CN114445634A - Sea wave height prediction method and system based on deep learning model - Google Patents
Sea wave height prediction method and system based on deep learning model Download PDFInfo
- Publication number
- CN114445634A CN114445634A CN202210189203.4A CN202210189203A CN114445634A CN 114445634 A CN114445634 A CN 114445634A CN 202210189203 A CN202210189203 A CN 202210189203A CN 114445634 A CN114445634 A CN 114445634A
- Authority
- CN
- China
- Prior art keywords
- wave height
- prediction
- sea
- data
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013136 deep learning model Methods 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000013135 deep learning Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000013528 artificial neural network Methods 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 26
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 18
- 210000002569 neuron Anatomy 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000013178 mathematical model Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 7
- 230000001413 cellular effect Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 15
- 230000007547 defect Effects 0.000 abstract description 8
- 238000005516 engineering process Methods 0.000 abstract description 7
- 239000000284 extract Substances 0.000 description 13
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 6
- 230000002457 bidirectional effect Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000002441 reversible effect Effects 0.000 description 4
- 230000003213 activating effect Effects 0.000 description 3
- 238000012876 topography Methods 0.000 description 3
- 239000002131 composite material Substances 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012950 reanalysis Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/045—Combinations of networks
Abstract
The invention discloses a sea wave height prediction method and a sea wave height prediction system based on a deep learning model, which relate to the technical field of ocean prediction and comprise the following steps: acquiring picture data of the wave height; preprocessing the picture data of the wave height of the sea waves, and taking the preprocessed picture data of the wave height of the sea waves as a training data set; constructing a sea wave effective wave height data prediction model, inputting the training data set into the sea wave effective wave height data prediction model for deep learning training until reaching a preset precision, and obtaining an optimal prediction model; and predicting the effective wave height of the sea wave through the optimal prediction model, and performing inverse normalization on a prediction result to obtain a predicted value of the effective wave height of the sea wave. The method can overcome the defects of large calculation amount, high cost, incapability of quick prediction, dependence on characteristic engineering and the like in the prior prediction technology, and further realize quick and accurate prediction with low cost.
Description
Technical Field
The invention relates to the technical field of ocean prediction, in particular to a sea wave height prediction method and a sea wave height prediction system based on a deep learning model.
Background
According to the mechanism of wave formation, waves can be divided into storms and swells. Since the sea surface wave is actually an irregular combination of a plurality of waves with different wave heights, periods and traveling directions, the wave height value of one wave is not representative. The effective wave height refers to an actual wave height value counted according to a certain rule, such as an average wave height, a root mean square wave height, a maximum wave height, a certain guarantee rate wave height, an average wave height of a certain part of waves and the like. The effective wave height has great influence on ocean engineering construction, marine navigation and transportation, and is also an important parameter for ocean disaster forecast and wave energy sustainable power generation evaluation. Currently, the most widely used forecasting models are the third generation numerical wave prediction models (WAM, WaveWatch III, SWAN, etc.), a computational model based on energy balance equations that considers various physical processes. The effective wave height of the sea waves is influenced by weather conditions, seasonal characteristics and terrain and topography factors, and has strong nonlinearity and strong asymmetry. According to different principles, the wave prediction models can be divided into the following types:
1) numerical calculation model
And the numerical calculation model represented by the WWWIII is used for calculating and analyzing physical data obtained by each sensor according to a natural rule through a real physical rule to obtain a prediction result. Because the effective wave height of the sea waves is influenced by multiple factors such as wind power, temperature, submarine topography and the like, the method has the defects of complex calculation process, long operation time, large amount of physical data, high prediction cost and the like, and can not realize quick and accurate prediction; and because the gridding discrete processing is adopted and the differential is replaced by the difference, numerical errors are inevitably introduced, and the problems of unconvergence and even instability of numerical calculation are faced.
2) Machine learning model
Machine learning is carried out by selecting a proper mathematical model, drawing up hyper-parameters, inputting sample data of characteristic engineering processing, and carrying out parameter tuning on the model by applying a proper learning algorithm according to a certain strategy, thereby obtaining a prediction model which has high fitting on the sample data and certain generalization capability. The machine learning method can fit a complex nonlinear process without system prior knowledge, and solves the nonlinear problem of complex physical mechanism. Therefore, the method is widely applied to prediction of the effective wave height, such as single-structure prediction models of Artificial Neural Networks (ANN), Support Vector Machines (SVM) and the like, and composite network prediction models of wavelet transformation neural networks (WLNN), EMD-SVR models, ICEEMDAN-ELM models, MLR-CWLS models and the like, and the method is good in short-term prediction and high in accuracy of a composite network prediction result. However, the machine learning method depends on the feature extraction technology, and the generalization capability needs to be improved.
3) Deep learning model
Different from a common machine learning model, the deep learning model learns more abstract feature data in the data by using a neural network containing a large number of hidden layers through a large number of vector calculations, performs higher-dimensionality feature division, and makes a decision according to the feature division. The deep learning method can automatically learn the internal rules and the representation levels of the sample data, and provides powerful support for ocean prediction. The existing deep learning method prediction effective wave height models comprise LSTM, CNN-LSTM and the like, but most of the models are simple structures, the relevance among data cannot be fully mined, and the data features can be fully extracted by applying a multilayer neural network. With the development of high-performance computing industry and deep learning models, deep learning has been widely applied to the prediction field with its strong learning ability, good nonlinear fitting ability and portability.
In the prior art, only time sequence prediction of a single observation point can be performed on sea wave prediction, and two-dimensional observation data are rarely and directly processed; in addition, the conventional prediction technology has the defects of large calculation amount, long calculation time, difficult convergence and the like. Therefore, how to solve the defects of large calculation amount, high cost, incapability of fast prediction, dependence on characteristic engineering and the like of the existing prediction technology and further realize fast and accurate prediction with low cost is a technical problem which needs to be solved by technical personnel in the field.
Disclosure of Invention
In view of the above, the invention provides a method and a system for predicting wave height of sea waves based on a deep learning model, which solve the problems in the background art.
In order to achieve the above purpose, the invention provides the following technical scheme:
a sea wave height prediction method based on a deep learning model comprises the following steps:
acquiring picture data of the wave height of the sea;
preprocessing the picture data of the wave height of the sea waves, and taking the preprocessed picture data of the wave height of the sea waves as a training data set;
constructing a sea wave effective wave height data prediction model, inputting the training data set into the sea wave effective wave height data prediction model for deep learning training until reaching a preset precision, and obtaining an optimal prediction model;
and predicting the effective wave height of the sea wave through the optimal prediction model, and performing inverse normalization on a prediction result to obtain a predicted value of the effective wave height of the sea wave.
The technical effect achieved by the technical scheme is as follows: the method solves the defects of large calculation amount, high cost, incapability of quick prediction, dependence on characteristic engineering and the like in the prior prediction technology, and realizes quick and accurate prediction with low cost.
Optionally, the preprocessing is performed on the image data of the sea wave height, and specifically includes the following steps:
eliminating abnormal wave values; wherein the abnormal wave value comprises a negative value, an abnormally high value;
filling missing values of the effective wave height by adopting a filling mode comprising average value filling and interpolation filling;
filling 0 values into different grid point data of land;
and normalizing the filled effective wave height data, wherein the formula is defined as follows:
in the formula, x is the data of the effective wave height of the original two-dimensional sea wave, xminIs the minimum value of input sample data x, xmaxIs the maximum value of the input sample data x.
Optionally, the constructed sea wave effective wave height data prediction model is a CNN-BilSTM-Attention deep learning prediction model, and a network structure combining a CNN neural network, a BilSTM neural network and an Attention mechanism is adopted.
The prediction model designed above has the following technical effects: the CNN model can fully extract the spatial local characteristics of data and is widely applied to image processing; the BilSTM model considers the function of bidirectional information flow, fully extracts the characteristics of time sequence data correlation information, improves the model prediction precision, and is suitable for the prediction of time sequence data; the Attention mechanism can evaluate and score the prediction result of the BilSTM and extract an important part in the prediction result.
Optionally, the mathematical model of the CNN neural network is:
ci＝f(wi*xi+bi)；
in the formula, xiRepresenting the input of the convolutional layer, ciIs the output characteristic map of the ith layer, wiWeight matrix representing convolution, representing dot product, biTo the variation vector, f (-) represents the activation function.
Optionally, the BilsTM neural network is formed by adding a reverse LSTM layer on the basis of the LSTM network, that is, the prediction result of the BilsTM neural network is the combination of the LSTM network and the reverse LSTM layer prediction result; wherein, the mathematical model of the LSTM model is as follows:
ft＝σ(Wf[ht-1,xt]+bf)；
it＝σ(Wi[ht-1,xt]+bi)；
at＝tanh(Wa·[ht-1,xt]+ba)；
Ct＝ft·Ct-1+it·at；
ot＝σ(Wo·[ht-1,xt]+bo)；
ht＝ot·tanh(Ct)；
in the formula (f)tA control vector representing a forgetting gate of the LSTM algorithm that determines the effect of a past state on a current state; σ (-) and tanh (-) represent Sigmoid and tanh activation functions, respectively; wf、Wi、WaAnd WoRepresenting a weight; h ist-1Is the output of the t-1 th neuron, xtInput to the t-th neuron; bf、bi、baAnd boIndicating a deviation; i.e. itControl vector representing input gate, which determines a of the current new inputt(ii) degree of acceptance; a istA new input state representing the current time; ctIndicating the cellular state of the t-th neuron, Ct-1Represents the cellular state of the t-1 th neuron, otRepresenting the output result of the output gate;
will otAnd tanh (C)t) Obtaining the final output result h after the dot product operationt。
Optionally, the mathematical model of the Attention mechanism is as follows:
M＝tanh(Y)；
A＝YαT；
wherein M represents the output of tanh function, Y represents the feature matrix captured by the BilSTM neural network, alpha represents the attention weight matrix obtained by Sigmoid function,
the transpose of the weight matrix is shown, and a is the output result after attention mechanism processing.
Optionally, the obtaining the optimal prediction model specifically includes the following steps:
performing two-dimensional convolution on the training data set, and extracting effective wave height characteristics; the two-dimensional convolution can take the actual spatial distribution of the wave height data of the sea waves into consideration, extract important features in the wave height data, reduce redundant information and improve prediction precision;
inputting the convolution result of the two-dimensional convolution into a BilSTM neural network, and predicting according to the change rule of data at the previous moment and the next moment; in the LSTM prediction process, important information is reserved, and non-important information is forgotten; the BilSTM considers the function of bidirectional information flow, fully extracts the characteristics of the time sequence data associated information, and improves the model prediction precision;
the neurons are temporarily deleted in the Dropout layer, and partial characteristics are randomly discarded, so that the complex co-adaptation relation among the neurons can be reduced, the overfitting of the model is prevented, and the generalization of the model is stronger;
an Attention mechanism is introduced, probability weight is distributed to output information of the BilSTM neural network, larger weight is given to important information in the characteristic diagram, and the capturing capability of the model to the important information is improved;
judging whether the model and the parameters meet the requirement of preset precision, if not, updating the parameters of the sea wave effective wave height data prediction model by using a back propagation algorithm, and continuing deep learning training; and if so, obtaining the optimal prediction model.
Optionally, the method further includes: carrying out error analysis on a predicted value of the effective wave height of the sea wave by adopting an average absolute error, a root mean square error and an average correlation coefficient of a sample space;
in the sample space, the average absolute error between the predicted value and the true value at each moment is as follows:
in the sample space, the mean of the root mean square errors of the predicted value and the true value at a single moment is as follows:
in the sample space, the average correlation coefficient of the predicted value and the true value at each moment is as follows:
in the formula, K is the time sequence length of the test samples, namely the number of the test samples; i represents the number of weft total lattice points; j represents the number of total lattice points in the warp direction; h isp(i, j) represents the value of the effective wave height predicted by a certain point in space based on CNN-BilSTM-Attention; h ism(i, j) represents a spatial correspondence hp(i, j) the actual value of the grid point location; n is the total number of examples.
The invention also provides a sea wave height prediction system based on the deep learning model, which comprises the following steps:
the acquisition module is used for acquiring picture data of the wave height of the sea waves;
the preprocessing module is used for preprocessing the picture data of the wave height of the sea waves and taking the preprocessed picture data of the wave height of the sea waves as a training data set;
the construction module is used for constructing a sea wave effective wave height data prediction model;
the training module is used for inputting the training data set into a sea wave effective wave height data prediction model for deep learning training until the preset precision is reached, and obtaining an optimal prediction model;
and the prediction module is used for predicting the wave height of the effective wave of the sea wave through the optimal prediction model and carrying out inverse normalization on the prediction result to obtain the prediction value of the wave height of the effective wave of the sea wave.
The present invention also provides a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for wave height prediction of sea waves.
Compared with the prior art, the invention discloses and provides a sea wave height prediction method and system based on a deep learning model, and the method and system have the following beneficial effects:
(1) the invention solves the defects of large calculation amount, high cost, incapability of quick prediction, dependence on characteristic engineering and the like in the prior prediction technology, and realizes quick and accurate prediction with low cost;
(2) the invention is suitable for the prediction of the wave height data of the sea waves, the designed prediction model adopts a network structure combining a CNN neural network, a BilSTM neural network and an Attention mechanism, and has the advantages that: the CNN model can fully extract the spatial local characteristics of data and is widely applied to image processing; the BilSTM model considers the function of bidirectional information flow, fully extracts the characteristics of time sequence data correlation information, improves the model prediction precision, and is suitable for the prediction of time sequence data; the Attention mechanism can evaluate and score the prediction result of the BilSTM and extract an important part in the prediction result.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic flow chart of a method for predicting wave height of sea waves in the present invention;
FIG. 2 is a block diagram of a CNN-BilSTM-Attention prediction model;
FIG. 3 is a diagram of the LSTM model architecture;
FIG. 4 is a diagram of a BilSTM network architecture;
FIG. 5 is an example prediction results presentation graph;
FIGS. 6(a) -6 (c) are graphs comparing the prediction effect of the present invention and the prior art of the significant wave height;
fig. 7 is a structural diagram of a wave height prediction system according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
The embodiment of the invention discloses a sea wave height prediction method based on a deep learning model, which comprises the following steps as shown in figure 1:
(1) acquiring the data of the effective wave height of original two-dimensional sea waves;
in the embodiment, the geographical range of the selection of the sea wave effective wave height data is 105 to 126 degrees from east longitude and 4 to 43 degrees from north latitude, the sea wave effective wave height data comprises two-dimensional map data consisting of observation values of 79 multiplied by 43 observation points, and the spatial resolution of the two-dimensional map data is 0.5 multiplied by 0.5.
(2) Preprocessing original two-dimensional sea wave effective wave height data, and taking the preprocessed sea wave effective wave height data as a training data set; the pretreatment process specifically comprises the following steps:
(2-1) eliminating abnormal wave values, such as negative values, abnormal high values and the like;
(2-2) filling missing values of the significant wave height, wherein the filling mode comprises the following steps: mean value filling, interpolation filling and the like;
(2-3) performing 0-value padding on the land-different lattice point data;
(2-4) normalizing the filled significant wave height data, wherein the formula is defined as follows:
in the formula, x is the data of the effective wave height of the original two-dimensional sea wave, xminIs the minimum value of input sample data x, xmaxIs the maximum value of the input sample data x.
And selecting the first 70% of the two-dimensional effective wave height data after data preprocessing as a training data set of the deep learning model, and taking the last 30% as a test data set of the model. To improve the prediction accuracy of the model, the training data set should be no less than 70% of the total sample data.
(3) Building a sea wave effective wave height data prediction model, inputting a training data set into the sea wave effective wave height data prediction model for deep learning training until reaching a preset precision, and obtaining an optimal prediction model;
(3-1) as the effective wave height of the sea wave is influenced by weather, has the characteristics of strong nonlinearity, strong asymmetry and the like, and is closely related to various factors such as weather conditions, seasonal characteristics, submarine topography advantages and the like, a deep learning prediction model of the effective wave height data prediction model CNN-BiLSTM-Attention is constructed, and as shown in FIG. 2, a network structure combining a CNN neural network, a BiLSTM neural network and an Attention mechanism is adopted, so that the method has the following advantages: the CNN model can fully extract the spatial local characteristics of data and is widely applied to image processing; the BilSTM model considers the function of bidirectional information flow, fully extracts the characteristics of time sequence data correlation information, improves the model prediction precision, and is suitable for the prediction of time sequence data; the Attention mechanism can evaluate and score the prediction result of the BilSTM and extract an important part in the prediction result.
(3-1-1) the mathematical model of the CNN neural network is:
ci＝f(wi*xi+bi)；
in the formula, xiRepresenting the input of the convolutional layer, ciIs the output characteristic map of the ith layer, wiWeight matrix representing convolution, representing dot product, biTo the variation vector, f (-) represents the activation function.
(3-1-2) adding a reverse LSTM layer on the basis of the LSTM network, namely the prediction result of the BiLSTM neural network is the combination of the LSTM network and the reverse LSTM layer prediction result; the mathematical model of the LSTM model (see fig. 3) is:
ft＝σ(Wf[ht-1,xt]+bf)；
it＝σ(Wi[ht-1,xt]+bi)；
at＝tanh(Wa·[ht-1,xt]+ba)；
Ct＝ft·Ct-1+it·at；
ot＝σ(Wo·[ht-1,xt]+bo)；
ht＝ot·tanh(Ct)；
in the formula (f)tA control vector representing a forgetting gate of the LSTM algorithm that determines the effect of a past state on a current state; σ (-) andtanh (·) represents Sigmoid and tanh activation functions, respectively; wf、Wi、WaAnd WoRepresenting a weight; h ist-1Is the output of the t-1 th neuron, xtInput to the t-th neuron; bf、bi、baAnd boIndicating a deviation; i.e. itControl vector representing input gate, which determines a of the current new inputt(ii) degree of acceptance; a istA new input state representing the current time; ctIndicating the cellular state of the t-th neuron, Ct-1Represents the cellular state of the t-1 th neuron, otRepresenting the output result of the output gate;
will otAnd tanh (C)t) Obtaining a final output result h after dot product operationt。
(3-1-3) the mathematical model of the Attention mechanism is:
M＝tanh(Y)；
A＝YαT；
wherein M represents the output of tanh function, Y represents the feature matrix captured by the BilSTM neural network, alpha represents the attention weight matrix obtained by Sigmoid function,
the transpose of the weight matrix is shown, and a is the output result after attention mechanism processing.
(3-2) completing model training by using a training data set to enable the model training to reach specified precision, grouping input two-dimensional effective wave height data according to a preset prediction time step, and generating a predicted value by each group of two-dimensional effective wave height data; setting a training label according to a preset prediction time interval, and determining the training direction of the model, wherein the specific training process comprises the following steps:
(3-2-1) performing two-dimensional convolution on the training data set, wherein the two-dimensional convolution can take the actual spatial distribution of the wave height data of the sea waves into consideration, extract important features, reduce redundant information and improve prediction precision. The two-dimensional convolution layer is three layers, so that the effective wave height characteristics which are abstract enough can be extracted, and the model prediction is convenient; the number of each layer of filters is 80, namely through one layer of convolution layer, the effective wave height data at each moment can obtain 80 effective wave height characteristic diagrams, and enough potential wave characteristics can be extracted; the activation function is a Relu function, and the function can set the convolution result smaller than 0 to be 0 and keep the convolution result larger than 0 unchanged, so that the convolution result does not have abnormal values and the neural network has good nonlinear fitting capability.
(3-2-2) inputting the convolution result of the two-dimensional convolution into a BilSTM neural network (see figure 4), and predicting according to the data change rule of the previous time and the next time; the BilSTM considers the function of bidirectional information flow, fully extracts the characteristics of the time sequence data associated information, and improves the model prediction precision. The BiLSTM layer is two layers and comprises two LSTM networks which are opposite to each other, the characteristic graphs obtained by two-dimensional convolution are respectively input into the LSTM networks in two directions to obtain two groups of predicted values, one group of predicted values is predicted from the past to the past, the other group of predicted values is predicted from the future to the past, and finally output results contain past and future information of a time sequence, so that the time sequence relation between effective wave height data can be better reflected, and accurate prediction results are obtained; the activating function selects a tanh function, the cyclic activating function is a Sigmoid function, and the activating function and the Sigmoid function provide nonlinear fitting capacity for the BilSTM network, so that a prediction program can adopt a GPU to accelerate calculation, and training time is shortened; through each layer of the BilSTM layer, 6 prediction characteristic maps are generated for the effective wave height at each moment.
(3-2-3) temporarily deleting the neurons at a Dropout layer, and randomly discarding part of the features; the Dropout layer parameter is set to be 0.5, neurons are deleted randomly and temporarily with the probability of 50%, so that the complex co-adaptive relation among the neurons can be reduced, and the updating of the weight parameter is independent of the co-action of implicit nodes with fixed relation; and overfitting of the model is prevented, so that the generalization of the model is stronger.
(3-2-4) introducing an Attention mechanism, distributing probability weight to output information of the BilSTM neural network, giving greater weight to important information in the characteristic diagram, and improving the grasping capability of the model to the important information.
(3-2-5) judging whether the model and the parameters meet the requirement of preset precision, if not, updating the parameters of the sea wave effective wave height data prediction model by using a back propagation algorithm, and continuing deep learning training; and if so, obtaining the optimal prediction model.
(4) And predicting the wave height of the sea wave by using the optimal prediction model, and performing inverse normalization on the prediction result to obtain the predicted value of the wave height of the sea wave.
In a further embodiment, the method further comprises: carrying out error analysis on a predicted value of the effective wave height of the sea wave by adopting an average absolute error, a root mean square error and an average correlation coefficient of a sample space;
in the sample space, the average absolute error between the predicted value and the true value at each moment is as follows:
in the sample space, the mean of the root mean square errors of the predicted value and the true value at a single moment is as follows:
in the sample space, the average correlation coefficient of the predicted value and the true value at each moment is as follows:
in the formula, K is the time sequence length of the test samples, namely the number of the test samples; i represents the number of weft total lattice points; j represents the number of total lattice points in the warp direction; h isp(i, j) represents the value of the effective wave height predicted by a certain point in space based on CNN-BilSTM-Attention; h ism(i, j) represents a spatial correspondence hp(i, j) the actual value of the grid point location; n is the total number of examples.
The embodiment of the invention also discloses a computer-storable medium on which a computer program is stored, which is characterized in that the computer program is executed by a processor to realize the steps of the wave height prediction method.
The method for predicting the height of the sea wave in example 1 is further understood by the following specific implementation examples.
The effective wave height data of the sea waves in the reanalysis product of the WW3 third-generation wave numerical mode published by NOAA in 2011-charge 2020 is adopted as an experimental data set. The geographic range of the selected data is the effective wave height data of the offshore area of China with east longitude from 105 degrees to 126 degrees and north latitude from 4 degrees to 43 degrees, the spatial resolution is 0.5 degrees multiplied by 0.5 degrees, and the selected time span is 2011-2020 years. In consideration of the timeliness of waves, the input data used in the present study are all wave field data at three consecutive time instants, and according to the difference of the predicted time step (e.g. 1h, 2h, 3h) and the difference of the data time span, the first 70% of the data set is used as a training sample and a verification sample, and the last 30% is used as a test sample. In the forecast of the effective wave height of the general sea state, the test data sets are effective wave height data of 12:00 in 14 months in 2018 to 23:59 in 31 months in 12 months in 2020, and the data are not involved in the training of the model so as to ensure the relative independence of the training and the testing. The predicted results are shown in the following table:
in order to further show the prediction results of the models, selecting and showing examples of the prediction results, as shown in fig. 5, (a-e) backward predicting the effective wave height spatial distribution maps of 1 hour, 3 hours, 6 hours, 12 hours and 24 hours for the effective wave height data at noon of 11:00, 12:00 and 13:00 of 5 month and 20 days of 2020; (f-j) is the effective wave height spatial profile of WW 3; and (k-o) is an absolute error space distribution diagram of the predicted values and the true values at five moments. As can be seen from fig. 5, the prediction error gradually increases as the prediction time span increases.
Fig. 6(a) -6 (c) are the comparison between the present invention and the prior art, and the present invention is superior to the prior art. As can be seen from FIG. 6(a), the predicted correlation between the CNN-BilSTM algorithm and the CNN-BilSTM-Attention algorithm is equivalent in the first 12h prediction, and the correlation between the CNN-BilSTM-Attention algorithm is higher in the 12h-24h prediction and better than that between the ConvLSTM algorithm and the CNN-BilSTM algorithm. As can be seen from FIGS. 6(b) and 6(c), overall, the CNN-BilSTM-Attention algorithm is optimal and superior to the ConvLSTM algorithm. The prediction within 6h, the CNN-BilSTM-Attention is obviously superior to the CNN-BilSTM algorithm, and the prediction performance of the CNN-BilSTM-Attention algorithm after 6h has certain advantages over the CNN-BilSTM algorithm and is superior to the ConvLSTM algorithm.
Example 2
The embodiment of the invention discloses a sea wave height prediction system based on a deep learning model, which is shown in figure 7 and comprises the following components:
the acquisition module is used for acquiring picture data of the wave height of the sea waves;
the preprocessing module is used for preprocessing the picture data of the wave height of the sea waves and taking the preprocessed picture data of the wave height of the sea waves as a training data set;
the construction module is used for constructing a sea wave effective wave height data prediction model;
the training module is used for inputting the training data set into the sea wave effective wave height data prediction model to carry out deep learning training until the preset precision is reached, and obtaining an optimal prediction model;
and the prediction module is used for predicting the wave height of the effective wave of the sea wave through the optimal prediction model and carrying out inverse normalization on the prediction result to obtain the prediction value of the wave height of the effective wave of the sea wave.
In the prior art, only time sequence prediction of a single observation point can be performed on sea wave prediction, and two-dimensional observation data are rarely and directly processed; in addition, the conventional prediction technology has the defects of large calculation amount, long calculation time, difficult convergence and the like. The technical scheme of the invention can overcome the defects of large calculation amount, high cost, incapability of quick prediction, dependence on characteristic engineering and the like of the conventional prediction technology, and further realize quick and accurate prediction with low cost.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A sea wave height prediction method based on a deep learning model is characterized by comprising the following steps:
acquiring picture data of the wave height of the sea;
preprocessing the picture data of the wave height of the sea waves, and taking the preprocessed picture data of the wave height of the sea waves as a training data set;
constructing a sea wave effective wave height data prediction model, inputting the training data set into the sea wave effective wave height data prediction model for deep learning training until reaching a preset precision, and obtaining an optimal prediction model;
and predicting the effective wave height of the sea wave through the optimal prediction model, and performing inverse normalization on a prediction result to obtain a predicted value of the effective wave height of the sea wave.
2. A sea wave height prediction method based on a deep learning model according to claim 1, wherein the image data of the sea wave height is preprocessed, and the method specifically comprises the following steps:
eliminating abnormal wave values; wherein the abnormal wave value comprises a negative value, an abnormally high value;
filling missing values of the effective wave height by adopting a filling mode comprising average value filling and interpolation filling;
filling 0 values into different grid point data of land;
and normalizing the filled effective wave height data, wherein the formula is defined as follows:
in the formula, x is the data of the effective wave height of the original two-dimensional sea wave, xminIs the minimum value of input sample data x, xmaxIs the maximum value of the input sample data x.
3. The deep learning model-based wave height prediction method of sea waves according to claim 1, characterized in that the constructed wave height data prediction model of effective waves is a CNN-BilSTM-Attention deep learning prediction model, and a network structure combining a CNN neural network, a BilSTM neural network and an Attention mechanism is adopted.
4. A deep learning model-based wave height prediction method as claimed in claim 3, wherein the mathematical model of the CNN neural network is:
ci＝f(wi*xi+bi)；
in the formula, xiRepresenting the input of the convolutional layer, ciIs the output characteristic map of the ith layer, wiWeight matrix representing convolution, representing dot product, biTo the variation vector, f (-) represents the activation function.
5. A sea wave height prediction method based on a deep learning model as claimed in claim 3, wherein the BilSTM neural network is formed by adding an inverse LSTM layer on the basis of the LSTM network, that is, the prediction result of the BilSTM neural network is the combination of the LSTM network and the inverse LSTM layer prediction result; wherein, the mathematical model of the LSTM model is as follows:
ft＝σ(Wf[ht-1,xt]+bf)；
it＝σ(Wi[ht-1,xt]+bi)；
at＝tanh(Wa·[ht-1,xt]+ba)；
Ct＝ft·Ct-1+it·at；
ot＝σ(Wo·[ht-1,xt]+bo)；
ht＝ot·tanh(Ct)；
in the formula (f)tA control vector representing a forgetting gate of the LSTM algorithm that determines the effect of a past state on a current state; σ (-) and tanh (-) represent Sigmoid and tanh activation functions, respectively; wf、Wi、WaAnd WoRepresenting a weight; h ist-1Is the output of the t-1 th neuron, xtInput to the t-th neuron; bf、bi、baAnd boIndicating a deviation; i.e. itControl vector representing input gate, which determines a of the current new inputt(ii) degree of acceptance; a istIndicating the new input state at the present moment, CtIndicating the cellular state of the t-th neuron, Ct-1Represents the cellular state of the t-1 th neuron, otRepresenting the output result of the output gate;
will otAnd tanh (C)t) Obtaining the final output result h after the dot product operationt。
6. A deep learning model-based wave height prediction method according to claim 3, characterized in that the mathematical model of the Attention mechanism is as follows:
M＝tanh(Y)；
A＝YαT；
wherein M represents the output of tanh function, Y represents the feature matrix captured by the BilSTM neural network, alpha represents the attention weight matrix obtained by Sigmoid function,
the transpose of the weight matrix is shown, and a is the output result after attention mechanism processing.
7. A sea wave height prediction method based on a deep learning model according to claim 3, wherein the obtaining of the optimal prediction model specifically comprises the following steps:
performing two-dimensional convolution on the training data set, and extracting effective wave height characteristics;
inputting the convolution result of the two-dimensional convolution into a BilSTM neural network, and predicting according to the change rule of data at the previous moment and the next moment;
temporarily deleting neurons in a Dropout layer, and randomly discarding partial features;
introducing an Attention mechanism, and distributing probability weight to output information of the BilSTM neural network;
judging whether the model and the parameters meet the requirement of preset precision, if not, updating the parameters of the sea wave effective wave height data prediction model by using a back propagation algorithm, and continuing deep learning training; and if so, obtaining the optimal prediction model.
8. A deep learning model based sea wave height prediction method according to claim 1, characterized in that the method further comprises: carrying out error analysis on a predicted value of the effective wave height of the sea wave by adopting an average absolute error, a root mean square error and an average correlation coefficient of a sample space;
in the sample space, the average absolute error between the predicted value and the true value at each moment is as follows:
in the sample space, the mean of the root mean square errors of the predicted value and the true value at a single moment is as follows:
in a sample space, the average correlation coefficient of the predicted value and the true value at each moment is as follows:
in the formula, K is the time sequence length of the test samples, namely the number of the test samples; i represents the number of weft total lattice points; j represents the number of total lattice points in the warp direction; h isp(i, j) represents the value of the effective wave height predicted by a certain point in space based on CNN-BilSTM-Attention; h ism(i, j) represents a spatial correspondence hp(i, j) the actual value of the grid point location; n is the total number of examples.
9. A sea wave height prediction system based on a deep learning model is characterized by comprising:
the acquisition module is used for acquiring picture data of the wave height of the sea waves;
the preprocessing module is used for preprocessing the picture data of the wave height of the sea waves and taking the preprocessed picture data of the wave height of the sea waves as a training data set;
the construction module is used for constructing a sea wave effective wave height data prediction model;
the training module is used for inputting the training data set into a sea wave effective wave height data prediction model for deep learning training until the preset precision is reached, and obtaining an optimal prediction model;
and the prediction module is used for predicting the wave height of the effective wave of the sea wave through the optimal prediction model and carrying out inverse normalization on the prediction result to obtain the prediction value of the wave height of the effective wave of the sea wave.
10. A computer-storable medium having a computer program stored thereon, wherein the computer program, when being executed by a processor, carries out the steps of the wave height prediction method according to any one of the claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210189203.4A CN114445634A (en) | 2022-02-28 | 2022-02-28 | Sea wave height prediction method and system based on deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210189203.4A CN114445634A (en) | 2022-02-28 | 2022-02-28 | Sea wave height prediction method and system based on deep learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114445634A true CN114445634A (en) | 2022-05-06 |
Family
ID=81373537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210189203.4A Pending CN114445634A (en) | 2022-02-28 | 2022-02-28 | Sea wave height prediction method and system based on deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114445634A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114912077A (en) * | 2022-05-27 | 2022-08-16 | 中国海洋大学 | Sea wave forecasting algorithm integrating random search and mixed decomposition error correction |
| CN115149529A (en) * | 2022-09-05 | 2022-10-04 | 国家海洋技术中心 | Wave energy power generation power prediction method and system |
| CN115169439A (en) * | 2022-06-16 | 2022-10-11 | 中国人民解放军国防科技大学 | Method and system for predicting effective wave height based on sequence-to-sequence network |
| CN115238602A (en) * | 2022-07-01 | 2022-10-25 | 中国海洋大学 | CNN-LSTM-based prediction method for contribution rate of transient liquefaction of wave-induced seabed to resuspension |
| CN116822253A (en) * | 2023-08-29 | 2023-09-29 | 山东省计算中心（国家超级计算济南中心） | Hybrid precision implementation method and system suitable for MANUM sea wave mode |
| CN117471575A (en) * | 2023-12-28 | 2024-01-30 | 河海大学 | Typhoon wave height forecasting method based on BO-LSTM neural network model |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102103708A (en) * | 2011-01-28 | 2011-06-22 | 哈尔滨工程大学 | Radial basis function neural network-based wave significant wave height inversion model establishment method |
| CN109359787A (en) * | 2018-12-06 | 2019-02-19 | 上海海事大学 | A kind of multi-modal wave forecasting system in small range sea area and its prediction technique |
| 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 |
| CN111053549A (en) * | 2019-12-23 | 2020-04-24 | 威海北洋电气集团股份有限公司 | Intelligent biological signal abnormality detection method and system |
| CN111199270A (en) * | 2019-12-30 | 2020-05-26 | 福建省海洋预报台 | Regional wave height forecasting method and terminal based on deep learning |
| CN111368086A (en) * | 2020-03-17 | 2020-07-03 | 昆明理工大学 | CNN-BilSTM + attribute model-based sentiment classification method for case-involved news viewpoint sentences |
| CN111950438A (en) * | 2020-08-10 | 2020-11-17 | 中国人民解放军国防科技大学 | Depth learning-based effective wave height inversion method for Tiangong No. two imaging altimeter |
| CN112711915A (en) * | 2021-01-08 | 2021-04-27 | 自然资源部第一海洋研究所 | Sea wave effective wave height prediction method |
| CN113051817A (en) * | 2021-03-19 | 2021-06-29 | 上海海洋大学 | Sea wave height prediction method based on deep learning and application thereof |
| CN113283588A (en) * | 2021-06-03 | 2021-08-20 | 青岛励图高科信息技术有限公司 | Near-shore single-point wave height forecasting method based on deep learning |
| CN113610945A (en) * | 2021-08-10 | 2021-11-05 | 西南石油大学 | Ground stress curve prediction method based on hybrid neural network |
-
2022
- 2022-02-28 CN CN202210189203.4A patent/CN114445634A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102103708A (en) * | 2011-01-28 | 2011-06-22 | 哈尔滨工程大学 | Radial basis function neural network-based wave significant wave height inversion model establishment method |
| CN109359787A (en) * | 2018-12-06 | 2019-02-19 | 上海海事大学 | A kind of multi-modal wave forecasting system in small range sea area and its prediction technique |
| 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 |
| CN111053549A (en) * | 2019-12-23 | 2020-04-24 | 威海北洋电气集团股份有限公司 | Intelligent biological signal abnormality detection method and system |
| CN111199270A (en) * | 2019-12-30 | 2020-05-26 | 福建省海洋预报台 | Regional wave height forecasting method and terminal based on deep learning |
| CN111368086A (en) * | 2020-03-17 | 2020-07-03 | 昆明理工大学 | CNN-BilSTM + attribute model-based sentiment classification method for case-involved news viewpoint sentences |
| CN111950438A (en) * | 2020-08-10 | 2020-11-17 | 中国人民解放军国防科技大学 | Depth learning-based effective wave height inversion method for Tiangong No. two imaging altimeter |
| CN112711915A (en) * | 2021-01-08 | 2021-04-27 | 自然资源部第一海洋研究所 | Sea wave effective wave height prediction method |
| CN113051817A (en) * | 2021-03-19 | 2021-06-29 | 上海海洋大学 | Sea wave height prediction method based on deep learning and application thereof |
| CN113283588A (en) * | 2021-06-03 | 2021-08-20 | 青岛励图高科信息技术有限公司 | Near-shore single-point wave height forecasting method based on deep learning |
| CN113610945A (en) * | 2021-08-10 | 2021-11-05 | 西南石油大学 | Ground stress curve prediction method based on hybrid neural network |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114912077A (en) * | 2022-05-27 | 2022-08-16 | 中国海洋大学 | Sea wave forecasting algorithm integrating random search and mixed decomposition error correction |
| CN114912077B (en) * | 2022-05-27 | 2023-06-30 | 中国海洋大学 | Sea wave forecasting method integrating random search and mixed decomposition error correction |
| CN115169439A (en) * | 2022-06-16 | 2022-10-11 | 中国人民解放军国防科技大学 | Method and system for predicting effective wave height based on sequence-to-sequence network |
| CN115238602A (en) * | 2022-07-01 | 2022-10-25 | 中国海洋大学 | CNN-LSTM-based prediction method for contribution rate of transient liquefaction of wave-induced seabed to resuspension |
| CN115238602B (en) * | 2022-07-01 | 2023-07-11 | 中国海洋大学 | Prediction method of contribution rate of wave-induced seabed transient liquefaction to resuspension |
| CN115149529A (en) * | 2022-09-05 | 2022-10-04 | 国家海洋技术中心 | Wave energy power generation power prediction method and system |
| CN116822253A (en) * | 2023-08-29 | 2023-09-29 | 山东省计算中心（国家超级计算济南中心） | Hybrid precision implementation method and system suitable for MANUM sea wave mode |
| CN116822253B (en) * | 2023-08-29 | 2023-12-08 | 山东省计算中心（国家超级计算济南中心） | Hybrid precision implementation method and system suitable for MANUM sea wave mode |
| CN117471575A (en) * | 2023-12-28 | 2024-01-30 | 河海大学 | Typhoon wave height forecasting method based on BO-LSTM neural network model |
| CN117471575B (en) * | 2023-12-28 | 2024-03-08 | 河海大学 | Typhoon wave height forecasting method based on BO-LSTM neural network model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114445634A (en) | Sea wave height prediction method and system based on deep learning model | |
| CN111199270B (en) | Regional wave height forecasting method and terminal based on deep learning | |
| CN111310889A (en) | Evaporation waveguide profile estimation method based on deep neural network | |
| CN103743402B (en) | A kind of underwater intelligent self adaptation Approach of Terrain Matching of topographic information based amount | |
| CN109146162A (en) | A kind of probability wind speed forecasting method based on integrated Recognition with Recurrent Neural Network | |
| CN111178304A (en) | High-resolution remote sensing image pixel level interpretation method based on full convolution neural network | |
| CN116449462B (en) | Method, system, storage medium and equipment for predicting effective wave height space-time sequence of sea wave | |
| CN113496310A (en) | Atmospheric pollutant prediction method and system based on deep learning model | |
| Wang et al. | Clouda: A ground-based cloud classification method with a convolutional neural network | |
| CN114004152B (en) | Multi-wind-field wind speed space-time prediction method based on graph convolution and recurrent neural network | |
| CN111811465B (en) | Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network | |
| CN111488974B (en) | Ocean wind energy downscaling method based on deep learning neural network | |
| CN111505706B (en) | Microseism P wave first arrival pickup method and device based on deep T-Net network | |
| CN113984198A (en) | Short wave radiation prediction method and system based on convolutional neural network | |
| CN115470957A (en) | System, computer equipment and storage medium for predicting wave height of offshore waves during typhoon based on deep learning | |
| CN110569871B (en) | Saddle point identification method based on deep convolutional neural network | |
| CN110648030A (en) | Method and device for predicting seawater temperature | |
| CN116467933A (en) | Storm surge water increasing prediction method and system based on deep learning | |
| CN115540832A (en) | Satellite altimetry submarine topography correction method and system based on VGGNet | |
| CN114742206A (en) | Rainfall intensity estimation method for comprehensive multi-space-time scale Doppler radar data | |
| CN210895535U (en) | Air temperature prediction system based on convolution cyclic neural network | |
| CN117408996B (en) | Surface defect detection method based on defect concentration and edge weight loss | |
| Rezvov et al. | Comparison of AI-based approaches for statistical downscaling of surface wind fields in the North Atlantic | |
| CN115204437A (en) | Sea surface temperature forecasting method based on memory map convolution network | |
| CN117518298A (en) | Tidal combination forecasting method with self-adaptive model structure |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220506 |