109581339B (en) Sonar identification method based on automatic adjustment self-coding network of brainstorming storm
CN110837915B (en) Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning
CN114676742A (en) Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network
CN111628494B (en) Low-voltage distribution network topology identification method and system based on logistic regression method
CN116937579B (en) Wind power interval prediction considering space-time correlation and interpretable method thereof
CN112149883A (en) Photovoltaic power prediction method based on FWA-BP neural network
CN114169445A (en) Day-ahead photovoltaic power prediction method, device and system based on CAE and GAN hybrid network
Li et al. Deep spatio-temporal wind power forecasting
CN114021483A (en) Ultra-short-term wind power prediction method based on time domain characteristics and XGboost
CN114022311A (en) Comprehensive energy system data compensation method for generating countermeasure network based on time sequence condition
CN112288157A (en) Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning
CN110163437B (en) Day-ahead photovoltaic power generation power prediction method based on DPK-means
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN117313795A (en) Intelligent building energy consumption prediction method based on improved DBO-LSTM
CN111488974A (en) Deep learning neural network-based ocean wind energy downscaling method
CN106960260B (en) Wind power prediction system convenient for power dispatching
CN115641231A (en) Typical scene extraction method and system for power system, electronic equipment and medium
CN112836876B (en) Power distribution network line load prediction method based on deep learning
CN114372640A (en) Wind power prediction method based on fluctuation sequence classification correction
CN110991743B (en) Wind power short-term combination prediction method based on cluster analysis and neural network optimization
CN114091782A (en) Medium-and-long-term power load prediction method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication