CN110298611A - Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning - Google Patents
Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning Download PDFInfo
- Publication number
- CN110298611A CN110298611A CN201910410132.4A CN201910410132A CN110298611A CN 110298611 A CN110298611 A CN 110298611A CN 201910410132 A CN201910410132 A CN 201910410132A CN 110298611 A CN110298611 A CN 110298611A
- Authority
- CN
- China
- Prior art keywords
- data
- variable
- optimal
- deep learning
- model
- 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
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Method is regulated and controled based on the cargo shipping efficiency of random forest and deep learning the invention proposes a kind of, reads multiple ship's navigation data samples；Optimal stochastic forest model is constructed, it is calculated and predicts error, and extract important feature variable；Optimal depth learning model is constructed, it is calculated and predicts error；It reads real-time ship's navigation data and obtains real-time aeronautical data characteristic variable, carry out first time prediction using real-time aeronautical data characteristic variable as input variable input optimal stochastic forest model, input in optimal depth learning model and predict for second of progress；First time prediction result and second of prediction result are weighted processing, obtain the final predicted value of cargo shipping efficiency, and seek confidence interval；Ship's navigation operation data is adjusted according to the corresponding relationship of the final predicted value of cargo shipping efficiency of acquisition and confidence interval.This method is simple and effective, can preferably cargo shipping efficiency be predicted and be regulated and controled, and realizes the optimization of cargo shipping efficiency.
Description
Technical field
The present invention relates to computer fields, and in particular to a kind of cargo shipping efficiency based on random forest and deep learning
Regulate and control method and system.
Background technique
Shipping is a kind of epochmaking means of transportation as water transportation, and status is very important.In shipping, ship
Shipping efficiency refers to that same model ship when transporting the cargo of equal unit, travels the displacement institute of unit length within the unit time
The oil mass of consumption, therefore cargo shipping efficiency is an important parameter index of the ship in navigational duty.
The many because being known as of cargo shipping efficiency are influenced, such as engine speed, gear oil pressure, left and right tailing axle revolving speed, GPS
Signal intensity data, GPS longitude and latitude data, weather data, drauht data, upper and lower water number evidence, oil tank liquid level data, start
Machine power, load-carrying draining, ship load information etc..Operate in ship can under optimal efficiency under different factors
It is the emphasis now studied at present.
Summary of the invention
In order to overcome above-mentioned defect existing in the prior art, the object of the present invention is to provide one kind based on random forest with
The cargo shipping efficiency of deep learning regulates and controls method and system.
In order to realize above-mentioned purpose of the invention, the present invention provides a kind of ship based on random forest and deep learning
Shipping efficiency regulates and controls method, includes the following steps S1-S6 or step S1-S5 or step S6:
S1, multiple ship's navigation data samples in reading database, each sample include one group of aeronautical data characteristic variable
And cargo shipping efficiency corresponding with this group of aeronautical data characteristic variable, random sampling, aeronautical data are carried out to these samples
Characteristic variable includes vessel motion data characteristics variable and ship status data characteristics variable；
S2 constructs optimal stochastic forest model, using the resulting aeronautical data characteristic variable of random sampling as random forest
The input variable of model carries out first time test, calculates the prediction error of optimal stochastic forest model, and obtained by the random sampling
Aeronautical data characteristic variable in extract important feature variable；
S3 constructs optimal deep learning model, using the important feature variable as the input of optimal deep learning model
Variable carries out second and tests, calculates the prediction error of optimal deep learning model；
S4 reads real-time ship's navigation data and obtains real-time aeronautical data characteristic variable, real-time aeronautical data feature is become
Amount carries out first time prediction as input variable input optimal stochastic forest model, carries out the in input optimal deep learning model
Re prediction；
First time prediction result and second of prediction result are weighted processing, it is final to obtain cargo shipping efficiency by S5
Predicted value, and seek confidence interval；
S6 adjusts ship's navigation fortune according to the corresponding relationship of the final predicted value of cargo shipping efficiency of acquisition and confidence interval
Row data are optimal vessel motion efficiency.
The present invention can cross and carry out the first prediction first with Random Forest model, then be carried out second by deep learning model
Prediction, the result predicted twice is weighted to obtain final cargo shipping efficiency and confidence interval, according to the ship of acquisition
The corresponding relationship of shipping efficiency and confidence interval adjusts ship's navigation operation data, is optimal vessel motion efficiency, this reinforcement
The accuracy of regulation cargo shipping efficiency.
Preferably, the vessel motion data characteristics variable includes ship power system data, the ship power system
Data include following type: one of engine speed, gear oil pressure, tailing axle revolving speed or any combination, the ship status number
Include following type according to characteristic variable: GPS signal delta data, weather data, drauht data, the upper and lower water number evidence of ship,
One of ship automatic control information or any combination；
The type of type and ship status data to vessel motion data is sampled, resulting vessel motion number of sampling
According to number of species be greater than ship status data number of species.
Since operation data can be adjusted in time by shipping work personnel, such as engine speed, and state variable is short
It is difficult to adjust in time, such as water number evidence, shipp. wt, this method stress to operation data and status data up and down
Property random sampling, preferably protrude the importance of operation data in final prediction result, this makes by improved random
Forest can be more prone to operation variable to the prediction of characteristic variable importance, for optimal cargo transport efficiency, allow shipping work people
Member is adjusted in time for operation variable.
A preferred embodiment of the present invention, the step S2 specifically:
S21, in Random Forest model, first by ship's navigation data sample be divided into training set, verifying collect and test
Collection；
S22, the default parameters using Random Forest model are trained with training set to Random Forest model training
Random Forest model；
S23, trained Random Forest model is verified with verifying collection, obtains the first model error；
S24, some or all of Random Forest model default parameters is combined with trellis search method, is obtained random
The multiple groups parameter combination of forest model；
S25, using cross validation method, the Random Forest model under each parameter combination is tested with verifying collection
Card, obtains the error of the Random Forest model of every group of parameter combination；
S26, by the error of the Random Forest model of the every group of parameter combination obtained after the first model error and cross validation
It is compared, error minimum is optimal stochastic forest model；
S27, first time test is carried out to test set using optimal stochastic forest model, and calculates optimal stochastic forest model
Prediction error, using optimal stochastic forest model to input data extract important feature variable.
A preferred embodiment of the present invention, the extracting method of the important feature variable are as follows: with gini index by variable weight
The property wanted scoring is indicated with VIM, it is assumed that has m characteristic variable X1, X2, X3..., Xm；
Each characteristic variable XjGini index score calculation formula are as follows:
Wherein, K indicates the species number of characteristic variable, PmkIndicate section
Ratio shared by k-th of type in point m, i.e., at will randomly select two samples from node m, inconsistent general of category flag
Rate, the supplementary set of k-th of type, P in the species number K of k ' expression characteristic variablemk'=1-Pmk；
Characteristic variable XjGini index variation amount in the importance of node m, i.e., before and after node m branch are as follows:Wherein GIlAnd GIrRespectively indicate the index of the Gini of branch latter two new node.
If characteristic variable XjWhether the node occurred in decision tree i or not in set M, then feature XjIn decision tree i
Without importance；
If characteristic variable XjThe node occurred in decision tree a in set M, set M be decision tree a in root node and
The set of leaf node, then feature XjIn the importance of decision tree a are as follows:
Assuming that have N tree in random forest, then
All importance acquired, which are done normalized, can be obtained important feature variables reordering, normalize calculation formula
Are as follows:C is characterized the total quantity of variable, to obtain the sequence of m characteristic variable, first according to sequence
After important feature variable can be obtained.
A preferred embodiment of the present invention, step S3 specifically:
S31, it is normalized using the important feature variable that random forest proposes as input variable, and to it, it will
Input variable and predictive variable normalized value 0 to 1, normalization mode are as follows:
Wherein,databRespectively indicate the important feature of input
The data maximums of each characteristic variable type in variable, data minimum value and data item are used as depth after being normalized
Learning model building data, b are 0 to the positive integer between B, and wherein B is the number of species of important characteristic variable；
S32, model training is carried out to keras deep neural network, constructs the network structure of kears deep neural network
Input layer, Dense layers, loss function, optimizer specifies monitor control index；
S33, the activation primitive of keras deep learning model is optimized；
The expression formula that activation primitive after optimization is is
S34, keras deep learning model parameter is adjusted using back-propagation algorithm, obtains optimal keras deep learning mould
Type, and tested by optimal keras deep learning model, and calculate the prediction error of optimal keras deep learning model.
The used activation primitive of deep learning method can preferably solve gradient and disappear and gradient explosion issues.
A preferred embodiment of the present invention, the weighting of first time prediction result and second of prediction result in the step S5
Processing method are as follows:
S51, using first time predicted value and second of predicted value as the input value of final prediction result；
S52, according to the prediction error of optimal stochastic forest model and optimal deep learning model, determine optimal stochastic forest
The weight of model and optimal deep learning model predicted value, first time predicted value and second of predicted value are weighted and are summed
To the final predicted value Y of cargo shipping efficiency.
The final predicted value Y of cargo shipping efficiency are as follows:
Wherein, RMSE4, RMSE5 respectively indicate prediction error to identical sample set under optimal stochastic forest model,
Prediction error under optimal deep learning model, Y1、Y2It respectively indicates to same group of input data under optimal stochastic forest model
Predicted value and the predicted value under deep learning model.
The method of weighting is simple and effective, can quickly obtain the final predicted value Y of cargo shipping efficiency.
A preferred embodiment of the present invention, confidence interval acquiring method are as follows:
To same group of real-time aeronautical data characteristic variable of reading respectively in optimal stochastic forest model and optimal depth
It practises and repeatedly being predicted in model, the predicted value of each optimal stochastic forest model and optimal deep learning model is weighted
Summation, then takes standard deviation δ to the result of each weighted sum again, and obtaining confidence interval is [Y- δ, Y+ δ]；
Or, carrying out every group of characteristic variable input optimal deep learning model to repeat p prediction in step s3, according to every
The output result of secondary prediction seeks the standard deviation for the predicted value that every group of characteristic variable input optimal deep learning model obtains, and is denoted as
δ1, δ1... δq, take mean value to obtain the standard deviation acquiredConfidence interval isWherein Y is ship goods
Transport the final predicted value of efficiency.Increase the accuracy of the final predicted value Y of cargo shipping efficiency.
A preferred embodiment of the present invention, the step S5 are as follows:
From being selected in historical data under identical ship status data closest to the ship of the final predicted value y of cargo shipping efficiency
Oceangoing ship aeronautical data characteristic variable sample, using the vessel motion data in the ship's navigation data characteristics variable sample as current ship
The adjustment target of the operation data of oceangoing ship.
The invention also provides a kind of cargo shipping efficiency regulator control systems, including vessel motion data acquisition unit, ship
State data acquisition unit and control unit, described control unit is by above-mentioned cargo shipping efficiency regulation method to cargo shipping
Efficiency is regulated and controled.The cargo shipping efficiency regulator control system structure is simple, and the personnel of steering a ship is enable fast and accurately to regulate and control ship
To under optimal cargo shipping efficiency, shipping efficiency is improved.
Additional aspect and advantage of the invention will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is building optimal stochastic forest model flow chart；
Fig. 3 is building deep learning model flow figure；
Fig. 4 is deep learning model output value calculation method schematic diagram.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, and for explaining only the invention, and is not considered as limiting the invention.
In the description of the present invention, unless otherwise specified and limited, it should be noted that term " installation ", " connected ",
" connection " shall be understood in a broad sense, for example, it may be mechanical connection or electrical connection, the connection being also possible to inside two elements can
, can also indirectly connected through an intermediary, for the ordinary skill in the art to be to be connected directly, it can basis
Concrete condition understands the concrete meaning of above-mentioned term.
As shown in Figure 1, the present invention provides a kind of cargo shipping efficiency regulation side based on random forest and deep learning
Method includes the following steps S1-S6 or step S1-S5 or step S6:
S1, multiple ship's navigation data samples in reading database, each sample include one group of aeronautical data characteristic variable
And cargo shipping efficiency corresponding with this group of aeronautical data characteristic variable, random sampling, aeronautical data are carried out to these samples
Characteristic variable includes vessel motion data characteristics variable and ship status data characteristics variable.Vessel motion data characteristics variable and
The difference of ship status data characteristics variable is that operation data characteristic variable can adjust in time by shipping work personnel, such as
Engine speed, status data are difficult to adjust in the characteristic variable short time, such as water number evidence up and down, shipp. wt.Here, it deletes
Carry out exceptional value and missing values in multiple ship's navigation data characteristics variable samples.
Here, vessel motion data characteristics variable includes ship power system data, and the ship power system data are excellent
It selects but is not limited to include following type: one of engine speed, gear oil pressure, tailing axle revolving speed or any combination.Engine turns
Speed includes left and right engine speed, and gear oil pressure includes left and right gear oil pressure, and tailing axle revolving speed turns comprising left and right tailing axle
Speed.Ship status data characteristics variable is preferably but not limited to include following type: GPS signal delta data, weather data, ship
Data are absorbed water, water number evidence, one of ship automatic control information or any combination above and below ship.GPS signal delta data includes GPS longitude and latitude
Degree evidence, weather data include wind speed, and ship automatic control information includes fuel tank liquid level data, engine power, load-carrying draining etc., ship
Oceangoing ship weight information includes cargo transport weight etc..
The type of type and ship status data characteristics variable to vessel motion data characteristics variable is sampled, sampling
The number of species of resulting vessel motion data characteristics variable are greater than the number of species of ship status data characteristics variable.For example,
6 variables in vessel motion data characteristics variable carry out high proportion sampling, choose 4 every time, ship status data characteristics becomes
Amount carries out low proportional sampling and extracts 2, to form the sample variable of a building one tree, the main purpose done so
The important of vessel motion data characteristics variable in final prediction result is preferably protruded also for than traditional random forest method
Property, convenient for regulating and controlling cargo shipping efficiency by adjusting controllable vessel motion data characteristics variable.
S2 constructs optimal stochastic forest model, using the resulting aeronautical data characteristic variable of random sampling as random forest
The input variable of model carries out first time test, calculates the prediction error of optimal stochastic forest model, and obtained by the random sampling
Aeronautical data characteristic variable in extract important feature variable.
Following steps are specifically included, as shown in Figure 2:
S21, in Random Forest model, first by ship's navigation data sample be divided into training set, verifying collect and test
Collection, the ration of division are preferably but not limited to as 7:2:1.
S22, the default parameters using Random Forest model are trained with training set to Random Forest model training
Random Forest model.Default parameters includes quantity, model the number of iterations, the model learning rate etc. of random forest building tree.
S23, trained Random Forest model is verified with verifying collection, obtains the first model error.
Specifically, the calculation method of the first model error are as follows:
Verifying concentrates each ship's navigation data sample to have one group of aeronautical data characteristic variable and cargo shipping efficiency y1,
In Random Forest model, aeronautical data characteristic variable is input variable, and Random Forest model output is the ship goods predicted
Transport efficiency yo, the input variable of verifying collection is input to training set and is trained in the Random Forest model come, one group will be obtained
As a result yo；yoWith y1The error function of model, i.e. the first model error, the first model error can be found outN is the number that ship's navigation data sample is concentrated in the verifying, i=1,2,3 ... n, yoiTable
Show the cargo shipping efficiency of the prediction of i-th of sample, y1iIndicate the practical cargo shipping efficiency of i-th of sample.
S24, some or all of Random Forest model default parameters is combined with trellis search method, is obtained random
The multiple groups parameter combination of forest model.
For example, by taking the part default parameters to Random Forest model is combined as an example, the important ginseng of Random Forest model
Number has n_estimators (n tree of building) and max_features (most independents variable that each tree randomly selects), this implementation
The two default parameters are just chosen in example to be combined.N_estimoators is always the bigger the better, for more setting mean value
Over-fitting can be reduced, is preferably integrated to obtain robustness, but income is successively decreased, and more trees can also expend
More memories.And max_features determines the randomness size of every number, lesser max_features can be reduced
Fitting, and carrying out trellis search method to Random Forest model is then that we attempt n_estimators and take 6 values, max_
Features also takes 6 values, has 6 different to take since we want the n_estimators attempted and max_features
Value, so just there is 36 kinds of parameter combinations.
S25, using cross validation method, the Random Forest model under each parameter combination is tested with verifying collection
Card, obtains the error of the Random Forest model of every group of parameter combination.
Concretely, cross validation method generallys use K folding, i.e. K is the number that user specifies, and usually goes 5 or 10.With 5
For folding, when executing 5 folding cross validation, verifying collection is divided into 5 roughly equal parts first, each section is called one
Folding, the following Random Forest model under one group of parameter combination of training, done using the first folding test set in cross validation, its
He does the training set in cross validation to train the Random Forest model under first group of parameter combination by folding, and obtaining should under this time verifying
The error of Random Forest model, the calculation method of error here are under group parameter combination
n1The number of ship's navigation data sample, i are concentrated for the verifying1=1,2,3 ... n1,Indicate i-th1A sample is at first group
The prediction cargo shipping efficiency of Random Forest model under parameter combination,Indicate i-th1The practical cargo shipping of a sample is imitated
Rate.
Second of test set made in cross validation of 2 foldings, other foldings do the training set in cross validation to train first
The error of Random Forest model under lower this group of parameter combination is verified in Random Forest model under group parameter combination, this time, here
Error calculation method is identical as the calculation method of RMSE2, and so on, take mean value as first group of ginseng the K error obtained
The error of Random Forest model under array conjunction, which is that the Random Forest model Generalization Capability under first group of parameter combination is good
Bad index.
All being verified using above-mentioned cross validation method to the Random Forest model under each group of parameter combination can
Obtain the error of the Random Forest model under each group of parameter combination.
S26, by the Random Forest model of the every group of parameter combination obtained after the first model error RMSE1 and cross validation
Error is compared, and error minimum is optimal stochastic forest model.
S27, first time prediction is carried out to test set using optimal stochastic forest model, obtains testing for the first time, such as 1 institute of table
Show, and calculate the prediction error of optimal stochastic forest model, important spy is extracted to input data using optimal stochastic forest model
Levy variable.
Table 1 first time test result
Here the extracting method of important feature variable are as follows:
The extracting method of the important feature variable are as follows: variable importance scoring is indicated with VIM with gini index, it is false
Equipped with m characteristic variable X1, X2, X3..., Xm；
Each characteristic variable XjGini index score calculation formula are as follows:
Wherein, K indicates the species number of characteristic variable, PmkIndicate section
Ratio shared by k-th of type in point m, i.e., at will randomly select two samples from node m, inconsistent general of category flag
Rate, this can be directly obtained in optimal stochastic forest model, the supplementary set of k-th of type in the species number K of k ' expression characteristic variable,
Pmk'=1-Pmk。
Characteristic variable XjGini index variation amount in the importance of node m, i.e., before and after node m branch are as follows:Wherein GIlAnd GIrRespectively indicate the index of the Gini of branch latter two new node.
If characteristic variable XjWhether the node occurred in decision tree i or not in set M, then feature XjIn decision tree i
Without importance；
If characteristic variable XjThe node occurred in decision tree a in set M, set M be decision tree a in root node and
The set of leaf node, then feature XjIn the importance of decision tree a are as follows:
Assuming that have N tree in random forest, then
Finally, all importance acquired, which are done normalized, can be obtained important feature variables reordering, normalization meter
Calculate formula are as follows:C is characterized the total quantity of variable, so that the sequence of m characteristic variable is obtained, according to
Important feature variable successively can be obtained in sequence.
S3 constructs optimal deep learning model, using the important feature variable as the input of optimal deep learning model
Variable carries out second and tests, calculates the prediction error of optimal deep learning model.
Step S3 specifically:
S31, it is normalized using the important feature variable that random forest proposes as input variable, and to it, it will
Input variable and predictive variable normalized value 0 to 1, normalization mode are as follows:
Wherein,databRespectively indicate each characteristic variable kind in the important feature variable of input
The data maximums of class, data minimum value and data item, after being normalized be used as deep learning modeling data, b be 0 to B it
Between positive integer, wherein B be important characteristic variable number of species.
The data flowchart of deep learning network modelling process as shown in Figure 3, S32, to keras deep neural network into
Row model training constructs the input layer of the network structure of kears deep neural network, and Dense layers, loss function, optimizer refers to
Determine monitor control index.
For deep learning model, with the continuous intensification of network layer, traditional activation primitive is more likely to produce gradient
Disappearance or gradient explosion phenomenon, so that neural network node inactivation be prevented to cause model from learning to input data and output data
Between characteristic relation.Solving the best approach that gradient disappearance is exploded with gradient is optimized to activation primitive.
S33, the activation primitive of keras deep learning model is optimized；
The expression formula that activation primitive after optimization is are as follows:
The activation primitive is made that limitation on the basis of original activation primitive, is not that the output valve of activation primitive is allowed to exist
It is directly 0 when input value is less than 0, but the output valve of activation primitive is arranged and is learnt in an a small range, for big
When 10, activation primitive is constant, and derivative 0, there is no the features of learning training data in this section for function.
For the data of input layer input, by taking d-th of hidden node in l layers as an example, the mode of calculating is as shown in Figure 4.It is first
First for B variable of input, random B weight w carries out summation operation:WhereinIt is l layers
The output of d-th of node,It is l-1 layers, i-th of input variable,It is l-1 layers, first input variable pair
The weight answered.After summarizing summation, after activation primitive, as output valve to next layer of transmitting.Each layer each section
Point is all to carry out operation in this manner, when data run can generate the reality of predicted value y and sample to the endCompare meeting
Generate prediction error.
The final purpose of neural network is to take minimum to the prediction error RMSE3 of g group ship's navigation data sample.
In formula: yhFor h group predicted value,For the actual efficiency value of h group.
S34, keras deep learning model parameter is adjusted using back-propagation algorithm, obtains keras deep learning model most
Optimal Parameters make neural network reach the condition for stopping learning and complete model training, it is deep to obtain optimal keras by successive ignition
Learning model is spent, there are two conditions described herein, first is that when the number of iterations is more than our customized parameters, second is that when missing
When difference reaches our permissible accuracies, which is existing method, and and will not be described here in detail；Pass through optimal keras depth again
Learning model is tested, and as shown in table 2, and calculates the prediction error of optimal keras deep learning model.
It should be noted that this programme is only to be changed to activation primitive in the building of optimal deep learning model
Into what remaining learning method was all made of is existing deep learning algorithm.Meanwhile in the present embodiment when deep learning, adopted
Ship's navigation data sample is identical as sample used in Random Forest model, only navigation employed in each sample
Data characteristics variable is different, the important feature variable that when deep learning extracts using Random Forest model, depth
It is also required to for ship's navigation data sample to be divided into training set, verifying collection and test set when habit, dividing method creation optimal stochastic
Consistent when forest model, i.e., training set, the verifying collection of two models are identical with test set.Ship's navigation data sample is divided into
The step of training set, verifying collection and test set, can be after the completion of step S1 carries out random sampling to multiple ship's navigation data samples
With regard to carrying out, optimal stochastic forest model, optimal deep learning model are constructed with training set and verifying collection respectively, then using most
Excellent Random Forest model carries out first time test to test set, carries out second to test set using optimal depth model and tests.
Second of the test result of table 2
S4 reads real-time ship's navigation data and obtains real-time aeronautical data characteristic variable, real-time aeronautical data feature is become
Amount carries out first time prediction as input variable input optimal stochastic forest model, inputs in optimal keras deep learning model
Second is carried out to predict.
First time prediction result and second of prediction result are weighted processing, it is final to obtain cargo shipping efficiency by S5
Predicted value, and seek confidence interval.
Specifically, in the step S4 first time prediction result and second of prediction result weighting processing method are as follows:
S51, using first time predicted value and second of predicted value as the input value of final prediction result；
S52, according to the prediction error of optimal stochastic forest model and optimal keras deep learning model, determine it is optimal with
The weight of machine forest model and optimal keras deep learning model predication value, to first time predicted value and second predicted value into
Row weighted sum obtains the final predicted value Y of cargo shipping efficiency.The final predicted value of cargo shipping efficiency
Wherein, RMSE4, RMSE5 are respectively indicated gloomy in optimal stochastic to identical sample set (the present embodiment middle finger test set)
The prediction error under prediction error, optimal keras deep learning model under woods model, Y1、Y2It respectively indicates defeated to same group
Enter predicted value of the data under optimal stochastic forest model and in deep learning model predication value, same group of input data here
Refer to and obtains real-time aeronautical data characteristic variable from the real-time ship's navigation data of reading.
Q indicate step S27 in using optimal stochastic forest model to test set into
When row is predicted for the first time, the number of ship's navigation data sample, r=1,2,3 ... q, y' in the test setrIndicate r-th of sample
The cargo shipping efficiency predicted under this optimal stochastic forest model, y "rIndicate the practical cargo shipping efficiency of r-th of sample.
After obtaining keras deep learning model the most optimized parameter in step S34,
It is predicted by the parameter model of optimization, is here still using the identical ship's navigation number with being used in step S27
It is predicted according to sample, that is, test set, therefore ship navigates in the number of the ship's navigation data sample in step S34 and step S27
The number of row data sample is identical.y"'rIndicate the ship predicted under the keras deep learning model of optimization of r-th of sample
Oceangoing ship shipping efficiency, y "rIndicate the practical cargo shipping efficiency of r-th of sample.
There are two types of confidence interval construction methods.
The first: is to same group of real-time aeronautical data characteristic variable of reading respectively in optimal stochastic forest model and optimal
Repeatedly predicted in deep learning model, to the predicted value of each optimal stochastic forest model and optimal deep learning model into
Row weighted sum, weighted sum method refer to step S51-S52, then take standard deviation δ to the result of each weighted sum again, obtain
It is [Y- δ, Y+ δ] to confidence interval.
Second: every group of characteristic variable being inputted into optimal keras deep learning model in step s3 and repeat p times in advance
It surveys, the mark for the predicted value that every group of characteristic variable input optimal deep learning model obtains is sought according to the output result predicted every time
It is quasi- poor, it is denoted as δ1, δ1... δq, take mean value to obtain the standard deviation acquiredConfidence interval isWherein Y
For the final predicted value of cargo shipping efficiency, as shown in table 3.
The 3 final prediction result of cargo shipping efficiency of table
S5 reads real-time ship's navigation data, according to the corresponding relationship tune of the cargo shipping efficiency of acquisition and confidence interval
Whole ship's navigation operation data, is optimal vessel motion efficiency.
In particular, being from being selected in historical data under identical ship status data closest to cargo shipping efficiency most
The ship's navigation data characteristics variable sample of whole predicted value Y, by the vessel motion in the ship's navigation data characteristics variable sample
Adjustment target of the data as the operation data of current ship.
For example: according to a timing of above-mentioned optimal stochastic forest model and optimal keras deep learning model prediction
Cargo shipping efficiency in section, with reference to 2, has chosen respectively with reference to 3 and imitates with the close shipping of shipping efficiency under current state with reference to 1
The ship historical state data of rate value.
It is described in detail: being modified if ship personnel carry out operation variable according to table 3, cargo shipping efficiency will herein
Optimization.Why not according to the final predicted value Y of cargo shipping efficiency, the reason of calculating the parameter combination of operation variable be exactly by
It is theoretical values in these parameter combinations, ship may not necessarily reach.For example, if in order to most seek optimal shipping efficiency, it can
Can calculate engine speed is 150rpm, and engine of boat and ship will not be travelled with this revolving speed.Therefore we select
Relevant history reference point is provided, in this way when optimal cargo transport efficiency can be according to historical storage data successive optimization here.
To some explanations of table 3:
For column, first is classified as table forefront, and second to be classified as " current state " be by shipping efficiency is database reality
When be calculated by come." prediction future time section efficiency " is by optimal stochastic forest model and optimal keras deep learning
Model combine predict come, although and it may be seen that prediction have error, as long as add confidence interval model
It encloses and contains the value 0.0339 of current state.And for being historic state with reference to 1,2,3, it is in our historical data base
The historical data of record, to vessel operation, personnel make reference.
Gear oil pressure in operation characteristic variable is classified variable, wherein 1 when representing normally travel, oil pressure is in normal model
In enclosing, 0 represents oil pressure lower than normal range (NR), and it is insufficient or the problems such as be short of power to will lead to engine oil at this time, so as to cause
Oil consumption is increased, and can increase oil pressure by injection lubricating oil.2 represent engine oil pressure higher than normal range (NR), this data we not
It is chosen, because needing the amendment that brings at this time.
The invention also provides a kind of cargo shipping efficiency regulator control systems, including vessel motion data acquisition unit, ship
State data acquisition unit and control unit, described control unit is by above-mentioned cargo shipping efficiency regulation method to cargo shipping
Efficiency is regulated and controled.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
Centainly refer to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be any
One or more embodiment or examples in can be combined in any suitable manner.
Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that: not
A variety of change, modification, replacement and modification can be carried out to these embodiments in the case where being detached from the principle of the present invention and objective, this
The range of invention is defined by the claims and their equivalents.
Claims (10)
1. a kind of regulate and control method based on the cargo shipping efficiency of random forest and deep learning, which is characterized in that including following step
Rapid S1-S6 or step S1-S5 or step S6:
S1, multiple ship's navigation data samples in reading database, each sample include one group of aeronautical data characteristic variable and
Cargo shipping efficiency corresponding with this group of aeronautical data characteristic variable carries out random sampling, aeronautical data feature to these samples
Variable includes vessel motion data characteristics variable and ship status data characteristics variable；
S2 constructs optimal stochastic forest model, using the resulting aeronautical data characteristic variable of random sampling as Random Forest model
Input variable carry out first time test, calculate the prediction error of optimal stochastic forest model, and from the resulting boat of random sampling
Important feature variable is extracted in row data characteristic variable；
S3 constructs optimal deep learning model, using the important feature variable as the input variable of optimal deep learning model,
It carries out second to test, calculates the prediction error of optimal deep learning model；
S4 reads real-time ship's navigation data and obtains real-time aeronautical data characteristic variable, real-time aeronautical data characteristic variable is made
Optimal stochastic forest model progress first time prediction is inputted for input variable, inputs second of progress in optimal deep learning model
Prediction；
First time prediction result and second of prediction result are weighted processing by S5, are obtained cargo shipping efficiency and are finally predicted
Value, and seek confidence interval；
S6 adjusts ship's navigation according to the corresponding relationship of the final predicted value of cargo shipping efficiency of acquisition and confidence interval and runs number
According to being optimal vessel motion efficiency.
2. according to claim 1 regulate and control method, feature based on the cargo shipping efficiency of random forest and deep learning
It is, the vessel motion data characteristics variable includes ship power system data, and the ship power system data include such as
Lower type: one of engine speed, gear oil pressure, tailing axle revolving speed or any combination, the ship status data characteristics variable
Including following type: GPS signal delta data, weather data, drauht data, ship up and down believe by water number evidence, ship automatic control
One of breath or any combination；
The type of type and ship status data to vessel motion data is sampled, resulting vessel motion data of sampling
Number of species are greater than the number of species of ship status data.
3. the cargo shipping EFFICIENCY PREDICTION method according to claim 1 based on random forest and deep learning, feature
It is, the step S2 specifically:
S21, in Random Forest model, first by ship's navigation data sample be divided into training set, verifying collection and test set；
S22, the default parameters using Random Forest model, with training set to Random Forest model training, obtain it is trained with
Machine forest model；
S23, trained Random Forest model is verified with verifying collection, obtains the first model error；
S24, some or all of Random Forest model default parameters is combined with trellis search method, obtains random forest
The multiple groups parameter combination of model；
S25, using cross validation method, the Random Forest model under each parameter combination is verified with verifying collection, is obtained
To the error of the Random Forest model of every group of parameter combination；
S26, the error of the Random Forest model of the every group of parameter combination obtained after the first model error and cross validation is carried out
Compare, error minimum is optimal stochastic forest model；
S27, first time test is carried out to test set using optimal stochastic forest model, and calculates the pre- of optimal stochastic forest model
Error is surveyed, important feature variable is extracted to input data using optimal stochastic forest model.
4. the cargo shipping EFFICIENCY PREDICTION method according to claim 1 or 3 based on random forest and deep learning, special
Sign is, the extracting method of the important feature variable are as follows: variable importance scoring is indicated with VIM with gini index, it is false
Equipped with m characteristic variable X1, X2, X3..., Xm；
Each characteristic variable XjGini index score calculation formula are as follows:
Wherein, K indicates the species number of characteristic variable, PmkIt indicates in node m
Ratio shared by k-th of type at will randomly selects two samples, the inconsistent probability of category flag, k ' from node m
Indicate the supplementary set of k-th of type in the species number K of characteristic variable, Pmk'=1-Pmk；
Characteristic variable XjGini index variation amount in the importance of node m, i.e., before and after node m branch are as follows:Wherein GIlAnd GIrRespectively indicate the index of the Gini of branch latter two new node.
If characteristic variable XjWhether the node occurred in decision tree i or not in set M, then feature XjDo not have in decision tree i
It makes a difference；
If characteristic variable XjFor the node occurred in decision tree a in set M, set M is root node and leaf segment in decision tree a
The set of point, then feature XjIn the importance of decision tree a are as follows:
Assuming that have N tree in random forest, then
All importance acquired, which are done normalized, can be obtained important feature variables reordering, normalize calculation formula are as follows:C is characterized the total quantity of variable, to obtain the sequence of m characteristic variable, successively may be used according to sequence
Obtain important feature variable.
5. the cargo shipping EFFICIENCY PREDICTION method according to claim 1 based on random forest and deep learning, feature
It is, step S3 specifically:
S31, it is normalized, will be inputted as input variable, and to it using the important feature variable that random forest proposes
Variable and predictive variable normalized value 0 to 1, normalization mode are as follows:
Wherein,databRespectively indicate the important feature variable of input
In each characteristic variable type data maximums, data minimum value and data item are used as deep learning after being normalized
Modeling data, b are 0 to the positive integer between B, and wherein B is the number of species of important characteristic variable；
S32, model training is carried out to keras deep neural network, constructs the input of the network structure of kears deep neural network
Layer, Dense layers, loss function, optimizer specifies monitor control index；
S33, the activation primitive of keras deep learning model is optimized；
The expression formula that activation primitive after optimization is is
S34, keras deep learning model parameter is adjusted using back-propagation algorithm, obtains optimal keras deep learning model,
And it is tested by optimal keras deep learning model, and calculate the prediction error of optimal keras deep learning model.
6. a kind of harmful influence cargo shipping efficiency regulation side based on random forest and deep learning according to claim 1
Method, which is characterized in that the weighting processing method of first time prediction result and second of prediction result in the step S5 are as follows:
S51, using first time predicted value and second of predicted value as the input value of final prediction result；
S52, according to the prediction error of optimal stochastic forest model and optimal deep learning model, determine optimal stochastic forest model
With the weight of optimal deep learning model predicted value, first time predicted value and second of predicted value are weighted summation and obtain ship
The final predicted value Y of oceangoing ship shipping efficiency.
7. according to claim 6 regulate and control method, feature based on the cargo shipping efficiency of random forest and deep learning
It is, the final predicted value of cargo shipping efficiency are as follows:
Wherein, RMSE4, RMSE5 respectively indicate prediction error to identical sample set under optimal stochastic forest model, optimal
Prediction error under deep learning model, Y1、Y2It respectively indicates pre- under optimal stochastic forest model to same group of input data
Measured value and the predicted value under optimal deep learning model.
8. according to claim 6 regulate and control method, feature based on the cargo shipping efficiency of random forest and deep learning
It is, confidence interval acquiring method are as follows:
Mould is learnt in optimal stochastic forest model and optimal depth respectively to same group of real-time aeronautical data characteristic variable of reading
It is repeatedly predicted in type, the predicted value of each optimal stochastic forest model and optimal deep learning model is weighted and is asked
With, standard deviation δ then is taken to the result of each weighted sum again, obtain confidence interval be [Y- δ, Y+ δ]；
Or, carrying out every group of characteristic variable input optimal deep learning model to repeat p prediction in step s3, according to pre- every time
The output result of survey seeks the standard deviation for the predicted value that every group of characteristic variable input optimal deep learning model obtains, and is denoted as δ1,
δ1... δq, take mean value to obtain the standard deviation acquiredConfidence interval isWherein Y is cargo shipping
The final predicted value of efficiency.
9. according to claim 1 regulate and control method, feature based on the cargo shipping efficiency of random forest and deep learning
It is, the step S5 are as follows:
It navigates under identical ship status data closest to the ship of the final predicted value Y of cargo shipping efficiency from being selected in historical data
Row data characteristic variable sample, using the vessel motion data in the ship's navigation data characteristics variable sample as current ship
The adjustment target of operation data.
10. a kind of cargo shipping efficiency regulator control system, which is characterized in that including vessel motion data acquisition unit, ship status
Data acquisition unit and control unit, described control unit is by the described in any item cargo shipping efficiency regulations of claim 1-9
Method regulates and controls cargo shipping efficiency.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910410132.4A CN110298611A (en) | 2019-05-16 | 2019-05-16 | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910410132.4A CN110298611A (en) | 2019-05-16 | 2019-05-16 | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110298611A true CN110298611A (en) | 2019-10-01 |
Family
ID=68026986
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910410132.4A Pending CN110298611A (en) | 2019-05-16 | 2019-05-16 | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110298611A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110798314A (en) * | 2019-11-01 | 2020-02-14 | 南京邮电大学 | A Quantum Key Distribution Parameter Optimization Method Based on Random Forest Algorithm |
| CN110826790A (en) * | 2019-10-31 | 2020-02-21 | 天津大学 | An intelligent prediction method for construction productivity of cutter suction dredger |
| CN111832599A (en) * | 2019-11-27 | 2020-10-27 | 北京中交兴路信息科技有限公司 | Gas station prediction method based on machine learning random forest |
| CN112149909A (en) * | 2020-09-28 | 2020-12-29 | 神华中海航运有限公司 | Ship oil consumption prediction method and device, computer equipment and storage medium |
| CN112836893A (en) * | 2021-02-26 | 2021-05-25 | 上海海事大学 | A method for predicting the fuel consumption of ships under severe sea conditions based on sea conditions and ship sailing conditions |
| CN113239025A (en) * | 2021-04-23 | 2021-08-10 | 四川大学 | Ship track classification method based on feature selection and hyper-parameter optimization |
| CN113375729A (en) * | 2021-07-15 | 2021-09-10 | 贵州电网有限责任公司 | Intelligent detection and early warning method for user transformer |
| CN113610453A (en) * | 2021-06-30 | 2021-11-05 | 宁波诺丁汉大学 | Multi-transportation-mode combined container transportation path selection method |
| CN113642800A (en) * | 2021-08-20 | 2021-11-12 | 林周县众陶联供应链服务有限公司 | Data analysis method and data analysis system for firing system of architectural ceramic kiln |
| CN113807606A (en) * | 2021-10-09 | 2021-12-17 | 上海交通大学 | Interpretable Ensemble Learning for Online Prediction of Intermittent Process Quality |
| CN114330895A (en) * | 2021-12-30 | 2022-04-12 | 大连海事大学 | A short-term ship speed prediction method based on time series random forest |
| CN114386697A (en) * | 2022-01-12 | 2022-04-22 | 合肥工业大学 | Ship main engine spare part prediction method based on improved random forest |
| CN114898805A (en) * | 2022-04-02 | 2022-08-12 | 山东大学 | A method and system for promoter prediction across multiple species |
| CN116424508A (en) * | 2022-12-31 | 2023-07-14 | 华中科技大学 | Ship stability prediction method and system based on GD weighted fusion RBFNN and random forest |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316501A (en) * | 2017-06-28 | 2017-11-03 | 北京航空航天大学 | A kind of SVMs Travel Time Estimation Method based on grid search |
| US20180211164A1 (en) * | 2017-01-23 | 2018-07-26 | Fotonation Limited | Method of training a neural network |
| CN108510741A (en) * | 2018-05-24 | 2018-09-07 | 浙江工业大学 | A kind of traffic flow forecasting method based on Conv1D-LSTM neural network structures |
| CN109214107A (en) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | A kind of ship's navigation behavior on-line prediction method |
| CN109376750A (en) * | 2018-06-15 | 2019-02-22 | 武汉大学 | A Remote Sensing Image Classification Method Integrating Mid-Wave Infrared and Visible Light |
| CN109447364A (en) * | 2018-11-08 | 2019-03-08 | 国网湖南省电力有限公司 | Power customer based on label complains prediction technique |
| CN109754122A (en) * | 2019-01-13 | 2019-05-14 | 胡燕祝 | A kind of Numerical Predicting Method of the BP neural network based on random forest feature extraction |
-
2019
- 2019-05-16 CN CN201910410132.4A patent/CN110298611A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180211164A1 (en) * | 2017-01-23 | 2018-07-26 | Fotonation Limited | Method of training a neural network |
| CN107316501A (en) * | 2017-06-28 | 2017-11-03 | 北京航空航天大学 | A kind of SVMs Travel Time Estimation Method based on grid search |
| CN108510741A (en) * | 2018-05-24 | 2018-09-07 | 浙江工业大学 | A kind of traffic flow forecasting method based on Conv1D-LSTM neural network structures |
| CN109376750A (en) * | 2018-06-15 | 2019-02-22 | 武汉大学 | A Remote Sensing Image Classification Method Integrating Mid-Wave Infrared and Visible Light |
| CN109214107A (en) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | A kind of ship's navigation behavior on-line prediction method |
| CN109447364A (en) * | 2018-11-08 | 2019-03-08 | 国网湖南省电力有限公司 | Power customer based on label complains prediction technique |
| CN109754122A (en) * | 2019-01-13 | 2019-05-14 | 胡燕祝 | A kind of Numerical Predicting Method of the BP neural network based on random forest feature extraction |
Non-Patent Citations (5)
| Title |
|---|
| 周华平: "《基于支持向量机的煤矿安全建模研究及应用》", 31 January 2015, 西安：西安电子科技大学出版社 * |
| 周平红: "《我国高等教育信息化水平测评与发展预测研究》", 31 December 2018, 武汉：华中师范大学出版社 * |
| 林开春等: "基于随机森林和神经网络的空气质量预测研究", 《青岛大学学报（工程技术版）》 * |
| 毋邦池等: "《物资经济管理知识问答》", 31 October 1986, 北京：中国财政经济出版社 * |
| 牟小辉等: "基于随机森林算法的内河船舶油耗预测模型", 《交通信息与安全》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826790A (en) * | 2019-10-31 | 2020-02-21 | 天津大学 | An intelligent prediction method for construction productivity of cutter suction dredger |
| CN110798314B (en) * | 2019-11-01 | 2023-02-24 | 南京邮电大学 | A Quantum Key Distribution Parameter Optimization Method Based on Random Forest Algorithm |
| CN110798314A (en) * | 2019-11-01 | 2020-02-14 | 南京邮电大学 | A Quantum Key Distribution Parameter Optimization Method Based on Random Forest Algorithm |
| CN111832599B (en) * | 2019-11-27 | 2024-02-09 | 北京中交兴路信息科技有限公司 | Gas station prediction method based on machine learning random forest |
| CN111832599A (en) * | 2019-11-27 | 2020-10-27 | 北京中交兴路信息科技有限公司 | Gas station prediction method based on machine learning random forest |
| CN112149909A (en) * | 2020-09-28 | 2020-12-29 | 神华中海航运有限公司 | Ship oil consumption prediction method and device, computer equipment and storage medium |
| 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 |
| CN112836893A (en) * | 2021-02-26 | 2021-05-25 | 上海海事大学 | A method for predicting the fuel consumption of ships under severe sea conditions based on sea conditions and ship sailing conditions |
| CN113239025A (en) * | 2021-04-23 | 2021-08-10 | 四川大学 | Ship track classification method based on feature selection and hyper-parameter optimization |
| CN113239025B (en) * | 2021-04-23 | 2022-08-19 | 四川大学 | Ship track classification method based on feature selection and hyper-parameter optimization |
| CN113610453A (en) * | 2021-06-30 | 2021-11-05 | 宁波诺丁汉大学 | Multi-transportation-mode combined container transportation path selection method |
| CN113375729A (en) * | 2021-07-15 | 2021-09-10 | 贵州电网有限责任公司 | Intelligent detection and early warning method for user transformer |
| CN113375729B (en) * | 2021-07-15 | 2023-10-31 | 贵州电网有限责任公司 | Intelligent detection and early warning method for user transformer |
| CN113642800B (en) * | 2021-08-20 | 2023-11-03 | 西藏众陶联供应链服务有限公司 | Data analysis method and data analysis system for firing system of building ceramic kiln |
| CN113642800A (en) * | 2021-08-20 | 2021-11-12 | 林周县众陶联供应链服务有限公司 | Data analysis method and data analysis system for firing system of architectural ceramic kiln |
| CN113807606B (en) * | 2021-10-09 | 2022-07-22 | 上海交通大学 | Interpretable Ensemble Learning for Online Prediction of Intermittent Process Quality |
| CN113807606A (en) * | 2021-10-09 | 2021-12-17 | 上海交通大学 | Interpretable Ensemble Learning for Online Prediction of Intermittent Process Quality |
| CN114330895A (en) * | 2021-12-30 | 2022-04-12 | 大连海事大学 | A short-term ship speed prediction method based on time series random forest |
| CN114330895B (en) * | 2021-12-30 | 2024-09-10 | 大连海事大学 | A short-term ship speed prediction method based on time series random forest |
| CN114386697A (en) * | 2022-01-12 | 2022-04-22 | 合肥工业大学 | Ship main engine spare part prediction method based on improved random forest |
| CN114898805A (en) * | 2022-04-02 | 2022-08-12 | 山东大学 | A method and system for promoter prediction across multiple species |
| CN114898805B (en) * | 2022-04-02 | 2024-06-18 | 山东大学 | Multi-species-crossing promoter prediction method and system |
| CN116424508A (en) * | 2022-12-31 | 2023-07-14 | 华中科技大学 | Ship stability prediction method and system based on GD weighted fusion RBFNN and random forest |
| CN116424508B (en) * | 2022-12-31 | 2024-01-26 | 华中科技大学 | Ship stability prediction method and system based on GD weighted fusion of RBFNN and random forest |
| US12093616B1 (en) | 2022-12-31 | 2024-09-17 | Huazhong University Of Science And Technology | Method and system for ship stability prediction by weighted fusion of radial basis function neural network and random forest based on gradient descent |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298611A (en) | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning | |
| CN109214107A (en) | A kind of ship's navigation behavior on-line prediction method | |
| US11459962B2 (en) | Electronic valve control | |
| CN107944648A (en) | A kind of accurate Forecasting Methodology of large ship speed of a ship or plane rate of fuel consumption | |
| CN103743402A (en) | Underwater intelligent self-adapted terrain matching method based on terrain information amount | |
| CN108563119A (en) | A kind of unmanned boat motion control method based on fuzzy support vector machine algorithm | |
| Prasanna et al. | An analysis on stock market prediction using data mining techniques | |
| CN109345296A (en) | Common people's Travel Demand Forecasting method, apparatus and terminal | |
| US12008294B2 (en) | Calibration of online real-world systems using simulations | |
| CN115751441A (en) | Heat supply system heating station heat regulation method and system based on secondary side flow | |
| Kim et al. | Real-time river-stage prediction with artificial neural network based on only upstream observation data | |
| US20220034753A1 (en) | Calibration of offline real-world systems using simulations | |
| CN118095600A (en) | Ship route optimization method based on oil consumption prediction | |
| CN113793220A (en) | Stock market investment decision method based on artificial intelligence model and related equipment | |
| CN113379151A (en) | Wind speed ultra-short term prediction method based on Bagging-CNN-GRU | |
| KR102497543B1 (en) | Military demand prediction model and practical system using machine learning | |
| CN118643277B (en) | A method for ship fuel consumption prediction and speed optimization based on stacking ensemble learning | |
| KR20240071996A (en) | A method and an apparatus for optimizing ship's navigation | |
| KR102708959B1 (en) | System for output of user interface | |
| US12165096B2 (en) | Apparatus and method for determining carbon emissions of a shipment | |
| Ilardi | Data-driven solutions to enhance planning, operation and design tools in Industry 4.0 context | |
| US20220170436A1 (en) | Method and device for ascertaining a closure point in time of an injector of an internal combustion engine with the aid of a machine learning system | |
| Pasaribu et al. | Comparative Analysis of ARIMA-based Models for Forecasting Pressure in Natural Gas Pipelines | |
| Lineberry | Estimating production cost while linking combat systems and ship design | |
| Tzoumezi | Parameters estimation in marine powertrain using neural networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191001 |
|
| RJ01 | Rejection of invention patent application after publication |