CN106599995A - Method for predicting ship navigation meteorological factors according to marine meteorological information - Google Patents
Method for predicting ship navigation meteorological factors according to marine meteorological information Download PDFInfo
- Publication number
- CN106599995A CN106599995A CN201611129832.9A CN201611129832A CN106599995A CN 106599995 A CN106599995 A CN 106599995A CN 201611129832 A CN201611129832 A CN 201611129832A CN 106599995 A CN106599995 A CN 106599995A
- Authority
- CN
- China
- Prior art keywords
- data
- ship
- model
- layer
- navigation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06N3/08—Learning methods
-
- 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"
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)
- Software Systems (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)
- Computational Linguistics (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)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a method of predicting ship navigation meteorological factors according to marine meteorological information. A sample data set is firstly obtained through ship navigation database screening query and data normalization; then, an alternating sparse auto-encoders (ASAE) deep learning regression prediction model is built, based on a large amount of historical ship navigation sample data sets, a least square method, an L2 weight decay constraint and a KL contrast sparsity constraint are adopted to define a cost function, and alternating unsupervised learning and supervised learning are combined for parameter tuning; and finally, to-be-predicted ship navigation data are substituted to the ASAE prediction model with the solved optimize4d parameters for meteorological factor prediction. The method of the invention can build a function relationship between the ship speed and diversified marine meteorological conditions in the ship navigation, the complicated nonlinear problem in meteorological factor calculation can be solved, the meteorological factor prediction precision is improved, and an important role is played in energy-saving emission-reduction navigation optimization.
Description
Technical field
The invention belongs to ship shipping field, and in particular to it is a kind of according to maritime meteorology information prediction ship's navigation it is meteorological because
The method of son.
Background technology
What shipping business energy resource consumption and goods operated increases substantially, and the maximization of ship causes shipping carbon emission speed
Degree increases, developed country's pay attention to day by day shipping business low-carbon emission reduction problem, and the trend that international carriage carbon emission is limited is irreversible.
The International Maritime Organization of January 1 in 2013 (IMO) has enforced " ship Energy design index (EEDI) ", " efficiency operation index
(EEOI) " and " ship energy efficiency management plan (SEEMP) ", this carries out the work with powerful motive force to what ship energy saving was reduced discharging.
" navigation optimization " is one of means that ship energy saving is reduced discharging, and wherein impact of the maritime meteorology to ship's navigation is pole in navigation optimization
For important Consideration, therefore Accurate Prediction meteorological factor becomes the technological difficulties of navigation optimization.During ship's navigation
Ship's speed is affected by various random factors such as ship self performance, ship navigation state and milder metocean conditions, prediction
Impact of the maritime meteorology to ship speed has great importance.Accurate prediction meteorological factor preferably can be led for meteorology
Boat platform provides data supporting so as to more reasonably can recommend optimal route and optimum speed for vessel underway oceangoing ship, not be only
The sailing decision-making of captain provides scientific guidance, and crewman can be instructed reasonably to manipulate ship's navigation, it is ensured that vessel underway oceangoing ship
Ship green navigation is realized under high energy efficiency, the navigation optimisation strategy of low emission.However, maritime meteorology influence factor's is random more
The difference of degeneration and ship self performance causes the functional relationship between ship's speed and each influence factor accurately cannot determine, utilizes
It is pre- that the physical model of tank experiments and hydromechanical computation model also cannot provide real-time meteorological factor for vessel underway oceangoing ship
Survey.
Deep learning (Deep Learning, DL) is implicit from inside angle of statistics analysis with mining data
Relation, based on the theory of machine learning, the distribution using a large amount of historical datas the bionic mechanism with reference to artificial neural network, leads to
Cross the neural network model of setting up many hidden layers and constantly train tuning model parameter, excavate the pattern implied between historical data
With rule, the nonlinear function for finally giving a solving practical problems and then a kind of research side for realizing data prediction
Method, it is advantageous that the set that can not only express the bigger more higher-dimension of data volume, and training method more compact.It is sparse from
Coding (Sparse Auto-Encoders, SAE) is one of the most frequently used deep learning method, and it is led to by multiple self-encoding encoders
Crossing unsupervised learning trains the hidden layer for obtaining stacking to form, and the hidden layer of its preceding layer self-encoding encoder is exported as later layer certainly
The input of encoder, with successively greedy coaching method training whole network, i.e., each layer of training network further trains whole successively
Neutral net.The advantage of sparse own coding is to map the deep learning forecast model of composition than shallow-layer nerve using multilayered nonlinear
Network is more effective, has the disadvantage there is " gradient disperse ", i.e., bottom-layer network can not be learnt comprehensively, and gradient descent method is only
The parameter of higher level is effectively corrected.
The content of the invention
In order to overcome disadvantage mentioned above, the invention provides one kind is according to maritime meteorology information prediction ship's navigation meteorological factor
Method, can solve the problem that maritime meteorology on ship's navigation impact in meteorological factor prediction problem, consider ship itself property
The factors such as energy, ship's navigation data and maritime meteorology change, go out from machine learning, big data technology and statistical angle
Send out, model training and parameter optimization are carried out by building sparse own coding (ASAE) the deep learning regressive prediction model of alternating, most
Training eventually obtains a meteorological factor forecast model for solving the Accurate Prediction that maritime meteorology factor is affected on ship speed.
The sparse own coding (ASAE) of alternating is the deep learning side that one kind proposed by the present invention can alleviate " gradient disperse "
Method.The advantage of the algorithm is to realize study and the tuning of bottom parameter with supervised learning by being alternately performed unsupervised learning,
Autoencoder network training is carried out first with unsupervised learning in pre-training, followed by the gradient descent method of supervised learning
Carry out arameter optimization.ASAE models only train a layer network every time, and the hiding layer data that layer network training is obtained as
The input data of next layer network, i.e., successively wolfishly train hidden layer, until all layers of network has been trained, finally will alternately
The parameter that sparse autoencoder network learns carries out transferring weights, in model top additional layer BP neural network from top to bottom
All layers of parameter is adjusted back.ASAE algorithms are by introducing L in cost function2Weight decay constraint and KL divergences
Sparsity constraints minimize residual sum of squares (RSS), i.e. L2Weight decay constraint can prevent Expired Drugs, KL divergences from ensure that
Parameter it is openness, improve the generalization of model and shorten the training time.
A kind of method according to maritime meteorology information prediction ship's navigation meteorological factor that the present invention is provided, primary object
It is to replace sparse own coding (ASAE) model, and the ship's navigation meteorological factor of sparse own coding (ASAE) deep learning of alternating
Forecast model framework.
Sparse own coding (ASAE) model of alternating can realize bottom with supervised learning by being alternately performed unsupervised learning
The study of layer parameter and tuning, alleviate " gradient disperse " problem of deep learning method generally existing.The sparse own coding of alternating
(ASAE) the ship's navigation meteorological factor forecast model framework of deep learning make use of substantial amounts of ship self performance data, history
Aeronautical data and maritime meteorology data, initially set up sparse own coding (ASAE) regressive prediction model of alternating, then according to minimum
Square law, L2Weight decay constraint and KL define a cost function to sdpecific dispersion sparsity constraints, and final goal is to solve for L2
Weight decay constraint and convex quadratic programming problems of the KL to square error minimization of loss under sdpecific dispersion sparsity constraints, with reference to
Unsupervised learning and supervised learning realize successively greedy training and parameter optimization, finally give the ship's navigation gas after optimization
As factor forecast model.
In order to realize object above, the present invention includes according to the method for maritime meteorology information prediction ship's navigation meteorological factor
Following steps：
Step one：Set up ship's navigation data base.The foundation of ship's navigation data base.Ship's navigation data base mainly include
Ship self performance data, three part of ship's navigation historical data and maritime meteorology data.Obtained by ship visit report
Ship performance data；Ship's navigation historical data is obtained by history log record；Collect from U.S.National Oceanic and
The meteorology that Atmospheric Administration (National Oceanic and Atmospheric Administration, NOAA) is issued is pre-
Breath is notified as maritime meteorology data；Finally transferring data to database server carries out building ship's navigation data base.For
The prediction of ship's navigation meteorological factor is carried out, when needing the data of collection to include captain, the beam, ship in ballast weight, navigation
Between, longitude, latitude, draft, cargo dead-weight, Angle of Trim, main frame maximum-continuous rating (MCR), ship rated speed, speed on the ground, oil
Consumption, meteorological factor, wind direction, wind speed, wave are high, wave to, gush height, pour into, gush cycle, wave height, wave direction, period of wave, sea pressure, its
Output variable of the middle meteorological factor as model, input variable of its dependent variable as model.
Step 2：Data prediction.Ship's navigation data base for setting up in step one carries out data selection, is data
The middle feature that there is shortage of data carries out linear interpolation supplement, and the ship's navigation data of integration are carried out unified normalization then
Process.As numerical value difference is big between aeronautical data feature, the characteristics of feature internal numeric difference is little, deep learning has been had a strong impact on
The complexity of the execution efficiency and model of algorithm, it is therefore desirable to be normalized to eliminate what dimension was brought to aeronautical data
Difference.For the minute differences inside prominent features in the present invention, the result that ship's navigation data in step one are selected is carried out
After normalization, be converted into fixed digit binary representation, fixed digit can be eight-digit binary number, sixteen bit system, three
12 binary systems.
Step 3：Set up the meteorological factor forecast model of sparse own coding (ASAE) deep learning of alternating.According to step 2
Complex nonlinear relation between the characteristic distributions and characteristic variable and meteorological factor of middle training dataset, sets up based on depth
Alternating sparse own coding (ASAE) the meteorological factor forecast model of study, the input matrix X of the model is normalization in step 2
Training dataset after conversion, predicts that output matrix isThe quantity of sparse own coding (ASAE) the model hidden layer of alternating is set
And the neuronal quantity of every layer of hidden layer, the weight matrix W and bias matrix b of network, per layer of net are initialized using random number
The output matrix computing formula of network isWhereinI tables
I-th layer of representation model, miI-th layer of neuron number for including is represented, z represents the weighted sum of i-th layer of neuron, and f (z) is represented and added
The activation value of power sum,Represent the output of the 1st neuron of i-th layer of hidden layer.Using the output of last layer network as next
The input of layer network, will last layer network output matrixAs the input matrix X of next layer network, successively iteration until
Calculate the output matrix of last layer of the model, model last layer is output as the meteorological factor predicted.
Step 4：Model training and parameter optimization.It is based on the training data obtained in step 2, sparse self-editing according to alternating
Code one cost function of (ASAE) net definitions, the target of the cost function are to solve for L2Weight decay constraint
And KL to sdpecific dispersion sparsity constraints (Wherein ρ is that every layer network is openness
Binding occurrence,Be the neuronal activation degree of every layer network) under square error minimization of loss convex quadratic programming problem, that is, solveWherein L2Weight decay constraint is by setting
Determine weight decay factor coefficient lambda, weigh weight parameter and the proportion predicted the outcome between square error term, prevent to a certain extent
The phenomenon of model over-fitting is stopped, KL passes through to set openness factor beta to sdpecific dispersion constraint, to the dilute of parameter in model training
Thin property is defined, and makes model obtain openness parameter learning.It is final to combine alternately unsupervised learning and supervised learning side
Method is trained to model and carries out parameter optimization using gradient descent method, and wherein unsupervised learning is that input matrix X is carried out
It is also X that self study is output matrix, and the output matrix for the ease of distinguishing unsupervised learning is used hereinRepresent, unsupervised training
This layer of hidden layer output is obtained, supervised learning is that the output matrix of unsupervised learning network is changed to label data to be instructed
Practice, i.e., output matrix is true meteorological factor y, and for minimum model cost function, the algorithm is along generation to gradient descent method
The direction of valency function least squares error is adjusted to network weight variable and bias variable.It is in course of adjustment, cost
Function is become progressively smaller until that meeting minimum error constraint or continuous 6 cost functions does not drop anti-increasing and then stop adjustment, finally gives
Optimal model parameters.Will alternately unsupervised learning bring into the parameter that supervised learning is carried out after gradient descent method optimization replace it is dilute
In the forecast model that thin own coding (ASAE) deep learning regressive prediction model establishment step is set up, the ASAE gas after being trained
As factor forecast model.
Step 5：Model application.By aeronautical data to be predicted, i.e. ship self performance data, ship's navigation data with
And in the ASAE meteorological factor forecast models obtained after training optimization in maritime meteorology data input step four, it is possible to achieve output
Meteorological factor.
It is of the invention compared with existing artificial neural network (ANN), sparse autoencoder network (SAE) forecast model, have
Advantages below and effect：
1. alternating sparse own coding (ASAE) model based on deep learning is present invention employs, various seas have been considered
The impact that foreign meteorological factor is produced to ship speed, when reducing the model training to complicated pluralism nonlinear data regression forecasting
Between, while improve precision of prediction.
2. the present invention can realize that unsupervised learning is used alternatingly during pre-training to be carried out successively with supervised learning
Greedy training, alleviates " gradient disperse " problem of deep learning generally existing, has carried out abundant study to model bottom data.
3. the present invention has stronger generalization ability and has carried out variable sparsity constraints.The present invention is in cost functionIn introduce L2Weight decay constraint with
And KL is to sdpecific dispersion constraint, the respectively Section 2 and Section 3 of formula, wherein L2Weight decay constraint is declined by setting weight
Subtracting coefficient coefficient lambda, weighs weight parameter and the proportion predicted the outcome between square error term, prevent to some extent model
The phenomenon of over-fitting.KL, is carried out to the openness of parameter in model training by setting openness factor beta to sdpecific dispersion constraint
Limit, make model obtain openness parameter learning.
Description of the drawings
Fig. 1 is flow chart of the present invention according to the method for maritime meteorology information prediction ship's navigation meteorological factor；
Fig. 2 is the ASAE forecast model figures of the present invention successively greedy training.
Specific embodiment
The one kind for describing present invention offer below in conjunction with the accompanying drawings in detail is meteorological according to maritime meteorology information prediction ship's navigation
The method of the factor.Fig. 1 is the flow chart of the method according to maritime meteorology information prediction ship's navigation meteorological factor.
Such as Fig. 1, the method according to maritime meteorology information prediction ship's navigation meteorological factor of the invention is mainly including five steps
Suddenly：
Step one：The foundation of ship's navigation data base.Ship's navigation data base mainly includes ship self performance data, ship
Three part of oceangoing ship navigation historical data and maritime meteorology data.Ship performance data are obtained by ship visit report；By going through
History log record obtains ship's navigation historical data；Collect from U.S.National Oceanic and Atmospheric Administration (National
Oceanic and Atmospheric Administration, NOAA) Weather Forecast Information issued is used as maritime meteorology number
According to；Finally transferring data to database server carries out building ship's navigation data base.In order to carry out ship's navigation it is meteorological because
The prediction of son, needs the data of collection to include that captain, the beam, ship in ballast weight, hours underway, longitude, latitude, drinking water are deep
Degree, cargo dead-weight, Angle of Trim, main frame maximum-continuous rating (MCR), ship rated speed, speed on the ground, oil consumption, meteorological factor, wind direction, wind
Speed, wave are high, wave to, gush height, pour into, gush cycle, wave height, wave direction, period of wave, sea pressure, wherein meteorological factor is used as output
Response, input variable of its dependent variable as model.
Step 2：Data prediction.Ship's navigation data base for setting up in step one carries out data selection, is data
The middle feature that there is shortage of data carries out linear interpolation supplement, and the ship's navigation data of integration are carried out unified normalization then
Process.As numerical value difference is big between aeronautical data feature, the characteristics of feature internal numeric difference is little, deep learning has been had a strong impact on
The complexity of the execution efficiency and model of algorithm, it is therefore desirable to be normalized to eliminate what dimension was brought to aeronautical data
Difference.For the minute differences inside prominent features in the present invention, the result that ship's navigation data in step one are selected is carried out
After normalization, be converted into fixed digit binary representation, fixed digit can be eight-digit binary number, sixteen bit system, three
12 binary systems.
Step 3：Set up the meteorological factor forecast model of sparse own coding (ASAE) deep learning of alternating.According to step 2
Complex nonlinear relation between the characteristic distributions and characteristic variable and meteorological factor of middle training dataset, sets up based on depth
Alternating sparse own coding (ASAE) the meteorological factor forecast model of study, the input matrix X of the model is normalization in step 2
Training dataset after conversion, predicts that output matrix isThe quantity of sparse own coding (ASAE) the model hidden layer of alternating is set
And the neuronal quantity of every layer of hidden layer, meteorological factor forecast model may be used to lower formula and successively calculates：
In formula, i represents i-th layer of model, and X represents i-th layer of input matrix, and W, b represent last layer and this layer of net respectively
The connection weight matrix of network and bias matrix,It is this layer of corresponding prediction output matrix,mi
I-th layer of neuron number for including is represented,Represent i-th layer of hidden layer the 1st neuron output, z represent i-th layer it is refreshing
The weighted sum of Jing units, f (z) represent the activation value of weighted sum, the linear function of conventional activation primitive, inclined-plane function, threshold value letter
Number, sigmoid function, hyperbolic tangent function etc., the activation primitive that the present invention is selected are that sigmoid function is as follows：
Using the output of last layer network as next layer network input, will last layer network output matrixAs
The input matrix X of next layer network, iteration is until calculating the output matrix of last layer of the model successively, model last layer
It is output as the meteorological factor predicted.
Step 4：Model training and parameter optimization.The sparse own coding of alternating combines unsupervised learning and supervised learning
The characteristics of, the supervised learning of unsupervised learning and BP neural network that autoencoder network is used alternatingly carries out the pre- instruction of network
Practice, in each pre-training, BP neural network has carried out arameter optimization using gradient descent method, the model only trains one layer of net every time
Network, and train the output data for obtaining as the input data of next layer network the layer network, i.e., successively wolfishly train hidden
Layer is hidden, until having trained all layers of network, the parameter that alternating sparse autoencoder network learns transferring weights is carried out into finally,
Adjust back in the top-down parameter to all layers of top additional layer BP neural network.The generation of the network parameter tuning
Valency function is as follows：
In formula, λ is weight decay factor, balance squared and L2The proportion of weight decay, Wherein ρ is the openness binding occurrence of every layer network,It is the neuronal activation degree of every layer network, β is openness
Restriction factor, the openness proportion of controlling network parameter.Tuning is carried out using gradient descent method to network parameter in the present invention, ladder
Degree descent method be used for minimum model cost function, the algorithm be along cost function least squares error direction to network weight
Weight variable and bias variable are adjusted.It is in course of adjustment, cost function is become progressively smaller until and meets minimum error constraint
Or continuous 6 cost functions do not drop anti-increasings and then stop adjustment, finally give optimal model parameters.Unsupervised learning will be replaced and had
Supervised learning carries out the parameter after gradient descent method optimization and brings sparse own coding (ASAE) the deep learning regression forecasting mould of alternating into
In the forecast model that type establishment step is set up, the ASAE meteorological factor forecast models after being trained.
Step 5：Model application.Ship self performance data, ship's navigation data and maritime meteorology data input are walked
In the ASAE meteorological factor forecast models obtained after training in rapid four, it is possible to achieve output meteorological factor.
Such as the ASAE forecast model figures that Fig. 2 is successively greedy training.Successively greedy training method is in model pre-training mistake
That is, in journey, successively training obtains optimized parameter, train hiding layer parameter unsupervisedly, and utilize supervised learning by parameter adjustment
To optimal value, optimized parameter is each layer connection weight matrix and bias matrix when cost function reaches minima.Alternating is dilute
Thin own coding model is the autoencoder network hidden layer stacking after successively being trained with supervised learning by alternately unsupervised learning
, the model top additional layer BP neural network and using gradient descent algorithm implementation model univers parameter
Parameter after overall tuning is substituted into ASAE models and obtains meteorological factor forecast model, navigation number most to be predicted at last by tuning
According to the ASAE meteorological factor forecast models obtained after input training optimization, output meteorological factor realizes maritime meteorology to ship
The prediction of advance speed effect.Estimated performance contrast of the table 1 for ANN, SAE and ASAE models proposed by the present invention, these three models are adopted
Model training is carried out as training sample with 7000 same datas.There it can be seen that ASAE models proposed by the present invention
Training time is most short and its mean square error for predicting the outcome is minimum.
The estimated performance contrast of 1 ANN, SAE and ASAE model of table
Those skilled in the art can carry out various remodeling and change to the present invention.Therefore, present invention covers falling into
Various remodeling and change in the range of appending claims and its equivalent.
Claims (2)
1. a kind of method according to maritime meteorology information prediction ship's navigation meteorological factor, it is characterised in that walk including following five
Suddenly：
Set up ship's navigation data base：Ship performance data, including captain, the beam, ship sky are obtained by ship visit report
Loading capacity；Ship's navigation historical data is obtained by history log record, including hours underway, longitude, latitude, drinking water are deeply
Degree, cargo dead-weight, Angle of Trim, main frame maximum-continuous rating (MCR), ship rated speed, speed on the ground, oil consumption, meteorological factor；Collect and
The Weather Forecast Information issued from U.S.National Oceanic and Atmospheric Administration is used as maritime meteorology data, including wind direction, wind speed, wave
High, wave to, gush height, pour into, gush cycle, wave height, wave direction, period of wave, sea pressure；Finally ship is built using above initial data
Oceangoing ship aeronautical data storehouse；
Data prediction：Initial data for collecting in ship's navigation data base carries out data selection, is there is number in data
Linear interpolation supplement is carried out according to the feature of disappearance, the multi-source data of integration is carried out into unified normalized finally, and will be returned
Data after one change are converted into the binary number of fixed digit, obtain model training data set；
Set up the sparse own coding deep learning regressive prediction model of alternating：The input matrix X of the model is data prediction step
In the data entered after line translation, prediction output matrix isThe quantity of the sparse own coding model hidden layer of alternating and every is set
The neuronal quantity of layer hidden layer, sets up the sparse own coding meteorological factor forecast model of alternating based on deep learning；Using with
Machine number initializes the weight matrix W and bias matrix b of network, and the output matrix computing formula per layer network isWherein output matrixI represents i-th layer of model, mi
I-th layer of neuron number for including is represented, z represents the weighted sum of i-th layer of neuron, and f (z) represents the activation value of weighted sum,
Represent the output of the 1st neuron of i-th layer of hidden layer；Using the output of last layer network as next layer network input, i.e.,
By the output matrix of last layer networkUsed as the input matrix X of next layer network, iteration is until calculating the model most successively
The output matrix of later layer, model last layer are output as the meteorological factor predicted；
Model training and parameter optimization：Based on the training data obtained in data prediction step, according to the sparse own coding of alternating
One cost function of net definitions, the target of the cost function are to solve for L2Weight decay constraint and KL are openness to sdpecific dispersion
The convex quadratic programming problem of the lower square error minimization of loss of constraint, that is, solve Wherein L2Weight decay constraint weighs weight parameter by setting weight decay factor coefficient lambda
With the proportion predicted the outcome between square error term, the phenomenon of model over-fitting is prevent to some extent, KL is to sdpecific dispersion
Constraint is defined to the openness of parameter in model training, has obtained model openness by setting openness factor beta
Parameter learning.The final alternately unsupervised learning that combines is trained to model and utilizes gradient descent method with supervised learning method
Carry out parameter optimization, wherein unsupervised learning is that self study i.e. output matrix is carried out to input matrix X is also X, for the ease of area
The output matrix of unsupervised learning is divided to use hereinRepresent, unsupervised training obtains this layer of hidden layer output, supervised learning is
The output matrix of unsupervised learning network is changed to label data to be trained, i.e., output matrix be true meteorological factor y, gradient
Descent method be used for minimum model cost function, the algorithm be along cost function least squares error direction to network weight
Matrix and bias matrix are adjusted.Be in course of adjustment, cost function be become progressively smaller until meet minimum error constraint or
Continuous 6 cost functions do not drop anti-increasings and then stop adjustment, finally give optimal model parameters.Bring the parameter after optimization into foundation
In the forecast model of the sparse own coding deep learning regressive prediction model step of alternating, the ASAE meteorological factors after being trained are pre-
Survey model；
Model application：By aeronautical data to be predicted, i.e. ship self performance data, ship's navigation data and maritime meteorology number
According to the ASAE meteorological factor forecast models after training in input model training and parameter optimisation step, and export meteorological factor.
2. the method according to maritime meteorology information prediction ship's navigation meteorological factor as claimed in claim 1, it is characterised in that
Described fixed digit in the data prediction step is that eight-digit binary number, sixteen-bit binary or 32 two enter
System.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611129832.9A CN106599995B (en) | 2016-12-09 | 2016-12-09 | A method of according to maritime meteorology information prediction ship's navigation meteorological factor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611129832.9A CN106599995B (en) | 2016-12-09 | 2016-12-09 | A method of according to maritime meteorology information prediction ship's navigation meteorological factor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106599995A true CN106599995A (en) | 2017-04-26 |
| CN106599995B CN106599995B (en) | 2019-04-30 |
Family
ID=58598111
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611129832.9A Active CN106599995B (en) | 2016-12-09 | 2016-12-09 | A method of according to maritime meteorology information prediction ship's navigation meteorological factor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106599995B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292451A (en) * | 2017-07-18 | 2017-10-24 | 上海海阳气象导航技术有限公司 | A kind of ship speed optimization method and equipment |
| CN107967920A (en) * | 2017-11-23 | 2018-04-27 | 哈尔滨理工大学 | A kind of improved own coding neutral net voice enhancement algorithm |
| CN108259136A (en) * | 2018-01-12 | 2018-07-06 | 重庆邮电大学 | A kind of intelligence sojourns in the cross-module state Intellisense method of business touring car |
| CN109902885A (en) * | 2019-04-09 | 2019-06-18 | 中国人民解放军国防科技大学 | Typhoon prediction method based on deep learning mixed CNN-LSTM model |
| CN110525602A (en) * | 2019-08-13 | 2019-12-03 | 浙江海洋大学 | A kind of ship track speed of a ship or plane integrated planning method |
| CN110705797A (en) * | 2019-10-09 | 2020-01-17 | 浙江海洋大学 | Ship oil consumption data prediction method based on ship sensor network |
| CN111325402A (en) * | 2020-02-21 | 2020-06-23 | 东南大学 | Method for predicting charging behavior of electric vehicle user based on BP neural network |
| WO2020212973A1 (en) * | 2019-04-18 | 2020-10-22 | Orca Ai Ltd. | Marine data collection for marine artificial intelligence systems |
| CN112836893A (en) * | 2021-02-26 | 2021-05-25 | 上海海事大学 | Method for predicting ship oil consumption under severe sea conditions based on sea condition and ship navigation conditions |
| CN113516565A (en) * | 2021-04-08 | 2021-10-19 | 国家电网有限公司 | Intelligent alarm processing method and device for power monitoring system based on knowledge base |
| CN113739807A (en) * | 2021-11-08 | 2021-12-03 | 聊城中翔泰电子科技有限公司 | Navigation route navigation method and system for ship |
| US11315046B1 (en) * | 2018-06-28 | 2022-04-26 | Ashton Robinson Cook | Machine learning-based disaster modeling and high-impact weather event forecasting |
| CN114659503A (en) * | 2022-03-17 | 2022-06-24 | 广东蓝鲲海洋科技有限公司 | Marine ecological environment monitoring method based on artificial intelligence |
| US20220343221A1 (en) * | 2018-06-28 | 2022-10-27 | Ashton Robinson Cook | Machine learning-based disaster modeling and high-impact weather event forecasting |
| CN115675780A (en) * | 2022-12-30 | 2023-02-03 | 武汉理工大学 | Ship draught prediction method and system, electronic equipment and readable storage medium |
| CN117453751A (en) * | 2023-12-22 | 2024-01-26 | 中国海洋大学 | Ocean big data cache loading system, operation method, device and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778671A (en) * | 2015-04-21 | 2015-07-15 | 重庆大学 | Image super-resolution method based on SAE and sparse representation |
| CN104809469A (en) * | 2015-04-21 | 2015-07-29 | 重庆大学 | Indoor scene image classification method facing service robot |
| CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
-
2016
- 2016-12-09 CN CN201611129832.9A patent/CN106599995B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778671A (en) * | 2015-04-21 | 2015-07-15 | 重庆大学 | Image super-resolution method based on SAE and sparse representation |
| CN104809469A (en) * | 2015-04-21 | 2015-07-29 | 重庆大学 | Indoor scene image classification method facing service robot |
| CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| 吕远: "基于稀疏编码的极化SAR影像地物分类", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陈汉英: "基于稀疏编码的半监督图像分类研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292451A (en) * | 2017-07-18 | 2017-10-24 | 上海海阳气象导航技术有限公司 | A kind of ship speed optimization method and equipment |
| CN107292451B (en) * | 2017-07-18 | 2020-11-03 | 上海海阳气象导航技术有限公司 | Ship speed optimization method and equipment |
| CN107967920A (en) * | 2017-11-23 | 2018-04-27 | 哈尔滨理工大学 | A kind of improved own coding neutral net voice enhancement algorithm |
| CN108259136B (en) * | 2018-01-12 | 2020-10-16 | 重庆邮电大学 | Cross-mode intelligent sensing method for intelligent sojourn business motor home |
| CN108259136A (en) * | 2018-01-12 | 2018-07-06 | 重庆邮电大学 | A kind of intelligence sojourns in the cross-module state Intellisense method of business touring car |
| US20220343221A1 (en) * | 2018-06-28 | 2022-10-27 | Ashton Robinson Cook | Machine learning-based disaster modeling and high-impact weather event forecasting |
| US20220327433A1 (en) * | 2018-06-28 | 2022-10-13 | Ashton Robinson Cook | Machine learning-based disaster modeling and high-impact weather event forecasting |
| US11315046B1 (en) * | 2018-06-28 | 2022-04-26 | Ashton Robinson Cook | Machine learning-based disaster modeling and high-impact weather event forecasting |
| CN109902885B (en) * | 2019-04-09 | 2020-01-07 | 中国人民解放军国防科技大学 | Typhoon prediction method based on deep learning mixed CNN-LSTM model |
| CN109902885A (en) * | 2019-04-09 | 2019-06-18 | 中国人民解放军国防科技大学 | Typhoon prediction method based on deep learning mixed CNN-LSTM model |
| WO2020212973A1 (en) * | 2019-04-18 | 2020-10-22 | Orca Ai Ltd. | Marine data collection for marine artificial intelligence systems |
| CN110525602A (en) * | 2019-08-13 | 2019-12-03 | 浙江海洋大学 | A kind of ship track speed of a ship or plane integrated planning method |
| CN110705797B (en) * | 2019-10-09 | 2023-09-22 | 浙江海洋大学 | Ship fuel consumption data prediction method based on ship sensing network |
| CN110705797A (en) * | 2019-10-09 | 2020-01-17 | 浙江海洋大学 | Ship oil consumption data prediction method based on ship sensor network |
| CN111325402A (en) * | 2020-02-21 | 2020-06-23 | 东南大学 | Method for predicting charging behavior of electric vehicle user based on BP neural network |
| CN112836893A (en) * | 2021-02-26 | 2021-05-25 | 上海海事大学 | Method for predicting ship oil consumption under severe sea conditions based on sea condition and ship navigation conditions |
| CN112836893B (en) * | 2021-02-26 | 2024-05-14 | 上海海事大学 | Method for predicting ship oil consumption under severe sea conditions based on sea condition and ship navigation condition |
| CN113516565A (en) * | 2021-04-08 | 2021-10-19 | 国家电网有限公司 | Intelligent alarm processing method and device for power monitoring system based on knowledge base |
| CN113739807A (en) * | 2021-11-08 | 2021-12-03 | 聊城中翔泰电子科技有限公司 | Navigation route navigation method and system for ship |
| CN114659503A (en) * | 2022-03-17 | 2022-06-24 | 广东蓝鲲海洋科技有限公司 | Marine ecological environment monitoring method based on artificial intelligence |
| CN115675780A (en) * | 2022-12-30 | 2023-02-03 | 武汉理工大学 | Ship draught prediction method and system, electronic equipment and readable storage medium |
| CN117453751A (en) * | 2023-12-22 | 2024-01-26 | 中国海洋大学 | Ocean big data cache loading system, operation method, device and medium |
| CN117453751B (en) * | 2023-12-22 | 2024-03-26 | 中国海洋大学 | Ocean big data cache loading system, operation method, device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106599995B (en) | 2019-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106599995B (en) | A method of according to maritime meteorology information prediction ship's navigation meteorological factor | |
| Xu et al. | A research on coordination between economy, society and environment in China: A case study of Jiangsu | |
| Bui-Duy et al. | Utilization of a deep learning-based fuel consumption model in choosing a liner shipping route for container ships in Asia | |
| CN110378799A (en) | Aluminium oxide comprehensive production index decision-making technique based on multiple dimensioned depth convolutional network | |
| CN103049798B (en) | A kind of short-term power generation power Forecasting Methodology being applied to photovoltaic generating system | |
| CN102005135A (en) | Genetic algorithm-based support vector regression shipping traffic flow prediction method | |
| CN101782743A (en) | Neural network modeling method and system | |
| CN107292534A (en) | The yardstick competition evaluation method and device of urban power distribution network long term dynamics investment | |
| CN108074004A (en) | A kind of GIS-Geographic Information System short-term load forecasting method based on gridding method | |
| CN104992244A (en) | Airport freight traffic prediction analysis method based on SARIMA and RBF neural network integration combination model | |
| CN108596242A (en) | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines | |
| CN113033917A (en) | Sewage treatment plant prediction planning operation management method based on peripheral data | |
| CN117236199B (en) | Method and system for improving water quality and guaranteeing water safety of river and lake in urban water network area | |
| CN114862035B (en) | Combined bay water temperature prediction method based on transfer learning | |
| CN102298706A (en) | Inland waterway ship large-scale prediction method in restricted conditions | |
| Xu et al. | Evaluation of island tourism sustainable development in the context of smart tourism | |
| Li et al. | Estimating waterway freight demand at Three Gorges ship lock on Yangtze River by backpropagation neural network modeling | |
| CN107992967A (en) | Based on the ship lock dispatching method for improving multi-objective genetic algorithm | |
| Wei et al. | A time-varying ensemble model for ship motion prediction based on feature selection and clustering methods | |
| CN112183721B (en) | Construction method of combined hydrological prediction model based on self-adaptive differential evolution | |
| Li et al. | Optimization approach of berth-quay crane-truck allocation by the tide, environment and uncertainty factors based on chaos quantum adaptive seagull optimization algorithm | |
| CN109615142A (en) | A kind of wind farm wind velocity combination forecasting method based on wavelet analysis | |
| Shimray et al. | A new MLP–GA–Fuzzy decision support system for hydro power plant site selection | |
| CN104463683B (en) | A kind of Mid-long term load forecasting method of power grid containing multi-source | |
| Long | Analysis of the key factors of ecological environment protection in the national economic sustainable development goals |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |