Effective wave height prediction method and device and electronic equipment

Abstract

The invention provides a significant wave height prediction method and device and electronic equipment, and relates to the technical field of sea wave data processing. The method comprises the following steps: acquiring an effective wave height data set for wave height prediction; determining model input data based on the correlation; carrying out two-layer decomposition on the effective wave height; wherein the primary decomposition comprises empirical mode decomposition based on a self-adaptive noise complete set and K-means clustering of mode components; the secondary decomposition comprises the step of optimizing variational mode decomposition of parameters based on a crown porcupine optimization algorithm; inputting model input data into a bidirectional long-short-term memory network based on an attention mechanism for training; the method comprises the steps of firstly screening data with relatively high correlation with the significant wave height as input data to reduce the calculation amount, then performing two-layer decomposition on the significant wave height to reduce the data complexity, and finally fully extracting data features through a bidirectional long-short-term memory network based on an attention mechanism to comprehensively predict the data so as to improve the prediction accuracy of the significant wave height. And the accuracy of predicting the significant wave height is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sea wave data processing, and in particular to an effective wave height prediction method, device and electronic equipment. BACKGROUND

[0002] Compared with wind energy and solar energy, wave energy has higher energy density as a clean and renewable energy. Effective wave height is a key parameter in the process of wave energy generation. Waves are affected by wind, ocean current, topography and many other factors, and have the characteristics of high randomness and sudden changes, which brings great difficulty to effective wave height prediction.

[0003] In the existing effective wave height prediction method, the numerical model method has large calculation amount and high calculation complexity; a single prediction model is difficult to capture the long-term relationship in the data, and has poor generalization ability when processing complex data; the traditional data decomposition algorithm is difficult to solve the modal aliasing problem, and the prediction cannot fully extract the data features, which limits the accuracy of effective wave height prediction. SUMMARY

[0004] The purpose of the present application is to provide an effective wave height prediction method, device and electronic equipment. First, the data with high correlation with effective wave height is selected as input data to reduce the calculation amount. Then, the effective wave height is decomposed into two layers to weaken the nonlinearity of the effective wave height and reduce the data complexity. Finally, the data features are fully extracted by the bidirectional long short-term memory network based on the attention mechanism, the data is comprehensively predicted, and the accuracy of predicting the effective wave height is improved.

[0005] In a first aspect, the present application provides an effective wave height prediction method, comprising: Obtaining historical feature data for wave height prediction, preprocessing the historical feature data to obtain an effective wave height data set; wherein the effective wave height data set includes: effective wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature; The data in the effective wave height data set that meets the preset condition with the effective wave height is determined as model input data; The effective wave height after one-time decomposition and two-time decomposition is determined as input data; wherein the one-time decomposition includes: adaptive noise complete set empirical mode decomposition and K-means clustering of modal components; the two-time decomposition includes: variational modal decomposition; the decomposition number and the penalty factor of the variational modal decomposition are optimized based on the crow hog optimization algorithm; The model input data is input into the bidirectional long short-term memory network based on the attention mechanism until the preset condition is reached, and an effective wave height prediction model is obtained; The target feature data is input into the effective wave height prediction model, and an effective wave height prediction value is output.

[0006] In some preferred embodiments of the present application, the preprocessing comprises: outlier rejection, missing data imputation and normalization.

[0007] In some preferred embodiments of the present application, the correlation is determined based on the following formula: ; wherein p is the correlation; R(x i ) is the rank of the variable x i ; R is the average rank of all x values; R(y i ) is the rank of the variable y i ; R is the average rank of all y values; n is the number of data pairs; x i is the i-th sample data in the x variable; x is one kind of variable data in the correlation calculation; y i is the i-th sample data in the y variable; y is another kind of variable data in the correlation calculation.

[0008] In some preferred embodiments of the present application, the step of adaptive noise complete ensemble empirical mode decomposition comprises: Gaussian white noise is added to all original to-be-decomposed sequences of significant wave height to obtain to-be-decomposed data; ensemble average is performed on the to-be-decomposed data after empirical mode decomposition to obtain a first modal component and a residual term; Gaussian white noise is continuously added to all to-be-decomposed sequences of the residual term obtained after each empirical mode decomposition, and then empirical mode decomposition and ensemble average are performed until a preset condition is reached, to obtain a plurality of first modal components.

[0009] In some preferred embodiments of the present application, the step of K-means clustering of the modal components comprises: fuzzy entropy of the plurality of first modal components is calculated; K-means clustering is performed on the plurality of fuzzy entropies to obtain high-frequency partial components, medium-frequency partial components and low-frequency partial components; the high-frequency partial components, the medium-frequency partial components and the low-frequency partial components are respectively aggregated.

[0010] In some preferred embodiments of the present application, the step of secondary decomposition comprises: variational mode decomposition is performed on the aggregated high-frequency partial components to obtain a plurality of second modal components.

[0011] In some preferred embodiments of the present application, the method further comprises determining the decomposition number and the penalty factor of the variational mode decomposition by the following steps: The search range of the decomposition number, the search range of the penalty factor, the population size of the GHO algorithm and the maximum iteration number of the GHO algorithm of the initialization of the variational mode decomposition are initialized, and a convergence threshold is defined. The decomposition number and the penalty factor are iteratively optimized based on the GHO algorithm until the maximum iteration number is reached, and the final decomposition number and the final penalty factor are obtained.

[0012] In some preferred embodiments of the present application, in the attention mechanism-based bidirectional long short-term memory network, the attention mechanism layer is arranged between the bidirectional long short-term memory network and the fully connected layer.

[0013] In a second aspect, the present application provides an effective wave height prediction device, comprising: A data processing module is configured to obtain historical feature data for wave height prediction, and preprocess the historical feature data to obtain an effective wave height dataset; wherein the effective wave height dataset comprises effective wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature. A first input data determination module is configured to determine data in the effective wave height dataset that meets a preset condition in terms of correlation with effective wave height as model input data. A second input data determination module is configured to determine the effective wave height after first decomposition and second decomposition as input data; wherein the first decomposition comprises adaptive noise complete ensemble empirical mode decomposition and K-means clustering of mode components; the second decomposition comprises variational mode decomposition; and the decomposition number and the penalty factor of the variational mode decomposition are optimized based on the GHO algorithm. A prediction model processing module is configured to input the model input data into the attention mechanism-based bidirectional long short-term memory network until a preset condition is met, and obtain an effective wave height prediction model. A data prediction module is configured to input target feature data into the effective wave height prediction model, and output an effective wave height prediction value.

[0014] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the memory stores computer executable instructions executable by the processor, and the processor executes the computer executable instructions to implement the effective wave height prediction method provided in the first aspect.

[0015] The present application has the following beneficial effects: The present invention provides a significant wave height prediction method, device and electronic equipment, the method comprising: obtaining historical characteristic data for wave height prediction, pre-processing the historical characteristic data to obtain a significant wave height data set; wherein the significant wave height data set comprises: significant wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature; calculating the correlation between any two data in the significant wave height data set; determining the data in the significant wave height data set whose correlation with the significant wave height meets a preset condition as model input data; determining the significant wave height after primary decomposition and secondary decomposition as input data; wherein the primary decomposition comprises: based on the adaptive noise complete set empirical mode Decomposition and K-means clustering of modal components; secondary decomposition includes: variational mode decomposition; the number of decompositions and penalty factors of variational mode decomposition are optimized based on the crown porcupine optimization algorithm; the model input data is input into the bidirectional long short-term memory network based on the attention mechanism until the preset conditions are met to obtain the effective wave height prediction model; the target feature data is input into the effective wave height prediction model to output the effective wave height prediction value; first, data with high correlation with the effective wave height is screened as input data to reduce the amount of calculation, and then the effective wave height is decomposed into two layers to weaken the nonlinearity of the effective wave height and reduce the data complexity. Finally, the data features are fully extracted through the bidirectional long short-term memory network based on the attention mechanism, and the data is comprehensively predicted, thereby improving the accuracy of the predicted effective wave height. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A flow chart of a significant wave height prediction method provided by an embodiment of the present invention; Figure 2 A schematic diagram of correlation provided by an embodiment of the present invention; Figure 3 A schematic diagram of a primary decomposition result of wave height data provided by an embodiment of the present invention; Figure 4 A schematic diagram of a K-means clustering result of a first modal component provided by an embodiment of the present invention; Figure 5 A schematic diagram of fitness iteration curves of multiple optimization algorithms provided in an embodiment of the present invention; Figure 6A high-frequency partial component decomposition result schematic diagram provided for an embodiment of the present application; Figure 7 A significant wave height prediction result schematic diagram provided for an embodiment of the present application; Figure 8 A significant wave height prediction device structure schematic diagram provided for an embodiment of the present application; Figure 9 A structure schematic diagram of an electronic device provided for an embodiment of the present application.

[0018] Icon: 310 - data processing module; 320 - first input data determination module; 330 - second input data determination module; 340 - prediction model processing module; 350 - data prediction module; 400 - memory; 401 - processor; 402 - bus; 403 - communication interface. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0021] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0022] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship when the product of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" and the like are only used for differentiation in description, and cannot be understood as indicating or implying relative importance.

[0023] In addition, the terms "horizontal", "vertical", "suspended" and the like do not mean that the components must be absolutely horizontal or suspended, but can be slightly inclined. For example, "horizontal" only means that it is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0024] In the description of the present application, it should be further pointed out that, unless otherwise specified and limited, the terms "arrangement", "installation", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0025] Compared with wind energy and solar energy, wave energy has higher energy density as a clean and renewable energy. Significant wave height is a key parameter in the process of wave energy generation, and accurate prediction of significant wave height can provide reliable recommendations for power generation. Waves are affected by wind, ocean currents, topography and many other factors, and the characteristics of high randomness and sudden changes make it difficult to predict significant wave height. Therefore, with the development of wave energy as a renewable energy, it is particularly important to establish a new significant wave height prediction algorithm.

[0026] The existing significant wave height prediction methods are mainly divided into two categories: numerical model and statistical model. The numerical model based on the calculation model simulates the sea waves by solving the calculation equation based on the theoretical model. Common numerical models include: Wave Prediction Model (WAM, numerical wave model), Simulating Wave Nearshore (SWAN, nearshore wave model) and WAVEWATCH III (third generation wave monitoring model) and the like. Although the numerical wave prediction model can accurately simulate the height, period, direction and other parameters of the sea waves. However, it requires a large amount of meteorological data and ocean observation data, and has the problems of large amount of calculation and high calculation complexity. At the same time, its accuracy depends on the quality of the input, and if these data are inaccurate or incomplete, the prediction accuracy will be affected.

[0027] With the continuous improvement of computer computing performance and the development of ocean observation technology such as buoys and observation ships, machine learning (ML) and deep learning (DL) methods based on statistical theory provide a new feasible scheme for accurate prediction of sea wave significant wave height. However, machine learning models such as artificial neural network (ANN) and support vector machine (SVM) cannot capture the long-term dependence relationship in time series data, and have certain defects and poor generalization ability when dealing with complex data.

[0028] In recent years, many deep learning models have been used for significant wave height forecasting, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and gated recurrent unit networks (GRUs). LSTMs, through their unique gating mechanism, are able to effectively capture nonlinear characteristics and long-term dependencies in time series. While AI-based nonlinear models have advantages in processing time series, their predictive performance can be limited when dealing with highly nonstationary series due to their difficulty adapting to the time-varying nature of the data.

[0029] The generation mechanism of ocean waves is complex, and wave time series always show strong volatility, nonlinearity, and non-stationarity. To address this problem, using decomposition algorithms to smooth nonlinear data to improve prediction accuracy is an effective method. However, EMD-like algorithms are prone to modal aliasing during re-decomposition, and white noise may remain in the decomposition results. In contrast, variational mode decomposition (VMD) provides a different solution. VMD abandons the idea of ​​noise assistance and effectively overcomes modal aliasing and avoids the problem of residual noise through preset frequency band constraints and variational optimization framework. However, manually setting parameters cannot achieve the optimal decomposition effect, thus affecting prediction accuracy.

[0030] Traditional data decomposition algorithms (such as EMD and EEMD) have difficulty solving the problem of modal aliasing, which limits the accuracy of significant wave height prediction. Significant wave height data has strong volatility and nonlinearity, and the high-frequency components obtained by a single decomposition still have high complexity, which affects the prediction effect. Although VMD decomposition is better, the number of decompositions needs to be manually set. and penalty factor ,This method is subjective and limited and cannot achieve the best decomposition effect.,In addition, the LSTM unidirectional network ignores the impact of backward information on the ,prediction results, resulting in insufficient feature extraction of the significant wave height ,sequence.

[0031] In order to solve the above problems and improve the accuracy of significant wave height prediction, a significant wave height prediction method based on two-layer decomposition and BiLSTM-Attention is proposed. The Spearman correlation coefficient is used to select wave and meteorological characteristic parameters with strong correlation with significant wave height as the input sequence. The nonlinearity of the significant wave height sequence is weakened by two-layer decomposition, thereby reducing the data complexity. The long-term dependency of significant wave height is captured by BiLSTM, and the attention mechanism assigns weight to the time series state information output by the BiLSTM layer. The data is predicted by a fully connected layer, and the predicted subsequences are accumulated to obtain the prediction results.

[0032] Some embodiments of the present application will be described in detail with reference to the drawings. The following embodiments and features of the embodiments described below can be combined with each other in the case of no conflict.

[0033] Embodiment one The embodiment of the present application provides an effective wave height prediction method, referring to Figure 1 The embodiment of the present application provides an effective wave height prediction method flow chart, the method comprises: Step S102, obtaining historical feature data for wave height prediction, pre-processing the historical feature data to obtain an effective wave height data set; wherein the effective wave height data set includes: effective wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature.

[0034] Specifically, the stored historical buoy data is used as historical feature data, and the feature data includes multiple categories, such as: effective wave height (SWH), wind direction (WDIR), average wind speed (WSPD), instantaneous wind speed (GST), dominant wave period (DPD), average wave period (APD), average wave direction (MWD), sea level pressure (PRES), air temperature (ATMP), sea surface temperature (WTMP) and dew point temperature (DEWP); The data set determined after pre-processing is the effective wave height data set.

[0035] Further, in some preferred embodiments of the present application, the pre-processing includes: abnormal data rejection, missing data interpolation and normalization processing.

[0036] Specifically, the data collected in the natural environment has great randomness on a single data point, but is generally stable overall. Based on the stability of the overall data, the abnormal points in the time series data are corrected, and the correction means includes rejecting abnormal data and interpolating missing data. Generally, the commonly used abnormal data rejection methods include: standard deviation method and quartile range method, etc. The commonly used missing data interpolation methods include: mean method, median method, linear interpolation method and polynomial interpolation method, etc.

[0037] In order to reduce the influence of the dimension of different feature data, the normalization processing is used to convert multiple data into a unified dimension. Commonly used normalization processing methods include: Min-Max normalization and vector normalization.

[0038] Step S104, determining the data in the effective wave height data set that meets the preset condition in relation to the effective wave height as model input data.

[0039] First, the correlation between any two data in the significant wave height data set is calculated. There are many methods for calculating data correlation, including: Pearson correlation coefficient, Kendall rank correlation coefficient, etc.

[0040] Since the feature data is not a Zhengtai distribution, or has outliers or nonlinear relationships, in some preferred embodiments of the present invention, the Spearman correlation coefficient is determined based on the following formula as the correlation between the two data: ; Among them, ρ is the correlation, and its value range is [-1,1]. The larger its absolute value, the greater the correlation between x and y; R(x i ) is the variable x i rank; is the average rank of all x values; R(y i ) is the variable y i rank; is the average rank of all y values; n is the number of data pairs; x i is the i-th sample data in the x variable; x is a variable data in the correlation calculation; y i is the i-th sample data in the y variable; y is another variable data in the correlation calculation.

[0041] Then, the data other than the significant wave height in the significant wave height dataset is used as the ocean environment characteristic parameter, and the correlation coefficient between the significant wave height and the ocean environment characteristic parameter is determined based on the above formula, as follows: Figure 2 The embodiment of the present invention shown in the figure provides a correlation diagram. The preset condition is generally that the absolute value is greater than a preset value. In some preferred embodiments of the present invention, the preset value is 0.5. Based on this, MWD, GST, WSPD, and WDIR show a high positive correlation with the significant wave height, with Spearman correlation coefficients of 0.76, 0.68, 0.66, and 0.51, respectively, and serve as core model input data. In addition, DEWP and DPD show a high negative correlation with the significant wave height, with Spearman correlation coefficients of -0.75 and -0.63, respectively, and serve as auxiliary parameters in the model input data. In order to reduce the redundancy caused by the massive amount of data, APD, PRES, ATMP, and WTMP, which have a low correlation with the significant wave height, are removed and not used as model input data.

[0042] In step S106, the significant wave height after the primary decomposition and the secondary decomposition is determined as the input data; wherein the primary decomposition includes: empirical mode decomposition based on adaptive noise complete set and K-means clustering of modal components; the secondary decomposition includes: variational mode decomposition; the number of decompositions and the penalty factor of the variational mode decomposition are optimized based on the crested porcupine optimization algorithm.

[0043] Specifically, in some preferred embodiments of the present application, the step of adaptive noise complete ensemble empirical mode decomposition comprises: adding Gaussian white noise to all original to-be-decomposed sequences of effective wave height to obtain to-be-decomposed data; performing empirical mode decomposition on the to-be-decomposed data and then performing ensemble averaging to obtain a first modal component and a residual term; adding Gaussian white noise to all to-be-decomposed sequences of the residual term obtained each time the empirical mode decomposition is performed, and then performing empirical mode decomposition and ensemble averaging again until a preset condition is reached, thereby obtaining a plurality of first modal components.

[0044] The effective wave height data is decomposed in the first layer, and the CEEMDAN (a method based on the EMD theory, which solves the problem of mode aliasing in EMD decomposition by adding adaptive white noise) is used to decompose the original effective wave height data with strong fluctuation into a plurality of modal components with different complexities, and the data is smoothed.

[0045] The specific steps include: adding n times of Gaussian white noise to each original to-be-decomposed sequence of the original time sequence y(t) to obtain a to-be-decomposed sequence y i (t); performing EMD decomposition on each y i (t) to obtain a first modal component of each original to-be-decomposed sequence, and taking the average value of the first modal components as a first modal component IMF1 obtained by CEEMDAN decomposition and a residual term r1(t); continuing to add Gaussian white noise to each to-be-decomposed sequence of the residual term, and performing EMD decomposition again to obtain another first modal component IMF2 of the original sequence and a second residual term r2(t); repeating the above steps until a stop condition is met, and the original data is decomposed into a plurality of modal components and residual terms; and smoothing the first modal components, as shown in a wave height data one-time decomposition result diagram provided by an embodiment of the present application. Figure 3 The first modal components of the decomposed effective wave height gradually decrease in complexity from IMF1 to IMF11; due to the influence of weather and terrain on the sea wave height, the sea wave height has strong fluctuation, and the change trend of the effective wave height sequence can be fully mined through one-time decomposition.

[0046] However, the first modal components with different frequencies obtained through CEEMDAN decomposition still have a high level of fuzzy entropy of some modes. Since the higher the fuzzy entropy value of a mode is, the higher the complexity of the mode is, and the mode may contain a large amount of noise components, the complexity of the high-frequency components is reduced through two-time decomposition. If all the components are decomposed twice, the calculation scale will be increased, and errors will be accumulated. To reduce errors and calculation, the first component mode needs to be aggregated first.

[0047] In some preferable embodiments of the present application, the step of K-means clustering of the modal components comprises: calculating the fuzzy entropy of the plurality of first modal components; performing K-means clustering on the plurality of fuzzy entropies to obtain high-frequency partial components, medium-frequency partial components and low-frequency partial components; and aggregating the high-frequency partial components, the medium-frequency partial components and the low-frequency partial components respectively.

[0048] Specifically, all the first modal components are clustered by using K-means clustering. The aggregation of the modal components is shown in Figure 4 A first modal component K-means clustering result diagram provided by the embodiment of the present application is shown in the figure, which is divided into high-frequency partial components IMF1, medium-frequency partial components IMF2 and low-frequency partial components IMF3.

[0049] Further, in some preferable embodiments of the present application, the step of secondary decomposition comprises: performing variational modal decomposition on the aggregated high-frequency partial components to obtain a plurality of second modal components.

[0050] Specifically, VMD (variational modal decomposition) decomposes the time series into K modal components with central frequencies ω k modal components with limited bandwidth, thereby reducing the complexity of the effective wave height data. Generally, the decomposition number K and the penalty factor a need to be manually set, and the decomposition effect is different when the parameters are set differently.

[0051] However, manual determination of the parameters increases the influence of human factors and cannot achieve the best decomposition effect. Therefore, in some preferable embodiments of the present application, the method further comprises: determining the decomposition number and the penalty factor of the variational modal decomposition by the following steps: initializing the search range of the decomposition number of the variational modal decomposition, the search range of the penalty factor, the population size of the GHO optimization algorithm and the maximum iteration number of the GHO optimization algorithm, and defining a convergence threshold; gradually iterating and optimizing the decomposition number and the penalty factor based on the GHO optimization algorithm until the maximum iteration number is reached, to obtain the final decomposition number and the final penalty factor.

[0052] Specifically, the VMD hyperparameters are optimized by using the CPO optimization algorithm to find the optimal [K, a] parameter combination. First, the modal decomposition number K and the penalty factor a of the VMD are initialized and the penalty factor the search range of K and a, the population size of CPO and the maximum number of iterations, and define the convergence threshold; use VMD to decompose the significant wave height to obtain multiple modal components (IMFs), select the information entropy as the fitness function, and evaluate the decomposition effect of VMD on the significant wave height sequence under each set of [K, a] parameters; in the iteration, CPO updates the population parameter combination according to the four defense strategies of its core: visual intimidation, sound intimidation, smell attack and physical attack, combined with the current optimal individual position; if the maximum number of iterations is reached or the convergence condition is met, the globally optimal [K, a] parameter combination is output; otherwise, continue to decompose, evaluate and update the population until the termination condition is met.

[0053] Further, referring to Figure 5 The embodiment of the application shown in the adaptive iteration curve diagram of multiple optimization algorithms provides the iteration optimization curves of the Crowned Porcupine Optimization (CPO), the Northern Goshawk Optimization (NGO), the Sand Cat Swarm Optimization (SCSO), the Gold Rush Optimization (GRO) and the Grey Wolf Optimization (GWO). The population size of each optimization algorithm is set to 10, the number of iterations is set to 30, the search range of the decomposition number K is set to [2, 14], and the search range of the penalty factor a is set to [200, 3000]. In the diagram, the convergence speed of the GWO optimization algorithm is the slowest, and it converges after 27 iterations. The convergence speeds of the SCSO and GRO optimization algorithms are relatively good, and they converge after 12 and 16 iterations respectively, but the fitness values after decomposition are relatively high, which are 3.33189 and 3.33181 respectively. The NGO has obvious improvement in decomposition effect, and the fitness value after convergence is 3.32886. The CPO has better optimization effect, and can quickly reduce the fitness value in the early stage of iteration, showing strong global search ability. The fitness value of the CPO after convergence is 3.32275, and the optimal parameters are [8, 1987]. By adding the CPO in the two-layer decomposition for parameter optimization, the VMD decomposition capability is further improved, and the neural network is better helped to extract features from the original time series.

[0054] According to the optimization result, the CPO-VMD is used to perform secondary decomposition on the high-frequency component after primary decomposition. The secondary decomposition result is shown in Figure 6 The embodiment of the application shown in the high-frequency partial component decomposition result diagram decomposes the high-frequency component into eight second modal components with different frequencies, and the complexity of each second modal component after decomposition is lower than that of the high-frequency partial component after primary decomposition, which is more conducive to model prediction.

[0055] In step S108, the model input data is input into the bidirectional long short-term memory network based on the attention mechanism until a preset condition is reached, and an effective wave height prediction model is obtained.

[0056] Specifically, the first modal component corresponding to the decomposed intermediate frequency part, the first modal component corresponding to the low frequency part component, the second modal component and the ocean environment characteristic parameter with higher correlation with the significant wave height are combined and input into the BiLSTM network, and the complex relationship in the time series data is captured through the unique bidirectional gating mechanism of the BiLSTM network, and the output hidden layer features are obtained. ; wherein, is the output hidden layer feature, w y is the weight, is the forward output vector, is the reverse output vector, b y is the offset.

[0057] Further, in some preferred embodiments of the present application, the attention mechanism layer is arranged between the bidirectional long short-term memory network and the fully connected layer in the bidirectional long short-term memory network based on the attention mechanism.

[0058] Different weights are assigned to the output features of the BiLSTM layer, so that the model pays more attention to the key information in the input sequence. After the Attention layer, the fully connected layer maps the features to a single output node, and the continuous value prediction is completed through the regression layer. Finally, the prediction values of the modal components of the significant wave height are superimposed to obtain the overall prediction result. The BiLSTM combines the forward LSTM layer and the backward LSTM layer to fully mine the time series data dependency from the front and back directions, and has better prediction effect.

[0059] Specifically, the Attention assigns different weights to the output features of the BiLSTM layer. The specific steps are as follows: the input sequence is input into the BiLSTM layer to output hidden states , and a group of query vectors q is obtained, the similarity s i between the query vectors and the hidden states is calculated; then, the attention weights k i are obtained by normalizing through Softmax; finally, the hidden states c i with attention scores are obtained by weighting and summing the attention weights and the corresponding values.

[0060] ; ; ; wherein, s i is the similarity probability distribution; k i is the attention weight value of the current input, and c i is the BiLSTM hidden layer output value h iThe hidden state with attention scores; 、 、 、 The learning parameters of the model are constantly updated during the model training process.

[0061] The BiLSTM and attention mechanism (Attention) are combined to assign different weights to the output features of the BiLSTM layer, so that the model pays more attention to the key information in the input sequence. After the Attention layer, the fully connected layer maps the features to a single output node, and the regression layer completes the continuous value prediction. Finally, the predicted values of each modal component of the significant wave height are superimposed to obtain the overall prediction result. The BiLSTM network in the model uses the Adam optimizer, and the hyperparameters of the BiLSTM network such as time step, number of neurons, random dropout ratio, maximum number of iterations and learning rate are set as shown in Table 1.

[0062] Table 1

[0063] Referring to Figure 7 The schematic diagram of the significant wave height prediction result provided by the embodiment of the application is shown. By combining the decomposition algorithm and inputting the decomposition result into the neural network, the problem of the single prediction model of LSTM and BiLSTM having hysteresis is effectively solved, and the prediction at the peak is good, which can improve the prediction effect of the neural network. The prediction performance of the single model is obviously improved by integrating the Attention mechanism. The Attention mechanism further enhances the prediction performance of the model by integrating it into the hybrid model. The Attention enhances the importance of the key feature information generated by the BiLSTM in the hidden layer, thereby improving its contribution to the output of the overall model. Whether in the overall trend or in the local fluctuation, it shows a good fitting effect.

[0064] Further, the prediction effect of the model is verified by selecting LSTM, BiLSTM, BiLSTM-Attention and CEEMDAN-CPO-VMD-BiLSTM as comparative models to compare the prediction results of one-step prediction. The mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) are selected as evaluation indexes.

[0065] The evaluation index is constrained by the following formula: ; ; ; Wherein, X i is the true value, Y iN is the length of the time series for the predicted value.

[0066] The evaluation results are shown in Table 2, and the prediction results MAE, RMSE and MAPE of the model provided by the embodiment of the application are 0.0544, 0.0670 and 2.6051 respectively, which are better than those of the model.

[0067]

[0068] In step S110, the target feature data is input into the significant wave height prediction model, and a significant wave height prediction value is output.

[0069] Specifically, the target feature data is obtained, the target feature data is the to-be-predicted feature data, the target feature data is preprocessed, and after the significant wave height is continuously decomposed in two layers, the significant wave height prediction model is input, and a significant wave height prediction value is output.

[0070] The significant wave height prediction method provided by the embodiment of the application is a significant wave height prediction method based on double-layer decomposition and BiLSTM-Attention. The Spearman correlation coefficient is used to select sea wave and meteorological feature parameters with high correlation with the significant wave height as input sequences. The nonlinearity of the significant wave height sequence is weakened through two-layer decomposition, and the data complexity is reduced. The long-term dependence relationship of the significant wave height is captured through BiLSTM, and the attention mechanism assigns weights to the time sequence state information output by the BiLSTM layer. The data is predicted through the full connection layer, and the predicted sub-sequences are accumulated to obtain the prediction result.

[0071] The application provides an effective wave height prediction method, comprising the following steps: obtaining historical characteristic data for wave height prediction, and preprocessing the historical characteristic data to obtain an effective wave height data set; wherein the effective wave height data set comprises effective wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature; calculating the correlation of any two data in the effective wave height data set; determining the data in the effective wave height data set that meets a preset condition in terms of correlation with the effective wave height as model input data; determining the effective wave height after one-time decomposition and two-time decomposition as input data; wherein the one-time decomposition comprises adaptive noise complete ensemble empirical mode decomposition and K-means clustering of modal components; the two-time decomposition comprises variational mode decomposition; the decomposition number and the penalty factor of the variational mode decomposition are optimized based on a Guan Haogu optimization algorithm; inputting the model input data into a bidirectional long short-term memory network based on an attention mechanism until a preset condition is met, to obtain an effective wave height prediction model; inputting target characteristic data into the effective wave height prediction model to output an effective wave height prediction value; the data with high correlation with the effective wave height is first screened as input data, thereby reducing the calculation amount, then the effective wave height is decomposed in two layers, thereby weakening the nonlinearity of the effective wave height and reducing the data complexity, and finally the bidirectional long short-term memory network based on the attention mechanism is used to fully extract the data characteristics, thereby comprehensively predicting the data and improving the accuracy of the predicted effective wave height.

[0072] Embodiment two On the basis of the above-mentioned embodiments, the application provides an effective wave height prediction device, which is shown in Figure 8 The effective wave height prediction device provided by the application embodiment is shown in the structural schematic diagram of the effective wave height prediction device, and the device comprises: The data processing module 310 is configured to obtain historical characteristic data for wave height prediction, and preprocess the historical characteristic data to obtain an effective wave height data set; wherein the effective wave height data set comprises effective wave height, wind direction, average wind speed, instantaneous wind speed, dominant wave period, average wave period, average wave direction, sea level pressure, air temperature, sea surface temperature and dew point temperature.

[0073] The first input data determination module 320 is configured to determine the data in the effective wave height data set that meets a preset condition in terms of correlation with the effective wave height as model input data.

[0074] The second input data determination module 330 is configured to determine the effective wave height after one-time decomposition and two-time decomposition as input data; wherein the one-time decomposition comprises adaptive noise complete ensemble empirical mode decomposition and K-means clustering of modal components; the two-time decomposition comprises variational mode decomposition; the decomposition number and the penalty factor of the variational mode decomposition are optimized based on a Guan Haogu optimization algorithm.

[0075] The prediction model processing module 340 is configured to input the model input data into a bidirectional long short-term memory network based on an attention mechanism until a preset condition is reached, so as to obtain an effective wave height prediction model.

[0076] The data prediction module 350 is configured to input the target feature data into the effective wave height prediction model, and output an effective wave height prediction value.

[0077] Further, in some preferred embodiments of the present application, the data processing module 310 is configured to pre-process the data through abnormal data elimination, missing data interpolation and normalization processing.

[0078] Further, in some preferred embodiments of the present application, the first input data determination module 320 is configured to determine the correlation based on the following formula: ; wherein p is the correlation; R(x i ) is the rank of the variable x i ; R(x i ) is the rank of the variable y i ; R(y i ) is the i-th sample data in the x variable; x is one kind of variable data in the correlation calculation; y i is the i-th sample data in the y variable; y is another kind of variable data in the correlation calculation.

[0079] Further, in some preferred embodiments of the present application, the second input data determination module 330 is configured to add Gaussian white noise to all original to-be-decomposed sequences of the effective wave height to obtain to-be-decomposed data; perform empirical mode decomposition on the to-be-decomposed data and then perform ensemble averaging to obtain a first modal component and a residual term; continue to add Gaussian white noise to all to-be-decomposed sequences of the residual term obtained after each empirical mode decomposition, and then perform empirical mode decomposition and ensemble averaging until a preset condition is reached, to obtain a plurality of first modal components.

[0080] Further, in some preferred embodiments of the present application, the second input data determination module 330 is configured to calculate the fuzzy entropy of the plurality of first modal components; perform K-means clustering on the plurality of fuzzy entropies to obtain a high-frequency partial component, a medium-frequency partial component and a low-frequency partial component; and aggregate the high-frequency partial component, the medium-frequency partial component and the low-frequency partial component respectively.

[0081] Further, in some preferred embodiments of the present application, the second input data determination module 330 is configured to perform variational mode decomposition on the aggregated high-frequency partial component to obtain a plurality of second modal components.

[0082] Further, in some preferable embodiments of the present application, the second input data determination module 330 is configured to determine the decomposition number and the penalty factor of the variational mode decomposition by: initializing a search range of the decomposition number of the variational mode decomposition, a search range of the penalty factor, a population size of the Cuckoo Optimization Algorithm, and a maximum iteration number of the Cuckoo Optimization Algorithm, and defining a convergence threshold; and iteratively optimizing the decomposition number and the penalty factor based on the Cuckoo Optimization Algorithm until the maximum iteration number is reached, to obtain a final decomposition number and a final penalty factor.

[0083] Further, in some preferable embodiments of the present application, the attention mechanism layer is arranged between the bidirectional long short-term memory network and the fully connected layer in the bidirectional long short-term memory network based on the attention mechanism.

[0084] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the effective wave height prediction device described above can refer to the corresponding process in the foregoing embodiment of the effective wave height prediction method, which will not be described here.

[0085] Embodiment three The embodiment of the present application also provides an electronic device for running the effective wave height prediction method; see Figure 9 The electronic device provided by the embodiment of the present application shown in the structural schematic diagram of the electronic device comprises a memory 400 and a processor 401, wherein the memory 400 is used for storing one or more computer instructions, and the one or more computer instructions are executed by the processor 401 to realize the effective wave height prediction method described above.

[0086] Further, Figure 9 The electronic device shown in the figure further comprises a bus 402 and a communication interface 403, and the processor 401, the communication interface 403 and the memory 400 are connected through the bus 402.

[0087] The memory 400 can contain a high-speed random access memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. The communication between the system network element and at least one other network element is realized through at least one communication interface 403 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used. The bus 402 can be an ISA bus, a PCI bus, or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 9 Only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0088] The processor 401 can be an integrated circuit chip having a processing capability of signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 401 or the instruction in the form of software. The processor 401 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. Each method, step and logic block diagram disclosed in the embodiment of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiment of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read only memory, a programmable read only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the storage 400, and the processor 401 reads the information in the storage 400, and combines the hardware to complete the steps of the method of the above embodiment.

[0089] The embodiment of the present application further provides a computer readable storage medium, the computer readable storage medium stores computer executable instructions, when the computer executable instructions are called and executed by a processor, the computer executable instructions cause the processor to implement the effective wave height prediction method, and the specific implementation can be referred to the method embodiment, and will not be repeated here.

[0090] The computer program product of the effective wave height prediction method, device and electronic equipment provided by the embodiment of the present application includes a computer readable storage medium storing program codes, the instructions included in the program codes can be used to execute the method in the foregoing method embodiment, and the specific implementation can be referred to the method embodiment, and will not be repeated here.

[0091] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and / or device described above can refer to the corresponding process in the foregoing method embodiment, and will not be repeated here.

[0092] In addition, in the description of the embodiments of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, or can be electrically connected; can be directly connected, or indirectly connected through an intermediate medium; can be internal communication of two elements. For those skilled in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0093] The functions described above, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the part of the present application which essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0094] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A significant wave height prediction method, characterized in that: include: Acquiring historical characteristic data for wave height prediction, and preprocessing the historical characteristic data to obtain a significant wave height dataset; Determining data in the significant wave height data set whose correlation with the significant wave height meets a preset condition as model input data; The significant wave height after primary decomposition and secondary decomposition is determined as the input data; wherein the primary decomposition includes: adaptive noise complete set empirical mode decomposition and K-means clustering of modal components; the secondary decomposition includes: variational mode decomposition; the number of decompositions and penalty factor of the variational mode decomposition are optimized based on the crested porcupine optimization algorithm; Inputting the model input data into a bidirectional long short-term memory network based on an attention mechanism until a preset condition is met, thereby obtaining a significant wave height prediction model; The target characteristic data is input into the significant wave height prediction model, and a significant wave height prediction value is output.

2. The significant wave height prediction method according to claim 1, characterized in that: The preprocessing includes: abnormal data elimination, missing data interpolation and normalization processing.

3. The significant wave height prediction method according to claim 1, characterized in that: The correlation is determined based on the following formula: ； Wherein, ρ is the correlation; R(x i ) is the variable x i rank; is the average rank of all x values; R(y i ) is the variable y i rank; is the average rank of all y values; n is the number of data pairs; x i is the i-th sample data in the x variable; x is a variable data in the correlation calculation; y i is the i-th sample data in the y variable; y is another variable data in the correlation calculation.

4. The significant wave height prediction method according to claim 1, characterized in that: The adaptive noise complete set empirical mode decomposition is a step, including: Adding Gaussian white noise to all sequences to be decomposed of the significant wave height to obtain data to be decomposed; Performing empirical mode decomposition on the data to be decomposed and then performing ensemble averaging to obtain a first modal component and a residual term; After continuing to add Gaussian white noise to all original to-be-decomposed sequences of the residual terms obtained after each empirical mode decomposition, the empirical mode decomposition and ensemble averaging are performed again until a preset condition is met to obtain multiple first modal components.

5. The significant wave height prediction method according to claim 4, characterized in that: The step of K-means clustering of the modal components comprises: calculating fuzzy entropies of a plurality of the first modal components; Performing K-means clustering on the plurality of fuzzy entropies to obtain a high-frequency component, a mid-frequency component, and a low-frequency component; Aggregation is performed on the high-frequency component, the mid-frequency component, and the low-frequency component respectively.

6. The significant wave height prediction method according to claim 5, characterized in that: The secondary decomposition step comprises: The aggregated high-frequency components are subjected to variational modal decomposition to obtain a plurality of second modal components.

7. The significant wave height prediction method according to claim 6, characterized in that: The method further includes determining the number of decompositions and the penalty factor of the variational mode decomposition by the following steps: Initializing the search range of the number of decompositions of the variational mode decomposition, the search range of the penalty factor, the population size of the crested porcupine optimization algorithm, and the maximum number of iterations of the crested porcupine optimization algorithm, and defining a convergence threshold; The decomposition number and the penalty factor are optimized step by step iteratively based on the crested porcupine optimization algorithm until the maximum number of iterations is reached, thereby obtaining the final decomposition number and the final penalty factor.

8. The significant wave height prediction method according to claim 1, characterized in that: In a bidirectional long short-term memory network based on an attention mechanism, an attention mechanism layer is arranged between the bidirectional long short-term memory network and a fully connected layer.

9. A significant wave height prediction device, characterized in that: include: A data processing module is used to obtain historical characteristic data for wave height prediction, and pre-process the historical characteristic data to obtain an effective wave height data set; a first input data determination module, configured to determine data in the significant wave height data set whose correlation with the significant wave height satisfies a preset condition as model input data; A second input data determination module is configured to determine the significant wave height after primary decomposition and secondary decomposition as the input data; wherein the primary decomposition includes: adaptive noise complete set empirical mode decomposition and K-means clustering of modal components; the secondary decomposition includes: variational mode decomposition; the number of decompositions and penalty factor of the variational mode decomposition are optimized based on the crested porcupine optimization algorithm; A prediction model processing module is used to input the model input data into a bidirectional long short-term memory network based on an attention mechanism until a preset condition is met to obtain a significant wave height prediction model; The data prediction module is used to input target characteristic data into the significant wave height prediction model and output a significant wave height prediction value.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the significant wave height prediction method according to any one of claims 1 to 8.

Images