CN107316501A - A kind of SVMs Travel Time Estimation Method based on grid search - Google Patents
A kind of SVMs Travel Time Estimation Method based on grid search Download PDFInfo
- Publication number
- CN107316501A CN107316501A CN201710507146.9A CN201710507146A CN107316501A CN 107316501 A CN107316501 A CN 107316501A CN 201710507146 A CN201710507146 A CN 201710507146A CN 107316501 A CN107316501 A CN 107316501A
- Authority
- CN
- China
- Prior art keywords
- ship
- data
- value
- prediction
- parameter
- 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
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012706 support-vector machine Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 31
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 15
- 238000011156 evaluation Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 27
- 238000010586 diagram Methods 0.000 claims description 11
- 238000002790 cross-validation Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 4
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 3
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000004069 differentiation Effects 0.000 claims description 2
- 238000000205 computational method Methods 0.000 claims 1
- 238000011835 investigation Methods 0.000 abstract description 6
- 238000011161 development Methods 0.000 abstract description 3
- 238000007418 data mining Methods 0.000 abstract description 2
- 238000012544 monitoring process Methods 0.000 description 6
- 241001269238 Data Species 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 238000009333 weeding Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G3/00—Traffic control systems for marine craft
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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"
-
- G06Q50/40—
Abstract
The invention discloses a kind of SVMs Travel Time Estimation Method based on grid search, belong to intelligent transportation field, including：1st, based on ship automatic identification system data, investigation, the division of up-downgoing ship and the big step of the rejecting of redundant data three according to missing data carry out the pretreatment work of data；2nd, the inland navigation craft travel time prediction model based on historical period is built, and training dataset is obtained according to model；3rd, the optimized parameter of forecast model is found based on SVMs grid data service；4th, based on optimized parameter, the prediction to inland navigation craft journey time is realized；5th, predict the outcome evaluation.The present invention can be used for ship automatic identification system data application data mining theories method is excavated and analyzed, realize the prediction to inland navigation craft journey time, its application will be helpful to improve the managerial skills of management of navigational affairs department, promote the fast development of inland water transport.
Description
Technical field
The present invention relates to a kind of SVMs Travel Time Estimation Method based on grid search, belong to intelligent transportation neck
Domain.This method can be based on ship automatic identification system (Automatic Identification System, AIS) data, real
Now to the prediction of inland navigation craft journey time, theoretical and technical support is provided for management of navigational affairs department.
Background technology
Inland water transport is one of important means of transportation of China, is comprehensively utilize water resource and complex transportation system important
Member is participated in, and is connected to inland area and coastal region, is that China brings huge economic profit every year.But, in China
River shipping still there are the problems such as navigation safety, physical distribution optimizing management, port and harbor planning are dispatched, and directly affect inland water transport
Development.In recent years, bank AIS base station constructions in inland river were at full speed, the ships quantity rapid growth equipped with AIS, and AIS can be adopted
The abundant data message of collection, carries out a variety of applications.It therefore, it can carry out inland river AIS data application data mining theories method
Excavate and analyze, realize the accurate prediction of ship journey time, to improve the managerial skills of management of navigational affairs department, promote inland river boat
The fast development of fortune.
Forecasting of Travel Time algorithm at home and abroad has a more in-depth study, but now its mainly in city road
The prediction of road journey time.Meanwhile, in the ship behavioral study that at home and abroad scholar is carried out based on AIS ship oceangoing ship traffic data,
Through being reached when solving between ship collision and volume of vessel traffic, ship motor pattern, ship's navigation track, ship away from, ship
There is good method in terms of the prediction of time.But, for also no more deep in terms of inland navigation craft Forecasting of Travel Time
Research, this problems demand solve.
The content of the invention
The purpose of the present invention is overcomes above-mentioned technical problem, when proposing a kind of SVMs stroke based on grid search
Between Forecasting Methodology.The present invention is cut from the Forecasting Methodology of Urban Travel Time, and selection is applied to inland navigation craft and navigates by water mode
SVMs (Support Vector Machine, SVM) algorithm, not only compensate for it is existing be directed to inland navigation craft stroke when
Between in terms of prediction research deficiency, and realize and realize that up-downgoing ship such as divides at the data prediction work.This method is predicted
More accurately, stably, application field is extensive for effect.
The present invention is a kind of SVMs Travel Time Estimation Method based on grid search, including following step
Suddenly：
Step 1: AIS data are based on, the investigation, the division of up-downgoing ship and redundant data according to missing data
Reject the pretreatment work that three big steps carry out data；
Step 2: the inland navigation craft travel time prediction model based on historical period is built, and according to model training
Data set；
Step 3: finding the optimized parameter of forecast model based on SVM grid data services；
Step 4: based on optimized parameter, realizing the prediction to inland navigation craft journey time；
Step 5: the evaluation that predicts the outcome.
The advantage of the invention is that：
(1) the prominent advantage of the present invention is exactly the Forecasting Methodology incision from Urban Travel Time, by SVM algorithm certainly
Learning ability, the complicated letter between capture past period journey time and a variety of Random Effect factors and present period journey time
Number relation, realizes the inland navigation craft Forecasting of Travel Time based on AIS data.
(2) present invention creatively uses AIS data, and maintenance data excavates theoretical method and excavated and analyzed, and in fact
The work of the data predictions such as investigation, the division of up-downgoing ship and the rejecting of redundant data of missing data is showed；
(3) optimized parameter of forecast model is found based on SVM grid data services, and is realized under optimized parameter to riverboat
The prediction of oceangoing ship journey time, precision of prediction is greatly improved.
Brief description of the drawings
Fig. 1 is method flow schematic diagram of the invention；
Fig. 2 is institute's survey region schematic diagram in embodiment；
Fig. 3 is up-downgoing ship course schematic diagram in embodiment；
Fig. 4 is grid data service search result schematic diagram in embodiment；
Fig. 5 is predicted value scatterplot schematic diagram in embodiment；
Fig. 6 is predicted value and actual value contrast schematic diagram in embodiment.
Embodiment
Below in conjunction with drawings and examples, the present invention is described in further detail.
The present invention proposes a kind of SVMs Travel Time Estimation Method based on grid search, flow chart such as Fig. 1 institutes
Show, comprise the following steps：
Step 1: AIS data predictions
1), the investigation of missing data：The AIS data collected mainly include No. MMSI, time of filing, longitude, latitude, right
The information such as the ground speed of a ship or plane, course over ground, Ship Types.In initial data, a certain ship at a certain moment there may be above-mentioned
The incomplete situation of information, for such Incomplete information, it should investigated and removed.
2), the division of up-downgoing ship：According to the division of water above and below the difference progress of ship course over ground angle, by original AIS
Data are divided into up ship and descending ship two large divisions.It can be seen from inland navigation craft navigates by water feature, in the up ship sailed in inland river
Oceangoing ship, its headway can be by water be different is influenceed up and down, and lower water shipping sail speed, can be to stroke apparently higher than upriver boat oceangoing ship
Time produces influence, and then can produce in the forecast model of foundation different parameters.According to ship's navigation course over ground angle
Difference, initial data is imported in map, ship up-downgoing differentiation figure is obtained, so as to be drawn to up ship and descending ship
Point.
3), the rejecting of redundancy leg：In initial data, it may appear that such as ship is not travelled, longitude and latitude error in data, ship
The berth wrong data such as beyond Yangtze River waterway, in addition to rejecting above-mentioned wrong data.There is part ship to be located at tributary leg,
Also part ship is wheel ferryboat, and navigated by water not along the up-downlink direction in inland river, and therefore, above-mentioned similar ship also should be original
Rejected in data.
Step 2: building the inland navigation craft travel time prediction model based on historical period
1), the structure of forecast model：The present invention utilizes answering between SVM self-learning capability, self-teaching items factor
Miscellaneous function relation.The journey time of inland navigation craft present period and the journey time of preceding several periods of the period have certain
Functional relation, i.e.,：
TK=f (Tk-1,Tk-2,...,Tk-n) (1)
Wherein, TKRepresent the average value for meeting each ship traveled distance time of prediction leg in present period；Tk-nGeneration
Table prediction period k preceding several periods meet each inland navigation craft traveled distance time, n=1,2 ..., n of prediction leg.
It is prediction leg that fixed point A and fixed point B, the leg L between A, B are chosen in prediction leg.Wherein, it is every to ensure
The starting point of one group of training data and final position are essentially identical, it is necessary to do as defined below, that is, assume selected starting point A coordinate
For (x, y), then actual start A ' coordinate needs to meet certain required precision, i.e.,:| x-x'|≤ε, | y-y'|≤ε.Together
Reason, terminal B is also required to meet above required precision.
2) training dataset of forecast model, is built：In the forecast model based on historical period, because closer to current
The period of period and the degree of correlation of prediction period are bigger, therefore every group of data in training set will include the right quantity period
Ship journey time, and because shipping sail speed is to influence another key factor of inland navigation craft journey time, therefore will
Using the historical time sequence closely related with inland navigation craft journey time and the average speed conduct of the period each ship's navigation
The characteristic value of prediction.2 two kinds of situations of up scene one and descending scene are splitted data into first, then the first dimensional input vector
For T1Second latitude input vector is T2, third dimension input vector is T3, the n-th dimensional input vector be Tn, average speed vector is V, defeated
Outgoing vector is Tn+1.Afterwards, the training data of construction is imported in SVM programs, SVM carries out self-teaching by historical data, looks for
To the complicated function relation between input value and output valve, you can realize the prediction to inland navigation craft journey time.
3), the determination of model parameter：The inland navigation craft travel time prediction model that the present invention is set up is from Gauss radial direction
Base kernel functionWith ε-insensitive loss function.Wherein σ is nuclear parameter, and ε is insensitive loss function parameter.
σ represents the parameter gamma in LIBSVM software kits, reflects distribution or the range property of training dataset, office
Portion's width neighborhood is determined by it；
ε represents that the parameter epsilon in LIBSVM software kits controls the width of insensitive band, and can affect to branch
Hold the number of vector.When ε value is smaller, regression accuracy is higher, and supporting vector number is then more, on the contrary, when ε value is larger, returning
Precision is relatively low, but supporting vector number can tail off；
C represents the parameter cost in LIBSVM software kits, while also referred to as regularization parameter, it is to reaching in error
The punishment degree of the sample of limit is controlled, and when value increases, the punishment of sample increases therewith.
Step 3: finding the optimized parameter of forecast model based on SVM grid data services
For parameter C, σ, ε of Radial basis kernel function selection, the present invention is using the k folding cross validations in grid search
Method.K folding cross validations are that training sample is divided into k parts, and k-1 parts are taken out every time as training data, remaining a work
For test data, so repeat to be k times, obtain the average cross checking accuracy rate of k times as a result, carrying out k folding cross validations
An efficiency value can be returned afterwards, cost the and gamma parameters corresponding to maximum efficiency are exactly the optimal of Radial basis kernel function
Parameter.In prediction work of the next step using SVM functions, the optimal value that the value positioning optimizing of above-mentioned parameter is obtained, you can
Reach the relatively good small effect of prediction.Comprise the following steps that：
1) hunting zone, is set：The e1071 function bags of R statistical softwares are called first, select set up training data
Collection, setting input and output data, and scan for after setting the hunting zone of gamma parameter cost parameters, concurrently set
Set.seed (10) ensures to the division of training set and to choose consistent every time.
2), the judgement of optimal value：Gamma parameter cost parameter values obtained by after search are the critical of bound
During value, now search value is not necessarily optimal value, it is necessary to reset hunting zone, obtains new parameter value.
3) optimal value, is obtained：After above-mentioned search several times, the parameter value obtained after search is between bound
When, it was demonstrated that set hunting zone is correct, the parameter value now searched as optimal value.
Step 4: based on optimized parameter, realizing the prediction to inland navigation craft journey time
1), the foundation of SVM models：Program is write, training dataset is imported in R, and sets input data set x,
Output data set y.When calling svm functions, it is necessary to set the type of SVM SVMs kernel functions, and gamma and
Cost optimal value
2), the test of training data：In this step, it is necessary to test training data, using above-mentioned forecast model,
To input set x predictions, predicted value is obtained, and contrasted with actual value.
3), the determination of each Factor Weight：Object properties are determined by attr () function in this step, each factor power is drawn
Weight.
Step 5: the evaluation that predicts the outcome
1) visualization, predicted the outcome：In this step, the present invention is using actual observation as abscissa, to be set up
The predicted value that produces of SVM models as ordinate, correlation scatter diagram is drawn to be contrasted.For the ease of comparing, while this
A group observations and the identical situation of predicted value are set up in invention, are contrasted with above-mentioned scatter diagram, can be more intuitively
Find out the distribution of predicted value.
2), standard diagrams are calculated：The present invention, which have selected following four error criterion it is predicted the outcome, to be analyzed,
Its calculation formula is as follows：
1. mean absolute error (MAD) is：
2. average absolute relative error (MAPE) is：
3. maximum absolute error (MAE) is：
MAE=max | actual value-predicted value | (4)
4. maximum relative error (MRE) is：
Embodiment
A kind of SVMs Travel Time Estimation Method based on grid search, using the AIS data of Changjiang River in Wuhan as
Example, it is specific as follows：
Step 1: in initial data, monitoring leg scope is part within the loop wire of Wuhan City three, total length about 22km, such as
Shown in Fig. 2 marks, the monitoring time is August in 2014 11 days 00:00:00 to 2014 on Augusts 19,13:59:59, when monitoring altogether
Long 206 hours.Based on SQL Server 2008, after investigation, 167865 datas are produced altogether.
According to the difference at ship's navigation course over ground angle, above-mentioned initial data is imported, ship as shown in Figure 3 is can obtain
Figure is distinguished in up-downgoing, wherein the red navigation route for representing up ship, blueness represents the navigation route of descending ship, will be above-mentioned
167865 datas are divided into up ship and descending ship.
As shown in two black rectangles inframes that numbering is 1,2 in Fig. 3, wherein black box 1 represents tributary of Yangtze boat
Section, black box 3 represents ferry ship.Therefore, above-mentioned two classes ship should also be rejected in initial data.Divide herein above-mentioned
The up-downgoing ship of class carries out the Weeding more than data respectively.In up ship, such as numbering 1 in Fig. 3 need to be rejected and compiled
Two parts red area in numbers 2.Coordinate according to Baidu map coordinate picking up system is shown, it can be deduced that reject data below
Longitude and latitude scope：Black box 1 is 114.2955 ° of LON ＜ and LAT >=30.5661 °；Similarly, black box 2 be LON >=
114.2979 ° and LAT ＜ 30.5715 °, reject after two groups of data, up ships data amounts to 86589.In descending ship,
It need to reject such as the blue region in numbering 1 in Fig. 3.Equally, the coordinate show value according to Baidu map coordinate picking up system, is rejected
The longitude and latitude scope black box 1 of data is 114.2978 ° of LON ＜ and LAT >=30.5660 °, is rejected after one group of data above,
Descending ships data amounts to 57586.
Step 2: the present embodiment was used as a monitoring period of time using 2 hours.The monitoring time is August in 2014 11 days 00:00:
00 to 2014 on Augusts 19,13:59:59,206 hours of duration are monitored altogether, therefore can be divided into 103 periods.Together
When, in order to ensure to there are ship within each period by leg L, therefore leg L 2 points of origin coordinates A, B determination
It is most important.It is final to determine that monitoring leg is Wuhan City's Baishazhou bridge herein by the unified investigation statistics to each period
To one section of Qing Chuange primary schools of Wuhan City, i.e. A points latitude value is 30.4905 ° ± 0.005 °, the latitude values of B points for 30.5654 ° ±
0.005°。
Construct SVM algorithm program can recognize training set when, the present invention by journey time equivalent be using the second as
Unit, if the longitude and latitude span for having a plurality of ship to meet 2 points of A, B in certain time period, now needs to take this several ships
Average value as the period ship journey time.The data of training dataset first is by the journey time of period 1,2,3,4
As input, the journey time of period 5 is used as output；Article 2, regard the journey time of period 2,3,4,5 as input, period 6
Journey time be used as output；Therefore the training dataset amounts to 99 datas.
Step 3: setting k values are 10 to carry out cross validation herein, e1071 function bags are called first, what selection was set up
Training dataset, setting input and output data, and 2 are set as to the hunting zone of gamma parameters-5~2-1, cost parameters
Hunting zone is set as 2-2~22, search result is as shown in Figure 4.
By operation result visible gamma=0.0625 and cost=0.5, search parameter value is faced between upper critical value with
Between dividing value.Therefore, the optimized parameter of SVM RBFs has now been reached.Pass through the SVM parameters based on grid data service
Optimization, it is final to determine that optimized parameter is followed successively by C=0.5, ε=0.1, σ=0.0625.
Step 4: writing program, training dataset is imported in R, and set input data set x, output data set
Close y.When calling svm functions, it is necessary to set the types of SVM SVMs kernel functions as Radial basis kernel function " radial ",
Gamma values are that 0.0625, cost values are 0.5.It was found from operation result, the operative orientation of svm functions returns for eps-,
Epsilon value is 0.1, and supporting vector number is 87.
Step 5: using above-mentioned forecast model, to input set x predictions, obtain predicted value, and with actual value progress pair
Than the predicted value of the SVM models generation to be set up is contrasted as ordinate, as a result shows as shown in Figure 5.For the ease of
Compare, while setting up a group observations and the identical situation of predicted value herein, as shown in Figure 6.From the results, it was seen that pre-
Surveying result has the trend of obvious convergence observation, and the scatterplot on observation both sides is evenly distributed.Due to training data data
Collection only 99 groups, cause some deviations that predict the outcome larger, if increase training data will be had to more than 200 groups, as a result it is bright
It is aobvious to take on a new look.
Step 5：Mean absolute error (MAD), average absolute relative error (MAPE), maximum are calculated respectively definitely by mistake
Poor (MAE), maximum relative error (MRE), each index result of calculation are as shown in table 1, unit (second).
The evaluation index result of table 1
It is described in detail above it is of the invention be preferable to carry out case, but the invention is not limited in above-mentioned case study on implementation
The part steps of the present invention in the range of the overall structure of the present invention, can be carried out a variety of conversion and again group by detail
Close, the present invention is no longer enumerated to various possible combinations, these conversion combinations belong to protection scope of the present invention.
Claims (4)
1. a kind of SVMs Travel Time Estimation Method based on grid search, including following steps：
Step 1: ship automatic identification system (Automatic Identification System, AIS) data prediction
Investigated firstly the need of to the missing data in AIS data.The AIS data collected mainly include No. MMSI, filing
The information such as time, longitude, latitude, speed on the ground, course over ground, Ship Types.In initial data, a certain ship at a certain moment
Oceangoing ship there may be the incomplete situation of above- mentioned information, for such Incomplete information, it should is investigated and removed.
Secondly, according to ship course over ground angle difference carry out above and below water division, by original AIS data be divided into up ship and under
Navigate oceangoing ship two large divisions.It can be seen from inland navigation craft navigates by water feature, in the up ship sailed in inland river, its headway can be by upper
The different influence of lower water, lower water shipping sail speed can produce influence, and then building apparently higher than upriver boat oceangoing ship on journey time
Different parameters can be produced in vertical forecast model.According to the difference at ship's navigation course over ground angle, initial data is imported ground
In figure, ship up-downgoing differentiation figure is obtained, so as to be divided to up ship and descending ship.
Finally redundancy leg is rejected again.In initial data, it may appear that such as ship is not travelled, longitude and latitude error in data, ship
The wrong data such as beyond Yangtze River waterway, in addition to rejecting above-mentioned wrong data.There is part ship to be located at tributary leg, also
It is wheel ferryboat to have part ship, and is navigated by water not along the up-downlink direction in inland river, and therefore, above-mentioned similar ship also should be in original number
Rejected according to middle.
Step 2: building the inland navigation craft travel time prediction model based on historical period
The present invention utilizes the self-learning capability of SVMs (Support Vector Machine, SVM), self-teaching items
Complicated function relation between factor.When the journey time of inland navigation craft present period and the stroke of preceding several periods of the period
Between have certain functional relation, i.e.,：
TK=f (Tk-1,Tk-2,...,Tk-n) (1)
Wherein, TKRepresent the average value for meeting each ship traveled distance time of prediction leg in present period；Tk-nRepresent pre-
The preceding several periods for surveying period k meet each inland navigation craft traveled distance time, n=1,2 ..., n of prediction leg.
It is prediction leg that fixed point A and fixed point B, the leg L between A, B are chosen in prediction leg.Wherein, it is each group of guarantee
The starting point of training data and final position are essentially identical, it is necessary to do as defined below, that is, the coordinate for assuming selected starting point A is
(x, y), then actual start A ' coordinate needs to meet certain required precision, i.e.,:| x-x'|≤ε, | y-y'|≤ε.Similarly,
Terminal B is also required to meet above required precision.
In the forecast model based on historical period, because being got over closer to the period of present period and the degree of correlation of prediction period
Greatly, every group of data therefore in training set will include the ship journey time of right quantity period, and because shipping sail speed
It is to influence another key factor of inland navigation craft journey time, therefore uses closely related with inland navigation craft journey time
Historical time sequence and the average speed of the period each ship's navigation as prediction characteristic value.Market are splitted data into first
2 two kinds of situations of scape one and descending scene, then the first dimensional input vector is T1Second latitude input vector is T2, the third dimension input to
Measure as T3, the n-th dimensional input vector be Tn, average speed vector is V, and output vector is Tn+1.Afterwards, the training data of construction is led
Enter in SVM programs, SVM carries out self-teaching by historical data, finds the complicated function relation between input value and output valve,
The prediction of internal river steamer oceangoing ship journey time can be achieved.
Step 3: finding the optimized parameter of forecast model based on SVM grid data services
For parameter C, σ, ε of Radial basis kernel function selection, the present invention is using the k folding cross-validation methods in grid search.K rolls over
Cross validation is that training sample is divided into k parts, and k-1 parts are taken out every time as training data, and remaining a be used as is tested
Data, so repeat to be k times, obtain the average cross checking accuracy rate of k times as a result, carrying out meeting after k folding cross validations
An efficiency value is returned to, cost the and gamma parameters corresponding to maximum efficiency are exactly the optimized parameter of Radial basis kernel function.
Next step is using in the prediction work of SVM functions, by the optimal value of the value positioning optimizing acquisition of above-mentioned parameter, you can reach relative
Preferably predict small effect.
Step 4: based on optimized parameter, realizing the prediction to inland navigation craft journey time
Program is write first, training dataset is imported in R, and set input data set x, output data set y.Adjusting
With during svm functions, it is necessary to set the type of SVM SVMs kernel functions, and gamma and cost optimal value；Secondly need
Training data is tested, using above-mentioned forecast model, to input set x predictions, obtain predicted value, and enter with actual value
Row contrast；Finally, object properties are determined by attr () function, draws each Factor Weight.
Step 5: the evaluation that predicts the outcome
Visualized first to surveying result, using actual observation as abscissa, the prediction produced with the SVM models set up
Value draws correlation scatter diagram to be contrasted as ordinate.For the ease of comparing, a group observations and prediction can also be set up
It is worth identical situation, is contrasted with above-mentioned scatter diagram, can more intuitively finds out the distribution of predicted value.Finally, count
The evaluation index of forecast model is calculated, predicting the outcome for inland navigation craft travel time prediction model is quantified and evaluated.
2. a kind of SVMs Travel Time Estimation Method based on grid search according to claim 1, described
In step 2, the determination method of model parameter is as follows：
The inland navigation craft travel time prediction model that the present invention is set up selects gaussian radial basis functionWith
ε-insensitive loss function.Wherein σ is nuclear parameter, and ε is insensitive loss function parameter.
σ represents the parameter gamma in LIBSVM software kits, reflects distribution or the range property of training dataset, local adjacent
Field width degree is determined by it；
ε represents that the parameter epsilon in LIBSVM software kits controls the width of insensitive band, and can affect to support to
The number of amount.When ε value is smaller, regression accuracy is higher, and supporting vector number is then more, on the contrary, when ε value is larger, regression accuracy
It is relatively low, but supporting vector number can tail off；
C represents the parameter cost in LIBSVM software kits, while also referred to as regularization parameter, it is to reaching the error upper limit
The punishment degree of sample is controlled, and when value increases, the punishment of sample increases therewith.
3. a kind of SVMs Travel Time Estimation Method based on grid search according to claim 1, described
In step 3, the specific method for finding forecast model optimized parameter is as follows：
1) hunting zone, is set：The e1071 function bags of R statistical softwares are called first, select set up training dataset, if
Fixed input and output data, and scan for after setting the hunting zone of gamma parameter cost parameters, concurrently set set.seed
(10) ensure to the division of training set and to choose consistent every time.
2), the judgement of optimal value：Gamma parameter cost parameter values obtained by after search are the critical value of bound
When, now search value is not necessarily optimal value, it is necessary to reset hunting zone, obtains new parameter value.
3) optimal value, is obtained：After above-mentioned search several times, when the parameter value obtained after search is between bound, card
Bright set hunting zone is correct, the parameter value now searched as optimal value.
4. a kind of SVMs Travel Time Estimation Method based on grid search according to claim 1, described
In step 5, four kinds of error criterion computational methods are as follows：
1), mean absolute error (MAD) is：
2), average absolute relative error (MAPE) is：
3), maximum absolute error (MAE) is：
MAE=max | actual value-predicted value | (4)
4), maximum relative error (MRE) is：
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710507146.9A CN107316501B (en) | 2017-06-28 | 2017-06-28 | Grid search based travel time prediction method for support vector machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710507146.9A CN107316501B (en) | 2017-06-28 | 2017-06-28 | Grid search based travel time prediction method for support vector machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107316501A true CN107316501A (en) | 2017-11-03 |
| CN107316501B CN107316501B (en) | 2020-04-14 |
Family
ID=60179698
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710507146.9A Active CN107316501B (en) | 2017-06-28 | 2017-06-28 | Grid search based travel time prediction method for support vector machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107316501B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256924A (en) * | 2018-02-26 | 2018-07-06 | 上海理工大学 | A kind of product marketing forecast device |
| CN108305103A (en) * | 2018-02-26 | 2018-07-20 | 上海理工大学 | A kind of product marketing forecast method of the supporting vector machine model based on parameter optimization |
| 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 |
| CN109544920A (en) * | 2018-11-22 | 2019-03-29 | 广东岭南通股份有限公司 | The acquisition of bus trip cost, analysis method and system based on transaction data |
| CN110188804A (en) * | 2019-05-16 | 2019-08-30 | 武汉工程大学 | The method of support vector machines optimal classification model parameter search based on MapReduce frame |
| CN110298611A (en) * | 2019-05-16 | 2019-10-01 | 重庆瑞尔科技发展有限公司 | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning |
| CN110334839A (en) * | 2019-04-15 | 2019-10-15 | 北京航空航天大学 | Flight is delayed prediction technique, device, equipment and storage medium |
| CN111008791A (en) * | 2019-12-24 | 2020-04-14 | 重庆科技学院 | Bread production modeling and decision parameter optimization method based on support vector machine |
| CN111597195A (en) * | 2020-05-07 | 2020-08-28 | 交通运输部水运科学研究所 | Ship atmospheric pollution emission list missing information processing method and device |
| CN111738500A (en) * | 2020-06-11 | 2020-10-02 | 大连海事大学 | Navigation time prediction method and device based on deep learning |
| CN111815487A (en) * | 2020-06-28 | 2020-10-23 | 珠海中科先进技术研究院有限公司 | Health education assessment method, device and medium based on deep learning |
| CN112070529A (en) * | 2020-08-24 | 2020-12-11 | 贵州民族大学 | Passenger carrying hotspot parallel prediction method, system, terminal and computer storage medium |
| CN112085286A (en) * | 2020-09-15 | 2020-12-15 | 周远鹏 | Ship behavior prediction model establishing method and ship behavior prediction model |
| WO2023039894A1 (en) * | 2021-09-16 | 2023-03-23 | 上海船舶运输科学研究所有限公司 | Method and system for analyzing transport capacity of container liner route |
| CN117455066A (en) * | 2023-11-13 | 2024-01-26 | 哈尔滨航天恒星数据系统科技有限公司 | Corn planting accurate fertilizer distribution method based on multi-strategy optimization random forest, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102005135A (en) * | 2010-12-09 | 2011-04-06 | 上海海事大学 | Genetic algorithm-based support vector regression shipping traffic flow prediction method |
| KR20110103082A (en) * | 2010-03-12 | 2011-09-20 | 울산대학교 산학협력단 | Method for providing vessel seaway and providing system of vessel seaway |
| CN102819663A (en) * | 2012-07-17 | 2012-12-12 | 哈尔滨工程大学 | Method for forecasting ship wake based on optimized support vector regression parameter |
| JP2015072618A (en) * | 2013-10-03 | 2015-04-16 | パシフィックシステム株式会社 | Agitator vehicle operation control system |
-
2017
- 2017-06-28 CN CN201710507146.9A patent/CN107316501B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20110103082A (en) * | 2010-03-12 | 2011-09-20 | 울산대학교 산학협력단 | Method for providing vessel seaway and providing system of vessel seaway |
| CN102005135A (en) * | 2010-12-09 | 2011-04-06 | 上海海事大学 | Genetic algorithm-based support vector regression shipping traffic flow prediction method |
| CN102819663A (en) * | 2012-07-17 | 2012-12-12 | 哈尔滨工程大学 | Method for forecasting ship wake based on optimized support vector regression parameter |
| JP2015072618A (en) * | 2013-10-03 | 2015-04-16 | パシフィックシステム株式会社 | Agitator vehicle operation control system |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108305103A (en) * | 2018-02-26 | 2018-07-20 | 上海理工大学 | A kind of product marketing forecast method of the supporting vector machine model based on parameter optimization |
| CN108256924A (en) * | 2018-02-26 | 2018-07-06 | 上海理工大学 | A kind of product marketing forecast device |
| 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 |
| CN109214107B (en) * | 2018-09-26 | 2020-05-05 | 大连海事大学 | Ship navigation behavior online prediction method |
| CN109544920B (en) * | 2018-11-22 | 2021-10-22 | 广东岭南通股份有限公司 | Bus trip cost obtaining and analyzing method and system based on transaction data |
| CN109544920A (en) * | 2018-11-22 | 2019-03-29 | 广东岭南通股份有限公司 | The acquisition of bus trip cost, analysis method and system based on transaction data |
| CN110334839A (en) * | 2019-04-15 | 2019-10-15 | 北京航空航天大学 | Flight is delayed prediction technique, device, equipment and storage medium |
| US11501648B2 (en) | 2019-04-15 | 2022-11-15 | Beihang University | Method and apparatus for predicting flight delay, device and storage medium |
| CN110298611A (en) * | 2019-05-16 | 2019-10-01 | 重庆瑞尔科技发展有限公司 | Regulate and control method and system based on the cargo shipping efficiency of random forest and deep learning |
| CN110188804A (en) * | 2019-05-16 | 2019-08-30 | 武汉工程大学 | The method of support vector machines optimal classification model parameter search based on MapReduce frame |
| CN110188804B (en) * | 2019-05-16 | 2022-12-20 | 武汉工程大学 | Method for searching optimal classification model parameters of support vector machine based on MapReduce framework |
| CN111008791A (en) * | 2019-12-24 | 2020-04-14 | 重庆科技学院 | Bread production modeling and decision parameter optimization method based on support vector machine |
| CN111597195A (en) * | 2020-05-07 | 2020-08-28 | 交通运输部水运科学研究所 | Ship atmospheric pollution emission list missing information processing method and device |
| CN111597195B (en) * | 2020-05-07 | 2023-03-14 | 交通运输部水运科学研究所 | Ship atmospheric pollution emission list missing information processing method and device |
| CN111738500A (en) * | 2020-06-11 | 2020-10-02 | 大连海事大学 | Navigation time prediction method and device based on deep learning |
| CN111738500B (en) * | 2020-06-11 | 2024-01-12 | 大连海事大学 | Navigation time prediction method and device based on deep learning |
| CN111815487A (en) * | 2020-06-28 | 2020-10-23 | 珠海中科先进技术研究院有限公司 | Health education assessment method, device and medium based on deep learning |
| CN111815487B (en) * | 2020-06-28 | 2024-02-27 | 珠海中科先进技术研究院有限公司 | Deep learning-based health education assessment method, device and medium |
| CN112070529A (en) * | 2020-08-24 | 2020-12-11 | 贵州民族大学 | Passenger carrying hotspot parallel prediction method, system, terminal and computer storage medium |
| CN112085286B (en) * | 2020-09-15 | 2022-04-15 | 周远鹏 | Ship behavior prediction model establishing method and system |
| CN112085286A (en) * | 2020-09-15 | 2020-12-15 | 周远鹏 | Ship behavior prediction model establishing method and ship behavior prediction model |
| WO2023039894A1 (en) * | 2021-09-16 | 2023-03-23 | 上海船舶运输科学研究所有限公司 | Method and system for analyzing transport capacity of container liner route |
| CN117455066A (en) * | 2023-11-13 | 2024-01-26 | 哈尔滨航天恒星数据系统科技有限公司 | Corn planting accurate fertilizer distribution method based on multi-strategy optimization random forest, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107316501B (en) | 2020-04-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107316501A (en) | A kind of SVMs Travel Time Estimation Method based on grid search | |
| Wang et al. | Voyage optimization combining genetic algorithm and dynamic programming for fuel/emissions reduction | |
| CN110188093A (en) | A kind of data digging system being directed to AIS information source based on big data platform | |
| CN106546244B (en) | A kind of method for determining ship's navigation local optimum path for convex barrier boat thing | |
| CN106845716A (en) | A kind of unmanned surface vehicle local delamination paths planning method based on navigation error constraint | |
| CN101644572B (en) | Detection method of ocean eddy variation based on historical similarity cases | |
| CN104573116B (en) | The traffic abnormity recognition methods excavated based on GPS data from taxi | |
| Li et al. | Exploring spatial-temporal change and gravity center movement of construction land in the Chang-Zhu-Tan urban agglomeration | |
| CN105206041A (en) | Smart-phone track chain-cluster identification method considering sequential DBSCAN | |
| Wijaya et al. | Predicting ship behavior navigating through heavily trafficked fairways by analyzing AIS data on Apache HBase | |
| CN107220724A (en) | Passenger flow forecast method and device | |
| CN107193060A (en) | A kind of multipath Typhoon Storm Surge Over method for quick predicting and system | |
| CN110991262A (en) | Multi-bandwidth geographical weighted regression cellular automata method for ecological service value prediction | |
| Yang et al. | Maritime traffic flow clustering analysis by density based trajectory clustering with noise | |
| CN114662575A (en) | Wind power water area ship navigation risk estimation method and system and storage medium | |
| Wen et al. | Spatio-temporal route mining and visualization for busy waterways | |
| CN114139642A (en) | Ship route extraction method combining ship journey semantic object and graph theory | |
| Zhou et al. | Compressing AIS trajectory data based on the multi-objective peak douglas–peucker algorithm | |
| Zhang et al. | Trajectory similarity measure design for ship trajectory clustering | |
| Lee et al. | Generation of Ship’s passage plan using data-driven shortest path algorithms | |
| Zhang et al. | Strength-weighted flow cluster method considering spatiotemporal contiguity to reveal interregional association patterns | |
| Xue et al. | Ship collision avoidance based on brain storm optimization near offshore wind farm | |
| Shao et al. | A GIS-MCDM combination for the site assessment of wave energy power station in qingdao, eastern China | |
| Zhang et al. | Density-weighted ant colony algorithm for ship trajectory reconstruction | |
| Chang | Identify and visualize Dutch inland waterways vessel movement anomalies during low water levels |
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 |