CN108510132A - A kind of sea-surface temperature prediction technique based on LSTM - Google Patents
A kind of sea-surface temperature prediction technique based on LSTM Download PDFInfo
- Publication number
- CN108510132A CN108510132A CN201810717212.XA CN201810717212A CN108510132A CN 108510132 A CN108510132 A CN 108510132A CN 201810717212 A CN201810717212 A CN 201810717212A CN 108510132 A CN108510132 A CN 108510132A
- Authority
- CN
- China
- Prior art keywords
- sea
- surface temperature
- lstm
- data
- score
- 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
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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (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
A kind of sea-surface temperature prediction technique based on LSTM, including prediction model and forecast future time period sea-surface temperature two parts are generated, which includes:1) z score standardizations are carried out to the history sea-surface temperature data within the scope of setting longitude and latitude and generates corresponding time row;2) it is 1 (day) to utilize time row and the standardized history sea-surface temperature data of z score to train LSTM models, time step, and input dimension is 1, obtains the sea-surface temperature prediction model based on LSTM;The forecast future time period sea-surface temperature includes:The standardized same day sea-surface temperature data of z score are inputted into forecasting model, setting prediction step number obtains the output of future time period as a result, output dimension is 1, and carrying out anti-z score standardizations to output result obtains sea-surface temperature predicted value.The present invention can be between mining data length time-dependent relation, be more suitable for learning the long periodicity changing rule of sea-surface temperature, preferable prediction result can be obtained.
Description
Technical field
The present invention relates to marine meterologal prediction fields, more particularly to a kind of sea based on LSTM recurrent neural networks models
Surface temperature prediction method.
Background technology
Sea-surface temperature is a kind of and the closer maritime meteorology data of human relation, such as ship's navigation, sea fishery
The production operations such as production, offshore oil platform are closely bound up with sea-surface temperature, scientific and reasonable sea-surface temperature predicted value
There is certain directive significance for mankind's offshore production operation.The acquisition of sea-surface temperature, because far from inland, conventional observation hand
Section such as buoy is difficult to cover comprehensively, main at present to carry out inverting using remote sensing satellite, therefore the prediction of sea-surface temperature becomes
Relative difficulty.
LSTM (Long-Short Term Memory) recurrent neural network achieved in many fields great in recent years
It breaks through, which can go through input data autonomous learning with the length dependence between learning time sequence data
What contribution of the history data to prediction data, the especially neural network overcame that other neural networks can not learn for a long time lacks
Point is more suitable for the long periodicity changing rule of excavation event.Existing neural network model is complex, need to use a variety of nerves
Network, just can be achieved sea-surface temperature prediction, occur that the extra large surface temperature of LSTM recurrent neural network systems progress is used alone
The method of prediction is spent, and the warm forecast precision Standard General in sea, at 0.8 DEG C or so, forecast precision is relatively low.
Invention content
The present invention provides a kind of sea-surface temperature prediction technique based on LSTM, solves sea-surface temperature in the prior art
The not high technical problem of precision of prediction.
To solve the above-mentioned problems, the present invention adopts the following technical scheme that:
Include the following steps:Step A generates prediction model;Step B forecasts future time period sea-surface temperature;
The step A includes the following steps:
A1 extracts marine environment data, using the existing remote sensing satellite data in the whole world, is finally inversed by marine surface temperature SST numbers
According to precision is 15 ' × 15 ', daily;
A2 extraction times arrange, and according to history SST data, the time of first time extraction SST are set as 1, next nature day is
2, and so on, smoothly gradually add 1 according to natural day, generated time row;
The history SST data that A3 obtains step A1 carry out z-score standardizations, and the time that step A2 is obtained arranges
Input data with the standardized history SST of z-score as LSTM models, input dimension are 1;
A4 build and training step A3 obtained by LSTM models, continuous adjusting parameter, preferentially Selecting All Parameters be based on
The sea-surface temperature prediction model of LSTM;
The step B includes the following steps:
Same day SST data handle with the z-score standardized methods used in step A3 by B1, and the time on the same day is arranged
Enter fishing ground forecasting model with standardized SST data category, Prediction Parameters are set, obtains the output of future time period as a result, output dimension
Degree is 1;
The output result that B2 obtains B1 carries out anti-z-score standardizations, the anti-z-score standardized methods and
Z-score standardization is corresponding in step A3, and processing obtains the SST predicted values of future time period, and precision of prediction is 15 ' × 15 ', often
It.
As further preferably, the step A4 includes as follows:
A41:LSTM models are built under Linux system;
A42:The input dimension of LSTM and the time step of input data are set;
A43:LSTM input datas are set and read batch sizes and length of window;
A44:LSTM model optimizers and learning rate are set;
A45:Hidden layer neuromere is arranged to count;
A46:Model iterations are set;
A47:Continuous adjusting parameter checks model degree of convergence with model loss, preferentially chooses high degree of convergence parameter, formed
Fishing ground prediction model based on LSTM.
It is to be based on the beneficial effects of the invention are as follows a kind of sea-surface temperature prediction technique based on LSTM provided by the invention
Time series excavates the long-term Fluctuation of sea-surface temperature data, can obtain preferable simulation and forecast effect, and precision of prediction is high.
Description of the drawings
Fig. 1 is sea-surface temperature prediction flow chart of the present invention;
Fig. 2 is LSTM recurrent neural network location mode control figures;
Fig. 3 is LSTM Recursive Neural Network Structure figures;
Fig. 4 is LSTM sea-surface temperatures prediction model at 45 ° 7 ' 30 " marine sites E N, 155 ° 7 ' 30 " sea-surface temperature
Simulation and forecast design sketch.
Specific implementation mode
The present invention is described in detail below in conjunction with attached drawing and concrete example, it is noted that discussed below
Example is only the more convenient understanding present invention, does not play any restriction effect to invention itself.
A kind of sea-surface temperature prediction technique based on LSTM that the present invention provides, as shown in Figure 1, the method comprising the steps of
A generates two step of sea-surface temperature of prediction model and step B prediction future time periods.
Step A generates prediction model:
Step A1, marine environment data is extracted:Using the existing remote sensing satellite data in the whole world, it is finally inversed by marine surface temperature
SST data, precision are 15 ' × 15 ' (longitudes × latitude), daily;
Step A2, extraction time arranges:According to history SST data, the time (natural day) of first time extraction SST is set as 1,
Next nature day is 2, and so on, gradually add 1 according to natural day sequence, generated time row;
Step A3, z-score standardizations are carried out to the history SST that step A1 is obtained, will arranges the time and is marked with z-score
The SST of standardization is 1 as LSTM mode input data, input dimension;
Step A31:Extract the mean value and standard deviation of history SST in 15 ' × 15 ' ranges.
Step A32:History SST is standardized using z-score standardized methods, z-score standardization is public
Formula is:
X '=(x- μ)/σ (1)
Wherein, x ' is the z-score standardized values of x, and μ is x array mean values, and σ is x array standard deviations.
Step A4, LSTM models are built and trained, by continuous adjusting parameter, are preferentially chosen using the model degree of convergence as standard
Prediction model of the gain of parameter based on LSTM;
Step B predicts that the sea-surface temperature of future time period includes:
Step B1, the z-score standardized method consistent in step A3 of same day SST data is handled, by the same day
Time row and standardized SST input steps A4 obtain sea-surface temperature prediction model, setting prediction step number, obtain future
The output of period is as a result, output dimension is 1;LSTM models can utilize current t moment data prediction future t+1 time datas.It goes through
History data are mainly used to training pattern, are the data of a period of time sequence, same day data be then for predicting future, be it is current this
The data at one moment.Usual same day data are included in historical data, but while being not included in interior also can be predicted.
Step B2, the anti-z-score standardizations of result, anti-z-score standardized methods and step A3 are exported to model
Middle z-score standardized methods are corresponding, and processing obtains the SST predicted values of future time period, and precision of prediction is 15 ' × 15 ', often
It;
Step A4 further comprises the following contents:
Step A41:LSTM models are built under Linux system;
Step A42:The input dimension of LSTM and the time step of input data are set.It is 1 that dimension is inputted in the present invention, when
Spacer step a length of 1 (day);
Step A43:LSTM input datas are set and read batch sizes and length of window.It uses and criticizes in model training
(batch) processing mode reads data, randomly selects the sequence of length of window (window size) in input data, and wrap
Dress up the batch data of batch sizes (batch size).That is batchsize sequence is shared in a batch, each sequence
Length is windowsize.The setting of batch sizes and window size is mainly imitated for balance computer memory size and operation
Rate, it is related with computer operational performance itself;
Step A44:LSTM model optimizers and learning rate are set.This method uses Adam algorithm optimization devices, study speed
Rate is 0.001;
Step A45:Hidden layer neuromere is arranged to count;
Step A46:Model iterations are set;
Step A47:Hidden layer neuromere is adjusted under same iterations to count, is whole in the points downward of same hidden layer neuromere
Iterations check model degree of convergence with model whole loss and iteration loss, preferentially choose high degree of convergence parameter, form base
In the sea-surface temperature prediction model of LSTM.
The control of LSTM recurrent neural network location modes is as shown in Figure 2.The core of LSTM is control unit state c, control
System includes forgeing door ft, input gate it, out gate ot.Current t moment forgets door ftIt is responsible for the c of control last momentt-1How many
It is saved in the c at current timet;Input gate itIt is responsible for the immediate status at control current timeHow many is input to active cell shape
State ct;Out gate otIt is responsible for control active cell state ctHow many exports h as the hidden layer at current timet.Its calculation formula point
It is not:
ft=σ (wf·[xt,ht-1]+bf) (2)
it=σ (wi·[xt,ht-1]+bi) (3)
ot=σ (wo·[xt,ht-1]+bo) (4)
Wherein, wf、wi、woIt is the weight matrix for forgeing door, input gate and out gate, b respectivelyf、bi、boIt is to forget respectively
The bias term of door, input gate and out gate, σ are sigmoid functions
LSTM Recursive Neural Network Structures are as shown in Figure 3.The input of LSTM includes:The location mode c of last momentt-1, on
One moment LSTM hidden layer output valve ht-1, current t moment network input value xt;The output of LSTM includes:The unit at current time
State ctWith the hidden layer output valve h of current time LSTMt。
Wherein, current input unit stateBy the input x of current t moment networkt, last moment LSTM hidden layer output valve
ht-1It codetermines, calculation formula is:
Wherein, wcIt is the weight matrix of input unit state, bcThe bias term of input unit state, tanh be hyperbolic just
Cut function
Active cell state ctBy forgetting door ft, last moment location mode ct-1, input gate itThe unit currently inputted
StateIt codetermines, calculation formula is:
Wherein, symbol ⊙ expressions are multiplied by element.
The hidden layer output valve h of current time LSTMtBy out gate otWith active cell state ctIt codetermines, calculates public
Formula is:
ht=ot⊙tanh(ct) (7)
And LSTM neural networks export Its calculation formula is:
Using mean square error (MSE) as loss function (loss) in the present invention, calculation formula is:
Wherein, N is sample number, and x is observed value, and x ' is predicted value.
Experimental example
It will be exemplified below the sea-surface temperature prediction technique effect based on LSTM.With 45 ° 7 ' 30 " N, 155 ° 7 '
30 " for 15 ' × 15 ' range marine sites centered on E, which is located at northwest Pacific, and originally experience Kuroshio warm current and Oyashio are cold
Joint effect is flowed, within the scope of northwest Pacific saury fishing ground, has and centainly represents meaning.
The history SST data in above range are extracted, time range is 1982-01-01 to 2015-12-31, is amounted to 34 years
12418 days, wherein data available 12418 days, and corresponding time row are generated according to the history SST times, as shown in table 1.
1 45 ° 7 ' 30 of the table " marine sites E N, 155 ° 7 ' 30 " history SST data
Serial number | Time | Time arranges | SST/℃ |
1 | 19820101 | 1 | 3.36 |
2 | 19820102 | 2 | 3.48 |
3 | 19820103 | 3 | 3.57 |
4 | 19820104 | 6 | 3.52 |
5 | 19820105 | 8 | 3.31 |
6 | 19820106 | 9 | 3.00 |
7 | 19820107 | 11 | 3.04 |
… | … | … | … |
998 | 19840924 | 998 | 13.38 |
999 | 19840925 | 999 | 13.39 |
1000 | 19840926 | 1000 | 13.31 |
… | … | … | … |
12417 | 20151230 | 12417 | 3.50 |
12418 | 20151231 | 12418 | 3.36 |
Z-score standardizations are carried out to SST.The mean value of history SST data is 6.459004 DEG C, and standard deviation is
4.248163, using time row and the standardized SST of z-score as the input of LSTM models, input data is as shown in table 2.
2 LSTM mode input data of table
If batch sizes are 12, length of window 30.In iteration 300 times, the number of hidden nodes is constantly adjusted, with mould
Type whole loss checks the LSTM sea-surface temperature prediction model degrees of convergence, and the results are shown in Table 3.
LSTM whole loss when 3 difference the number of hidden nodes of table
As can be seen from Table 3, when the number of hidden nodes is 300 or more, LSTM model whole loss starts to tend towards stability.When
Whole loss is minimum when hidden layer number of nodes is 400, is 0.00054356.
If the number of hidden nodes is 400, iteration checks LSTM models convergent under different iterations 2500 times, with this time
The LSTM sea-surface temperature prediction model degrees of convergence are checked in iteration loss, and the results are shown in Table 4.
Secondary iteration loss when 4 difference iterations of table
As can be seen from Table 4, in the case of the number of hidden nodes is 400, when iterations are more than 300 times, iteration loses
In stabilization.When wherein iterations are 601, iteration loss reduction is 0.00130815.
It should be noted that convergence direction is inconsistent when LSTM models are trained every time, each run result is caused all to slightly have
Difference, but difference is little, and global convergence trend is consistent.LSTM hidden layer neuromeres points are set as 400, iteration 600
Secondary to be trained, setting prediction step number is 7, i.e. the SST data of following one week of prediction, modeling prediction effect as shown in figure 4,
Wherein black represents SST actual values, and green represents the SST analogues value, and red represents SST predicted values.The analogue value is namely based on existing
Truthful data, establish the simulation to data with existing after LSTM models, the analogue value with actual value registration is higher illustrates pattern die
Quasi- effect is better.In this experimental example, the analogue value is overlapped with actual value height, is achieved preferable simulation effect, is based on height weight
Reason is closed, uses colored picture of drawing with display model effect.As a result display model whole loss is 0.00041835, SST moulds
Quasi- precision is 94.72%, model export result be [- 0.7402, -0.7501, -0.7590, -0.7680, -0.7773, -
0.7865, -0.7958], obtained after anti-z-score standardizations one week i.e. 2016-01-01 to 2016-01-07 following
Sea-surface temperature predicted value be respectively 3.3145 DEG C, 3.2724 DEG C, 3.2346 DEG C, 3.1964 DEG C, 3.1569 DEG C, 3.1178
DEG C, 3.0783 DEG C.According to existing satellite data, this week SST actual value is respectively 3.36 DEG C, 3.24 DEG C, 3.11 DEG C, 3.16 DEG C,
3.23 DEG C, 3.01 DEG C, 2.89 DEG C, obtain following one day, three days, five days, one week precision of prediction be respectively 98.64%,
97.88%, 98.04%, 97.16%.
It should be noted that when predict step number be more than or equal to 2 when, the t+2 moment (predicted value of current time t) be
It is predicted on the basis of t+1 moment predicted values, but the t+1 moment not yet occurs, lacks actual value, therefore predicts step number
Longer, prediction error is bigger.Sea-surface temperature variational regularity itself is strong, 7 days following using LSTM model predictions in the present invention
Sea-surface temperature value can still keep degree of precision.
The above description is merely a specific embodiment is illustrated, but protection scope of the present invention is not
Be confined to this, the prediction of the sea-surface temperature of any ocean geography position based on LSTM, should all cover the scope of the present invention it
It is interior.
Claims (2)
1. a kind of sea-surface temperature prediction technique based on LSTM, which is characterized in that include the following steps:Step A production forecasts
Model;Step B forecasts future time period sea-surface temperature;
The step A includes the following steps:
A1 extracts marine environment data, using the existing remote sensing satellite data in the whole world, is finally inversed by marine surface temperature SST data,
Precision is 15 ' × 15 ', daily;
A2 extraction times arrange, and according to history SST data, the time of first time extraction SST is set as 1, and next nature day is 2, according to
It is secondary to analogize, smoothly gradually add 1 according to natural day, generated time row;
The history SST data that A3 obtains step A1 carry out z-score standardizations, the time row and z- that step A2 is obtained
Input datas of the standardized history SST of score as LSTM models, input dimension are 1;
A4 builds the LSTM models obtained by simultaneously training step A3, and continuous adjusting parameter, preferentially Selecting All Parameters acquisition is based on LSTM's
Sea-surface temperature prediction model;
The step B includes the following steps:
Same day SST data handle with the z-score standardized methods used in step A3 by B1, and the time on the same day is arranged and marked
The SST data categories of standardization enter fishing ground forecasting model, and Prediction Parameters are arranged, and obtain the output of future time period as a result, output dimension is
1;
B2 carries out anti-z-score standardizations, the anti-z-score standardized methods and step to the output result that B1 is obtained
Z-score standardization is corresponding in A3, and processing obtains the SST predicted values of future time period, and precision of prediction is 15 ' × 15 ', daily.
2. a kind of sea-surface temperature prediction technique based on LSTM according to claim 2, which is characterized in that the step
A4 includes as follows:
A41:LSTM models are built under Linux system;
A42:The input dimension of LSTM and the time step of input data are set;
A43:LSTM input datas are set and read batch sizes and length of window;
A44:LSTM model optimizers and learning rate are set;
A45:Hidden layer neuromere is arranged to count;
A46:Model iterations are set;
A47:Continuous adjusting parameter checks model degree of convergence with model loss, preferentially chooses high degree of convergence parameter, formation is based on
The fishing ground prediction model of LSTM.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810717212.XA CN108510132A (en) | 2018-07-03 | 2018-07-03 | A kind of sea-surface temperature prediction technique based on LSTM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810717212.XA CN108510132A (en) | 2018-07-03 | 2018-07-03 | A kind of sea-surface temperature prediction technique based on LSTM |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108510132A true CN108510132A (en) | 2018-09-07 |
Family
ID=63404117
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810717212.XA Pending CN108510132A (en) | 2018-07-03 | 2018-07-03 | A kind of sea-surface temperature prediction technique based on LSTM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108510132A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109376904A (en) * | 2018-09-18 | 2019-02-22 | 广东电网有限责任公司 | A kind of short-term Wind power forecasting method and system based on DWT and LSTM |
CN109472396A (en) * | 2018-10-17 | 2019-03-15 | 四川佳联众合企业管理咨询有限公司 | Mountain fire prediction technique based on depth e-learning |
CN109615226A (en) * | 2018-12-12 | 2019-04-12 | 焦点科技股份有限公司 | A kind of operation indicator method for monitoring abnormality |
CN109655828A (en) * | 2018-12-18 | 2019-04-19 | 中国人民解放军国防科技大学 | SST (stimulated Raman Scattering) depth learning inversion method for multi-frequency one-dimensional synthetic aperture microwave radiometer |
CN109708689A (en) * | 2018-11-14 | 2019-05-03 | 沈阳理工大学 | Based on the chain pharmacy temperature and humidity early warning system for improving long Memory Neural Networks in short-term |
CN109978279A (en) * | 2019-04-10 | 2019-07-05 | 青岛农业大学 | Marine surface temperature regional prediction method |
CN110197307A (en) * | 2019-06-03 | 2019-09-03 | 上海海洋大学 | A kind of domain type sea-surface temperature prediction technique of combination attention mechanism |
CN110648030A (en) * | 2019-10-31 | 2020-01-03 | 吉林大学 | Method and device for predicting seawater temperature |
CN111351664A (en) * | 2020-02-13 | 2020-06-30 | 成都运达科技股份有限公司 | Bearing temperature prediction and alarm diagnosis method based on LSTM model |
CN111829693A (en) * | 2020-07-30 | 2020-10-27 | 成都运达科技股份有限公司 | Bearing temperature detection method and system based on LSTM model of time and space dimensions |
CN113063737A (en) * | 2021-03-26 | 2021-07-02 | 福州大学 | Ocean heat content remote sensing inversion method combining remote sensing and buoy data |
CN113297801A (en) * | 2021-06-15 | 2021-08-24 | 哈尔滨工程大学 | Marine environment element prediction method based on STEOF-LSTM |
CN116822710A (en) * | 2023-05-24 | 2023-09-29 | 国家海洋环境预报中心 | Coral reef whitening hot spot prediction method, calcification rate prediction method and electronic equipment |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101846753A (en) * | 2010-04-29 | 2010-09-29 | 南京信息工程大学 | Climate time sequence forecasting method based on empirical mode decomposition and support vector machine |
CN105988146A (en) * | 2015-01-29 | 2016-10-05 | 中国科学院空间科学与应用研究中心 | Application data processing method of spaceborne microwave radiometer |
CN106096767A (en) * | 2016-06-07 | 2016-11-09 | 中国科学院自动化研究所 | A kind of link travel time prediction method based on LSTM |
CN106598917A (en) * | 2016-12-07 | 2017-04-26 | 国家海洋局第二海洋研究所 | Upper ocean thermal structure prediction method based on deep belief network |
CN107330541A (en) * | 2017-05-22 | 2017-11-07 | 上海海洋大学 | A kind of west and central Pacific ocean stripped tuna Predicting Center Fishery method |
CN107544904A (en) * | 2017-08-21 | 2018-01-05 | 哈尔滨工程大学 | A kind of prediction model of software reliability based on depth CG LSTM neutral nets |
CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
US20180166066A1 (en) * | 2016-12-14 | 2018-06-14 | International Business Machines Corporation | Using long short-term memory recurrent neural network for speaker diarization segmentation |
-
2018
- 2018-07-03 CN CN201810717212.XA patent/CN108510132A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101846753A (en) * | 2010-04-29 | 2010-09-29 | 南京信息工程大学 | Climate time sequence forecasting method based on empirical mode decomposition and support vector machine |
CN105988146A (en) * | 2015-01-29 | 2016-10-05 | 中国科学院空间科学与应用研究中心 | Application data processing method of spaceborne microwave radiometer |
CN106096767A (en) * | 2016-06-07 | 2016-11-09 | 中国科学院自动化研究所 | A kind of link travel time prediction method based on LSTM |
CN106598917A (en) * | 2016-12-07 | 2017-04-26 | 国家海洋局第二海洋研究所 | Upper ocean thermal structure prediction method based on deep belief network |
US20180166066A1 (en) * | 2016-12-14 | 2018-06-14 | International Business Machines Corporation | Using long short-term memory recurrent neural network for speaker diarization segmentation |
CN107330541A (en) * | 2017-05-22 | 2017-11-07 | 上海海洋大学 | A kind of west and central Pacific ocean stripped tuna Predicting Center Fishery method |
CN107544904A (en) * | 2017-08-21 | 2018-01-05 | 哈尔滨工程大学 | A kind of prediction model of software reliability based on depth CG LSTM neutral nets |
CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
Non-Patent Citations (3)
Title |
---|
QIN ZHANG等: "Prediction of Sea Surface Temperature Using Long Short-Term Memory", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
王鑫等: "基于LSTM循环神经网络的故障时间序列预测", 《北京航空航天大学学报》 * |
陆泽楠等: "基于 LSTM 神经网络模型的钢铁价格预测", 《科技视界》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109376904A (en) * | 2018-09-18 | 2019-02-22 | 广东电网有限责任公司 | A kind of short-term Wind power forecasting method and system based on DWT and LSTM |
CN109472396A (en) * | 2018-10-17 | 2019-03-15 | 四川佳联众合企业管理咨询有限公司 | Mountain fire prediction technique based on depth e-learning |
CN109472396B (en) * | 2018-10-17 | 2023-06-20 | 成都卡普数据服务有限责任公司 | Mountain fire prediction method based on deep network learning |
CN109708689A (en) * | 2018-11-14 | 2019-05-03 | 沈阳理工大学 | Based on the chain pharmacy temperature and humidity early warning system for improving long Memory Neural Networks in short-term |
CN109615226A (en) * | 2018-12-12 | 2019-04-12 | 焦点科技股份有限公司 | A kind of operation indicator method for monitoring abnormality |
CN109655828B (en) * | 2018-12-18 | 2020-06-26 | 中国人民解放军国防科技大学 | SST (stimulated Raman Scattering) depth learning inversion method for multi-frequency one-dimensional synthetic aperture microwave radiometer |
CN109655828A (en) * | 2018-12-18 | 2019-04-19 | 中国人民解放军国防科技大学 | SST (stimulated Raman Scattering) depth learning inversion method for multi-frequency one-dimensional synthetic aperture microwave radiometer |
CN109978279A (en) * | 2019-04-10 | 2019-07-05 | 青岛农业大学 | Marine surface temperature regional prediction method |
CN109978279B (en) * | 2019-04-10 | 2023-05-02 | 青岛农业大学 | Ocean surface temperature area prediction method |
CN110197307A (en) * | 2019-06-03 | 2019-09-03 | 上海海洋大学 | A kind of domain type sea-surface temperature prediction technique of combination attention mechanism |
CN110648030A (en) * | 2019-10-31 | 2020-01-03 | 吉林大学 | Method and device for predicting seawater temperature |
CN111351664A (en) * | 2020-02-13 | 2020-06-30 | 成都运达科技股份有限公司 | Bearing temperature prediction and alarm diagnosis method based on LSTM model |
CN111829693A (en) * | 2020-07-30 | 2020-10-27 | 成都运达科技股份有限公司 | Bearing temperature detection method and system based on LSTM model of time and space dimensions |
CN111829693B (en) * | 2020-07-30 | 2021-10-22 | 成都运达科技股份有限公司 | Bearing temperature detection method and system based on LSTM model of time and space dimensions |
CN113063737A (en) * | 2021-03-26 | 2021-07-02 | 福州大学 | Ocean heat content remote sensing inversion method combining remote sensing and buoy data |
CN113297801A (en) * | 2021-06-15 | 2021-08-24 | 哈尔滨工程大学 | Marine environment element prediction method based on STEOF-LSTM |
CN116822710A (en) * | 2023-05-24 | 2023-09-29 | 国家海洋环境预报中心 | Coral reef whitening hot spot prediction method, calcification rate prediction method and electronic equipment |
CN116822710B (en) * | 2023-05-24 | 2024-07-12 | 国家海洋环境预报中心 | Coral reef whitening hot spot prediction method, calcification rate prediction method and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108510132A (en) | A kind of sea-surface temperature prediction technique based on LSTM | |
Song et al. | Predicting TEC in China based on the neural networks optimized by genetic algorithm | |
CN108665104A (en) | A kind of fishing ground forecasting procedure based on LSTM | |
CN111199270B (en) | Regional wave height forecasting method and terminal based on deep learning | |
Thibaud et al. | Threshold modeling of extreme spatial rainfall | |
Wang et al. | Simulating multiple class urban land-use/cover changes by RBFN-based CA model | |
Rees et al. | Evolving integral projection models: evolutionary demography meets eco‐evolutionary dynamics | |
US20190230913A1 (en) | Method of Predicting Central Fishing Ground of Flying Squid Family Ommastrephidae | |
CN110322009A (en) | Image prediction method based on the long Memory Neural Networks in short-term of multilayer convolution | |
CN109978279B (en) | Ocean surface temperature area prediction method | |
CN112651665A (en) | Surface water quality index prediction method and device based on graph neural network | |
CN104809479A (en) | Fish HIS (habitat suitability index) modeling method based on SVM (support vector machine) | |
WO2006120724A1 (en) | Geographic information system using neural networks | |
CN114693005B (en) | Three-dimensional underground oil reservoir dynamic prediction method based on convolution Fourier neural network | |
Maddu et al. | Short-range reservoir inflow forecasting using hydrological and large-scale atmospheric circulation information | |
CN114152257A (en) | Ship prediction navigation method based on attention mechanism and environment perception LSTM | |
CN111811465B (en) | Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network | |
CN111027249A (en) | Machine learning-based inter-well connectivity evaluation method | |
Fix et al. | Simultaneous autoregressive models for spatial extremes | |
CN112183721B (en) | Construction method of combined hydrological prediction model based on self-adaptive differential evolution | |
Ferrero-Guillén et al. | Optimal Wing and Horizontal Tail plane design for maximizing the aircraft performance in cruise flight | |
CN110852415B (en) | Vegetation index prediction method, system and equipment based on neural network algorithm | |
CN110648030A (en) | Method and device for predicting seawater temperature | |
CN116401939A (en) | North sea ice short-term forecasting method based on gradient constraint neural network | |
Gerey et al. | Groundwater Single‐and Multiobjective Optimization Using Harris Hawks and Multiobjective Billiards‐Inspired Algorithm |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180907 |