CN114492217A - Typhoon and wave height prediction method based on mixed time series perceptron model - Google Patents

Typhoon and wave height prediction method based on mixed time series perceptron model Download PDF

Info

Publication number
CN114492217A
CN114492217A CN202210406972.5A CN202210406972A CN114492217A CN 114492217 A CN114492217 A CN 114492217A CN 202210406972 A CN202210406972 A CN 202210406972A CN 114492217 A CN114492217 A CN 114492217A
Authority
CN
China
Prior art keywords
typhoon
data
layer
wave height
perceptron 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.)
Granted
Application number
CN202210406972.5A
Other languages
Chinese (zh)
Other versions
CN114492217B (en
Inventor
王智峰
巩艺杰
董胜
陶山山
张日
黄炜楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ocean University of China
Original Assignee
Ocean University of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ocean University of China filed Critical Ocean University of China
Priority to CN202210406972.5A priority Critical patent/CN114492217B/en
Publication of CN114492217A publication Critical patent/CN114492217A/en
Application granted granted Critical
Publication of CN114492217B publication Critical patent/CN114492217B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A typhoon wave height prediction method based on a mixed time series perceptron model relates to the technical field of big data application and safety risk assessment, and comprises the following steps: step 1, data acquisition; step 2, preprocessing data; step 3, establishing a perceptron model based on a mixed time sequence; and 4, predicting the wave height of the typhoon sea waves according to the sensor model. The invention can forecast the change of the wave height of the sea with time under the typhoon weather which changes violently in time, rapidly and accurately, in particular to the following steps: the sensor model established by the invention realizes timely, rapid and accurate prediction of the wave height of the sea wave in typhoon weather by searching the nonlinear physical relationship between the physical quantities of the wind speed, the wind direction, the air pressure, the typhoon center position and the typhoon center maximum wind speed and the wave height during typhoon and considering the relationship of the wave height changing along with time during the typhoon process.

Description

Typhoon and wave height prediction method based on mixed time series perceptron model
Technical Field
The invention relates to the technical field of big data application and safety risk assessment, in particular to a typhoon and wave height prediction method based on a mixed time series sensor model.
Background
According to statistics, typhoon disasters are second oceanic disasters in China, and strong wind and extreme sea waves brought by typhoons can damage coastal protection structures and cause great harm to life safety and property safety of human society. The billow caused by typhoon is one of the main factors for destroying the coastal buildings. During the typhoon, the typhoon wind field changes continuously along with the movement of the typhoon center, and the wave brought by the typhoon also changes continuously along with the change of the wind field. In this case, it is important to predict the wave change condition during typhoon in time, and it is particularly important for wave energy acquisition, marine fishery, marine traffic, and marine engineering construction. Therefore, there is a need to develop a tool capable of predicting wave conditions in typhoon period in time to predict the generation and development of billows in advance, so as to achieve the effect of early warning.
At present, the wave is predicted in time mainly by an artificial neural network model, and the artificial neural network model searches for the change rule of the wave height by training a large amount of wave height data, so that the occurrence and the change of the future wave height are predicted. The long-term and short-term memory network is an improved recurrent neural network, and can solve the problem that the recurrent neural network cannot handle long-distance dependence. The long-short term memory network is mainly used for classifying, processing and predicting data based on time series. However, neither the ordinary artificial neural network nor the long-short term memory network can deal well with the problem of nonlinearity with complex physical quantities, which cannot be well captured for the complex physical relationship between input and output quantities.
Because the ocean condition changes rapidly and violently during the typhoon, the wave changes caused by the typhoon are difficult to capture, simulate and predict, and the research on the timely prediction of the wave height during the typhoon is limited at present. The existing wave spectrum numerical model is usually used for simulating the development and the change of waves, however, the wave spectrum model must be driven by a complex input wind field, and simultaneously needs boundary conditions for constraint, and the calculation amount is large, the calculation time is long, and the rapid and timely prediction is not facilitated. In addition, machine learning algorithms such as neural networks have been applied to wave height prediction as a fast and efficient tool, but the current wave height prediction application only stays in common weather conditions, the wave height changes are not severe, and the dynamics of the wave height changes are relatively easy to capture and simulate. In typhoon weather with severe change, the wave spectrum model is difficult to realize the timely prediction of the wave height due to the limitation of the input conditions of the wave spectrum model, and the existing neural network model for predicting the wave height is difficult to capture the severe wave height change due to the lack of calculation of the typhoon physical background, so the wave height prediction in the typhoon weather is not accurate.
Disclosure of Invention
The invention provides a typhoon wave height prediction method based on a mixed time series sensor model, aiming at establishing a mixed time series sensor model to predict wave height change in a typhoon period in time under severe typhoon weather conditions, so as to achieve the timely, rapid and accurate prediction effect. The model establishes the physical relation under the typhoon weather by searching the relation between the wave height and the physical quantities such as the wind speed, the wind direction, the air pressure, the typhoon center position and the typhoon center maximum wind speed during the typhoon, and simultaneously considers the relation of the wave height changing along with time in the typhoon process, thereby not only ensuring the real-time property, but also ensuring the accuracy, and therefore, the real-time change of the wave height under the typhoon weather can be predicted in real time and accurately.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a typhoon wave height prediction method based on a mixed time series perceptron model comprises the following steps: step 1, data acquisition; step 2, preprocessing data; step 3, establishing a perceptron model based on a mixed time sequence; and 4, predicting the wave height of the typhoon sea waves according to the sensor model.
Preferably, in step 1: the data to be acquired comprises typhoon center information and marine environment information of the measuring points; the typhoon center information is obtained from the optimal path data of the typhoon of the central meteorological station typhoon network, and comprises the longitude and latitude of the typhoon center position, the typhoon center air pressure and the typhoon center maximum wind speed; the marine environment information of the measuring points comprises wind speed, wind direction and effective wave height of the measuring points.
Preferably, in step 1, the data of one typhoon process is used as one data set, and a plurality of data sets are obtained.
Preferably, in step 2, the data in each data set are sorted according to a time sequence, the effective wave height is used as an output parameter of the perceptron model to be built, and other data information is used as an input parameter.
Preferably, the step 3 comprises the following specific steps:
A. calculation of time series conversion layer: converting the preprocessed input data into input neurons containing time series information with a specific time delay, for each input neuron
Figure 691285DEST_PATH_IMAGE001
Using corresponding translation vectorsS i Performing a transformation operation as shown in formula (1):
Figure 377481DEST_PATH_IMAGE002
(1)
in the formula (1), the reaction mixture is,
Figure 381209DEST_PATH_IMAGE003
input vector for the first n time instants
Figure 822555DEST_PATH_IMAGE004
If the information at that moment is taken into account, then
Figure 13365DEST_PATH_IMAGE005
If the information at that moment is ignored, then
Figure 565569DEST_PATH_IMAGE006
；
B. Data normalization processing: input parameters with different orders of magnitude are normalized, and the calculation method is shown as formula (2):
Figure 158224DEST_PATH_IMAGE007
(2)
in the formula (2), the reaction mixture is,Datais the sample data of the sample data,Data minis the minimum value of the sample data,Data maxis the maximum value of the sample data,Data norin order to normalize the processed data, the data is normalized,Data nor,maxin order to normalize the upper limit of the interval,Data norminis the lower limit of the normalization interval;
wherein the different orders of magnitude refer to: for example, the wind speed is about 20-60, the wind direction is 0-360, the wave height is 0-15, the values of the variables are greatly different, and the variables need to be normalized to the same order of magnitude;
C. establishing a multilayer perceptron model: determining the number of input layer sensing units according to the input parameters and the output parameter types
Figure 708154DEST_PATH_IMAGE008
Implicit number of layersNum h Number of sensing unitsE h And the number of output layersE out ；
D. Determining boundary conditions and initial conditions of a perceptron model: the training algorithm selects a gradient descent back propagation algorithm, the training learning rate is 0.1, the training iteration number is 200, the training target error is 1e-5, the activation function of each unit of the hidden layer is a tansig function and is defined as
Figure 448577DEST_PATH_IMAGE009
The activation function of each unit of the output layer is a purelin function and is defined asf(x)=x；
E. Initializing a perceptron model weight threshold: randomly generating a weight matrix for the entire perceptron modelWAnd threshold matrixb；
F. The signal (information) of each neuron is gradually transferred from the conversion layer to the hidden layer and the output layer according to the topological structure of the perceptron model to describe the nonlinear behavior thereof, which is expressed as formula (3):
Figure 742155DEST_PATH_IMAGE010
(3)
in formula (3):
Figure 454896DEST_PATH_IMAGE011
is a firstl-1Layer onekA unit andllayer onenThe weight between the individual units is such that,
Figure 706886DEST_PATH_IMAGE012
is as followsl-1Layer onekThe output of the individual cells is then,
Figure 934605DEST_PATH_IMAGE013
is as followslLayer onenThe threshold value of each of the cells is,f(. h) is an activation function;
G. calculating an error cost functionC(w,b) Computing output of output layera NumhVector and target vectory expAnd (4) the error of each layer is reversely transferred layer by layer, and the gradient of each layer of cost function about the weight and the threshold is estimated through a gradient descent algorithm, as shown in formulas (4) to (8):
Figure 31874DEST_PATH_IMAGE014
(4)
Figure 599122DEST_PATH_IMAGE015
(5)
Figure 287592DEST_PATH_IMAGE016
(6)
Figure 940290DEST_PATH_IMAGE017
(7)
Figure 637988DEST_PATH_IMAGE018
(8)
in formulae (4) to (8):
Figure 614338DEST_PATH_IMAGE019
in order to output the layer error vector(s),a Numhfor the output of the vectors at the output layer,y expin order to be the target vector,δ l is as followslThe error vector of the layer is determined,δ l+1is as followslThe error vector of the +1 layer is,w l+1is composed oflLayer and the firstlA connection weight matrix between +1 and,
Figure 942551DEST_PATH_IMAGE020
is the gradient of the cost function with respect to the weight,
Figure 816966DEST_PATH_IMAGE021
a gradient of the cost function with respect to a threshold;
H. modifying the weight matrix of the whole perceptron model according to the gradient of the cost function about the weight and the threshold valueWAnd threshold matrixb：
Figure 521617DEST_PATH_IMAGE022
(9)
Figure 860194DEST_PATH_IMAGE023
(10)
In formulas (9) and (10):
Figure 93730DEST_PATH_IMAGE024
the efficiency of the network learning is improved,
Figure 721020DEST_PATH_IMAGE025
first, thelLayer onenThe bias of the individual neurons is such that,
Figure 494941DEST_PATH_IMAGE026
is the firstl-1 layer ofkA neuron to the firstlLayer onenThe connection weights of the individual neurons are,Cis an error cost function.
Preferably, in the step 4, the specific steps F-H are repeated until the output layer outputsa NumhVector and target vectory expError of (2)n e Less than a predetermined errore q I.e. satisfye q >n e And ending the training to obtain the predicted wave height sequence of the typhoon sea waves.
The typhoon wave height prediction method based on the mixed time series sensor model has the beneficial effects that:
the invention can forecast the change of the wave height of the sea with time under the typhoon weather which changes violently in time, rapidly and accurately, in particular to the following steps: the sensor model established by the invention realizes timely, rapid and accurate prediction of the wave height of the sea wave in typhoon weather by searching the nonlinear physical relationship between the physical quantities of the wind speed, the wind direction, the air pressure, the typhoon center position and the typhoon center maximum wind speed and the wave height during typhoon and considering the relationship of the wave height changing along with time during the typhoon process.
Drawings
Fig. 1 and 2, flow charts of the present invention.
Detailed Description
In the following, embodiments of the present invention are described in detail in a stepwise manner, which is merely a preferred embodiment of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of the present invention.
In the description of the present invention, it should be noted that the terms "upper", "lower", "left", "right", "top", "bottom", "inner", "outer", and the like indicate orientations and positional relationships based on the orientations and positional relationships shown in the drawings, and are only used for describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation and a specific orientation configuration and operation, and thus, the present invention is not to be construed as being limited thereto.
Please refer to fig. 1:
a typhoon wave height prediction method based on a mixed time series perceptron model comprises the following steps: step 1, data acquisition; step 2, preprocessing data; step 3, establishing a perceptron model based on a mixed time sequence; step 4, predicting the wave height of the typhoon sea waves according to a sensor model;
in the step 1: the data to be acquired comprises typhoon center information and marine environment information of the measuring points; the typhoon center information is obtained from the optimal path data of the typhoon of the central meteorological station typhoon network, and comprises the longitude and latitude of the typhoon center position, the typhoon center air pressure and the typhoon center maximum wind speed; the marine environment information of the measuring points comprises wind speed, wind direction and effective wave height of the measuring points;
in the step 1, data of a typhoon process is used as a data set to obtain a plurality of data sets;
in the step 2, data in each data set are sorted according to a time sequence, the effective wave height is used as an output parameter of a perceptron model to be built, and other data information is used as an input parameter;
the step 3 comprises the following specific steps:
A. calculation of time series conversion layer: converting the preprocessed input data into input neurons containing time series information with a specific time delay, for each input neuron
Figure 688025DEST_PATH_IMAGE027
Using corresponding translation vectors
Figure 92461DEST_PATH_IMAGE028
Performing a transformation operation as shown in formula (1):
Figure 207048DEST_PATH_IMAGE002
(1)
in the formula (1), the reaction mixture is,
Figure 909294DEST_PATH_IMAGE003
input vector for the first n time instants
Figure 160146DEST_PATH_IMAGE004
If the information at that moment is taken into account, then
Figure 797801DEST_PATH_IMAGE005
If the information at that moment is ignored, then
Figure 134105DEST_PATH_IMAGE006
；
B. Data normalization processing: input parameters with different orders of magnitude are normalized, and the calculation method is shown as formula (2):
Figure 515407DEST_PATH_IMAGE007
(2)
in the formula (2), the reaction mixture is,Datais the sample data of the sample data,Data minis the minimum value of the sample data,Data maxis the maximum value of the sample data,Data norin order to normalize the processed data, the data is normalized,Data nor,maxin order to normalize the upper limit of the interval,Data norminis the lower limit of the normalization interval;
wherein the different orders of magnitude refer to: for example, the wind speed is about 20-60, the wind direction is 0-360, the wave height is 0-15, the values of the variables are greatly different, and the variables need to be normalized to the same order of magnitude;
C. establishing a multilayer perceptron model: determining the number of input layer sensing units according to the input parameters and the output parameter types
Figure 886346DEST_PATH_IMAGE029
Number of hidden layersNum h Number of sensing unitsE h And the number of output layersE out ；
D. Determining boundary conditions and initial conditions of a perceptron modelThe initial conditions are as follows: the training algorithm selects a gradient descent back propagation algorithm, the training learning rate is 0.1, the training iteration number is 200, the training target error is 1e-5, the activation function of each unit of the hidden layer is a tansig function and is defined as
Figure 632585DEST_PATH_IMAGE009
The activation function of each unit of the output layer is a purelin function and is defined asf(x)=x；
E. Initializing a perceptron model weight threshold: randomly generating a weight matrix for the entire perceptron modelWAnd threshold matrixb；
F. The signal (information) of each neuron is gradually transferred from the conversion layer to the hidden layer and the output layer according to the topological structure of the perceptron model to describe the nonlinear behavior thereof, which is expressed as formula (3):
Figure 252922DEST_PATH_IMAGE010
(3)
in formula (3):
Figure 641178DEST_PATH_IMAGE011
is as followsl-1Layer onekA unit andllayer onenThe weight between the individual units is such that,
Figure 397782DEST_PATH_IMAGE012
is as followsl-1Layer onekThe output of the individual cells is then,
Figure 314922DEST_PATH_IMAGE013
is as followslLayer onenThe threshold value of each of the cells is,f(. h) is an activation function;
calculating an error cost functionC(w,b) Computing output layer outputa NumhVector and target vectory expAnd (3) transmitting the error in reverse layer by layer, and estimating the gradient of each layer of cost function relative to the weight and the threshold value by a gradient descent algorithm, as shown in formulas (4) to (8):
Figure 688134DEST_PATH_IMAGE014
(4)
Figure 614502DEST_PATH_IMAGE015
(5)
Figure 428874DEST_PATH_IMAGE016
(6)
Figure 844812DEST_PATH_IMAGE017
(7)
Figure 643004DEST_PATH_IMAGE018
(8)
in formulae (4) to (8):
Figure 107483DEST_PATH_IMAGE019
in order to output the layer error vector(s),a Numhfor the output of the vectors at the output layer,y expin order to be the target vector,δ l is as followslThe error vector of the layer is determined,δ l+1is a firstlThe error vector of the +1 layer is,w l+1is composed oflLayer and the firstlA connection weight matrix between +1 and,
Figure 838679DEST_PATH_IMAGE020
is the gradient of the cost function with respect to the weight,
Figure 363201DEST_PATH_IMAGE021
a gradient of the cost function with respect to the threshold;
H. correcting the weight matrix of the whole perceptron model according to the gradient of the cost function about the weight and the threshold valueWAnd threshold matrixb：
Figure 445427DEST_PATH_IMAGE022
(9)
Figure 713597DEST_PATH_IMAGE023
(10)
In formulas (9) and (10):
Figure 502562DEST_PATH_IMAGE024
the efficiency of the network learning is improved,
Figure 260302DEST_PATH_IMAGE025
first, thelLayer onenThe bias of the individual neurons is such that,
Figure 501927DEST_PATH_IMAGE026
is the firstl-1 layer ofkThe first neuron pairlLayer onenThe connection weights of the individual neurons are,Cis an error cost function.
In the step 4, the specific steps F-H are repeated until the output layer outputsa NumhVector and target vectory expError of (2)n e Less than a predetermined errore q That is to say satisfye q >n e And ending the training to obtain the predicted wave height sequence of the typhoon sea waves.
The following describes embodiments of the present invention in conjunction with practical applications:
the typhoon Taili is formed in 9 months in 2017, is upgraded to strong typhoon in 13 days in 9 months, and then enters the middle of the east sea of China and is close to coastal areas of Fujian province. Under the influence of a typhoon Taili wind field, the wind power is gradually strengthened, the great waves impact the seawall, the wave height reaches 3-4 floors, the maximum wind speed of the typhoon center is 52 m/s, and casualties and huge property loss are caused. In order to test the performance of the model provided by the invention, the typhoon Taili is subjected to prediction simulation.
Corresponding to the steps 1 and 2, the example obtained a data set of 55 typhoon courses passing through the sea area near taiwan area in Fujian in 1991-2015. The data set comprises the wind speed, the wind direction and the effective wave height of an experimental point, the longitude and latitude of a typhoon center, the air pressure of the typhoon center and the maximum wind speed of the typhoon center. Since the data set is large in data volume, specific data is omitted here.
The method in step 3 of the invention is used for predicting the effective wave height of a certain measuring point in the Taili period of typhoon. Meanwhile, three widely applied machine learning models, namely a Radial Basis Function (RBF) neural network, a Multilayer Perceptron (MLP) neural network and a Long Short-Term Memory (LSTM) neural network, are selected for comparison. The predictions are divided into 3-hour predictions, 6-hour predictions, 12-hour predictions, and 24-hour predictions. For comparing the prediction error, 3 indexes are selected to evaluate the error between the prediction result and the actually measured effective wave height, which are respectively as follows: coefficient of correlation (R)2) Absolute value of relative error (RAE) and relative root error (RRSE). Wherein the correlation coefficient is in the range of 0-1, the larger the numerical value is, the smaller the error is represented, and the smaller the numerical values of RAE and RRSE are, the smaller the error is represented. Table 1 shows the error between the predicted result of the present invention and the measured effective wave height (last column) and the error between the predicted results of the other three models. The data show that the invention predicts the R of the result under the same lead period condition2Consistently greater than the results of the other three models, and RAE and RRSE consistently less than the results of the other three models. The results demonstrate the effectiveness and accuracy of the present invention in typhoon weather.
TABLE 1 prediction error of other prediction models of the invention
Figure 370526DEST_PATH_IMAGE030

Claims (6)

1. A typhoon wave height prediction method based on a mixed time series perceptron model is characterized in that: the method comprises the following steps: step 1, data acquisition; step 2, preprocessing data; step 3, establishing a perceptron model based on a mixed time sequence; and 4, predicting the wave height of the typhoon sea waves according to the sensor model.
2. The method for predicting the typhoon wave height based on the mixed time series perceptron model according to claim 1, wherein: in the step 1: the data to be acquired comprises typhoon center information and marine environment information of the measuring points; the typhoon center information is obtained from the optimal path data of the typhoon of the central meteorological station typhoon network, and comprises the longitude and latitude of the typhoon center position, the typhoon center air pressure and the typhoon center maximum wind speed; the marine environment information of the measuring points comprises wind speed, wind direction and effective wave height of the measuring points.
3. The method for predicting the typhoon wave height based on the mixed time series perceptron model according to claim 2, characterized in that: in the step 1, data of a typhoon process is used as a data set, and a plurality of data sets are obtained.
4. A typhoon wave height prediction method based on a mixed time series perceptron model according to claim 3, characterized by: in the step 2, the data in each data set are sorted according to the time sequence, the effective wave height is used as an output parameter of the perceptron model to be built, and other data information is used as an input parameter.
5. The method for predicting the height of a typhoon wave based on the mixed time series sensor model according to claim 4, wherein: the step 3 comprises the following specific steps:
A. calculation of time series conversion layer: converting the preprocessed input data into input neurons containing time series information with a specific time delay, for each input neuron
Figure 981137DEST_PATH_IMAGE001
Using corresponding translation vectorsS i Performing a transformation operation as shown in formula (1):
Figure 429435DEST_PATH_IMAGE002
(1)
in the formula (1), the reaction mixture is,
Figure 740331DEST_PATH_IMAGE003
input vector for the first n time instants
Figure 401119DEST_PATH_IMAGE004
If the information at that moment is taken into account, then
Figure 746650DEST_PATH_IMAGE005
If the information at that moment is ignored, then
Figure 897009DEST_PATH_IMAGE006
；
B. Data normalization processing: input parameters with different orders of magnitude are normalized, and the calculation method is shown as formula (2):
Figure 695201DEST_PATH_IMAGE007
(2)
in the formula (2), the reaction mixture is,Dataas the sample data, the data is,Data minis the minimum value of the sample data,Data maxis the maximum value of the sample data,Data norin order to normalize the processed data, the data is normalized,Data nor,maxin order to normalize the upper limit of the interval,Data norminis the lower limit of the normalization interval;
C. establishing a multilayer perceptron model: determining the number of input layer sensing units according to the input parameters and the output parameter types
Figure 159680DEST_PATH_IMAGE008
Implicit number of layersNum h Number of sensing unitsE h And the number of output layersE out ；
D. Determining boundary conditions and initial conditions of a perceptron model: training algorithm selection gradient descent back propagation algorithmThe training learning rate is 0.1, the training iteration number is 200, the training target error is 1e-5, and the activation function of each unit of the hidden layer is a tansig function defined as
Figure 359717DEST_PATH_IMAGE009
The activation function of each unit of the output layer is a purelin function and is defined asf(x)=x；
E. Initializing a weight threshold of a perceptron model: randomly generating a weight matrix for the entire perceptron modelWAnd a threshold matrixb；
F. The signal (information) of each neuron is gradually transferred from the conversion layer to the hidden layer and the output layer according to the topological structure of the perceptron model to describe its nonlinear behavior, which can be expressed as formula (3):
Figure 680977DEST_PATH_IMAGE010
(3)
in formula (3):
Figure 966465DEST_PATH_IMAGE011
is as followsl-1Layer onekA unit andllayer onenThe weight between the individual units is such that,
Figure 234635DEST_PATH_IMAGE012
is as followsl-1Layer onekThe output of the individual cells is then,
Figure 23600DEST_PATH_IMAGE013
is as followslLayer onenThe threshold value of each of the cells is,f(. h) is an activation function;
G. calculating an error cost functionC(w,b) Computing output of output layera NumhVector and target vectory expAnd (4) the error of each layer is reversely transferred layer by layer, and the gradient of each layer of cost function about the weight and the threshold is estimated through a gradient descent algorithm, as shown in formulas (4) to (8):
Figure 984602DEST_PATH_IMAGE014
(4)
Figure 554124DEST_PATH_IMAGE015
(5)
Figure 360406DEST_PATH_IMAGE016
(6)
Figure 269456DEST_PATH_IMAGE017
(7)
Figure 135781DEST_PATH_IMAGE018
(8)
in formulae (4) to (8):
Figure 192599DEST_PATH_IMAGE019
in order to output the layer error vector(s),a Numhfor the output of the vectors at the output layer,y expis a target vector, and is a target vector,δ l is as followslThe error vector of the layer is determined,δ l+1is as followslThe error vector of the +1 layer is,w l+1is composed oflLayer and the firstlA connection weight matrix between +1 and,
Figure 802572DEST_PATH_IMAGE020
is the gradient of the cost function with respect to the weight,
Figure 589566DEST_PATH_IMAGE021
a gradient of the cost function with respect to the threshold;
H. modifying the weight matrix of the whole perceptron model according to the gradient of the cost function about the weight and the threshold valueWAnd threshold matrixb：
Figure 892371DEST_PATH_IMAGE022
(9)
Figure 170906DEST_PATH_IMAGE023
(10)
In formulas (9) and (10):
Figure 318990DEST_PATH_IMAGE024
the efficiency of the network learning is improved,
Figure 202633DEST_PATH_IMAGE025
first, thelLayer onenThe bias of the individual neurons is such that,
Figure 207498DEST_PATH_IMAGE026
is the firstl-1 layer ofkThe first neuron pairlLayer onenThe connection weights of the individual neurons are,Cis an error cost function.
6. The method for predicting the typhoon wave height based on the mixed time series perceptron model according to claim 5, characterized in that: in the step 4, the specific steps F-H are repeated until the output layer outputsa NumhVector and target vectory expError of (2)n e Less than a predetermined errore q I.e. satisfye q >n e And ending the training to obtain the wave height sequence of the typhoon sea waves.
CN202210406972.5A 2022-04-19 2022-04-19 Typhoon and wave height prediction method based on mixed time series perceptron model Active CN114492217B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210406972.5A CN114492217B (en) 2022-04-19 2022-04-19 Typhoon and wave height prediction method based on mixed time series perceptron model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210406972.5A CN114492217B (en) 2022-04-19 2022-04-19 Typhoon and wave height prediction method based on mixed time series perceptron model

Publications (2)

Publication Number Publication Date
CN114492217A true CN114492217A (en) 2022-05-13
CN114492217B CN114492217B (en) 2022-07-22

Family

ID=81489277

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210406972.5A Active CN114492217B (en) 2022-04-19 2022-04-19 Typhoon and wave height prediction method based on mixed time series perceptron model

Country Status (1)

Country Link
CN (1) CN114492217B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116147587A (en) * 2023-04-17 2023-05-23 南开大学 Wave prediction method and wave measurement system
CN116611270A (en) * 2023-07-20 2023-08-18 国家海洋局北海预报中心((国家海洋局青岛海洋预报台)(国家海洋局青岛海洋环境监测中心站)) Typhoon wave real-time aggregate forecasting method, medium and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005306188A (en) * 2004-04-21 2005-11-04 Kawasaki Heavy Ind Ltd Waveform predicting method of arrival ocean wave, and operation control method of sailing body in ocean wave
KR101736918B1 (en) * 2016-03-29 2017-05-17 한국해양과학기술원 System and Method for Predicting Sea Wave Using Sea Surface Wind Numerical Model Forecast Data
CN109886217A (en) * 2019-02-26 2019-06-14 上海海洋大学 A method of it is high that wave being detected from Nearshore Wave video based on convolutional neural networks

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005306188A (en) * 2004-04-21 2005-11-04 Kawasaki Heavy Ind Ltd Waveform predicting method of arrival ocean wave, and operation control method of sailing body in ocean wave
KR101736918B1 (en) * 2016-03-29 2017-05-17 한국해양과학기술원 System and Method for Predicting Sea Wave Using Sea Surface Wind Numerical Model Forecast Data
CN109886217A (en) * 2019-02-26 2019-06-14 上海海洋大学 A method of it is high that wave being detected from Nearshore Wave video based on convolutional neural networks

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
ABDUSSELAM ALTUNKAYNAK: ""Prediction of significant wave height using geno-multilayer perceptron"", 《ELSEVIER》 *
刘刚 等: "《人工智能导论》", 31 July 2020, 北京：北京邮电大学出版社 *
吕艳: ""基于深度学习的网络入侵检测方法研究"", 《万方数据库》 *
周水华 等: ""基于人工神经网络的台风浪高快速计算方法"", 《热带海洋学报》 *
唐闻: ""基于深度学习的计算机图像识别技术研究"", 《电脑编程技巧与维护》 *
武装: "《京津冀地区PM2.5及其他空气污染物的时空分布特征研究》", 30 September 2018, 北京：科学技术文献出版社 *
陈希 等: ""人工神经网络技术在台风浪预报中的应用"", 《海洋科学》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116147587A (en) * 2023-04-17 2023-05-23 南开大学 Wave prediction method and wave measurement system
CN116611270A (en) * 2023-07-20 2023-08-18 国家海洋局北海预报中心((国家海洋局青岛海洋预报台)(国家海洋局青岛海洋环境监测中心站)) Typhoon wave real-time aggregate forecasting method, medium and system
CN116611270B (en) * 2023-07-20 2023-10-03 国家海洋局北海预报中心((国家海洋局青岛海洋预报台)(国家海洋局青岛海洋环境监测中心站)) Typhoon wave real-time aggregate forecasting method, medium and system

Also Published As

Publication number Publication date
CN114492217B (en) 2022-07-22

Similar Documents

Publication Publication Date Title
Alemany et al. Predicting hurricane trajectories using a recurrent neural network
CN114626512B (en) High-temperature disaster forecasting method based on directed graph neural network
CN114492217B (en) Typhoon and wave height prediction method based on mixed time series perceptron model
CN109142171B (en) Urban PM10 concentration prediction method based on feature expansion and fusing with neural network
Lian et al. A novel data-driven tropical cyclone track prediction model based on CNN and GRU with multi-dimensional feature selection
CN104616078B (en) Photovoltaic system electricity generation power Forecasting Methodology based on Spiking neutral nets
CN112558185A (en) Bidirectional GRU typhoon track intelligent prediction and forecast system based on attention mechanism, computer equipment and storage medium
CN114154401B (en) Soil erosion modulus calculation method and system based on machine learning and observation data
Yao et al. Wave height forecast method with multi-step training set extension LSTM neural network
CN115877483A (en) Typhoon path forecasting method based on random forest and GRU
Rohaimi et al. 3 Hours ahead of time flood water level prediction using NNARX structure: Case study pahang
Mar et al. Optimum neural network architecture for precipitation prediction of Myanmar
Moghaddamnia et al. Evaporation estimation using support vector machines technique
CN116822716A (en) Typhoon prediction method, system, equipment and medium based on space-time attention
Bechrakis et al. Wind speed prediction using artificial neural networks
Jallal et al. Half-hour global solar radiation forecasting based on static and dynamic multivariate neural networks.
Yao et al. Extreme motion prediction and early-warning assessment of semisubmersible platform based on deep learning method
Mandal et al. Prediction of Wind Speed using Machine Learning
Zhang et al. Correcting predictions from simulating wave nearshore model via Gaussian process regression
LU504816B1 (en) Interpretable deep learning model for extreme climate prediction
Elbisy The use of group method of data handling and multilayer perceptron neural network for the prediction of significant wave height
CN118246348B (en) Track reconstruction method for marine buoy
Zelinka et al. Computer intelligence in modeling, prediction, and analysis of complex dynamical systems
Ladjal et al. Application of machine learning techniques for predicting the WQI for water quality monitoring: a case study in Algeria
Lin et al. Three-dimensional temporal-spatial attention for tropical cyclone forecast

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