CN116933152A - Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network - Google Patents
Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network Download PDFInfo
- Publication number
- CN116933152A CN116933152A CN202310673979.8A CN202310673979A CN116933152A CN 116933152 A CN116933152 A CN 116933152A CN 202310673979 A CN202310673979 A CN 202310673979A CN 116933152 A CN116933152 A CN 116933152A
- Authority
- CN
- China
- Prior art keywords
- imf
- wave
- data
- value
- wave information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 35
- 239000002245 particle Substances 0.000 claims abstract description 41
- 238000005457 optimization Methods 0.000 claims abstract description 33
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 15
- 230000009467 reduction Effects 0.000 claims abstract description 12
- 238000004364 calculation method Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims description 22
- 238000012360 testing method Methods 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 5
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 4
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000013210 evaluation model Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 239000013589 supplement Substances 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims description 2
- 230000001186 cumulative effect Effects 0.000 claims description 2
- 238000000513 principal component analysis Methods 0.000 description 12
- 230000000694 effects Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 101100136092 Drosophila melanogaster peng gene Proteins 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C13/00—Surveying specially adapted to open water, e.g. sea, lake, river or canal
- G01C13/002—Measuring the movement of open water
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Hydrology & Water Resources (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wave information prediction method and system based on a multidimensional EMD-PSO-LSTM neural network, relates to the technical field of ocean information prediction, and aims to solve the problems that the prior art cannot comprehensively predict wave information by utilizing a multidimensional dataset, and the calculation time and the calculation amount of the predicted wave information are long, the calculation amount is large and the cost is high. According to the invention, three-dimensional information data of effective wave height, peak direction and peak period of sea waves are collected, three-dimensional sea wave information data are decomposed by using an Empirical Mode Decomposition (EMD) method to obtain intrinsic mode components and residual components under different time scales, PCA dimension reduction is carried out on a three-dimensional data sequence obtained by decomposition, key factors influencing the effective wave height, peak direction and peak period of sea wave information are screened out, parameter optimization is carried out on an LSTM model by using a particle swarm optimization algorithm (PSO), and the LSTM model is constructed; the final wave information prediction model can realize the rapid and high-precision prediction of the effective wave height, the peak direction and the peak period of the wave.
Description
Technical Field
The invention relates to the technical field of ocean information prediction, in particular to a method and a system for predicting ocean wave information based on a multidimensional EMD-PSO-LSTM neural network.
Background
The method has very important significance in developing high-end marine equipment and tamping the capability foundation of marine strong national construction by constructing a national marine comprehensive test field. The monitoring and forecasting of the sea wave environment parameters and the sea wave information are vital to the relevant activities such as intelligent unmanned system test verification, experimental debugging of the offshore sensor instrument, test operation of the test platform and the like in the test field, and the smooth proceeding of the test activities can be effectively ensured by timely and accurate wave monitoring and forecasting in the sea area of the test field, so that the loss of personnel and economy is reduced and even avoided.
Wave parameters such as wave height, wave direction, wave period and the like are important factors influencing the test tasks of the marine test field. The traditional method for measuring the sea wave parameters mainly utilizes relevant instruments such as wave buoys and the like to observe in place, has relatively high real-time performance and accuracy of observation, has a small observation range, can only realize fixed-point observation of the sea wave parameters at a throwing position, and needs to arrange a large number of wave buoys in a target sea area to realize full-coverage observation. In recent years, with the continuous development of ocean remote sensing technologies, emerging observation means such as a synthetic aperture radar, a high-frequency ground wave radar, an X-band marine radar and the like for carrying out ocean wave observation are developed, so that full coverage type ocean wave monitoring of a radar observation area is possible, wherein the ocean remote sensing technology based on the X-band radar becomes an important means for ocean wave observation. After the sea wave parameters are acquired, the forecasting of the sea wave parameters is necessary for the equipment testing activity in the sea area of the test field. With the continuous development of deep learning technology and the continuous development of neural network methods, the prediction method based on the neural network is gradually applied to sea wave analysis, and many results are achieved.
In recent years, deep learning has also been rapidly developed as an important branch of machine learning, wherein convolutional neural networks and cyclic neural networks are early developed and widely applied, and improved LSTM has higher prediction accuracy for time series based on the cyclic neural networks. In 2020, zhao Yong et al used LSTM for predictive analysis of four typical wave test data, and the results showed that LSTM prediction accuracy was significantly better than that of the support vector machine model [1] . Fan et al predict effective wave heights of ten sites under different environmental conditions for 1 hour and 6 hours and compare the effective wave heights with other prediction models, and the result shows that LSTM can realize stable prediction effect [2] .2021, lu Peng et al predict effective wave height by introducing attention mechanism based on LSTM based on variation modal decomposition, and compare with models such as support vector regression and artificial neural network to show that the model has the best prediction effect [3] .2023 Minuzzi used LSTM for effective wave height prediction of seven different locations on Brazil coast for multiple time intervals, and the results showed higher prediction accuracy than the fifth generation atmospheric re-analysis data of the middle weather forecast center in Europe [4] 。
In the current research on the sea wave information prediction technology, LSTM has great advantages in the aspect of prediction precision, but the problems of low prediction precision, large calculated amount, long calculation time, difficult convergence and the like still exist. And the known multidimensional wave information data set and the neural network cannot be fully utilized, and the essential characteristics of effective wave height, wave crest period, wave crest peak direction and time variation of the multidimensional data set are mined.
Disclosure of Invention
The invention aims to solve the technical problems that:
the existing wave information prediction method mostly adopts single-dimensional data to predict wave information, cannot achieve higher prediction accuracy, cannot comprehensively predict wave information by utilizing a multi-dimensional data set, and has the disadvantages of long calculation time, large calculation amount and high cost.
The invention adopts the technical scheme for solving the technical problems:
the invention provides a wave information prediction method based on a multidimensional EMD-PSO-LSTM neural network, which comprises the following steps:
step 1, acquiring three-dimensional information data of effective wave height, peak direction and peak period of sea waves, preprocessing the data, and dividing a data set into a training set, a verification set and a test set;
step 2, decomposing the three-dimensional sea wave information data by using an Empirical Mode Decomposition (EMD) method to obtain intrinsic mode components and residual components under different time scales, and obtaining data sequences { IMF (inertial measurement Filter) with different feature scales 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 ；
Step 3, performing PCA dimension reduction on the three-dimensional data sequence obtained in the step 2, and screening key factors influencing the effective wave height, the peak direction and the peak period of wave information in the data sequence to form a new three-dimensional data sequence;
step 4, performing parameter optimization on the LSTM model by adopting a particle swarm optimization algorithm PSO, and constructing the LSTM model;
and 5, training the constructed LSTM model by adopting a three-dimensional data sequence of a training set to obtain a sea wave information prediction model, optimizing model parameters by adopting a verification set, and adopting a test set evaluation model function to realize prediction of the effective wave height, the peak-to-peak direction and the peak period of sea waves.
Further, in the step 1, three-dimensional Fourier transform is carried out on the radar sequence image to be converted into a wave number frequency spectrum, noise is filtered through a dispersion relation band-pass filter, and the wave signal to noise ratio is calculated, so that a three-dimensional data set with combined effective wave height, wave crest peak direction and wave crest period is obtained;
the data preprocessing mainly comprises missing value processing and normalization processing, wherein the missing value processing is used for processing missing values and wild values of data in a front-back average value filling mode, namely, taking an average value of two non-empty sea wave parameter values which are closest to each other before and after the point to supplement the sea wave parameter value at the missing moment, and the normalization processing is as follows:
wherein ,as normalized result, x i X is the input data min Is the minimum value of data, x max Data maximum.
Further, in the step 2, the three-dimensional sea wave information time sequence is decomposed by adopting an EMD algorithm, and the method comprises the following steps:
(1) for an original sea wave data sequence x (t), finding all maximum value points and minimum value points in the sequence, adopting a 3-time spline interpolation function to fit an upper envelope line and a lower envelope line, and solving the average value m of the upper envelope line and the lower envelope line 1 (t) let h 1 (t)＝x(t)-m 1 (t), wherein x (t) and h 1 (t) are m-dimensional vectors;
(2) judging h 1 (t) whether the IMF component condition is satisfied, if so, h 1 (t) is an eigenvector of the original signal, if not, h 1 (t) assigning x (t) in step (1), repeating step (1) until h 1 (t) satisfying IMF component conditions; the IMF component condition is that the difference value between the number of zero crossing points and the number of extreme crossing points of the whole data segment is less than or equal to 1, and the average value of the upper envelope line and the lower envelope line is 0;
(3) let r 1 (t)＝x(t)-h 1 (t) r is to 1 (t) as raw data, iterating steps (1) - (2) until r n (t) when it is small or becomes a monotonic function, the iteration ends, yielding h 1 (t)、h 2 (t)、…、h n(t) and rn (t)；
(4) Decomposing x (t) into n intrinsic variables h by the above steps 1 (t)、h 2 (t)、…、h n (t) and one residual variable r n (t), i.e
(5) The steps are respectively carried out on the three wave parameter sequences, and the wave information data are decomposed into eigenmode components with different frequencies and residual components { IMF } 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 。
Further, in the step 3, PCA dimension reduction is performed on the sea wave data sequence, which comprises the following steps:
(1) for a group of sea wave parameter feature sequences { IMF } 1 ,IMF 2 ,…,IMF n Forming a matrix X of m rows and n columns according to columns, and normalizing the matrix to obtain a matrix Y= { IMF' 1 ,IMF' 2 ,…,IMF' n }；
(2) Calculating a correlation coefficient matrix r= (R) of Y according to ij ) m×n ；
in the formula ,xk Is IMF i The value of the kth data point in'; y is k Is IMF j The value of the kth data point in' x and y are IMF respectively i' and IMFj ' mean;
(3) calculating a characteristic value matrix of the matrix R and a characteristic vector corresponding to the characteristic value;
(4) calculating the contribution rate tau of the parameter i ：
Calculating the cumulative contribution rate eta i ：
Wherein lambda is a characteristic value;
respectively carrying out dimension reduction on the three sea wave parameter decomposition sequences through the steps (1) - (4);
(5) according to eta i And determining the number of main components to be selected, and further determining key factors influencing the effective wave height, the peak direction and the peak period of the wave information to form a new three-dimensional data sequence.
Further, in the step 4, a particle swarm optimization algorithm PSO is adopted to perform parameter optimization on the LSTM model, and the method comprises the following steps:
(1) initializing a particle swarm optimization algorithm by taking the number of neural network hidden layer units of the LSTM model, the batch processing size, the initial learning rate and the maximum iteration number as optimization objects;
(2) dividing subgroups, constructing an LSTM model by using each particle parameter, training and predicting, taking the average absolute percentage error of a predicted result to represent the fitness of particles, and calculating the fitness of each particle;
(3) determining a global optimal particle position Gbest and a local optimal particle position Prest through a population division result and a particle fitness value, and updating the speeds and positions of the common particles and the local optimal particles;
(4) and repeating the above processes to perform iterative computation until the maximum iterative times are reached, and obtaining the optimal parameter value.
Further, the sea is evaluated using mean absolute error MAE, mean absolute percent error, MAPE and root mean square error RMSEThe deviation between the predicted value and the measured value of the quantity information prediction model adopts a determination coefficient R 2 The fitting goodness of the predicted value to the measured value is measured, and the calculation formulas of all evaluation indexes are respectively as follows:
in the formula ,yi ' is a predicted value; y is i Is the actual measurement value;is the average value of the samples; n is the number of samples.
A wave information prediction system based on a multidimensional EMD-PSO-LSTM neural network is provided with a program module corresponding to the steps of any one of the technical schemes, and the steps in the wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network are executed in running.
A computer readable storage medium storing a computer program configured to implement the steps of the multidimensional EMD-PSO-LSTM neural network-based sea wave information prediction method of any of the above technical solutions when invoked by a processor.
Compared with the prior art, the invention has the beneficial effects that:
according to the wave information prediction method and system based on the multidimensional EMD-PSO-LSTM neural network, a prediction model combining EMD (empirical mode decomposition), PCA (principal component analysis) and PSO (particle swarm optimization) with the LSTM (long-short-term memory) is established, EMD decomposition is firstly carried out on a three-dimensional data set consisting of effective wave height, peak-to-peak direction and peak period, the non-stationarity of a wave information sequence is reduced, the feature diversity is increased, key factors influencing estimated wave information are screened out by using the PCA, and the operand of a subsequent neural network is reduced; and then optimizing parameters such as the number of hidden layer units of the neural network, the batch processing size, the initial learning rate, the maximum iteration number and the like by using PSO, and finally realizing high-quality prediction of effective wave height, peak-to-peak direction and peak period by using LSTM.
Compared with a model with single-dimensional input, the multi-input model provided by the invention has obvious superiority, can improve the prediction accuracy, can realize simultaneous prediction of effective wave height, wave crest period and wave crest peak direction, can predict the effective wave height, wave crest peak direction and wave crest period in real time by utilizing wave information data obtained by inversion, and has obvious improvement on the prediction model accuracy compared with the current common prediction model.
Drawings
FIG. 1 is a flow chart of a method for predicting sea wave information based on a multidimensional EMD-PSO-LSTM neural network in an embodiment of the invention;
FIG. 2 is a flowchart of the EMD algorithm in the embodiment of the invention for decomposing the time series of sea wave information;
FIG. 3 is a flow chart of key influencing factors of principal component analysis PCA extracted feature sequences in an embodiment of the present invention;
FIG. 4 is a flowchart of parameter optimization of the LSTM model by the particle swarm optimization algorithm PSO according to an embodiment of the present invention;
fig. 5 is a schematic diagram of LSTM nerve unit structure in an embodiment of the present invention.
Detailed Description
In the description of the present invention, it should be noted that the terms "first," "second," and "third" mentioned in the embodiments of the present invention are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", or a third "may explicitly or implicitly include one or more such feature.
In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
The first embodiment is as follows: referring to fig. 1 to 5, the invention provides a wave information prediction method based on a multidimensional EMD-PSO-LSTM neural network, as shown in fig. 1, comprising the following steps:
step 1, acquiring three-dimensional information data of effective wave height, peak direction and peak period of sea waves, preprocessing the data, and dividing a data set into a training set, a verification set and a test set;
step 2, decomposing the three-dimensional sea wave information data by using an Empirical Mode Decomposition (EMD) method to obtain intrinsic mode components and residual components under different time scales, and obtaining data sequences { IMF (inertial measurement Filter) with different feature scales 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 ；
Step 3, performing PCA dimension reduction on the three-dimensional data sequence obtained in the step 2, and screening key factors influencing the effective wave height, the peak direction and the peak period of wave information in the data sequence to form a new three-dimensional data sequence;
step 4, performing parameter optimization on the LSTM model by adopting a particle swarm optimization algorithm PSO, and constructing the LSTM model;
and 5, training the constructed LSTM model by adopting a three-dimensional data sequence of a training set to obtain a sea wave information prediction model, optimizing model parameters by adopting a verification set, and adopting a test set evaluation model function to realize prediction of the effective wave height, the peak-to-peak direction and the peak period of sea waves.
Step 1, performing three-dimensional Fourier transform on a radar sequence image to convert the radar sequence image into a wave number frequency spectrum, filtering noise through a dispersion relation band-pass filter, and calculating the wave signal to noise ratio to obtain a three-dimensional data set with combined effective wave height, wave crest peak direction and wave crest period;
the data preprocessing mainly comprises missing value processing and normalization processing, wherein the missing value processing is used for processing missing values and wild values of data in a front-back average value filling mode, namely, taking an average value of two non-empty sea wave parameter values which are closest to each other before and after the point to supplement the sea wave parameter value at the missing moment, and the normalization processing is as follows:
wherein ,as normalized result, x i X is the input data min Is the minimum value of data, x max Data maximum.
As shown in fig. 2, in step 2, the three-dimensional sea wave information time sequence is decomposed by adopting an EMD algorithm, which includes the following steps:
(1) for an original sea wave data sequence x (t), finding all maximum value points and minimum value points in the sequence, adopting a 3-time spline interpolation function to fit an upper envelope line and a lower envelope line, and solving the average value m of the upper envelope line and the lower envelope line 1 (t) let h 1 (t)＝x(t)-m 1 (t), wherein x (t) and h 1 (t) are m-dimensional vectors;
(2) judging h 1 (t) whether the IMF component condition is satisfied, if so, h 1 (t) is an eigenvector of the original signal, if not, h 1 (t) assigning x (t) in step (1), repeating step (1) until h 1 (t) satisfying IMF component conditions; the IMF component condition is that the difference value between the number of zero crossing points and the number of extreme crossing points of the whole data segment is less than or equal to 1, and the average value of the upper envelope line and the lower envelope line is 0;
(3) let r 1 (t)＝x(t)-h 1 (t) Will r 1 (t) as raw data, iterating steps (1) - (2) until r n (t) when it is small or becomes a monotonic function, the iteration ends, yielding h 1 (t)、h 2 (t)、…、h n(t) and rn (t)；
(4) Decomposing x (t) into n intrinsic variables h by the above steps 1 (t)、h 2 (t)、…、h n (t) and one residual variable r n (t), i.e
(5) The steps are respectively carried out on the three wave parameter sequences, and the wave information data are decomposed into eigenmode components with different frequencies and residual components { IMF } 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 。
As shown in fig. 3, the step 3 of PCA dimension reduction of the sea wave data sequence includes the following steps:
(1) for a group of sea wave parameter feature sequences { IMF } 1 ,IMF 2 ,…,IMF n Forming a matrix X of m rows and n columns according to columns, and normalizing the matrix to obtain a matrix Y= { IMF' 1 ,IMF' 2 ,…,IMF' n }；
(2) Calculating a correlation coefficient matrix r= (R) of Y according to ij ) m×n ；
in the formula ,xk Is IMF i The value of the kth data point in'; y is k Is IMF j The value of the kth data point in' x and y are IMF respectively i' and IMFj ' mean;
(3) calculating a characteristic value matrix of the matrix R and a characteristic vector corresponding to the characteristic value;
(4) calculating the contribution rate tau of the parameter i ：
Calculating the accumulated contribution rate eta:
wherein lambda is a characteristic value;
respectively carrying out dimension reduction on the three sea wave parameter decomposition sequences through the steps (1) - (4);
(5) according to eta i And determining the number of main components to be selected, and further determining key factors influencing the effective wave height, the peak direction and the peak period of the wave information to form a new three-dimensional data sequence.
The step eliminates redundancy and correlation of different time series data obtained by EMD decomposition, and reduces the operation amount.
As shown in fig. 4, in step 4, parameter optimization is performed on the LSTM model by using a particle swarm optimization algorithm PSO, including the following steps:
(1) initializing a particle swarm optimization algorithm by taking the number of neural network hidden layer units of the LSTM model, the batch processing size, the initial learning rate and the maximum iteration number as optimization objects;
(2) dividing subgroups, constructing an LSTM model by using each particle parameter, training and predicting, taking the average absolute percentage error of a predicted result to represent the fitness of particles, and calculating the fitness of each particle;
(3) determining a global optimal particle position Gbest and a local optimal particle position Prest through a population division result and a particle fitness value, and updating the speeds and positions of the common particles and the local optimal particles;
(4) and repeating the above processes to perform iterative computation until the maximum iterative times are reached, and obtaining the optimal parameter value.
Using average absolute error MAE. The average absolute percentage error, MAPE and Root Mean Square Error (RMSE) evaluate the deviation between the predicted value and the actual measured value of the mass information prediction model, and a decision coefficient R is adopted 2 The fitting goodness of the predicted value to the measured value is measured, and the calculation formulas of all evaluation indexes are respectively as follows:
in the formula ,yi ' is a predicted value; y is i Is the actual measurement value;is the average value of the samples; n is the number of samples.
As shown in fig. 5, the LSTM neural network in this embodiment includes: a sequence data input layer, an LSTM neural network layer, a Dropout layer, a full connection layer and a regression layer;
the sequence input layer is used for inputting the sea wave information parameters into the neural network in sequence according to time sequence, the LSTM neural network layer is used for mining time sequence characteristics of the sea wave information parameters and using the time sequence characteristics of the sea wave information parameters for predicting the sea wave information parameters, the Dropout layer is a random inactivation layer, the full connection layer is used for multiplying the input by a weight matrix, then a bias vector is added, each neuron in one layer is connected to each neuron in the other layer, the regression layer is used as an output layer of data and used for outputting a structure dynamic response predicted value and an error measurement which is convenient for forward propagation.
Evaluation criteria and model comparison
And acquiring a sea wave three-dimensional data set with a time span of 6 being one month, dividing the data set, taking the first 4 months of data as a training set, then taking 1 month of data as a verification set, and taking the last 1 month of data as a test set.
The invention adopts an EMD algorithm to decompose the original sequence data to obtain IMF components and residual components of effective wave height, peak-to-peak direction and peak period, so as to gradually decompose different scale fluctuation or trend of the original wave parameter sequence. The number of IMF components and the number of residual components obtained by EMD-decomposing three sea wave parameters are shown in table 1, and 46-dimensional IMF component numbers and 3 residual components can be obtained by decomposition, and a total of 49-dimensional feature sequences are taken as a new feature sequence set.
TABLE 1
In order to remove noise existing in the feature sequence data, redundancy and correlation of the feature sequence data are reduced, and principal component analysis is performed on the obtained feature sequence data. By PCA dimension reduction, the dimension and complexity of data input are reduced. As shown in table 2, the feature sequence dimension obtained by PCA dimension reduction of the 49-dimensional feature sequence was taken as a new feature sequence set, which is 27-dimensional feature sequence in total.
TABLE 2
Adopting the parameter beta as 1 ＝0.9，β 2 Adam optimizer of fused, second order momentum =0.999 trains neural networks. Parameters such as learning rate, iteration times, regularization and the like of neural network training are dynamically adjusted by adopting a cross-validation method, so that the neural network is prevented from being limitedThe problem of overfitting occurs in the dataset of (a) to improve the generalization ability of the neural network.
In the prediction of the effective wave height, the particle swarm optimization algorithm performs parameter optimization to obtain 256 LSTM hidden layer units, the initial learning rate is 0.0156, the batch processing size is 128, and the training iteration number is 487. In the prediction of peak-to-peak prediction, the particle swarm optimization algorithm performs parameter optimization to obtain the LSTM hidden layer unit number of 275, the initial learning rate of 0.0136, the batch processing size of 64 and the training iteration number of 404. In the prediction of the peak period, the particle swarm optimization algorithm performs parameter optimization to obtain the number of LSTM hidden layer units of 194, the initial learning rate of 0.0150, the batch processing size of 179 and the training iteration number of 500.
Historical data is used as input in all model training. Training and testing of the multi-input model in the sea wave parameter prediction experiment, namely using a multi-input model with historical effective wave height, wave crest period and wave crest peak to three-dimensional data. And simultaneously, the prediction precision before and after parameter optimization by the particle swarm optimization algorithm is compared.
The average absolute error MAE, the average absolute percentage error MAPE and the root mean square error RMSE are adopted to measure the deviation between the predicted value and the measured value of the sea wave parameter, and the smaller the value is, the higher the accuracy of the prediction model is. Determining the coefficient R 2 The fitting goodness of the sea wave parameter predicted value to the measured value is measured, and the value range is [0,1]The closer the value is to 1, the better the prediction effect of the model is.
The result of the effective wave height prediction is shown in table 3.
TABLE 3 Table 3
The peak period prediction results are shown in table 4.
TABLE 4 Table 4
The peak-to-peak prediction results are shown in table 5.
TABLE 5
The result shows that in the prediction of the effective wave height, the wave crest period and the wave crest peak direction, along with the improvement of the input dimension, the predicted value scattering points are more concentrated, the prediction is obviously improved on MAE, MAPE and RMSE, and the combination of the empirical mode decomposition, the particle swarm optimization algorithm parameter optimization and the multidimensional data set obviously improves the precision of the prediction model.
Although the present disclosure is disclosed above, the scope of the present disclosure is not limited thereto. Various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the disclosure, and such changes and modifications would be within the scope of the disclosure.
The literature cited in the present invention:
[1] zhao Yong Sudan Li, et al Protect on malformed wave prediction based on LSTM neural network [ J ]. University of science and technology, nature science edition, 2020,48 (7): 47-51.
[2]Fan S,Xiao N,Dong S.A novel model to predict significant wave height based on long short-term memory network[J].Ocean Engineering,2020,205:107298.
[3] Lu Peng, saint, guoliang, etc. wave height prediction based on variational modal decomposition and attention mechanisms [ J ]. Marine mapping, 2021,41 (2): 34-39.
[4]Minuzzi F C,Farina L.A deep learning approach to predict significant wave height using long short-term memory[J].Ocean Modelling,2023,181:102151.
Claims (8)
1. The wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network is characterized by comprising the following steps of:
step 1, acquiring three-dimensional information data of effective wave height, peak direction and peak period of sea waves, preprocessing the data, and dividing a data set into a training set, a verification set and a test set;
step 2, decomposing the three-dimensional sea wave information data by using an Empirical Mode Decomposition (EMD) method to obtain intrinsic mode components and residual components under different time scales, and obtaining data sequences { IMF (inertial measurement Filter) with different feature scales 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 ；
Step 3, performing PCA dimension reduction on the three-dimensional data sequence obtained in the step 2, and screening key factors influencing the effective wave height, the peak direction and the peak period of wave information in the data sequence to form a new three-dimensional data sequence;
step 4, performing parameter optimization on the LSTM model by adopting a particle swarm optimization algorithm PSO, and constructing the LSTM model;
and 5, training the constructed LSTM model by adopting a three-dimensional data sequence of a training set to obtain a sea wave information prediction model, optimizing model parameters by adopting a verification set, and adopting a test set evaluation model function to realize prediction of the effective wave height, the peak-to-peak direction and the peak period of sea waves.
2. The wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network, which is characterized in that in the step 1, radar sequence images are subjected to three-dimensional Fourier transform to be converted into wave number frequency spectrum, noise is filtered through a dispersion relation band-pass filter, and the wave signal to noise ratio is calculated, so that a three-dimensional data set with combined effective wave height, peak-to-peak direction and peak period is obtained;
the data preprocessing mainly comprises missing value processing and normalization processing, wherein the missing value processing is used for processing missing values and wild values of data in a front-back average value filling mode, namely, taking an average value of two non-empty sea wave parameter values which are closest to each other before and after the point to supplement the sea wave parameter value at the missing moment, and the normalization processing is as follows:
wherein ,as normalized result, x i X is the input data min Is the minimum value of data, x max Data maximum.
3. The wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network according to claim 1, wherein the step 2 is characterized in that an EMD algorithm is adopted to decompose a three-dimensional wave information time sequence, and the method comprises the following steps:
(1) for an original sea wave data sequence x (t), finding all maximum value points and minimum value points in the sequence, adopting a 3-time spline interpolation function to fit an upper envelope line and a lower envelope line, and solving the average value m of the upper envelope line and the lower envelope line 1 (t) let h 1 (t)＝x(t)-m 1 (t), wherein x (t) and h 1 (t) are m-dimensional vectors;
(2) judging h 1 (t) whether the IMF component condition is satisfied, if so, h 1 (t) is an eigenvector of the original signal, if not, h 1 (t) assigning x (t) in step (1), repeating step (1) until h 1 (t) satisfying IMF component conditions; the IMF component condition is that the difference value between the number of zero crossing points and the number of extreme crossing points of the whole data segment is less than or equal to 1, and the average value of the upper envelope line and the lower envelope line is 0;
(3) let r 1 (t)＝x(t)-h 1 (t) r is to 1 (t) as raw data, iterating steps (1) - (2) until r n (t) when it is small or becomes a monotonic function, the iteration ends, yielding h 1 (t)、h 2 (t)、…、h n(t) and rn (t)；
(4) Decomposing x (t) into n intrinsic variables h by the above steps 1 (t)、h 2 (t)、…、h n (t) and one residual variable r n (t), i.e
(5) The steps are respectively carried out on the three wave parameter sequences, and the wave information data are decomposed into eigenmode components with different frequencies and residual components { IMF } 1 ,IMF 2 ,…,IMF n } 1 、{IMF 1 ,IMF 2 ,…,IMF n } 2 、{IMF 1 ,IMF 2 ,…,IMF n } 3 And residual component r n1 、r n2 、r n3 。
4. The wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network as claimed in claim 1, wherein the step 3 is to perform PCA dimension reduction on the wave data sequence, and comprises the following steps:
(1) for a group of sea wave parameter feature sequences { IMF } 1 ,IMF 2 ,…,IMF n Forming a matrix X of m rows and n columns according to columns, and normalizing the matrix to obtain a matrix Y= { IMF' 1 ,IMF' 2 ,…,IMF' n }；
(2) Calculating a correlation coefficient matrix r= (R) of Y according to ij ) m×n ；
in the formula ,xk Is IMF i The value of the kth data point in'; y is k Is IMF j The value of the kth data point in', and />IMF respectively i' and IMFj ' mean;
(3) calculating a characteristic value matrix of the matrix R and a characteristic vector corresponding to the characteristic value;
(4) calculating the contribution rate tau of the parameter i ：
Calculating the cumulative contribution rate eta i ：
Wherein lambda is a characteristic value;
respectively carrying out dimension reduction on the three sea wave parameter decomposition sequences through the steps (1) - (4);
(5) according to eta i And determining the number of main components to be selected, and further determining key factors influencing the effective wave height, the peak direction and the peak period of the wave information to form a new three-dimensional data sequence.
5. The wave information prediction method based on the multidimensional EMD-PSO-LSTM neural network according to claim 1, wherein the step 4 is characterized in that a particle swarm optimization algorithm PSO is adopted to perform parameter optimization on the LSTM model, and the method comprises the following steps:
(1) initializing a particle swarm optimization algorithm by taking the number of neural network hidden layer units of the LSTM model, the batch processing size, the initial learning rate and the maximum iteration number as optimization objects;
(2) dividing subgroups, constructing an LSTM model by using each particle parameter, training and predicting, taking the average absolute percentage error of a predicted result to represent the fitness of particles, and calculating the fitness of each particle;
(3) determining a global optimal particle position Gbest and a local optimal particle position Prest through a population division result and a particle fitness value, and updating the speeds and positions of the common particles and the local optimal particles;
(4) and repeating the above processes to perform iterative computation until the maximum iterative times are reached, and obtaining the optimal parameter value.
6. The method for predicting sea wave information based on multidimensional EMD-PSO-LSTM neural network as recited in claim 1, wherein the deviation between the predicted value and the measured value of the mass information prediction model is evaluated by mean absolute error MAE, mean absolute percentage error, MAPE and root mean square error RMSE, and the decision coefficient R is adopted 2 The fitting goodness of the predicted value to the measured value is measured, and the calculation formulas of all evaluation indexes are respectively as follows:
in the formula ,yi ' is a predicted value; y is i Is the actual measurement value;is the average value of the samples; n is the number of samples.
7. A wave information prediction system based on a multidimensional EMD-PSO-LSTM neural network is characterized in that: the system having program modules corresponding to the steps of any of the preceding claims 1 to 6, the steps of the method for predicting sea wave information based on the multidimensional EMD-PSO-LSTM neural network being performed at run-time.
8. A computer-readable storage medium, characterized by: the computer readable storage medium stores a computer program configured to implement the steps of the multidimensional EMD-PSO-LSTM neural network based sea wave information prediction method of any one of claims 1 to 6 when invoked by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310673979.8A CN116933152B (en) | 2023-06-07 | 2023-06-07 | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310673979.8A CN116933152B (en) | 2023-06-07 | 2023-06-07 | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116933152A true CN116933152A (en) | 2023-10-24 |
| CN116933152B CN116933152B (en) | 2024-05-03 |
Family
ID=88374557
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310673979.8A Active CN116933152B (en) | 2023-06-07 | 2023-06-07 | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116933152B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117647808A (en) * | 2024-01-25 | 2024-03-05 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110516844A (en) * | 2019-07-25 | 2019-11-29 | 太原理工大学 | Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method |
| CN114549925A (en) * | 2022-01-18 | 2022-05-27 | 大连理工大学 | Sea wave effective wave height time sequence prediction method based on deep learning |
| CN114742307A (en) * | 2022-04-20 | 2022-07-12 | 中山大学 | Wave element prediction method and system |
| CN115096357A (en) * | 2022-06-07 | 2022-09-23 | 大连理工大学 | Indoor environment quality prediction method based on CEEMDAN-PCA-LSTM |
| CN115761596A (en) * | 2022-12-07 | 2023-03-07 | 武汉浩谱海洋探测系统有限公司 | Video monitoring wave measuring device and system based on neural network and model training method |
-
2023
- 2023-06-07 CN CN202310673979.8A patent/CN116933152B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110516844A (en) * | 2019-07-25 | 2019-11-29 | 太原理工大学 | Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method |
| CN114549925A (en) * | 2022-01-18 | 2022-05-27 | 大连理工大学 | Sea wave effective wave height time sequence prediction method based on deep learning |
| CN114742307A (en) * | 2022-04-20 | 2022-07-12 | 中山大学 | Wave element prediction method and system |
| CN115096357A (en) * | 2022-06-07 | 2022-09-23 | 大连理工大学 | Indoor environment quality prediction method based on CEEMDAN-PCA-LSTM |
| CN115761596A (en) * | 2022-12-07 | 2023-03-07 | 武汉浩谱海洋探测系统有限公司 | Video monitoring wave measuring device and system based on neural network and model training method |
Non-Patent Citations (2)
| Title |
|---|
| 詹可 等: ""基于EMD-LSTM的波高时间序列预测模型"", 《万方数据库》, 27 May 2022 (2022-05-27), pages 842 - 849 * |
| 赵明伟 等: ""基于EMD-PSO-LSTM组合模型的城市轨道交通短时客流预测"", 《铁道运输与经济》, 17 July 2022 (2022-07-17), pages 110 - 118 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117647808A (en) * | 2024-01-25 | 2024-03-05 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
| CN117647808B (en) * | 2024-01-25 | 2024-04-30 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116933152B (en) | 2024-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109947086B (en) | Mechanical fault migration diagnosis method and system based on counterstudy | |
| Fu et al. | A hybrid approach for multi-step wind speed forecasting based on two-layer decomposition, improved hybrid DE-HHO optimization and KELM | |
| Luo et al. | Design of a combined wind speed forecasting system based on decomposition-ensemble and multi-objective optimization approach | |
| CN109885907B (en) | Cloud model-based satellite attitude control system health state assessment and prediction method | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| Hu et al. | Research and application of a hybrid model based on Meta learning strategy for wind power deterministic and probabilistic forecasting | |
| CN110689179A (en) | Water bloom prediction method based on space-time sequence mixed model | |
| CN114912077B (en) | Sea wave forecasting method integrating random search and mixed decomposition error correction | |
| CN112966722A (en) | Regional landslide susceptibility prediction method based on semi-supervised random forest model | |
| CN116933152B (en) | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network | |
| CN116380445B (en) | Equipment state diagnosis method and related device based on vibration waveform | |
| CN111931983A (en) | Precipitation prediction method and system | |
| CN113051817A (en) | Sea wave height prediction method based on deep learning and application thereof | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Dong et al. | Ensemble wind speed forecasting system based on optimal model adaptive selection strategy: Case study in China | |
| CN117009814A (en) | Engine fault diagnosis method and system based on multi-source information fusion | |
| Sun et al. | Pruning Elman neural network and its application in bolt defects classification | |
| CN113361782A (en) | Photovoltaic power generation power short-term rolling prediction method based on improved MKPLS | |
| CN113010954A (en) | Bridge structure damage identification method and device and terminal equipment | |
| CN117194926A (en) | Method and system for predicting hoisting window period of land wind power base | |
| CN116127833A (en) | Wind power prediction method, system, device and medium based on VMD and LSTM fusion model | |
| CN116258269A (en) | Ultra-short-term load dynamic prediction method based on load characteristic decomposition | |
| CN116108976A (en) | Ultra-short-term wind speed prediction method | |
| CN116029419A (en) | Deep learning-based long-term new energy daily average generation power prediction method and system | |
| CN113449920A (en) | Wind power prediction method, system and computer readable medium |
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 |