CN111199270B - Regional wave height forecasting method and terminal based on deep learning - Google Patents
Regional wave height forecasting method and terminal based on deep learning Download PDFInfo
- Publication number
- CN111199270B CN111199270B CN201911389448.6A CN201911389448A CN111199270B CN 111199270 B CN111199270 B CN 111199270B CN 201911389448 A CN201911389448 A CN 201911389448A CN 111199270 B CN111199270 B CN 111199270B
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- wave height
- input
- output
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a regional wave height forecasting method and a terminal based on deep learning, which are used for acquiring data generated by a numerical mode in a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components; dividing the data into training data and test data; training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer; testing the different neural network models by using the test data, determining a neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be forecasted according to the neural network model with the best prediction effect; the LSTM layer and the convolution layer are superposed together, so that not only can the time sequence relation of data be obtained, but also the spatial characteristics can be extracted, and the space-time characteristics of wave height data can be learned, so that the accuracy of regional wave height prediction is improved.
Description
Technical Field
The invention relates to the field of ocean forecasting, in particular to a regional wave height forecasting method and a terminal based on deep learning.
Background
The wave height is an important factor for marine activities such as mariculture, maritime search and rescue, maritime transportation and ocean engineering, and has practical value for military operations such as maritime military exercises, fleet navigation and ocean ownership. Wave height forecasting has been a hotspot and difficult problem in marine science all the time. Since the wave height is a very complex physical phenomenon, the research on the physical mechanism has not made a breakthrough progress, and a theoretically strict and perfect wave height forecasting method has not been provided so far.
The current Wave height prediction methods mainly comprise two types, one type is a numerical prediction model, and the numerical prediction model is based on Wave generation and dissipation physical processes, such as SWAN, WAM, wave Watch III and the like. The other is to adopt an artificial Neural Network (NN) method, establish a model by using the prior real-time or quasi-real-time data experience of weather and wave data or buoys, and carry out wave height prediction in an artificial intelligence mode. The numerical model requires deep knowledge of the physical mechanism of wave generation, especially the main factors affecting the wave height and the operation thereof, and the main input factors are used as the input items of the numerical model to predict the wave height at the future time in a computer simulation mode, but the change of the wave cannot be accurately simulated in the long term due to the limitation of the understanding level of people on the physical mechanism of wave generation at present. And the artificial intelligence methods such as the neural network and the like have better self-adaptive learning and nonlinear mapping capabilities and are more suitable for being applied to wave height prediction. In the only some regional forecast data, the neural network with a simple structure is mostly used for regional wave height forecast, deep association between the neural network structure and the wave height forecast is not deeply researched, the existing research result of the physical mechanism of the wave height is neglected, the unique space-time characteristic is not considered, and the forecast effect is good.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the regional wave height forecasting method and the terminal based on deep learning are provided, and the regional wave height forecasting precision is improved.
In order to solve the technical problems, the invention adopts a technical scheme that:
a regional wave height forecasting method based on deep learning comprises the following steps:
s1, acquiring data generated by a numerical mode in a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
and S4, testing the different neural network models by using the test data, determining a neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect.
In order to solve the technical problem, the invention adopts another technical scheme as follows:
a deep learning based regional wave height forecasting terminal, comprising a memory, a processor and a computer program stored on the memory and operable on the processor, the processor implementing the following steps when executing the computer program:
s1, acquiring data generated by a numerical mode in a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
and S4, testing the different neural network models by using the test data, determining the neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect.
The invention has the beneficial effects that: the neural network architecture overlapped with the LSTM layer and the convolutional layer is adopted to forecast the regional wave height, because the regional wave height data has both a temporal relation and a spatial relation, the neural network needs to learn the two relations simultaneously in the training process, the LSTM is a long-term and short-term memory network which is good at processing the data with a time dependence relation, the convolutional neural network has good performance in the aspect of extracting regional characteristics, the number of parameters needing to be learned is reduced by utilizing the spatial relation, and the training performance of a deep learning network model is improved. Therefore, the LSTM layer and the convolution layer are superposed together, so that not only can the time sequence relation of data be obtained, but also the spatial characteristics can be extracted, and the space-time characteristics of the wave height data can be learned, so that the accuracy of the regional wave height prediction is improved.
Drawings
Fig. 1 is a flowchart illustrating steps of a method for forecasting regional wave height based on deep learning according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a terminal for forecasting regional wave height based on deep learning according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an LSTM layer according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the internal structure of neuronal cells in the LSTM layer according to an embodiment of the present invention;
description of reference numerals:
1. a regional wave height forecasting terminal based on deep learning; 2. a memory; 3. a processor.
Detailed Description
In order to explain technical contents, achieved objects, and effects of the present invention in detail, the following description is made with reference to the accompanying drawings in combination with the embodiments.
Referring to fig. 1, a method for forecasting regional wave height based on deep learning includes the steps of:
s1, acquiring data generated in a numerical mode within a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
and S4, testing the different neural network models by using the test data, determining the neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect.
From the above description, the beneficial effects of the present invention are: the neural network architecture overlapped with the LSTM layer and the convolutional layer is adopted to forecast the regional wave height, because the regional wave height data has both a temporal relation and a spatial relation, the neural network needs to learn the two relations simultaneously in the training process, the LSTM is a long-term and short-term memory network which is good at processing the data with a time dependence relation, the convolutional neural network has good performance in the aspect of extracting regional characteristics, the number of parameters needing to be learned is reduced by utilizing the spatial relation, and the training performance of a deep learning network model is improved. Therefore, the LSTM layer and the convolution layer are superposed together, not only can the time sequence relation of data be obtained, but also the spatial characteristic can be extracted, so that the space-time characteristic of wave height data can be learned, and the accuracy of regional wave height prediction is improved.
Further, the method between the steps S1 and S2 further includes the steps of:
preprocessing the acquired data;
the pretreatment comprises the following steps:
when the continuous preset number of data are missing, eliminating the data in the time period corresponding to the missing data;
and when the data at a single time point is missing, filling the missing data with the mean values of the data corresponding to the time points before and after the missing data.
According to the above description, the data with serious deletion is directly removed through the preprocessing of the data, and the missing data on individual time points is filled with the mean values of the data corresponding to the time points before and after the missing data, so that the accuracy of the data for training and testing is ensured, and the accuracy of regional report forecasting can be further improved.
Further, the equation of the LSTM layer in step S3 is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
in the formula (f) t 、i t 、o t Respectively representing a forgetting gate, an input gate and an output gate, W f 、W i 、W o Weight parameters of the forgetting gate, the input gate and the output gate, respectively, b f 、b i 、b o Respectively, offset terms of the forgetting gate, the input gate and the output gate, b f Is to calculate
Bias term of time, C t-1 、C t Respectively the last timeNeuronal cell state at the moment of time, h t-1 、h t The outputs of the last time and the current time are adopted as sigma, a tan h function is adopted as x t Representing the input data.
Further, the training the neural network by using the training data in the step S3 to obtain different neural network models includes:
dividing the training data into first input data and first output data, wherein the first input data comprise air pressure, air temperature and wind field components, and the first output data comprise actual regional wave heights corresponding to the first input data;
inputting the first input data into a neural network to obtain first prediction data predicted by the neural network, wherein the first prediction data comprises predicted regional wave height;
calculating a loss function f from the first predicted value and the first output data lost ：
Wherein y is the first output data,
n is the batch number of the input data;
according to the loss function f lost And adjusting the parameters of the neural network to obtain different neural network models.
According to the above description, the training data is divided into the input data and the corresponding output data, the prediction data obtained by passing the input data through the neural network is compared with the output data, the loss function is calculated, and the parameters of the neural network are adjusted through the loss function to obtain different neural network models, so that the reliability of the trained neural network model can be ensured.
Further, the step S4 of testing the different neural network models by using the test data, and determining the neural network model with the best prediction effect includes:
dividing the test data into second input data and second output data, wherein the second input data comprise air pressure, air temperature and wind field components, and the second output data comprise actual regional wave height corresponding to the second input data;
inputting the second input data into the different neural network models respectively for testing to obtain corresponding second prediction data, wherein the second prediction data comprise predicted regional wave heights;
and respectively comparing second prediction data obtained by predicting different neural network models with the second output data, and determining the neural network model corresponding to the second prediction data with the minimum difference with the second output data as the neural network model with the best prediction effect.
According to the above description, after different neural network models are trained through training data, the neural network model with the best prediction effect is selected through testing data, and through the combination of training and testing, the adaptability of the finally determined neural network model can be ensured, and the accuracy of regional wave height prediction is further improved.
Referring to fig. 2, a terminal for forecasting regional wave height based on deep learning includes a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements the following steps when executing the computer program:
s1, acquiring data generated in a numerical mode within a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
and S4, testing the different neural network models by using the test data, determining the neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect.
From the above description, the beneficial effects of the present invention are: the neural network architecture overlapped with the LSTM layer and the convolutional layer is adopted to forecast the regional wave height, because the regional wave height data has both a temporal relation and a spatial relation, the neural network needs to learn the two relations simultaneously in the training process, the LSTM is a long-term and short-term memory network which is good at processing the data with a time dependence relation, the convolutional neural network has good performance in the aspect of extracting regional characteristics, the number of parameters needing to be learned is reduced by utilizing the spatial relation, and the training performance of a deep learning network model is improved. Therefore, the LSTM layer and the convolution layer are superposed together, so that not only can the time sequence relation of data be obtained, but also the spatial characteristics can be extracted, and the space-time characteristics of the wave height data can be learned, so that the accuracy of the regional wave height prediction is improved.
Further, the method between the steps S1 and S2 further includes the steps of:
preprocessing the acquired data;
the pretreatment comprises the following steps:
when the continuous preset number of data are missing, removing the data in the time period corresponding to the missing data;
and when the data at a single time point is missing, filling the missing data with the mean values of the data corresponding to the time points before and after the missing data.
According to the above description, the data with serious deletion is directly removed through the preprocessing of the data, and the missing data on individual time points is filled with the mean values of the data corresponding to the time points before and after the missing data, so that the accuracy of the data for training and testing is ensured, and the accuracy of regional report forecasting can be further improved.
Further, the equation of the LSTM layer in step S3 is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
in the formula (f) t 、i t 、o t Respectively representing a forgetting gate, an input gate and an output gate, W f 、W i 、W o Weight parameters of the forgetting gate, the input gate and the output gate, respectively, b f 、b i 、b o Respectively, offset terms of the forgetting gate, the input gate and the output gate, b f Is a calculation of
Bias term of time, C t-1 、C t The states of the neuron cells at the previous moment and the current moment, h t-1 、h t The outputs of the previous time and the current time are respectively, sigma adopts a tanh function, x t Representing the input data.
Further, the training the neural network by using the training data in the step S3 to obtain different neural network models includes:
dividing the training data into first input data and first output data, wherein the first input data comprise air pressure, air temperature and wind field components, and the first output data comprise actual regional wave heights corresponding to the first input data;
inputting the first input data into a neural network to obtain first prediction data predicted by the neural network, wherein the first prediction data comprises predicted regional wave height;
calculating a loss function f from the first predicted value and the first output data lost ：
Wherein y is the first output data,
n is the batch number of the input data;
according to the loss function f lost And adjusting the parameters of the neural network to obtain different neural network models.
According to the above description, the training data is divided into the input data and the corresponding output data, the prediction data obtained by passing the input data through the neural network is compared with the output data, the loss function is calculated, and the parameters of the neural network are adjusted through the loss function to obtain different neural network models, so that the reliability of the trained neural network model can be ensured.
Further, the step S4 of testing the different neural network models by using the test data, and determining the neural network model with the best prediction effect includes:
dividing the test data into second input data and second output data, wherein the second input data comprise air pressure, air temperature and wind field components, and the second output data comprise actual regional wave height corresponding to the second input data;
inputting the second input data into the different neural network models respectively for testing to obtain corresponding second prediction data, wherein the second prediction data comprise predicted regional wave heights;
and respectively comparing second prediction data obtained by predicting different neural network models with the second output data, and determining the neural network model corresponding to the second prediction data with the minimum difference with the second output data as the neural network model with the best prediction effect.
According to the above description, after different neural network models are trained through training data, the neural network model with the best prediction effect is selected through testing data, and through the combination of training and testing, the adaptability of the finally determined neural network model can be ensured, and the accuracy of regional wave height prediction is further improved.
Example one
Referring to fig. 1, a method for forecasting regional wave height based on deep learning includes the steps of:
s1, acquiring data generated in a numerical mode within a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
wherein, the data is derived from the numbers generated by the numerical mode, and comprises the regional wave height, the air pressure, the air temperature, and the latitude component u and longitude component v of the wind field, and the data are the data of all grid points on the region, but not the single-point data on a certain point;
after the data are obtained, carrying out data standardization preprocessing on the data, including noise data filtering, normalization and other processing;
wherein, the filtration of noise data is including the rejection and the filling to missing data, and is specific:
when the continuous preset number of data are missing, eliminating the data in the time period corresponding to the missing data;
when data on a single time point is missing, filling the missing data with the mean values of the data corresponding to the time points before and after the missing data;
s2, dividing the data into training data and testing data;
specifically, the preprocessed data are divided according to a certain proportion, one part is used as training data, and the other part is used as test data, wherein the proportion can be set according to the actual situation, and optionally 8:2, 80% of the training data and 20% of the test data are selected;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
the equation for the LSTM layer in step S3 is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
in the formula, f t 、i t 、o t Respectively representing a forgetting gate, an input gate and an output gate, W f 、W i 、W o Weight parameters of the forgetting gate, the input gate and the output gate, respectively, b f 、b i 、b o Respectively, offset terms of the forgetting gate, the input gate and the output gate, b f Is a calculation of
A bias term of time; c t-1 、C t The states of the neuron cells at the previous moment and the current moment, h t-1 、h t The outputs of the previous time and the current time are respectively, sigma adopts a tanh function, x t Represents input data, [ h ] t-1 ,x t ]Is represented by h t-1 And x t An array of compositions;
wherein, the activating function of the LSTM layer uses a tanh function, and the activating function of the convolutional layer uses a relu function;
specifically, the schematic structure of the LSTM network is shown in fig. 3, and fig. 4 is a schematic internal structure of the neuron cell in fig. 3;
the formula for the convolutional layer is as follows:
convolutional layer calculation formula: assuming the l-th layer is the convolutional layer and the l-1 th layer is the input layer, with x l-1 Indicating that the weight matrix corresponding to the jth convolution kernel in the ith layer is used
Represents, convolution kernel->
The corresponding bias is used>
Indicates that the jth feature map calculated by the convolutional layer is used>
Expressed, the convolution calculation formula is as follows:
wherein f denotes an activation function, M j Representing the number of convolution kernels; x in the equation for the LSTM layer t Namely the output of the convolution layer;
specifically, the training of the neural network by using the training data to obtain different neural network models includes:
dividing the training data into first input data and first output data, wherein the first input data comprise air pressure, air temperature and wind field components, and the first output data comprise actual regional wave heights corresponding to the first input data;
inputting the first input data into a neural network to obtain first prediction data predicted by the neural network, wherein the first prediction data comprises predicted regional wave height;
specifically, the air pressure, the air temperature and the wind field components u and v are taken as multichannel data to serve as input data of a neural network, and the output data is wave height data of all grid points corresponding to a prediction region;
calculating a loss function f from the first predicted value and the first output data lost ：
Wherein y is the first output data,
n is the batch number of the input data;
according to the loss function f lost Adjusting parameters of the neural network to obtain different neural network models;
s4, testing the different neural network models by using the test data, determining a neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect;
in step S4, the different neural network models are tested by using the test data, and determining the neural network model with the best prediction effect includes:
dividing the test data into second input data and second output data, wherein the second input data comprise air pressure, air temperature and wind field components, and the second output data comprise actual regional wave height corresponding to the second input data;
inputting the second input data into the different neural network models respectively for testing to obtain corresponding second prediction data, wherein the second prediction data comprise predicted regional wave heights;
respectively comparing second prediction data obtained by predicting different neural network models with the second output data, and determining the neural network model corresponding to the second prediction data with the minimum difference with the second output data as the neural network model with the best prediction effect;
preferably, because the deep learning network model has large calculation amount and numerous connection layer parameters, the calculation of one deep learning network is executed in parallel in a multi-GPU scheduling mode, the model training is accelerated in a data parallel and algorithm model parallel pipeline mode, and the training result is quickly obtained.
Example two
Referring to fig. 2, a terminal 1 for forecasting regional wave height based on deep learning includes a memory 2, a processor 3, and a computer program stored on the memory 2 and executable on the processor 3, where the processor 3 implements the steps in the first embodiment when executing the computer program.
In summary, according to the regional wave height forecasting method and terminal based on deep learning provided by the invention, different neural network models are obtained through training of training data, then different neural network models are tested through test data, an optimal neural network model is determined, and the regional wave height is forecasted by adopting a neural network architecture on which an LSTM layer and a convolutional layer are superimposed. Therefore, the LSTM layer and the convolution layer are superposed together, not only can the time sequence relation of data be obtained, but also the spatial characteristic can be extracted, so that the space-time characteristic of wave height data can be learned, and the accuracy of regional wave height prediction is improved.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent changes made by using the contents of the present specification and the drawings, or applied directly or indirectly to the related technical fields, are included in the scope of the present invention.
Claims (4)
1. A regional wave height forecasting method based on deep learning is characterized by comprising the following steps:
s1, acquiring data generated by a numerical mode in a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
s4, testing the different neural network models by using the test data, determining a neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect;
the equation of the LSTM layer in step S3 is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
in the formula (f) t 、i t 、o t Respectively representing a forgetting gate, an input gate and an output gate, W f 、W i 、W o Weight parameters of the forgetting gate, the input gate and the output gate, respectively, b f 、b i 、b o Respectively, offset terms of the forgetting gate, the input gate and the output gate, b f Is to calculate
Bias term of time, C t-1 、C t The states of the neuron cells at the previous and current times, h t-1 、h t The outputs of the last time and the current time are adopted as sigma, a tan h function is adopted as x t Representing input data;
in the step S3, training the neural network by using the training data to obtain different neural network models includes:
dividing the training data into first input data and first output data, wherein the first input data comprise air pressure, air temperature and wind field components, and the first output data comprise actual regional wave height corresponding to the first input data;
inputting the first input data into a neural network to obtain first prediction data predicted by the neural network, wherein the first prediction data comprises predicted regional wave height;
calculating a loss function f from the first prediction data and the first output data lost ：
Wherein y is the first output data,
n is the batch number of the input data;
according to the loss function f lost Adjusting parameters of the neural network to obtain different neural network models;
in step S4, the different neural network models are tested by using the test data, and determining the neural network model with the best prediction effect includes:
dividing the test data into second input data and second output data, wherein the second input data comprise air pressure, air temperature and wind field components, and the second output data comprise actual regional wave height corresponding to the second input data;
inputting the second input data into the different neural network models respectively for testing to obtain corresponding second prediction data, wherein the second prediction data comprise predicted regional wave heights;
and respectively comparing second prediction data obtained by predicting different neural network models with the second output data, and determining the neural network model corresponding to the second prediction data with the minimum difference with the second output data as the neural network model with the best prediction effect.
2. The method for forecasting the regional wave height based on deep learning of claim 1, wherein the steps between S1 and S2 further comprise the steps of:
preprocessing the acquired data;
the pretreatment comprises the following steps:
when the continuous preset number of data are missing, removing the data in the time period corresponding to the missing data;
when data at a single time point is missing, filling the missing data with the mean values of data corresponding to time points before and after the missing data.
3. A deep learning-based regional wave height forecasting terminal, comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the following steps:
s1, acquiring data generated in a numerical mode within a preset time period, wherein the data comprises regional wave height, air pressure, air temperature and wind field components;
s2, dividing the data into training data and testing data;
s3, training a neural network by using the training data to obtain different neural network models, wherein the neural network comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises an LSTM layer and a convolutional layer;
s4, testing the different neural network models by using the test data, determining a neural grid model with the best prediction effect, and forecasting the regional wave height within the time to be predicted according to the neural network model with the best prediction effect;
the equation of the LSTM layer in step S3 is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
in the formula (f) t 、i t 、o t Respectively representing a forgetting gate, an input gate and an output gate, W f 、W i 、W o Weight parameters of the forgetting gate, the input gate and the output gate, respectively, b f 、b i 、b o Respectively, offset terms of the forgetting gate, the input gate and the output gate, b f Is a calculation of
Bias term of time, C t-1 、C t The states of the neuron cells at the previous moment and the current moment, h t-1 、h t The outputs of the previous moment and the current moment are respectively adopted by sigmatan h function, x t Representing input data;
in the step S3, the training of the neural network by using the training data to obtain different neural network models includes:
dividing the training data into first input data and first output data, wherein the first input data comprise air pressure, air temperature and wind field components, and the first output data comprise actual regional wave heights corresponding to the first input data;
inputting the first input data into a neural network to obtain first prediction data predicted by the neural network, wherein the first prediction data comprises predicted regional wave heights;
calculating a loss function f from the first prediction data and the first output data lost ：
Wherein y is the first output data,
n is the batch number of input data for the first prediction data;
according to the loss function f lost Adjusting parameters of the neural network to obtain different neural network models;
in step S4, the different neural network models are tested by using the test data, and determining the neural network model with the best prediction effect includes:
dividing the test data into second input data and second output data, wherein the second input data comprise air pressure, air temperature and wind field components, and the second output data comprise actual regional wave height corresponding to the second input data;
inputting the second input data into the different neural network models respectively for testing to obtain corresponding second prediction data, wherein the second prediction data comprise predicted regional wave heights;
and respectively comparing second prediction data obtained by predicting different neural network models with the second output data, and determining the neural network model corresponding to the second prediction data with the minimum difference with the second output data as the neural network model with the best prediction effect.
4. The deep learning-based regional wave height forecasting terminal according to claim 3, further comprising, between the steps S1 and S2, the steps of:
preprocessing the acquired data;
the pretreatment comprises the following steps:
when the continuous preset number of data are missing, eliminating the data in the time period corresponding to the missing data;
and when the data at a single time point is missing, filling the missing data with the mean values of the data corresponding to the time points before and after the missing data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911389448.6A CN111199270B (en) | 2019-12-30 | 2019-12-30 | Regional wave height forecasting method and terminal based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911389448.6A CN111199270B (en) | 2019-12-30 | 2019-12-30 | Regional wave height forecasting method and terminal based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111199270A CN111199270A (en) | 2020-05-26 |
| CN111199270B true CN111199270B (en) | 2023-04-18 |
Family
ID=70746436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911389448.6A Active CN111199270B (en) | 2019-12-30 | 2019-12-30 | Regional wave height forecasting method and terminal based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111199270B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111811465B (en) * | 2020-07-01 | 2022-03-18 | 南方海洋科学与工程广东省实验室(湛江) | Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network |
| CN112307676B (en) * | 2020-11-04 | 2022-10-14 | 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)) | Wave height numerical prediction model result correction method |
| CN112558185A (en) * | 2020-11-19 | 2021-03-26 | 中国石油大学(华东) | Bidirectional GRU typhoon track intelligent prediction and forecast system based on attention mechanism, computer equipment and storage medium |
| CN112785034B (en) * | 2020-12-22 | 2022-08-02 | 河海大学 | Typhoon path forecasting method, system, medium and terminal based on fusion neural network |
| CN112711915B (en) * | 2021-01-08 | 2022-02-25 | 自然资源部第一海洋研究所 | Sea wave effective wave height prediction method |
| CN113283588B (en) * | 2021-06-03 | 2022-04-19 | 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 |
| CN113344275B (en) * | 2021-06-15 | 2022-10-14 | 上海交通大学 | Floating platform wave climbing online forecasting method based on LSTM model |
| CN114445634A (en) * | 2022-02-28 | 2022-05-06 | 南京信息工程大学 | Sea wave height prediction method and system based on deep learning model |
| CN114519311B (en) * | 2022-04-21 | 2022-07-22 | 中国海洋大学 | Prediction method, system, storage medium and application of total harbor basin wave effective wave height |
| CN117471575B (en) * | 2023-12-28 | 2024-03-08 | 河海大学 | Typhoon wave height forecasting method based on BO-LSTM neural network model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104050379A (en) * | 2014-06-25 | 2014-09-17 | 东南大学 | Sea wave height prediction method based on ARMA model |
| CN108334987A (en) * | 2018-02-08 | 2018-07-27 | 山东汇贸电子口岸有限公司 | A kind of sea wave height prediction technique based on wavelet decomposition-neural network |
| CN109460874A (en) * | 2018-11-19 | 2019-03-12 | 厦门大学 | A kind of ariyoshi wave height prediction technique based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6384065B2 (en) * | 2014-03-04 | 2018-09-05 | 日本電気株式会社 | Information processing apparatus, learning method, and program |
-
2019
- 2019-12-30 CN CN201911389448.6A patent/CN111199270B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104050379A (en) * | 2014-06-25 | 2014-09-17 | 东南大学 | Sea wave height prediction method based on ARMA model |
| CN108334987A (en) * | 2018-02-08 | 2018-07-27 | 山东汇贸电子口岸有限公司 | A kind of sea wave height prediction technique based on wavelet decomposition-neural network |
| CN109460874A (en) * | 2018-11-19 | 2019-03-12 | 厦门大学 | A kind of ariyoshi wave height prediction technique based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111199270A (en) | 2020-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111199270B (en) | Regional wave height forecasting method and terminal based on deep learning | |
| CN109978228B (en) | PM2.5 concentration prediction method, device and medium | |
| WO2018001338A1 (en) | Grey system-based pelagic squid resource richness forecasting method | |
| CN106022954B (en) | Multiple BP neural network load prediction method based on grey correlation degree | |
| CN110689179A (en) | Water bloom prediction method based on space-time sequence mixed model | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
| Braakmann-Folgmann et al. | Sea level anomaly prediction using recurrent neural networks | |
| CN109920248B (en) | Bus arrival time prediction method based on GRU neural network | |
| CN112712214B (en) | Method, system, device and storage medium for predicting track of maritime search and rescue object | |
| CN114445634A (en) | Sea wave height prediction method and system based on deep learning model | |
| CN112001110A (en) | Structural damage identification monitoring method based on vibration signal space real-time recursive graph convolutional neural network | |
| CN112862178A (en) | Ocean station water level space-time prediction method and device based on deep learning | |
| CN111428419A (en) | Suspended sediment concentration prediction method and device, computer equipment and storage medium | |
| Li et al. | Leveraging LSTM for rapid intensifications prediction of tropical cyclones | |
| CN115758876A (en) | Method, system and computer equipment for forecasting accuracy of wind speed and wind direction | |
| CN112036598A (en) | Charging pile use information prediction method based on multi-information coupling | |
| CN114418930A (en) | Underwater whale target detection method based on light YOLOv4 | |
| CN116976227B (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
| CN111811465B (en) | Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network | |
| CN116933152B (en) | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network | |
| CN112949944A (en) | Underground water level intelligent prediction method and system based on space-time characteristics | |
| CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
| CN115759291B (en) | Spatial nonlinear regression method and system based on ensemble learning | |
| CN114021445B (en) | Ocean vortex mixing non-locality prediction method based on random forest model |
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 |