Collaborative prediction method and system for Yantang river eagle based on reconstruction and decomposition

Abstract

The invention discloses a reconstruction and decomposition Yantang river eagle collaborative prediction method and system. The method comprises the following steps: firstly, performing data reconstruction on an original irregular tidal event flow, and generating a regularized multivariate time sequence containing tide level and time dynamics by constructing a flood tide event pair and converting time difference between tidal events into numerical characteristics; then, inputting the sequence into a deep learning model integrated with adaptive periodic decomposition and multi-scale time sequence change learning; the model outputs predicted values of future high and low tide levels, rising tide duration and tide periods at a time through a collaborative prediction decoder, and finally restores the predicted values into accurate rising tide height and arrival time. According to the method, through data reconstruction, an irregular tide prediction problem is converted into a regular multivariate time series prediction task, and collaborative prediction of the tidal bore height and the arrival time is realized.

Description

Technical Field

[0001] The present invention relates to the field of Qiantang River tidal bore prediction, and in particular to a method for collaboratively predicting the height and arrival time of the Qiantang River tidal bore through data reconstruction and a deep learning model. Background Art

[0002] The Qiantang River tidal bore, a world-renowned natural wonder, is of irreplaceable importance in ensuring safe tidal viewing, coastal engineering construction, shipping management, and water resource management. However, the Qiantang River tidal bore's extremely complex dynamics pose significant challenges to accurate prediction. The formation and evolution of the tidal bore are influenced by multiple periodic factors, including the twice-daily semidiurnal cycle, the twice-monthly spring tide cycle, and the annual astronomical spring tide cycle, exhibiting a pronounced multi-periodic nature. Furthermore, the actual arrival times of tides are not fixed. Due to the natural drift of the tidal cycle (a lunar day is approximately 24 hours and 50 minutes), tidal events occur at irregular times when observed based on a standard natural day, posing difficulties in modeling. Existing prediction methods primarily focus on predicting tide height, often employing traditional time series models or conventional deep learning models. However, these models exhibit three drawbacks when faced with the simultaneous task of predicting both the height and arrival time of the tidal bore.

[0003] First, existing methods typically assume that data is sampled at evenly spaced time points, making it difficult to directly process data from unevenly spaced event streams like the Qiantang River tidal bore. These methods typically treat the tidal bore height and arrival time as two separate tasks, ignoring the inherent and strong physical coupling between them—for example, stronger tidal bores are often accompanied by faster propagation speeds, resulting in earlier arrival times. This separate forecasting approach fails to capture the dynamics of coordinated changes, limiting the accuracy and reliability of the overall forecast.

[0004] Secondly, at the feature extraction level, existing technologies primarily fall into two approaches, both of which have limitations. One approach employs an end-to-end prediction model, directly feeding the raw sequence into a complex neural network for fitting. While this "black-box" approach eliminates manual feature engineering, the lack of explicit extraction and modeling of the inherent periodicity and trend within the sequence makes it difficult for the model to learn clear and interpretable temporal patterns. This makes it prone to overfitting on non-stationary data, resulting in poor generalization. Another approach draws inspiration from classic time series analysis and employs a "decomposition first, prediction later" strategy, first decomposing the raw sequence into a trend term, a period term, and a residual term. However, the decomposition modules in these approaches, such as moving averages or Fourier transforms, are typically static and non-adaptive. This means that the decomposition method is pre-set, and its parameters cannot adapt to the specific characteristics of the input data or changes in the prediction task. This can lead to suboptimal decomposition results, loss of critical information, or the introduction of decomposition artifacts.

[0005] Finally, regarding prediction methods, traditional models typically use a self-regressive prediction model, generating future predictions at each time point. The bottleneck of this approach lies in error accumulation: in multi-step predictions, small errors in the first step are passed as input to the second step, causing the errors to be amplified step by step. As the prediction time increases, the model output degrades rapidly, and the prediction curve becomes overly smooth, losing a large amount of real detailed fluctuations, making it difficult to meet the needs of long-term or high-precision predictions.

[0006] Therefore, how to invent a novel data representation method to transform the irregular tidal event flow into a regular sequence that can be processed by advanced models, and build a collaborative prediction model that can avoid error accumulation, adaptively decompose time series dynamics, and deeply learn its inherent patterns, is a technical challenge that needs to be solved urgently in this field. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the defects of the existing technology in processing complex sequences with irregular time and multiple superposition of multiple periods, such as the Qiantang River tidal bore, such as poor model applicability, serious error accumulation, static decomposition method, and inability to coordinate prediction of height and time. To this end, the present invention proposes a collaborative prediction method for the Qiantang River tidal bore based on event pair reconstruction and adaptive period decomposition. The core idea of ​​this method is: first, through a novel data reconstruction technology, the irregular tidal event prediction problem is cleverly transformed into a regularized multivariate time series prediction task containing deep physical meaning; second, an advanced model that deeply integrates adaptive decomposition and multi-scale time series learning is used to perform deep feature extraction and collaborative prediction on the multivariate sequence, thereby achieving accurate and reliable synchronous prediction of tidal bore height and arrival time.

[0008] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0009] In a first aspect, the present invention provides a collaborative prediction method for Qiantang River tidal bore based on reconstruction and decomposition, which comprises:

[0010] S1. Reconstruct the original time series data of historical tidal events in the Qiantang River. Pair each low tide event with the first high tide event that follows it to form a high tide event pair. Extract the features of each high tide event pair in six variable dimensions: low tide level, high tide level, high tide duration, tidal cycle, day within the year, and lunar day. Normalize the features and obtain a multivariate time series consisting of six univariate time series.

[0011] S2. Input the multivariate time series into the trained tidal collaborative prediction model, and pass through multiple stacked dynamic local-global learning modules and prediction decoders in sequence. Each dynamic local-global learning module first inputs the univariate time series of each variable dimension into the adaptive period decomposition module to obtain multiple periodic components representing different frequency characteristics, and then inputs each periodic component into the multi-scale temporal change learning module. The global features and local features are extracted from each periodic component in parallel and then fused. The fused feature vectors of all periodic components are then input into the adaptive aggregation module for two-dimensional convolution and reweighting to obtain the final feature representation of each variable dimension. Finally, the final feature representation of the six variable dimensions is input into the multivariate interactive aggregation module for interactive aggregation between the variable dimensions, and the interactive aggregation features are used as the module output; the prediction decoder receives the interactive aggregation features output by the last dynamic local-global learning module, and maps them to the low tide level, high tide level, tide duration and tidal period of the tide event pair in the prediction window through the fully connected layer, and then calculates the tide level and time of each tidal event in the prediction window in sequence.

[0012] As a preferred embodiment of the first aspect above, the feature extraction method for each high tide event pair of six variable dimensions is as follows: extract the low tide level and high tide level from the low tide event and high tide event in the current high tide event pair respectively, calculate the difference between the timestamps of the high tide event and the low tide event in the current high tide event pair as the high tide duration, calculate the difference between the timestamps of the low tide event in the next high tide event pair and the low tide event in the current high tide event pair as the tidal cycle, take the day number of the timestamp of the low tide event in the current high tide event pair in the Gregorian calendar year as the day of the year, and take the day number of the timestamp of the low tide event in the current high tide event pair in the lunar month as the lunar day.

[0013] As a preferred embodiment of the above-mentioned first aspect, in the adaptive periodic decomposition module, a maximum overlap discrete wavelet transform (MODWT) operator with a learnable parameter matrix as a filter is first used to perform multi-level wavelet decomposition on the univariate time series of each variable dimension, and adaptively deconstruct it into multiple initial periodic components. Then, a neural network is used to generate a mask for each initial periodic component and the mask is used to adaptively adjust each initial periodic component. Finally, the adjusted periodic components are reconstructed through multi-resolution analysis to obtain multiple periodic components representing different frequency characteristics as the final output.

[0014] As a preferred embodiment of the first aspect, the multi-scale temporal change learning module includes a global change learning branch and a local change learning branch:

[0015] In the global change learning branch, the input periodic component is first linearly mapped, and then the linearly mapped periodic component is divided into multiple overlapping time segments. All time segments are then combined into a two-dimensional matrix and sequentially passed through a self-attention layer and a feedforward network to obtain a two-dimensional spatial feature. The two-dimensional spatial feature is flattened and linearly mapped back to a one-dimensional time series space to obtain a global feature.

[0016] In the local variation learning branch, the input periodic component is passed through multiple separable causal convolutional layers with different dilation rates in parallel, and then the convolution results are fused and superimposed on the input periodic component in a residual connection manner to form local features;

[0017] Finally, the global features and local features are further fused to obtain the fused feature vector of the current input periodic component.

[0018] As a preferred embodiment of the above-mentioned first aspect, in the adaptive aggregation module, the fused feature vectors of all periodic components of the current variable dimension are first combined into a two-dimensional tensor and then input into the two-dimensional convolutional network, and then the two-dimensional convolution result is re-decomposed into the convolution feature vectors corresponding to each periodic component, and then weighted aggregation is performed with the component variance of each periodic component as the weight to obtain the final feature representation of the current variable dimension.

[0019] As a preferred embodiment of the above-mentioned first aspect, in the multivariable interaction aggregation module, the final feature representations of the six variable dimensions are spliced ​​into a two-dimensional matrix, and then compressed into a vector by averaging along the time dimension, and then converted into the weighted weights of the final feature representations of the six variable dimensions through a linear layer and weighted aggregation is performed to obtain the interactive aggregation features as the final output of the current dynamic local-global learning module.

[0020] As a preferred embodiment of the first aspect, the tidal bore collaborative prediction model needs to be pre-supervised trained before being used for actual reasoning, and its loss function adopts mean square error loss, root mean square error loss or mean absolute error loss.

[0021] In a second aspect, the present invention provides a Qiantang River tidal bore collaborative prediction system based on reconstruction and decomposition, which includes:

[0022] Data input module, used to input the original time series data of historical tidal events in Qiantang River;

[0023] A collaborative prediction module, configured to generate a prediction result based on the input original time series data of historical tidal events of the Qiantang River and in accordance with the collaborative prediction method for the Qiantang River tidal bore based on reconstruction and decomposition as described in any one of the solutions of the first aspect above;

[0024] The data output module is used to output the generated prediction results according to the preset output method.

[0025] In a third aspect, the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, can implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described in any of the solutions in the first aspect above.

[0026] In a fourth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0027] The memory is used to store computer programs;

[0028] The processor is configured to implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described in any one of the solutions of the first aspect when executing the computer program.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] This paper proposes a collaborative prediction method for the Qiantang River tidal bore based on event-pair reconstruction and adaptive periodic decomposition. First, through an innovative "tidal event pair" data reconstruction method, the unevenly spaced tidal event stream in the physical world is successfully converted into a well-structured, temporally discretized multivariate time series. This fundamentally addresses the technical difficulty of traditional time series models in processing irregular event data, laying a solid foundation for accurate prediction using advanced deep learning models.

[0031] Secondly, this method no longer treats tide level and time as isolated prediction targets. Instead, it incorporates the "high tide duration" and "tidal period" that determine time as endogenous variables, along with tide level characteristics, into a unified model for learning and prediction. This design captures the inherent physical coupling between the two, enabling predictions of tidal height and arrival time to be mutually verified and optimized, resulting in predictions that are more consistent with real physical dynamics.

[0032] Furthermore, the collaborative prediction model employed in this paper integrates a learnable, adaptive periodic decomposition module, which dynamically separates periodic patterns of different scales based on data characteristics, overcoming the static and non-adaptive limitations of existing decomposition methods. Furthermore, the model employs a non-autoregressive decoding approach, generating predictions for multiple future features simultaneously. This significantly improves prediction efficiency and mechanically avoids the critical error accumulation problem inherent in traditional autoregressive models, ensuring the reliability of long-term predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of the steps of the collaborative prediction method of Qiantang River tidal bore based on reconstruction and decomposition;

[0034] Figure 2 It is a structural schematic diagram of the tidal bore collaborative prediction model of the present invention.

[0035] Figure 3 Schematic diagram of the structure of the dynamic local-global learning module.

[0036] Figure 4 This is a structural diagram of the adaptive periodic decomposition operator of the present invention.

[0037] Figure 5 This is the structural diagram of our multi-scale temporal relationship learning module.

[0038] Figure 6 Schematic diagram of the module composition of the Qiantang River tidal bore collaborative prediction system based on reconstruction and decomposition.

[0039] Figure 7 It is a structural diagram of computer electronic equipment.

[0040] Figure 8 This is a comparison chart of the prediction model and the Transformer model in an embodiment of the present invention.

[0041] Figure 9 Graph showing the error distribution of the prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in the various embodiments of the present invention can be combined accordingly without conflicting with each other.

[0043] In the description of the present invention, it should be understood that the terms "first" and "second" are used solely for descriptive purposes and are not to be construed as indicating or implying relative importance or implicitly specifying the number of technical features being described. Therefore, features defined as "first" or "second" may explicitly or implicitly include at least one of such features.

[0044] The present invention provides a collaborative prediction method for the Qiantang River tidal bore based on reconstruction and decomposition. This method converts irregular tidal event streams into regularized multivariate time series and adopts an advanced model that can adaptively decompose and deeply learn its internal patterns to achieve collaborative prediction of tidal bore height and arrival time. Figure 1 As shown, in a preferred embodiment of the present invention, the method includes the following steps:

[0045] S1. Obtain and reconstruct raw tidal data to generate regularized multivariate time series.

[0046] In step S1, the original time series data of the historical tidal events of the Qiantang River need to be reconstructed. Each low tide event is paired with the first high tide event that follows it to form a high tide event pair. The features of each high tide event pair, including low tide level, high tide level, high tide duration, tidal cycle, day within the year, and lunar day, a total of six variable dimensions, are extracted and normalized to obtain a multivariate time series consisting of six univariate time series.

[0047] It should be noted that the source of raw time series data for historical tidal events on the Qiantang River is not limited; it can generally be obtained from hydrological observation stations along the river. Because tidal events occur at varying times on different dates, the raw time series data for historical tidal events is a stream of events with uneven time intervals. Each record contains the arrival timestamp, tide type (high tide H / low tide L), and tide level of a tidal event. Directly inputting this data into the model will not yield accurate predictions, so the raw data must be reconstructed to generate a regularized, model-friendly multivariate time series.

[0048] It should be noted that one of the core innovations of the present invention lies in this data reconstruction step. This step first constructs a series of "high tide event pairs" by pairing each low tide event with the first high tide event that follows it in time. Subsequently, the eigenvalues ​​of the six variable dimensions are extracted for each high tide event pair to generate a six-dimensional numerical feature vector. For ease of description, the high tide event pair whose features need to be extracted is called the current high tide event pair, and the high tide event pair that follows this high tide event pair is called the next high tide event pair. The calculation methods of the six dimensional features of the current high tide event pair are as follows:

[0049] The first feature (low tide level): extract the low tide level directly from the low tide events within the current high tide event pair;

[0050] The second feature (high tide level): extracts the high tide level directly from the high tide events within the current high tide event pair;

[0051] The third feature (high tide duration): Calculate the difference between the timestamps of the high tide event and the low tide event within the current high tide event pair.

[0052] The fourth feature (tidal cycle): Calculate the difference between the timestamps of the low tide event in the next high tide event pair and the low tide event in the current high tide event pair.

[0053] The fifth feature (day within the year): extract the timestamp of the low tide event within the current high tide event, and calculate the ordinal number of the timestamp in the Gregorian calendar year, which is recorded as the day within the year.

[0054] The sixth feature (Lunar Day): Extract the timestamp of the low tide event within the current high tide event pair, and calculate the day number of the timestamp in the lunar month, which is recorded as the lunar day.

[0055] It should be noted that days within the year are accumulated across different Gregorian calendar months, while lunar days are counted from the beginning across different lunar months and do not accumulate. For example, if the low tide event timestamp is January 1, 2013, its day number in the 2013 year is the 1st day, so its day within the year is 1. Its lunar calendar date is November 20th, so its day number in the 11th lunar month is the 20th day, so its lunar day number is 20. For another example, if the low tide event timestamp is February 1, 2013, its day number in the 2013 year is the 32nd day (because January has 31 days), so its day within the year is 32. Its lunar calendar date is December 21st, so its day number in the 12th lunar month is the 21st day, so its lunar day number is 21.

[0056] It should be noted that before data reconstruction, the raw data can be preprocessed, including using linear or mean interpolation to complete missing tidal event records and smoothing or removing anomalous tidal values ​​that exceed preset physical thresholds. Furthermore, for each high tide event pair after data reconstruction, a six-dimensional numerical feature vector is extracted from these six variable dimensions. These six-dimensional numerical feature vectors are recombined in the order of the events to obtain a multivariate time series. This multivariate time series can be considered as consisting of six univariate time series: low tide level, high tide level, high tide duration, tidal cycle, day of the year, and lunar day, each corresponding to a univariate time series. The length T of all univariate time series is the same, equal to the number of high tide event pairs in the historical data. Furthermore, the univariate time series need to be normalized before input into the model. For example, using maximum-minimum normalization to scale all eigenvalues ​​to the range [0, 1] or [-1, 1] can improve the efficiency and stability of subsequent model training.

[0057] S2. Obtain a pre-trained tidal bore collaborative prediction model, and input the multivariate time series into the tidal bore collaborative prediction model for collaborative prediction.

[0058] In step S2, a deep learning model for collaborative prediction needs to be built and trained in advance, which is called the tidal bore collaborative prediction model. The multivariate time series is then input into the trained tidal bore collaborative prediction model, and passes through multiple stacked dynamic local-global learning blocks (DLGBlock) and prediction decoders within the model. Each dynamic local-global learning block first inputs the univariate time series of each variable dimension into an adaptive period decomposition module to obtain multiple periodic components representing different frequency characteristics. Each periodic component is then input into a multi-scale temporal variation learning module, and global and local features are extracted from each periodic component in parallel and fused. The fused feature vectors of all periodic components are then input into an adaptive aggregation module for two-dimensional convolution and reweighting to obtain the final feature representation of each variable dimension. Finally, the final feature representation of the six variable dimensions is input into a multivariate interactive aggregation module for interactive aggregation between the variable dimensions, with the interactive aggregated features as the module output. The prediction decoder receives the interactive aggregated features output by the last dynamic local-global learning module and maps them into the low tide level, high tide level, tide duration and tidal period of the tide event pair in the prediction window through a fully connected layer, thereby calculating the tide level and time of each tide event in the prediction window in sequence.

[0059] like Figure 2As shown in FIG, the overall architecture of the tidal bore collaborative prediction model adopted by the present invention is composed of multiple stacked DLG modules, and the information flow is enhanced by residual connection. The last DLG module is connected to the prediction decoder. The internal structure of each DLG module is as follows Figure 3 As shown in Figure 2, the model consists of an adaptive period decomposition module, a multiscale temporal variation learning module, an adaptive aggregation module, and a multivariate interaction aggregation module. The model input is a multivariate time series, and the output is a four-dimensional core feature prediction vector for one or more high tide event pairs within the future prediction window.

[0060] The following describes in detail the specific functional implementations of the adaptive period decomposition module, multi-scale temporal change learning module, adaptive aggregation module and multivariate interaction aggregation module within the above-mentioned DLG module.

[0061] In an embodiment of the present invention, the detailed structure of the adaptive period decomposition module is as follows: Figure 4 Its core is a maximum overlap discrete wavelet transform (MODWT) operator with a learnable parameter matrix as the filter, that is, a pair of high-pass wavelet filters used in traditional MODWT. and low-pass scaling filters The parameters of the wavelet transform are fixed, while the present invention sets them as a learnable parameter matrix during the model training process. This distinguishes it from the traditional fixed-filter wavelet transform in that it can adaptively adjust the filter parameters during model training through the backpropagation algorithm, thereby specifically decomposing the input multivariate time series into multiple periodic components that are most suitable for the current prediction task.

[0062] The specific process within the above-mentioned adaptive periodic decomposition module is as follows: the module takes the univariate time series of the six variable dimensions in the multivariate time series as input, first uses the above-mentioned MODWT operator to perform multi-level wavelet decomposition on the univariate time series of each variable dimension, and adaptively deconstructs it into multiple initial periodic components. Then, a neural network is used to generate a mask for each initial periodic component, and the initial periodic component is multiplied by the mask. The mask is used to adaptively adjust each initial periodic component. Finally, all adjusted periodic components are reconstructed through multi-resolution analysis (MRA) to obtain the final output of multiple periodic components representing different frequency characteristics.

[0063] It should be noted that in the adaptive periodic decomposition module, the level of wavelet decomposition and the number of periodic components after the final MRA reconstruction are parameters that can be optimized and adjusted according to the actual results.

[0064] In an embodiment of the present invention, the detailed structure of the multi-scale temporal variation learning module is as follows: Figure 5As shown. This module uses a parallel dual-branch structure to deeply extract the temporal features of each periodic component. Specifically, the multi-scale temporal change learning module includes a global change learning branch and a local change learning branch: in the global change learning branch, the input periodic component is first linearly mapped, and then the linearly mapped periodic component is divided into multiple overlapping time segments. All time segments are then combined into a two-dimensional matrix and sequentially passed through the self-attention layer and the feed-forward network (FFN) to obtain two-dimensional spatial features. The two-dimensional spatial features are flattened and linearly mapped back to the one-dimensional time series space to obtain global features; in the local change learning branch, the input periodic component is passed through multiple separable causal convolution layers with different expansion rates in parallel, and then the convolution results are fused and superimposed on the input periodic component in the form of residual connection to form local features; finally, the global features and local features are further fused to obtain the fused feature vector of the current input periodic component. In the dual-learning branch structure of this multi-scale temporal variation learning module, the global variation learning branch efficiently captures long-range dependencies in the sequence through overlapping blocks and block self-attention mechanisms. Meanwhile, the local variation learning branch uses multi-scale separable causal convolutions with different dilation rates to capture detailed local dynamic patterns in different receptive fields. Ultimately, the output features of the two branches at different scales are fused, and the resulting fused feature vector is a comprehensive feature representation of the periodic component.

[0065] In an embodiment of the present invention, the adaptive aggregation module first combines the fused feature vectors of all periodic components of the current variable dimension into a two-dimensional tensor and then inputs it into a two-dimensional convolutional network (2D-CNN). This produces a two-dimensional convolution result of the same dimension as the input tensor. The post-convolution feature vectors in each row of the two-dimensional convolution result still correspond one-to-one to each periodic component. The two-dimensional convolution result is then re-decomposed into the post-convolution feature vectors corresponding to each periodic component. Weighted aggregation is then performed using the component variance of each periodic component as a weight to obtain the final feature representation for the current variable dimension. It should be noted that the weight used for the post-convolution feature vector of each periodic component during this weighted aggregation is the component variance of that periodic component (output by the adaptive periodic decomposition module for the current variable dimension). A periodic component can be viewed as a vector, and its weight can be obtained by calculating the variance of all values ​​in the vector. Thus, the post-convolution feature vectors corresponding to all periodic components in the current variable dimension can be weightedly aggregated using their respective weights to obtain the final feature representation for the current variable dimension.

[0066] In an embodiment of the present invention, the design of the above-mentioned multivariable interactive aggregation module is borrowed from the field of computer vision, and is intended to learn the numerical features of the six different variable dimensions described in the present invention as feature groups, and learn the intrinsic correlation between variables. Specifically, in the multivariable interactive aggregation module, the final feature representations of the six variable dimensions are spliced ​​into a two-dimensional matrix, and then compressed into a vector by averaging along the time dimension, and then converted into the weighted weights of the final feature representations of the six variable dimensions through a linear layer and weighted aggregation is performed to obtain the interactive aggregation features as the final output of the current dynamic local-global learning module. The multivariable interactive aggregation module performs adaptive weighted aggregation on each feature dimension by calculating the global information within the feature group and generating importance weights. In this way, the model can explicitly learn complex cross-dimensional dependencies such as "when the tidal week shortens, the importance of high tide features should increase", thereby enhancing the expressive power of the model.

[0067] Therefore, after the multivariate time series of the present invention is input into the model, it needs to pass through multiple cascaded adaptive periodic decomposition modules in sequence. Each adaptive periodic decomposition module can dynamically separate periodic patterns of different scales according to data characteristics, which can overcome the static and non-adaptive defects of existing decomposition methods.

[0068] Furthermore, in the collaborative tidal bore prediction model, the number of adaptive period decomposition modules can be adjusted appropriately based on actual conditions. In addition to the cascaded stack of multiple adaptive period decomposition modules, the model also requires a collaborative prediction decoder at the end. The collaborative prediction decoder primarily consists of one or more fully connected layers. It receives the interactively aggregated features, which incorporate both temporal and feature dimension information, from the final adaptive period decomposition module and, using a non-autoregressive direct prediction method, outputs a four-dimensional prediction vector in one go. This prediction vector contains predicted values ​​for four core dynamic features of one or more future high tide event pairs: low tide level, high tide level, high tide duration, and tidal period.

[0069] However, since the original tidal data is reconstructed into a pair of high tide events in the present invention, the low tide level, high tide level, high tide duration and tidal period of the high tide event pair are also predicted. Therefore, the present invention also needs to restore the information of the high tide event pair into the tide level and time information of the tidal event. Since tidal events are a series of event streams with a sequential dependency, it is necessary to calculate the tide level and time of each tidal event within the prediction window in chronological order based on the timestamp of the last tidal event in the historical data, combined with the predicted low tide level, high tide level, high tide duration and tidal period of the high tide event pair. By performing this result restoration operation, the tidal height and arrival time within the future prediction window range with clear physical meaning can be obtained.

[0070] In an embodiment of the present invention, the specific calculation logic of the above-mentioned result restoration operation is: first, the prediction result is denormalized, and then the arrival timestamp of the last known low tide event in the historical data is added to the predicted tidal cycle duration in the four-dimensional prediction vector to calculate the precise arrival time of the future low tide event; then, based on the arrival time of the future low tide event, the predicted high tide duration in the prediction vector is added to calculate the precise arrival time of the future high tide event; at the same time, the low tide level and high tide level prediction values ​​in the prediction vector are directly used as the tidal height values ​​of the future event.

[0071] It should be noted that steps S1 and S2 above describe the process of collaborative prediction of the Qiantang River tidal bore during the actual reasoning phase. The collaborative tidal bore prediction model is pre-trained, meaning that the model must be pre-trained before being used for reasoning. The model can be supervised trained using a labeled dataset constructed from historical data. During the training process, the loss function is to minimize the mean absolute error (MAE), mean square error (MSE), or root mean square error (RMSE) between the four-dimensional prediction vector and the four true core numerical feature vectors. AdamW is used as the model optimizer to adjust all learnable parameters within the network, and an early stopping mechanism can be introduced to control the iterative cycle of training.

[0072] It should be noted that the steps of the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition shown in S1 to S2 above can essentially be implemented in the form of a computer program or software function module.

[0073] Therefore, based on the same inventive concept, Figure 6 As shown, the present invention further provides a Qiantang River tidal bore collaborative prediction system based on reconstruction and decomposition corresponding to the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition provided in the above embodiment, which includes:

[0074] Data input module, used to input the original time series data of historical tidal events in Qiantang River;

[0075] The collaborative prediction module is used to generate prediction results based on the input original time series data of historical tidal events of the Qiantang River and the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described above;

[0076] The data output module is used to output the generated prediction results according to the preset output method.

[0077] It should be noted that the above-mentioned data input module and data output module can provide data input and output control functions in the form of a GUI interface. The specific prediction result output method can be designed according to actual functional needs. For example, the results can be stored locally in the form of a document, or uploaded to a cloud server, or can be directly displayed locally.

[0078] In addition, based on the same inventive concept, Figure 7 As shown, the present invention also provides a computer electronic device corresponding to the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition provided in the above embodiment, which includes a memory and a processor;

[0079] The memory is used to store computer programs;

[0080] The processor is configured to implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described above when executing the computer program;

[0081] Furthermore, the logic instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention.

[0082] Therefore, based on the same inventive concept, the present invention provides a computer-readable storage medium corresponding to the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition, and the storage medium stores a computer program. When the computer program is executed by the processor, it can implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described above.

[0083] Therefore, based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described above.

[0084] Specifically, in the computer-readable storage medium of the above three embodiments, the stored computer program is executed by the processor to perform the above steps S1 to S2.

[0085] It is understood that the storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage medium may be any medium capable of storing program code, such as a USB flash drive, a mobile hard drive, a magnetic disk, or an optical disk.

[0086] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it 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 gate or transistor logic devices, and discrete hardware components.

[0087] It should also be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division. In actual implementation, there may be other division methods, for example, multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0088] The present invention will further illustrate, through a specific embodiment, the application of the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition shown in steps S1 to S2 above to a specific example, so as to demonstrate the construction, training process and test results of its specific tidal bore collaborative prediction model.

[0089] Example

[0090] In this embodiment, the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition includes the following steps:

[0091] Step 1: Obtain and reconstruct the Qiantang River tidal bore dataset.

[0092] First, we obtain raw time series data recording historical tidal events on the Qiantang River. This data includes the arrival timestamp, tide type, and tide level of each tidal event. In this example, we use years of tidal event records from the Cangqian Hydrological Station on the Qiantang River as the raw dataset. First, we clean the data and use linear interpolation to complete a small number of missing events.

[0093] Then, a series of high tide event pairs are constructed by pairing each low tide event in the cleaned and completed time series data with the first high tide event that follows it in time.

[0094] Subsequently, for each high tide event pair in the series, a multidimensional numerical feature vector is calculated and generated. The six variable dimensions contained in this multidimensional numerical feature vector are [low tide level, high tide level, high tide duration, tidal cycle, day within the year, and lunar day]. The respective feature extraction methods are as follows: the low tide level and high tide level of the high tide event pair are used as the first and second features respectively; the difference between the timestamps of the high tide event and the low tide event in the high tide event pair is used as the third feature to characterize the high tide duration; the difference between the timestamps of the low tide event of the next high tide event pair and the low tide event of the current high tide event pair is used as the fourth feature to characterize the tidal cycle; and the day within the year and the lunar day corresponding to the timestamp of the low tide event of the current high tide event pair are used as the fifth and sixth features respectively.

[0095] Finally, the multidimensional numerical feature vectors calculated for all high tide event pairs are arranged in the chronological order of their corresponding event pairs, thereby generating a regularized multivariate time series that can be processed by the model. In a multivariate time series, there is a univariate time series corresponding to each variable dimension. Because the characteristic numerical ranges of different variable dimensions vary greatly, in order to eliminate the dimensionality effect, the maximum and minimum value normalization method is used to preprocess each univariate time series. The processing formula is as follows:

[0096]

[0097] in, To normalize the data, is the original data of different input factors, are the maximum and minimum values ​​of the corresponding input factors, respectively.

[0098] Finally, to construct the dataset for model training, we used conventional training sample construction methods to construct a series of training samples. Each training sample contains a multivariate time series within the historical window and a four-dimensional prediction vector for each high tide event pair within the prediction window. The four-dimensional prediction vector includes the low tide level, high tide level, high tide duration, and tidal period. The training sample dataset is divided into training, validation, and test sets in a ratio of 7:2:1.

[0099] Step 2: Build a collaborative tidal bore prediction model

[0100] The tidal bore collaborative prediction model constructed in this embodiment has a detailed structure as follows: Figure 2 As shown in Figure 2, the input of the model is a multivariate time series. The model consists of multiple stacked dynamic local-global learning blocks (DLG Blocks). The internal structure of each DLG block is as follows: Figure 3 As shown in Figure 2, each consists of an adaptive period decomposition module, a multi-scale temporal variation learning module, an adaptive aggregation module, and a multivariate interaction aggregation module. The final DLG module is then connected to a prediction decoder, which outputs a four-dimensional core feature prediction vector for one or more high tide event pairs within the future prediction window.

[0101] In this embodiment, the function of the adaptive period decomposition module is to adaptively decompose the input multivariate time series into multiple period components. Its detailed structure and calculation process are as follows: Figure 4 As shown in Figure 2. The core of this module is an organic combination of a learnable maximum overlap discrete wavelet transform (MODWT) and multi-resolution analysis (MRA). The MODWT operator uses a learnable parameter matrix as a filter to receive the input multivariate time series and adaptively deconstruct each univariate time series into multiple periodic components representing different frequency characteristics through multi-level wavelet decomposition. Specifically, the data processing flow within the MODWT operator can be described as follows:

[0102] First, for an input univariate time series , MODWT uses a pair of high-pass wavelet filters and low-pass scaling filters Perform wavelet decomposition step by step to obtain detail coefficients (high frequency part) and scale coefficients (low frequency part), and then use the coefficients at each level and the univariate time series Generate initial periodic components representing different frequency characteristics. The total number of wavelet decomposition levels is denoted as J, and the specific value can be optimized according to actual conditions. In this embodiment, J=6. In the level decomposition, the detail coefficient and scale coefficient The calculation process can be expressed as:

[0103]

[0104]

[0105] in, is the original input univariate time series , It is a filter and length, and Filter and No. dimension, is a sequence length, is the time step. The difference between this invention and the traditional MODWT is that the original fixed filter and The coefficients of are set as the learnable parameter matrix, that is, the filter and They are respectively set as a learnable parameter matrix, and the parameters in these two matrices are automatically optimized through back propagation during model training, so that the decomposition method can be specialized for the unique periodicity of Qiantang River tidal data.

[0106] After obtaining the initial periodic components at each level, this embodiment introduces a mask generation mechanism to adaptively enhance or suppress the information of certain periodic components. This mechanism uses a small neural network to generate a mask based on the initial periodic components. Its own characteristics generate a corresponding mask for each decomposed periodic component , and using the mask Adaptively adjust each initial period component. Generate mask The process of adjusting the periodic component can be expressed as:

[0107]

[0108]

[0109] in, are the learnable weight matrices and bias vectors in the neural network, is the Gaussian error linear unit activation function, is the Sigmoid activation function, represents the element-wise product of matrices, Represents the adjusted periodic component.

[0110] Finally, the adjusted periodic components are reconstructed by multi-resolution analysis (MRA), and the adjusted wavelet coefficients are reconstructed back to the time domain. and scale coefficient Perform inverse transform and reconstruct the detail component aligned with the original sequence. and Smooth Component .These That is, the final output is multiple periodic components representing different frequency characteristics, which represents the reconstruction result , and are sent to subsequent modules for learning.

[0111] In this embodiment, the structure of the multi-scale temporal variation learning module is as follows: Figure 5 As shown, its input is each periodic component (in is the sequence length), this module can extract deep temporal features from each periodic component in parallel, thereby performing global and local feature learning in parallel. Therefore, the internal structure of this module includes a global variation learning branch (Global Variations Learning) and a local variation learning branch (Local Variations Learning).

[0112] The global change learning branch aims to efficiently capture long-range dependencies. The core steps within the branch include linear mapping, overlapping blocks, two-dimensional stacking, block self-attention, and linear mapping reduction. The specific processes are as follows:

[0113] Linear mapping: convert the periodic component of the input Use a learnable weight matrix to map to the feature space and implement the mapping through the input linear layer;

[0114] Overlapping Patching: First, the time dimension of the mapped periodic component sequence is divided into two parts according to the preset patch length. and stride , divided into Overlapping time slices (Patches). The calculation formula is .

[0115] 2D stacking: The length is The time segments are stacked along the new dimension to form a two-dimensional feature matrix .

[0116] Patch Attention: Next, the two-dimensional matrix It is sent to a standard self-attention module, fused based on the self-attention mechanism, and then input into the feedforward network (FFN). Note that the self-attention mechanism here calculates this The correlation between the "blocks", rather than the original The core formula for calculating the correlation between time points is:

[0117]

[0118] in, They are respectively The query, key and value matrices are obtained after different linear mappings. , the computational complexity of this mechanism is reduced from Significantly reduced to , which can efficiently process long historical series and capture long-range dependencies such as annual periodicity.

[0119] Linear mapping restoration: After the output of attention calculation and feed-forward network (FFN), the output features are flattened and finally mapped through a linear layer, thereby mapping it from the two-dimensional block space back to the one-dimensional time series space to obtain the global features .

[0120] The above local variation learning branch directly uses the original one-dimensional periodic component The core idea of ​​this method is to use multiple parallel separable causal convolution layers with different dilation rates to extract local temporal dependencies from different receptive field scales.

[0121] The specific processing flow is as follows:

[0122] In order to capture local patterns of different ranges, a multi-scale strategy is introduced. In the local change learning branch, A causal convolutional layer that transforms the periodic component of the input Pass multiple separable causal convolutional layers with different dilation rates in parallel. The dilation rate of each layer is Increasing exponentially, for example ,in Lower dilation rates (such as 2, 4) focus on very local changes, while higher dilation rates (such as 8, 16) can capture temporal dependencies over a wider range. The convolution operation at each scale uses a one-dimensional separable causal convolution, which follows two properties:

[0123] Causality: To ensure the rigor of time series prediction, that is, the output at time step t depends only on the input at time t and before, causal convolution is used. This is achieved by asymmetric padding of the input sequence during the convolution operation.

[0124] Separability: To improve the efficiency of the model and reduce the number of parameters, this embodiment adopts the idea of ​​depthwise separable convolution. That is, the standard convolution is decomposed into depthwise convolution and pointwise convolution. Here, this embodiment allows each input channel to use its own convolution kernel for calculation independently (that is, the number of groups is equal to the number of channels), which greatly reduces the computational complexity. For any scale of causal convolution layer, its input The detailed processing steps are as follows:

[0125]

[0126] in, : Represents a one-dimensional dilated separable causal convolution operation. : It is a layer normalization operation used to stabilize the training process and accelerate convergence. : is a key pruning operation. Since the asymmetric padding of causal convolution will produce redundant outputs at the end of the sequence, The operation will precisely trim these redundant values ​​to ensure that the length of the output sequence is the same as the input sequence. Strict equality. : is the Gaussian error linear unit activation function. : represents the expansion rate The extracted features.

[0127] This embodiment will The features obtained after processing with convolutional layers of different scales (i.e. different expansion rates) The fusion is performed to form the final local features. The fusion method is not direct addition, but an averaging strategy: first, the features extracted from all scale branches are stacked in a new dimension. Then, the mean of these stacked features in the scale dimension is calculated to obtain a more robust local change representation that integrates all scale information. .

[0128]

[0129] Finally, in order to integrate the original period information and enhance the generalization ability of the model, the fused local features Pass through a Dropout layer and then add back to the original input using a residual connection superior.

[0130]

[0131] in: : It is a regularization technique used to prevent the model from overfitting. : It is the final output periodic component sequence that contains rich local dynamic information.

[0132] For each periodic component under each variable dimension, its output in both global and local branches is and Finally, they need to be fused by element-by-element addition to obtain the fused feature vector of the periodic component. .

[0133] In the above adaptive aggregation module, the fusion feature vector of all periodic components of each variable dimension is first Combined into a two-dimensional tensor Then input the two-dimensional convolutional network to learn the interaction pattern between different periodic components in 2D space and obtain the same tensor as the input Two-dimensional convolution results of the same dimension , two-dimensional convolution results The convolution feature vectors of each row in the _____ still correspond to each periodic component one by one, so the two-dimensional convolution result can be decomposed row by row into the convolution feature vectors corresponding to each periodic component, and then the convolution feature vectors corresponding to all periodic components are weighted and aggregated with the component variance of each periodic component as the weight to obtain the final feature representation of the current variable dimension. . It represents the time dependence pattern of a single series. It can be seen that when the MODWT operator decomposes the multivariate time series into multiple periodic components, it also needs to calculate the variance of each periodic component, which is used as a weight to reweight each periodic component to enhance the influence of important periodic components in subsequent processing.

[0134] It should be noted that the six variable dimensions of low tide level, high tide level, high tide duration, tidal cycle, day of the year, and lunar day all need to pass their corresponding single variable time series through the above-mentioned adaptive period decomposition module, multi-scale time series change learning module and adaptive aggregation module to obtain the final feature representation of each variable dimension. The final feature representation of 6 variable dimensions It will be further input into the multivariate interaction aggregation module for interactive learning.

[0135] In this embodiment, the multivariate interaction aggregation module represents the final feature of the six variable dimensions Splice into a two-dimensional matrix For this feature group containing 6 variable dimensions , can be averaged along the time dimension Compress to vector , the vector The linear layer converts the vector into a 6-dimensional weighted vector, which represents the weighted weights of the final feature representations of the six variable dimensions. The final feature representations of the six variable dimensions are then weighted based on the weighted weights. The weighting process can be expressed as:

[0136]

[0137]

[0138] in, is the interactive aggregation feature finally output by the dynamic local-global learning module, are the learnable weights and biases corresponding to the group, is the activation function.

[0139] The last module of the tide-boom collaborative prediction model in this embodiment is the prediction decoder, which can be implemented by a fully connected network. It receives the interactive aggregation features output by the last dynamic local-global learning module and maps them into a four-dimensional prediction vector for each tide event pair in the prediction window through a fully connected layer. , that is, it predicts the low tide level, high tide level, high tide duration and tidal period for each high tide event pair.

[0140] The above-mentioned tidal bore collaborative prediction model needs to be supervised and trained before being used for actual reasoning. The specific training and reasoning methods are described below.

[0141] Step 3: Model training and inference

[0142] During the training phase, this example uses data from 96 historical event pairs (approximately 1.5 months) as input to predict the four core features of the next 24 event pairs (approximately 12 days). The mean squared error (MSE) loss function is used to calculate the model's loss to assess the difference between the model's predictions and the real data. The calculation formula is as follows:

[0143]

[0144] in, is the number of future high tide events predicted in a single time (in this embodiment ), Traverse the C (C=4) core feature dimensions that need to be predicted (low tide level, high tide level, high tide duration, tidal cycle), Represents the future prediction window The first high tide event The real data of the core features is the label value, It represents The corresponding model prediction data.

[0145] At the same time, this embodiment uses AdamW as the model optimizer to adjust all learnable parameters within the network to minimize the loss function value, thereby optimizing prediction performance. In addition, to prevent model overfitting and ensure the effectiveness of training, early stopping technology is adopted. This technology controls the training process by monitoring the loss value on the validation set. If the loss on the validation set does not decrease significantly over a certain number of consecutive training cycles (epochs), it can be considered that the model has begun to overfit or is no longer learning new patterns. At this time, training will be stopped. This method not only saves computing resources and avoids meaningless training time, but also improves the generalization ability of the model to a certain extent and more accurately controls the training process, ensuring that the model achieves the best prediction effect without failure due to overfitting.

[0146] After training, the tidal bore collaborative prediction model can be used for inference on a test set or a time series that actually needs to be predicted. During inference, a sequence of historical "high tide event pairs" prior to the target period can be constructed using the same data reconstruction method as the training samples. This sequence is then input into the pre-trained tidal bore collaborative prediction model to obtain a four-dimensional feature prediction vector for the future event pair. Finally, the future tidal bore height and arrival time are obtained through the restoration method described in S2.

[0147] Step 4: Performance comparison and evaluation

[0148] This example compares the performance of this collaborative tidal bore prediction model (referred to as the inventive model) with a traditional Transformer model on the same test set. Table 1 (to be generated later) shows the comparison of the two models across several key metrics, including mean absolute error (MAE) and root mean square error (RMSE).

[0149] Table 1 Comparison of indicators between the proposed model and the Transformer model

[0150]

[0151] and Figure 8 and Figure 9The superiority of the model of the present invention is intuitively demonstrated from the two perspectives of prediction examples and error distribution. Figure 8 A specific prediction example is shown. By comparison, it can be seen that the prediction curve of the model of the present invention can closely follow the fluctuation of the true value, while the prediction curve of the Transformer model has obvious error accumulation phenomenon. Figure 9 The error distribution of the model of the present invention is shown in the histogram. By comparison, the error distribution of the model of the present invention is highly concentrated near 0, indicating that most predicted values ​​are very close to the true values. The shape of the error distribution is close to the normal distribution, indicating that the prediction errors have good symmetry and no obvious skewness.

[0152] In summary, the Qiantang River tidal bore collaborative prediction method based on event-pair reconstruction and adaptive periodic decomposition, developed in this paper, transforms irregular tidal event streams into regularized multivariate time series through novel "tidal event pair" data reconstruction. Combined with an advanced model capable of adaptively decomposing periodic patterns, deeply learning multiscale temporal dynamics, and performing multivariate interaction, it ultimately achieves collaborative prediction of tidal bore height and arrival time. This method not only fundamentally addresses the difficulty traditional models have in processing irregular event data, but its non-autoregressive prediction mechanism also avoids error accumulation, demonstrating significant superiority in both prediction accuracy and reliability, providing important decision-making support for disaster prevention and mitigation, engineering management, and other related work.

[0153] The embodiment described above is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Persons skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent substitution or equivalent transformation falls within the scope of protection of the present invention.

Claims

1. A collaborative prediction method for Qiantang River tidal bore based on reconstruction and decomposition, characterized by: include: S1. Reconstruct the original time series data of historical tidal events in the Qiantang River. Pair each low tide event with the first high tide event that follows it to form a high tide event pair. Extract the features of each high tide event pair in six variable dimensions: low tide level, high tide level, high tide duration, tidal cycle, day within the year, and lunar day. Normalize the features and obtain a multivariate time series consisting of six univariate time series. S2. Input the multivariate time series into the trained tidal collaborative prediction model, and pass through multiple stacked dynamic local-global learning modules and prediction decoders in sequence. Each dynamic local-global learning module first inputs the univariate time series of each variable dimension into the adaptive period decomposition module to obtain multiple periodic components representing different frequency characteristics, and then inputs each periodic component into the multi-scale temporal change learning module. The global features and local features are extracted from each periodic component in parallel and then fused. The fused feature vectors of all periodic components are then input into the adaptive aggregation module for two-dimensional convolution and reweighting to obtain the final feature representation of each variable dimension. Finally, the final feature representation of the six variable dimensions is input into the multivariate interactive aggregation module for interactive aggregation between the variable dimensions, and the interactive aggregation features are used as the module output; the prediction decoder receives the interactive aggregation features output by the last dynamic local-global learning module, and maps them to the low tide level, high tide level, tide duration and tidal period of the tide event pair in the prediction window through the fully connected layer, and then calculates the tide level and time of each tidal event in the prediction window in sequence.

2. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: The feature extraction method for the six variable dimensions of each high tide event pair is as follows: extract the low tide level and high tide level from the low tide event and high tide event in the current high tide event pair respectively, calculate the difference between the timestamps of the high tide event and the low tide event in the current high tide event pair as the high tide duration, take the difference between the timestamps of the low tide event in the next high tide event pair and the low tide event in the current high tide event pair as the tidal period, take the ordinal number of the low tide event timestamp in the current high tide event pair in the Gregorian calendar year as the day of the year, and take the ordinal number of the low tide event timestamp in the current high tide event pair in the lunar month as the lunar day.

3. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: In the adaptive periodic decomposition module, a maximum overlap discrete wavelet transform (MODWT) operator with a learnable parameter matrix as a filter is first used to perform multi-level wavelet decomposition on the univariate time series of each variable dimension, adaptively deconstructing it into multiple initial periodic components. A neural network is then used to generate a mask for each initial periodic component and the mask is used to adaptively adjust each initial periodic component. Finally, the adjusted periodic components are reconstructed through multi-resolution analysis to obtain multiple periodic components representing different frequency characteristics as the final output.

4. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: The multi-scale temporal change learning module includes a global change learning branch and a local change learning branch: In the global change learning branch, the input periodic component is first linearly mapped, and then the linearly mapped periodic component is divided into multiple overlapping time segments. All time segments are then combined into a two-dimensional matrix and sequentially passed through a self-attention layer and a feedforward network to obtain a two-dimensional spatial feature. The two-dimensional spatial feature is flattened and linearly mapped back to a one-dimensional time series space to obtain a global feature. In the local variation learning branch, the input periodic component is passed through multiple separable causal convolutional layers with different dilation rates in parallel, and then the convolution results are fused and superimposed on the input periodic component in a residual connection manner to form local features; Finally, the global features and local features are further fused to obtain the fused feature vector of the current input periodic component.

5. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: In the adaptive aggregation module, the fused feature vectors of all periodic components of the current variable dimension are first combined into a two-dimensional tensor and then input into the two-dimensional convolutional network. The two-dimensional convolution result is then re-decomposed into the convolution feature vectors corresponding to each periodic component, and then weighted aggregation is performed using the component variance of each periodic component as the weight to obtain the final feature representation of the current variable dimension.

6. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: In the multivariate interaction aggregation module, the final feature representations of the six variable dimensions are spliced ​​into a two-dimensional matrix, then compressed into a vector by averaging along the time dimension, and then converted into the weighted weights of the final feature representations of the six variable dimensions through a linear layer and weighted aggregation is performed to obtain the interactive aggregation features as the final output of the current dynamic local-global learning module.

7. The Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition according to claim 1, characterized in that: The tidal bore collaborative prediction model needs to be pre-supervised trained before being used for actual reasoning, and its loss function adopts mean square error loss, root mean square error loss or mean absolute error loss.

8. A Qiantang River tidal bore collaborative prediction system based on reconstruction and decomposition, characterized by: include: Data input module, used to input the original time series data of historical tidal events in Qiantang River; A collaborative prediction module, configured to generate a prediction result according to the reconstruction and decomposition-based collaborative prediction method for the Qiantang River tidal bore according to any one of claims 1 to 7, based on the input original time series data of the historical tidal events of the Qiantang River; The data output module is used to output the generated prediction results according to the preset output method.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as claimed in any one of claims 1 to 7 can be implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the Qiantang River tidal bore collaborative prediction method based on reconstruction and decomposition as described in any one of claims 1 to 7 when executing the computer program.

Images