CN106338274A - Wave measurement apparatus for correcting wave characteristic parameters based on multi-parameters, and wave measurement method thereof - Google Patents

Abstract

The invention relates to a wave measurement apparatus for correcting wave characteristic parameters based on multi-parameters, and a wave measurement method thereof. According to the present invention, the sensor with characteristics of mature technology and low cost and other modules are used, and other complicated-design mechanical structures are not required, such that the structure design is simplified, and the cost is low; 1-4 groups of the independent and different multi-parameter sensors are used to perform the synchronous detection, wherein other sensors can be continuously used when one of the sensors fails while the whole detection data collection is not affected, such that the detection reliability and the fault tolerance of the whole equipment are improved; and the derivation is performed by using the traditional formula, the integrating and the secondary integrating are required to be used, and the derivation is non-linear so as to easily produce the error accumulation, while the support vector regression model is used, the target function is fitted by using the non-linear kernel function, the non-linear kernel function is non-linear, the input error is subjected to compatibility, and the calculation is the one calculation, such that the assumptions and the accumulative errors in the derivation are avoided, and the problems of the nonlinearity and the error accumulation of the wave characteristic data derivation are solved.

Description

A kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter and method

Technical field

The present invention relates to a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter and method.

Background technology

Congenerous product mainly adopts pressure type to survey wave technology at present, acoustics surveys wave technology, optics surveys wave technology, remote sensing is surveyed Wave technology.

The pressure oscillation that pressure type subaqueous survey instrument is caused using high resolution sensor measurement surface ripple, according to linear Ripple is theoretical, obtains the relation between corrugated and surge pressure, and the relation of this pressure wave and surface wave quality inspection is still in half theory half Experience state, the surface wave degree of accuracy being finally inversed by still is queried by each side, and observational study even shows, its conversion of different sea areas Coefficient is different, and its rating method is repeatedly to move up and down simulated pressure change and wave in the water of laboratory by popping one's head in , it is impossible to reflect the relation between pressure wave and the surface wave in actual ocean completely, accuracy is low for relation.

Sonic wave gauge strangles principle according to dupp, and instrument looks up under water, thus recording the fluctuation of Free Surface, but Under the conditions of harsh weather and breaker, aqueous vapor intersection surface and interface is not very clear, leads to waverecord to produce more serious Noise, thus affect testing result.

Optics surveys the parameter that Pohle measures wave with the correlation of underwater emission field and sea wave height, surveys ripple using optics The wave instrument that technology is researched and developed is affected by Water quality, and in tested water body, change of water quality and silt debris all will lead to larger detection Trueness error.

Laser remote sensing image derivation wave information is complicated, and in the application of offshore, such as harbour, bay etc. is also limited , the sea image of generation will be transformed into accurate wave spectrum and also differ greatly, so needing to do considerably complicated post processing work Make to obtain quantitative wave information, and laser remote sensing image scattering mechanism is also not exclusively understood backward, its accuracy is also It is poor.

Part acceleration of gravity formula wave instrument can only adapt to single buoy specification or can only a kind of special buoy specification.

Due to existing wave wave height, the isoparametric derivation method of wave direction is more is using derivation of equation method, so Actual algorithm consider upper more be algorithm in the ideal situation, and the ocean wave of reality is affected by many factors, and The factor impact of float device itself.

Existing product in same domain, due to using complicated frame for movement, increased design with the complexity manufacturing with become Basis and volume weight；Due to by the way of traditional, cost being difficult to control, and range is made to be greatly affected； Because volume is big, protrude water surface area greatly, big by wind-force and other ambient influnences, easily produce data deviation；Due to using multiple Miscellaneous electronics and plant equipment, power consumption is big, and the power supply of solar energy is limited, can only timing acquiring data.

Because oceanographic equipment maintenance difficulties are big, same domain product using traditional wave detection mode, take larger volume and Weight, needs to be installed using heavy mechanical equipment and safeguard, leads to install and O＆M cost is high, difficulty is big, and safeguards Frequency is high, and such as sensor experiences failure is keeped in repair it is necessary to remove at once, and workload and difficulty are big；

Content of the invention

It is an object of the invention to provide a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter and Method, detects the wave characteristics parameters such as wave wave height, cycle, wave direction on same device simultaneously；The buoy of foundation institute carry Diameter and density and ocean depth, as parameter, revise wave wave height, to adapt to buoy and the ocean depth of different specification size Degree；Using multigroup, detection is synchronized based on the symmetrically arranged sensor of center of gravity, single sensing is reduced by Multi-sensor Fusion The noise problem of device, can improve the spatial resolution of system, system resistance is transsexual and precision, simplify structure design, reduce Cost.

A kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter of the present invention, including a buoy main body, Three multi-parameter sensors are set up on the vertical center axis of this buoy main body separately, the wherein first multi-parameter sensor is located at centre bit Put, the second multi-parameter sensor is located at the upper summit of buoy main body, the 3rd multi-parameter sensor is located at the lower summit of buoy main body； Separately have four multi-parameter sensors be divided into the first multi-parameter sensor by the center of a horizontal plane constituted foursquare Four vertex positions, are so constituted symmetrical two-by-two three multi-parameter sensor group centered on the first multi-parameter sensor, place Reason device is located at the center of buoy bottom part body, and above-mentioned all of parameter sensors are respectively connecting to this processor, and pass through Data communication interface externally exports the data of this multi-parameter sensor collection, as the input ginseng of support vector regression model Number.

The data of described multi-parameter sensor collection includes: the angle of x, y, z axle, acceleration and angular velocity data.

Further, a kind of survey wave method of the survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter, bag Include following steps:

Step 1, in same time and same site, the data gathering all multi-parameter sensors in buoy forms one and adopts Sample data x:

Wherein, x_ijFor j-th parameter of i-th multi-parameter sensor, i=1 ..., n, this n is Multi-parameter sensing in buoy The total number of device, j=1 ..., m, the parameter sum that this m provides for multi-parameter sensor, using sampled data x as support vector The |input paramete of regression model, the output valve of the vector regression model that is supported in real time, that is, with regard to wave wave height, cycle, wave direction Wave characteristics data predicted value y:

Step 2, the sampled data of collection is randomly divided into some groups, each group of sampled data is built a belt sag because The support vector regression model of son, on the basis of Real-time Collection sensor parameters signal, the constructed support vector of training returns Return model, carry out the reckoning of minimal error, after inspection data input system, returned with support vector constructed above respectively Model is predicted respectively, all predicted values is averaged and obtains the predicted value of wave characteristics data, if all inspection data collection When the goodness of fit of the predicted value in conjunction and actual value reaches required precision, terminate training, be supported vector regression model；

Step 3, in same time and same site, the data gathering all multi-parameter sensors in buoy forms one and adopts Sample data x, using sample data x as support vector regression model |input paramete, be supported vector regression model in real time Output valve, that is, with regard to wave wave height, the cycle, the wave characteristics data of wave direction predicted value y.

Described step 2 specifically includes following steps:

Step 21, select Polynomial kernel function: k (x, x ')=(1+<x | x '>)^d

Wherein x is the |input paramete of support vector regression model, and x ' is the acceleration parameter of the known sample having recorded, D is preset positive integer；

To each group of sampled data x, support vector regression model is trained to be equivalent to following optimization problem is solved:

$min\frac{1}{2}k(\mathrm{\&omega;},\mathrm{\&omega;})+c\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})$

$subjectto.\left\{\begin{array}{c}{y}_i-k(\mathrm{\&omega;},x_i)\&le;\&element;+{\mathrm{\&xi;}}_i\\ k(\mathrm{\&omega;},x_i)-y_i\&le;\&element;+{\mathrm{\&xi;}}_i^{*}\\ {\mathrm{\&xi;}}_i,{\mathrm{\&xi;}}_i^{*}\&greaterequal;0\end{array}\right.$

Wherein, ξ_i、It is relaxation factor, i represents i-th sample, ω is regression variable, and c is the punishment for relaxation factor Coefficient, p is the total sample number of measurement, and ∈ is soft margin coefficient, you can the error of the permission of prediction wave characteristics data of acceptance, y_iIt is the wave characteristics index of i-th sample, corresponding wave height, cycle or wave direction respectively, x_iAcceleration for i-th sample of measurement Degree parameter；

Step 22, above optimization problem is converted into its lagrange duality problem solve:

$-\frac{1}{2}\underset{i,j=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})k(x_i,x_j)-\&element;\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i+{\mathrm{\&alpha;}}_i^{*})+\underset{i=0}{\overset{p}{\&sigma;}}{y}_i({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})$

$\begin{array}{ccc}subjectto.\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})=0& and& {\mathrm{\&alpha;}}_i,{\mathrm{\&alpha;}}_i^{*}\&element;\&lsqb;0,c\&rsqb;\end{array}$

This optimization problem is the planning problem in a convex space, by the iterative predetermined times of random initial value, can Obtain Approximate Global Optimal Solution, reduction formula is expressed as:

$y=f\left(x\right)=\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})k(x_i,x)$

Step 23, by the 10 support vector regression models obtaining corresponding reduction formula f respectively₁,f₂,…,f₁₀, will check After data entry system, it is predicted with above 10 support vector regression models respectively, obtains the prediction of wave characteristics data It is worth and be:

$y={\mathrm{\&sigma;}}_1^{10}{f}_i/10;$

When the goodness of fit of the predicted value in step 24, all inspection data set and actual value reaches required precision, then recognize For above reduction formula f₁,f₂,…,f₁₀Can be as final solution, if the goodness of fit is not up to requiring, return to step 21 Again random packet sampling parameter, and repeat above step, until predicted value in all inspection data set and actual value The goodness of fit reaches requirement, terminates training, be supported vector regression model.

The predicted value high with regard to wave wave calculated in step 3 is modified, according to institute's carry buoy with baud Property, carry out estimation in conjunction with law of buoyancy etc. and derive:

πr²hρ_Water=(h_Actual-h_Survey)πr²ρ_Buoy；

$t=\frac{2\mathrm{\&pi;}}{\mathrm{\&omega;}};$

It is derived from:

Wherein: r is the radius of buoy, buoy contacts the half of the diameter of water surface part, and h is the height of buoy duty wave Degree, the height that is, wave is flattened or raised by buoy, ρ_WaterFor the density of water, ρ_BuoyFor the density of buoy, λ is wavelength, h_ActualFor reality The wave wave on border is high, h_SurveyFor the high predicted value of step 3 calculated wave wave, t is the cycle of ripple, h_{The depth of water}For the depth of water, ω is Angular frequency.

Because oceanographic equipment maintenance difficulties are big, same domain product using traditional wave detection mode, take larger volume and Weight, needs to be installed using heavy mechanical equipment and safeguard, leads to install and O＆M cost is high, difficulty is big, and safeguards Frequency is big, and such as sensor experiences failure is keeped in repair it is necessary to remove at once, and workload and difficulty are big.The present invention uses technology The modules such as the ripe sensor of low cost, it is not necessary to other design complicated frame for movement, simplify structure design, cost is relatively Low.Synchronize detection because invention employs the independently different multi-parameter sensor of 1 to 4 groups, when wherein certain sensor goes out During existing fault, other sensors can be continuing with, and does not affect the collection of whole detection data, improves the detection of integral device Reliability and fault-tolerance；The present invention uses traditional derivation of equation, and due to needing using integration and quadratic integral, deriving, it is non-to have Linearly, easily produce deviation accumulation, the present invention uses support vector regression model, carrys out matching mesh by using Non-linear Kernel function Scalar functions, itself have a nonlinear characteristic, compatible error originated from input and be once to calculate, it is to avoid a variety of hypothesis in reasoning And cumulative errors, solve the problems, such as the non-linear and deviation accumulation that wave characteristics data is derived.

Brief description

Fig. 1 is the structural representation of the survey wave apparatus of the present invention.

Below in conjunction with the drawings and specific embodiments, the present invention is further described.

Specific embodiment

As shown in figure 1, a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter, including a buoy master Body 9, sets up three multi-parameter sensors 5,6,7, wherein multi-parameter sensor 5 on the vertical center axis of this buoy main body 9 separately In center, multi-parameter sensor 6 is located at the upper summit of buoy main body 9, and multi-parameter sensor 7 is located under buoy main body 9 Summit；Multi-parameter sensor 1,2,3,4 be divided into multi-parameter sensor 5 by the center of a horizontal plane constituted foursquare Four vertex positions, are so constituted symmetrical two-by-two three multi-parameter sensor group centered on multi-parameter sensor 5, i.e. many ginsengs Count sensor 6 and 7,2 and 4,1 and 3 three group；Processor 8 is located at the center of buoy main body 9 bottom, above-mentioned all of parameter Sensor is respectively connecting to this processor 8, and externally exports, by data communication interface, the data that this multi-parameter sensor gathers, The data of above-mentioned multi-parameter sensor collection includes: the angle of x, y, z axle, acceleration and angular velocity data, as support The |input paramete of vector regression model, then the data of any one multi-parameter sensor collection is x_i, x_i=(x_i1, x_i2, x_i3..., x_i9), wherein, i=1 ..., n, n are the total number of multi-parameter sensor；By high for the wave wave, cycle (wave period) With the wave characteristics data of wave direction as support vector regression model output valve y, this wave characteristics data y=(y₁, y₂, y₃), y₁High, the y for wave₂For cycle and y₃For wave direction；Specifically include following steps:

Step 1, in same time and same site, the data gathering all multi-parameter sensors in buoy forms one and adopts Sample data x:

Wherein, x_ijFor j-th parameter of i-th multi-parameter sensor, i=1 ..., n, this n is Multi-parameter sensing in buoy The total number of device, j=1 ..., m, the parameter sum that this m provides for multi-parameter sensor, using sample data x as support vector The |input paramete of regression model, the output valve of the vector regression model that is supported in real time, that is, with regard to wave wave height, cycle, wave direction Wave characteristics data predicted value y:

Step 2, the sampled data of collection is randomly divided into some groups, each group of sampled data is built a belt sag because The support vector regression model of son, on the basis of Real-time Collection sensor parameters signal, the constructed support vector of training returns Return model, carry out the reckoning of minimal error, after inspection data input system, returned with support vector constructed above respectively Model is predicted respectively, all predicted values is averaged and obtains the predicted value of wave characteristics data, if all inspection data collection When the goodness of fit of the predicted value in conjunction and actual value reaches required precision, terminate training, be supported vector regression model:

Step 21, select Polynomial kernel function: k (x, x ')=(1+<x | x '>)^d

Wherein x is the |input paramete of support vector regression model, and x ' is the acceleration parameter of the known sample having recorded, D is preset positive integer；

To each group of sampled data x, support vector regression model is trained to be equivalent to following optimization problem is solved:

$min\frac{1}{2}k(\mathrm{\&omega;},\mathrm{\&omega;})+c\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})$

$subjectto.\left\{\begin{array}{c}{y}_i-k(\mathrm{\&omega;},x_i)\&le;\&element;+{\mathrm{\&xi;}}_i\\ k(\mathrm{\&omega;},x_i)-y_i\&le;\&element;+{\mathrm{\&xi;}}_i^{*}\\ {\mathrm{\&xi;}}_i,{\mathrm{\&xi;}}_i^{*}\&greaterequal;0\end{array}\right.$

Wherein, ξ_i、It is relaxation factor, i represents i-th sample, ω is regression variable, and c is the punishment for relaxation factor Coefficient, p is the total sample number of measurement, and ∈ is soft margin coefficient, you can the error of the permission of prediction wave characteristics data of acceptance, y_iIt is the wave characteristics index of i-th sample, corresponding wave height, cycle or wave direction respectively, x_iAcceleration for i-th sample of measurement Degree parameter；

Step 22, above optimization problem is converted into its lagrange duality problem solve:

$-\frac{1}{2}\underset{i,j=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})k(x_i,x_j)-\&element;\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i+{\mathrm{\&alpha;}}_i^{*})+\underset{i=0}{\overset{p}{\&sigma;}}{y}_i({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})$

$\begin{array}{ccc}subjectto.\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})=0& and& {\mathrm{\&alpha;}}_i,{\mathrm{\&alpha;}}_i^{*}\&element;\&lsqb;0,c\&rsqb;\end{array}$

This optimization problem is the planning problem in a convex space, by the iterative of random initial value 1000 times, can obtain Obtain Approximate Global Optimal Solution, reduction formula is expressed as:

$y=f\left(x\right)=\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})k(x_i,x)$

Step 23, by the 10 support vector regression models obtaining corresponding reduction formula f respectively₁,f₂,…,f₁₀, will check After data entry system, it is predicted with above 10 support vector regression models respectively, obtains the prediction of wave characteristics data It is worth and be:

$y={\mathrm{\&sigma;}}_1^{10}{f}_i/10;$

Step 24, when the goodness of fit of the predicted value in all inspection data set and actual value reaches requirement (r- Square is more than 0.9) it is believed that reducing formula f above₁,f2,…,f₁₀Can be as final solution, if the goodness of fit is not up to Require, then return to step 21 random packet sampling parameter again, and repeat above step, until in all inspection data set The goodness of fit of predicted value and actual value reaches requirement, terminates training, be supported vector regression model, so far solves wave special Levy the non-linear and deviation accumulation problem of data derivation；

Step 3, in same time and same site, the data gathering all multi-parameter sensors in buoy forms one and adopts Sample data x, using sample data x as support vector regression model |input paramete, be supported vector regression model in real time Output valve, that is, with regard to wave wave height, the cycle, the wave characteristics data of wave direction predicted value y.

Using 1 to 4 groups, symmetrically arranged multi-parameter sensor centered on buoy main body carries out data fusion to the present invention, carries The high spatial resolution of system, resistance be transsexual and precision；Reduce the noise problem of single-sensor by multiple sensors.Should The basic process of multiple Data Fusion of Sensor technology is exactly to obtain the local message of measurand by multiple sensors, these Local message has complementarity and redundancy in room and time, then according to certain blending algorithm by the local of measurand Information is reasonably combined and is cooperated, and expands coverage on room and time for the multisensor measurement data, eliminates many Redundancy in individual sensor measurement data and error message, thus improve accuracy and the reliability of measurand information.

After wave wave height is calculated by algorithm above, for adapting to buoy and the ocean depth of different specification size, The diameter of the buoy of institute's carry and density and ocean depth as parameter, are gone automatically to repair by the present invention by these relevant parameters Positive wave wave high parameter, is carried out on the basis of high for the data being gathered by each sensor calculated wave wave predicted value Revise.

According to institute's carry buoy with wave property, carry out estimation in conjunction with law of buoyancy etc. and derive:

πr²hρ_Water=(h_Actual-h_Survey)πr²ρ_Buoy；

$t=\frac{2\mathrm{\&pi;}}{\mathrm{\&omega;}};$

It is derived from:

Wherein: r is the radius of buoy, buoy contacts the half of the diameter of water surface part, h is the height of buoy duty wave Degree, the height that is, wave is flattened or raised by buoy, ρ_WaterFor the density of water, ρ_BuoyFor the density of buoy, λ is wavelength, h_ActualFor reality The wave wave on border is high, h_SurveyFor the high predicted value of step 3 calculated wave wave, t is the cycle of ripple, h_{The depth of water}For the depth of water, ω is Angular frequency.

Because oceanographic equipment maintenance difficulties are big, same domain product using traditional wave detection mode, take larger volume and Weight, needs to be installed using heavy mechanical equipment and safeguard, leads to install and O＆M cost is high, difficulty is big, and safeguards Frequency is big, and such as sensor experiences failure is keeped in repair it is necessary to remove at once, and workload and difficulty are big.The present invention uses technology The modules such as the ripe sensor of low cost, it is not necessary to other design complicated frame for movement, simplify structure design, cost is relatively Low.Synchronize detection because invention employs the independently different multi-parameter sensor of 1 to 4 groups, when wherein certain sensor goes out During existing fault, other sensors can be continuing with, and does not affect the collection of whole detection data, improves the detection of integral device Reliability and fault-tolerance.

Claims (5)

1. a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter it is characterised in that: include a buoy master Body, sets up three multi-parameter sensors separately, during the wherein first multi-parameter sensor is located on the vertical center axis of this buoy main body Heart position, the second multi-parameter sensor is located at the upper summit of buoy main body, and the 3rd multi-parameter sensor is located under buoy main body Summit；Four multi-parameter sensors are separately had to be divided into the pros being constituted by the center of a horizontal plane with the first multi-parameter sensor Four vertex positions of shape, are so constituted symmetrical two-by-two three multi-parameter sensor centered on the first multi-parameter sensor Group, processor is located at the center of buoy bottom part body, and above-mentioned all of parameter sensors are respectively connecting to this processor, and Externally export the data of this multi-parameter sensor collection by data communication interface, defeated as support vector regression model Enter parameter.

2. a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter according to claim 1, it is special Levy and be that the data of described multi-parameter sensor collection includes: the angle of x, y, z axle, acceleration and angular velocity data.

3. the survey ripple of a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter according to claim 1 Method is it is characterised in that comprise the steps:

Step 1, in same time and same site, the data gathering all multi-parameter sensors in buoy forms a hits According to x:

Wherein, x_ijFor j-th parameter of i-th multi-parameter sensor, i=1 ..., n, this n is multi-parameter sensor in buoy Total number, j=1 ..., m, the parameter sum that this m provides for multi-parameter sensor, sampled data x is returned as support vector The |input paramete of model, the output valve of the vector regression model that is supported in real time, that is, with regard to wave wave height, the cycle, wave direction ripple Predicted value y of unrestrained characteristic:

Step 2, the sampled data of collection is randomly divided into some groups, a belt sag factor is built to each group of sampled data Support vector regression model, on the basis of Real-time Collection sensor parameters signal, the constructed support vector of training returns mould Type, carries out the reckoning of minimal error, after inspection data input system, uses support vector regression model constructed above respectively It is predicted respectively, all predicted values is averaged and obtains the predicted value of wave characteristics data, if in all inspection data set Predicted value and the goodness of fit of actual value when reaching required precision, terminate training, be supported vector regression model；

Step 3, in same time and same site, the data gathering all multi-parameter sensors in buoy forms a hits According to x, using sample data x as support vector regression model |input paramete, the output of the vector regression model that is supported in real time Value, that is, with regard to wave wave height, the cycle, the wave characteristics data of wave direction predicted value y.

4. the survey ripple of a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter according to claim 3 Method is it is characterised in that described step 2 specifically includes following steps:

Step 21, select Polynomial kernel function: k (x, x ')=(1+<x | x '>)^d

Wherein x is the |input paramete of support vector regression model, and x ' is the acceleration parameter of the known sample having recorded, and d is Preset positive integer；

To each group of sampled data x, support vector regression model is trained to be equivalent to following optimization problem is solved:

$min\frac{1}{2}k(\mathrm{\&omega;},\mathrm{\&omega;})+c\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})$

$subjectto.\left\{\begin{array}{c}{y}_i-k(\mathrm{\&omega;},x_i)\&le;\&element;+{\mathrm{\&xi;}}_i\\ k(\mathrm{\&omega;},x_i)-y_i\&le;\&element;+{\mathrm{\&xi;}}_i^{*}\\ {\mathrm{\&xi;}}_i,{\mathrm{\&xi;}}_i^{*}\&greaterequal;0\end{array}\right.$

Wherein, ξ_i、It is relaxation factor, i represents i-th sample, ω is regression variable, and c is the punishment system for relaxation factor Number, p is the total sample number of measurement, and ε is soft margin coefficient, you can the error of the permission of prediction wave characteristics data of acceptance, y_i It is the wave characteristics index of i-th sample, corresponding wave height, cycle or wave direction respectively, x_iAcceleration for i-th sample of measurement Parameter；

Step 22, above optimization problem is converted into its lagrange duality problem solve:

$-\frac{1}{2}\underset{i,j=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})({\mathrm{\&alpha;}}_j-{\mathrm{\&alpha;}}_j^{*})k(x_i,x_j)-\&element;\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i+{\mathrm{\&alpha;}}_i^{*})+\underset{i=0}{\overset{p}{\&sigma;}}{y}_i({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})$

$subjectto.\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})=0{\mathrm{and\&alpha;}}_i,{\mathrm{\&alpha;}}_i^{*}\&element;\&lsqb;0,c\&rsqb;$

This optimization problem is the planning problem in a convex space, by the iterative predetermined times of random initial value, can obtain Approximate Global Optimal Solution, reduction formula is expressed as:

$y=f\left(x\right)=\underset{i=1}{\overset{p}{\&sigma;}}({\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_i^{*})k(x_i,x)$

Step 23, by the 10 support vector regression models obtaining corresponding reduction formula f respectively₁,f₂,…,f₁₀, by inspection data After input system, it is predicted with above 10 support vector regression models respectively, the predicted value obtaining wave characteristics data is:

$y={\mathrm{\&sigma;}}_1^{10}{f}_i/10;$

The goodness of fit of the predicted value in step 24, all inspection data set and actual value reach during required precision then it is assumed that with Upper reduction formula f₁,f₂,…,f₁₀Can be used as final solution, if the goodness of fit is not up to requiring, return to step 21 is again Random packet sampling parameter, and repeat above step, until the coincideing of predicted value in all inspection data set and actual value Degree reaches requirement, terminates training, be supported vector regression model.

5. the survey ripple of a kind of survey wave apparatus based on multi-parameter sensor correction wave characteristics parameter according to claim 3 Method is it is characterised in that be modified to the predicted value high with regard to wave wave calculated in step 3, according to institute's carry buoy With wave property, carry out estimation in conjunction with law of buoyancy etc. and derive:

πr²hρ_Water=(h_Actual-h_Survey)πr²ρ_Buoy；

$t=\frac{2\mathrm{\&pi;}}{\mathrm{\&omega;}};$

It is derived from:

Wherein: r is the radius of buoy, buoy contacts the half of the diameter of water surface part, and h is the height of buoy duty wave, that is, The height that wave is flattened or raised by buoy, ρ_WaterFor the density of water, ρ_BuoyFor the density of buoy, λ is wavelength, h_ActualFor actual ripple Wave wave is high, h_SurveyFor the high predicted value of step 3 calculated wave wave, t is the cycle of ripple, h_{The depth of water}For the depth of water, ω is angular frequency Rate.