Intelligent prediction method and device for tidal level of tidal river section

Abstract

This application discloses an intelligent method and device for predicting tide levels in tidal river sections, relating to the field of water conservancy engineering technology. The method includes: constructing a non-steady-state harmonic analysis model and a DRSN-LSTM model; coupling the non-steady-state harmonic analysis model and the DRSN-LSTM model to predict the tide level trend in a study area, obtaining the target prediction results for the study area. Utilizing the advantage of the non-steady-state harmonic analysis model in simulating long-term tide level changes and the characteristic of DRSN-LSTM in capturing the influence of multiple factors on the magnitude of tide level changes, this method improves the insufficient prediction accuracy of traditional methods under conditions of abnormal weather, artificial prediction influences, and long lead times. Furthermore, considering the characteristics of the non-steady-state harmonic analysis model and the DRSN-LSTM model, the coupling method between the two is optimized, effectively improving the accuracy of tide level prediction.

Description

Technical Field

[0001] This application relates to the field of water conservancy engineering technology, and in particular to a method and device for intelligent prediction of tide levels in tidal river sections. Background Technology

[0002] Tidal river sections are located in the most economically vibrant areas along the Yangtze River and coast, serving as both engines of economic development and key targets for ecological and environmental protection. Accurate forecasting of tidal river levels is crucial for ensuring regional flood control, water supply, ecological stability, and navigation. The Jiangsu section of the Yangtze River exhibits typical characteristics of tidal river sections. East China Sea tides propagate upstream from the Wusongkou, but are influenced by topography, changes in river width, artificial water diversion, and runoff from the upper reaches of the Yangtze, making the tides less stable than ocean tides. During the flood season, tidal river sections are affected by a combination of factors, including local floods, torrential rains, and typhoon-induced water level increases, potentially leading to sudden rises and falls in water levels. Due to the complex mechanisms and variable influencing factors of the hydrological processes in tidal river sections, tidal level changes exhibit irregular amplitudes on top of periodicity, increasing the difficulty of forecasting.

[0003] Harmonic analysis is a commonly used method for tide level forecasting. It assumes that tides are a linear combination of sine and cosine functions, but actual tides are influenced by various nonlinear factors, such as the interaction between tides and river runoff, and the interaction between tides and wind. By constructing hydrodynamic models, the nonlinear interactions between multiple influencing factors and tides can be comprehensively considered. This allows for the forecasting of not only estuary water levels under the influence of runoff and tides, but also storm tides under extreme weather conditions such as typhoons. However, constructing hydrodynamic models requires comprehensive meteorological, hydrological, topographical, and hydraulic engineering data, and fine-grained grid partitioning is needed to improve simulation accuracy, thus increasing computational resource requirements. Machine learning, with its advantages in handling large-scale data and complex nonlinear relationships, is widely used in hydrological time series forecasting. Methods such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs) can effectively mine and utilize patterns and relationships in historical data, providing an efficient data-driven approach for tide level forecasting. However, machine learning lacks consideration of the physical mechanisms of tides and runoff, making it difficult to learn related seasonal and periodic characteristics, which makes it difficult for this type of method to simulate the long-term complex changes in tide levels. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and device for intelligent prediction of tidal levels in tidal river sections that can improve the accuracy of tidal level prediction, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides an intelligent method for predicting tide levels in tidal river sections. The method includes:

[0006] Based on real-time tidal level data of the study area, upstream runoff, and offshore tides, a nonsteady harmonic analysis model is constructed.

[0007] Based on tidal data, water level data, meteorological data, and engineering data of the study area, DRSN-LSTM models for non-flood season and flood season are constructed.

[0008] The unsteady harmonic analysis model and the DRSN-LSTM model are coupled to predict the tidal trend in the study area, and the target prediction results for the study area are obtained.

[0009] Among them, the tide trend prediction includes the prediction of the current time node, the prediction of the previous time node, and the change prediction determined by the prediction of the current time node and the previous time node.

[0010] Obtaining target prediction results for the study area includes:

[0011] Based on the previous time point predictions of the unsteady harmonic analysis model and the DRSN-LSTM model, obtain the tidal baseline value; determine the tidal change trend based on the change prediction of the unsteady harmonic analysis model, and obtain the tidal change amplitude based on the change prediction of the DRSN-LSTM model; determine the target prediction result based on the tidal baseline value, tidal change trend, and tidal change amplitude.

[0012] In one embodiment, the method further includes:

[0013] A baseline coefficient and an amplitude coefficient are introduced. The baseline coefficient is used to adjust the influence of the previous time node prediction of the unsteady harmonic analysis model and the DRSN-LSTM model on the tidal baseline value. The amplitude coefficient is used to adjust the influence of the tidal level change amplitude on the target prediction result.

[0014] In one embodiment, the target prediction result is represented as:

[0015]

[0016] Where Wt represents the target prediction result at time t. This represents the prediction for the current time point based on the non-steady harmonic analysis model. This indicates the prediction at the previous time point based on the non-steady-state harmonic analysis model. This represents the prediction for the current time point based on the DRSN-LSTM model. This indicates the prediction at the previous time point based on the DRSN-LSTM model, where α and β represent the baseline coefficient and amplitude coefficient, respectively.

[0017] In one embodiment, the method further includes:

[0018] An improved genetic algorithm was used to optimize the baseline coefficient and amplitude coefficient to obtain the optimal baseline coefficient and optimal amplitude coefficient for different periods.

[0019] In one embodiment, the improved genetic algorithm includes:

[0020] Based on the genetic algorithm, an adaptive operator is introduced to optimize the crossover probability and mutation probability in the genetic algorithm;

[0021] The adaptive operator is determined based on the number of iterations, population size, fitness value of individuals in the population, and average fitness value of individuals in the population.

[0022] In one embodiment, the method further includes:

[0023] Based on the harmonic components decomposed by the nonsteady harmonic model, combined with spatial autocorrelation analysis, the spatial correlation between each harmonic component and the study area is captured, and the overall spatial distribution characteristics of tides in the study area are obtained according to the spatial correlation of each harmonic component.

[0024] The decay rate of tidal levels is obtained based on the overall spatial distribution characteristics, and continuous target prediction results are generated based on the decay rate.

[0025] In one embodiment, after obtaining the tidal decay rate based on the overall spatial distribution characteristics, the method further includes:

[0026] The decay rate is evaluated and corrected using the DRSN-LSTM model, and continuous target prediction results are generated based on the corrected decay rate.

[0027] Secondly, this application also provides an intelligent tidal level prediction device for tidal river sections, the device comprising:

[0028] The first model building module is used to construct a nonsteady harmonic analysis model based on real-time tidal level data of the study area, upstream runoff, and offshore tides.

[0029] The second model building module is used to build DRSN-LSTM models for non-flood season and flood season based on tidal data, water level data, meteorological data and engineering data of the study area.

[0030] The prediction output module is used to couple the unsteady harmonic analysis model and the DRSN-LSTM model to predict the tidal trend of the study area and obtain the target prediction results of the study area.

[0031] Among them, the tide trend prediction includes the prediction of the current time node, the prediction of the previous time node, and the change prediction determined by the prediction of the current time node and the previous time node.

[0032] Obtaining target prediction results for the study area includes:

[0033] Based on the previous time point predictions of the unsteady harmonic analysis model and the DRSN-LSTM model, obtain the tidal baseline value; determine the tidal change trend based on the change prediction of the unsteady harmonic analysis model, and obtain the tidal change amplitude based on the change prediction of the DRSN-LSTM model; determine the target prediction result based on the tidal baseline value, tidal change trend, and tidal change amplitude.

[0034] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described intelligent tidal level prediction method for tidal river sections.

[0035] Fourthly, this application also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps in the above-described intelligent tidal level prediction method for tidal river sections.

[0036] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in the above-described intelligent tidal level prediction method for tidal river sections.

[0037] The aforementioned intelligent tidal level prediction method and device for tidal river sections constructs a non-steady-state harmonic analysis model based on real-time tidal level data from the study area, upstream runoff, and offshore tides. Based on tidal level data, water level data, meteorological data, and engineering data from the study area, DRSN-LSTM models for both the non-flood season and the flood season are constructed. The non-steady-state harmonic analysis model and the DRSN-LSTM model are coupled to predict tidal level trends in the study area, obtaining the target prediction results. By leveraging the advantage of the non-steady-state harmonic analysis model in simulating long-term tidal level trends and the ability of DRSN-LSTM to capture the influence of multiple factors on the amplitude of tidal level changes, this method improves the insufficient forecast accuracy of traditional methods under conditions of abnormal weather, artificial prediction influences, and long lead times. Furthermore, considering the characteristics of the unsteady harmonic analysis model and the DRSN-LSTM model, the coupling method between the two was optimized. Based on the prediction of the previous time node corresponding to the unsteady harmonic analysis model and the DRSN-LSTM model, the tidal level reference value is obtained; the tidal level change trend is determined based on the change prediction of the unsteady harmonic analysis model; and the tidal level change amplitude is obtained based on the change prediction of the DRSN-LSTM model. Based on the tidal level reference value, the tidal level change trend, and the tidal level change amplitude, the target prediction result is determined. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of a DRSN in one embodiment;

[0039] Figure 2 This is a schematic diagram of an LSTM implementation;

[0040] Figure 3 This is a flowchart illustrating an intelligent tidal level prediction method for tidal river sections in one embodiment.

[0041] Figure 4 This is a schematic diagram showing the distribution of the study area and hydrological stations in one embodiment;

[0042] Figure 5 This is a comparison chart of the predicted and measured values ​​of the T_TIDE and NS_TIDE models in one embodiment;

[0043] Figure 6 This is a comparison chart of the NS_TIDE model forecast and the measured values ​​during a typhoon in one embodiment.

[0044] Figure 7 The prediction error of the DRSN-LSTM model under different lead times in one embodiment;

[0045] Figure 8 This is a comparison chart of the DRSN-LSTM model forecast and the measured values ​​during a typhoon in one embodiment.

[0046] Figure 9 This represents the prediction error of the coupled model under different lead times in one embodiment;

[0047] Figure 10 This is a comparison chart of the predicted and measured values ​​of the coupled model under a 24-hour forecast period in one embodiment. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] This application provides an intelligent method for predicting the tide level in a tidal river section, including the following steps:

[0050] Step 102: Based on real-time tidal level data of the study area, upstream runoff, and offshore tides, construct a non-steady-state harmonic analysis model.

[0051] The harmonic analysis model, based on physical and mathematical principles, identifies and quantifies the main periodic components constituting tidal phenomena, corresponding to the tidal cycle caused by the relative motion of the Earth, Moon, and Sun. It decomposes the complex periodic tidal signal into a series of tidal constituents, transforming it into a simple sum of sine and cosine waves. Using historical tidal data, the harmonic constant of each tidal constituent is obtained and used to predict the tides on any given date. The nonstationary tidal harmonic analysis model (NS_TIDE) is an improved model based on the traditional harmonic analysis model (T_TIDE), considering the nonlinear effects of runoff and open-sea tidal range. NS_TIDE, based on the framework of the T_TIDE model, considers the changes in mean water level, tidal constituent amplitude, and phase over time in its tidal expression.

[0052] The formula for the unsteady harmonic analysis model is expressed as follows:

[0053]

[0054] In the formula, η(t) is the tidal level, n is the total number of tidal fractions, and σ k Let ηk be the angular frequency of the k-th tidal constituent, and t be the time node; η0 is the subtidal module, representing the average water level fluctuation of the low-frequency tidal constituent, and the second term on the right is the runoff tidal module; η0, C k and S k These are variables related to runoff and tidal range, and their corresponding formulas are as follows:

[0055]

[0056] In the formula, Q R (t) represents the runoff value after low-pass filtering at the upstream runoff hydrological station, R S (t) represents the tidal range at the downstream reference tide station; (p) s ,q s ,r s ) and (p f ,q f ,r f ) represent the unknown index values ​​in the subtidal and runoff tidal modules of the NS_TIDE model, respectively; t Q and t R S represents the estimated time for the runoff from the upstream hydrological station and the tidal wave from the downstream tidal station to propagate to the target station, respectively; i c k i and s k ,i (i=0~2) are unknown coefficients.

[0057] Step 104: Based on the tidal data, water level data, meteorological data, and engineering data of the study area, construct DRSN-LSTM models for the non-flood season and the flood season.

[0058] To address the spatiotemporal characteristics of multiple influencing factors on tide levels, a data-driven tide level forecasting model, DRSN-LSTM, is constructed based on DRSN (Deep Residual Shrinkage Network) and LSTM (Long Short-Term Memory). This model possesses the ability to extract spatial features using convolutional networks and temporal features using recurrent neural networks.

[0059] To address the issues of gradient vanishing and inability to extract key features inherent in Convolutional Neural Networks (CNNs), this embodiment employs DRSN. Its Residual Shrinkage Building Unit (RSBU) enhances network performance through residual learning and the Squeeze-and-Excitation Networks (SENet). The principle is as follows: Figure 1 As shown, residual learning uses identity mapping operations to directly input features from shallow networks into deep networks, solving the gradient vanishing and network degradation problems caused by increasing network depth. SENet uses a soft thresholding function to set redundant feature information to 0, retaining important features and improving the feature extraction capability for high-dimensional data.

[0060] LSTM is a special type of recurrent neural network (RNN), and its principle is as follows: Figure 2 As shown. LSTM uses the basic RNN structure, based on the input x at time t. t and the hidden state h at time t-1 t-1 To calculate the output y at time t t and hidden state h t Unlike RNNs, LSTMs add gating structures to the hidden layers. By combining short-term and long-term memories, they mitigate the problems of vanishing or exploding gradients, enabling them to learn the short-periodic temporal patterns of multiple influencing factors of tide levels.

[0061] Step 106: Couple the unsteady harmonic analysis model and the DRSN-LSTM model to predict the tidal trend in the study area, and obtain the target prediction results for the study area.

[0062] After analyzing the characteristics of the nonsteady harmonic analysis model and the DRSN-LSTM model under flood season, non-flood season and abnormal weather scenarios, the prediction results of the two models on tide level are coupled to obtain the target prediction result.

[0063] The tidal trend prediction includes the prediction at the current time node, the prediction at the previous time node, and the change prediction determined by the prediction at the current time node and the prediction at the previous time node. In step 106, obtaining the target prediction results for the study area includes: obtaining the tidal baseline value based on the previous time node prediction corresponding to the unsteady harmonic analysis model and the DRSN-LSTM model; determining the tidal change trend based on the change prediction of the unsteady harmonic analysis model; obtaining the tidal change magnitude based on the change prediction of the DRSN-LSTM model; and determining the target prediction result based on the tidal baseline value, the tidal change trend, and the tidal change magnitude.

[0064] Using time t as the current time node and time t-1 as the previous time node, the tide level predictions of the unsteady harmonic analysis model and the DRSN-LSTM model at times t and t-1 are obtained. The change in tide level prediction from time t to time t-1 is taken as the change prediction of the unsteady harmonic analysis model and the DRSN-LSTM model. Here, the difference between the tide level predictions of the unsteady harmonic analysis model at time t and time t-1 can be taken as the change prediction of the unsteady harmonic analysis model at time t, and the difference between the tide level predictions of the DRSN-LSTM model at time t and time t-1 can be taken as the change prediction of the DRSN-LSTM model at time t.

[0065] The tidal level predictions at time t-1 are combined using the unsteady harmonic analysis model and the DRSN-LSTM model to obtain a baseline tidal level. This combination can be a direct addition or a weighted addition, with the weights determined based on the prediction performance of the unsteady harmonic model and the DRSN-LSTM model.

[0066] The tidal level changes are combined using the nonsteady harmonic analysis model and the DRSN-LSTM model to obtain the magnitude and trend of the changes. Given the better long-term performance of the nonsteady harmonic model, the tidal level change trend is obtained based on its predictions; similarly, given the better short-term performance of the DRSN-LSTM model, the tidal level change magnitude is obtained based on its predictions.

[0067] Finally, by combining the tidal baseline value, tidal trend, and tidal range, the target prediction result is determined.

[0068] Traditional tide prediction methods combine the predictions of a mechanism-driven unsteady harmonic model and a data-driven DRSN-LSTM model at the current time point, or use the output of the unsteady harmonic model as the input of the DRSN-LSTM model. Traditional methods combine the advantages of the unsteady harmonic model (strong physical foundation and good interpretability) with the advantages of the DRSN-LSTM model (strong nonlinear modeling ability and adaptability), allowing the two models to complement each other and improve prediction performance. However, traditional tide prediction methods do not consider the prediction differences between the two models over time. This application analyzes and identifies the characteristic that the DRSN-LSTM model has slightly better short-term prediction accuracy than the unsteady harmonic model, but lower prediction accuracy over long-term forecasts than the NS_TIDE model. Therefore, this application specifically improves the combination method of the unsteady harmonic model and the DRSN-LSTM model. The tide baseline value is determined by the predictions of both models at the previous time point, the long-term tide trend is determined by the unsteady harmonic model, and the short-term tide magnitude is determined by the DRSN-LSTM model, thus achieving a more accurate tide prediction effect.

[0069] In one embodiment, the method further includes: introducing a baseline coefficient and an amplitude coefficient, using the baseline coefficient to adjust the influence of the previous time node prediction of the unsteady harmonic analysis model and the DRSN-LSTM model on the tidal baseline value, and using the amplitude coefficient to adjust the influence of the tidal level change amplitude on the target prediction result.

[0070] Different models typically exhibit varying performance; one model may perform well on certain datasets, while another may outperform it on others. By introducing baseline and amplitude coefficients to appropriately allocate the influence of the non-stationary harmonic model and the DRSN-LSTM model on the target prediction results, and by assigning higher weights to the better-performing model in specific scenarios, the overall predictive power of the model can be improved.

[0071] In one embodiment, the target prediction result is represented as:

[0072]

[0073] Where Wt represents the target prediction result at time t. This represents the prediction for the current time point based on the non-steady harmonic analysis model. This indicates the prediction at the previous time point based on the non-steady-state harmonic analysis model. This represents the prediction for the current time point based on the DRSN-LSTM model. This indicates the prediction at the previous time point based on the DRSN-LSTM model, where α and β represent the baseline coefficient and amplitude coefficient, respectively.

[0074] In one embodiment, the method further includes: optimizing the baseline coefficient and amplitude coefficient using an improved genetic algorithm to obtain the optimal baseline coefficient and optimal amplitude coefficient for different periods.

[0075] Genetic Algorithm (GA) is an optimization method derived from the evolutionary process of "survival of the fittest" in nature. In practical applications, the GA algorithm suffers from problems such as being prone to getting trapped in local optima, immature convergence, and slow convergence speed, making it difficult to find the global optimum. To address this problem, a novel adaptive operator is proposed, constructing an improved genetic algorithm. By improving the processes of initial population generation, selection, crossover, and mutation, the performance of the genetic algorithm is effectively improved. The expression of the adaptive operator γ is as follows:

[0076]

[0077] In the formula, t represents the number of generations in the current population, and f k (k = 1, 2, ..., p) represents the fitness value of an individual in the current population, and p is the population size. This represents the average fitness of individuals in the current population. Based on the adaptive adjustment requirements, the crossover probability P during the mutation process is optimized using the adaptive operator γ-crossover. c And the probability of mutation P m The formula is as follows:

[0078]

[0079] In the formula, f' is the larger fitness value among the two individuals to be crossed, and f represents the fitness of the mutant individual. max f represents the maximum fitness of the population. min Let γ be the minimum fitness of the population, γ0 represent the adaptive operator of the initial state, and k1 to k4 be the adaptive control parameters. Studies have shown that through the above optimization, population diversity can be guaranteed and premature convergence avoided in the early stages of population evolution, and the destruction of superior genes can be avoided in the later stages of population evolution, thus improving the convergence and stability of the genetic algorithm.

[0080] This embodiment utilizes an improved genetic algorithm to optimize the baseline coefficient and amplitude coefficient, with the objective function being the root mean square error (MSE) of the prediction result.

[0081] In one embodiment, such as Figure 3 As shown, the intelligent prediction method for tidal levels in tidal river sections mainly includes data preprocessing, NS_TIDE model construction, DRSN-LSTM model construction, and model coupling optimization based on an improved genetic algorithm. The specific steps are as follows:

[0082] (1) The collected hydrological, engineering and meteorological data are checked for consistency and completeness, outliers are deleted, missing data are imputed, and after normalization, the dataset is divided into training set and test set.

[0083] (2) Based on the measured tidal level data of the study area, upstream runoff and offshore tides, the NS_TIDE model was constructed to calibrate and forecast the tidal process; based on tidal level, water level, meteorological and engineering data, the DRSN-LSTM model was constructed. Since the flood season of the river section hub is mainly for flood control and drainage prediction, and ecological water flow prediction is secondary, while the hub gates are kept closed for a long time during the non-flood season, it is necessary to further divide the dataset into two types of datasets: non-flood season and flood season, and train and forecast the DRSN-LSTM model respectively.

[0084] (3) Tide trend prediction by coupling NS_TIDE model and DRSN-LSTM model: The baseline value of tide prediction at time t is obtained by weighted sum of the two model predictions at time t-1; For tide change, the trend of tide change is predicted by NS_TIDE model and the absolute value of tide change amplitude is predicted by DRSN-LSTM model.

[0085] (4) In the process of predicting the tide level trend in step (3), a benchmark coefficient and an amplitude coefficient are introduced. The improved genetic algorithm is used to optimize the benchmark coefficient and the amplitude coefficient, and the optimal solutions of the benchmark coefficient and the amplitude coefficient for different periods are obtained. The final tide level target prediction results are then output.

[0086] After repeated debugging, the DRSN module in the DRSN-LSTM model has a convolution kernel size of 3×1, a stride of 2, and 32 neurons; the LSTM module has 64 neurons, a learning rate of 0.05, all activation functions are ReLU, the number of iterations is 200, and the batch size is 32. The adaptive control parameters of the genetic algorithm are set to k1 = 0.5, k2 = 0.4, k3 = 0.01, and k4 = 0.09.

[0087] This study focuses on the tidal changes at the confluence of the Qinhuai New River and the Yangtze River in Nanjing. The Qinhuai New River Hydropower Project, located at this point, comprises control gates, pumping stations, fishways, and locks, and is responsible for flood control, drainage, irrigation, water environment improvement, and navigation. Therefore, accurate tidal forecasting on the Yangtze River side of the project is crucial for improving its efficiency and guiding refined prediction. Tidal changes are influenced by various factors, including runoff from the upper reaches of the Yangtze River, tides from the outer sea in the lower reaches of the Yangtze River, water diversion and drainage from the Qinhuai New River project, wind direction, wind speed, and rainfall. Using the Qinhuai New River Hydropower Project as a benchmark, this paper uses the tidal level data and the distribution of hydrological stations, as well as the distances between stations, as shown below. Figure 4As shown, Datong Station and Wusongkou Station are approximately 205 km and 350 km from the target area, respectively, with average runoff and tidal propagation times of 19 h and 34 h, respectively. According to the requirements of the NS_TIDE model, upstream runoff hydrological stations and downstream tidal stations need to be determined. Considering that Datong Station is largely unaffected by tides, it was selected as the hydrological reference station. Wusong Station, located at the Yangtze River estuary, is least affected by runoff and was chosen as the tidal reference station. Tiexinqiao Hydrological Station, located approximately [number missing] km upstream of the Qinhuai New River, is an important hydrological control section, directly affecting the discharge flow of the hydropower station during the flood season's flood control and drainage phase.

[0088] The hydrological data measured at Datong Station, Wusongkou Station, Nanjing Station, and Tiexinqiao Station, as well as the Wusongkou tide level forecast data and Tiexinqiao water level forecast data during the flood season and storm surge, were obtained from the Jiangsu Provincial Hydrological Bureau. The wind direction, wind speed, and rainfall data of the research target area were obtained from the Nanjing Meteorological Bureau. The tide level data at the research target location, the engineering data of the Qinhuai River New River Hub, the prediction rules, and other data were obtained from the Jiangsu Provincial Qinhuai River Water Conservancy Project Management Office. The above data mainly uses hourly data from January 2019 to September 2024. The collected data were preprocessed to remove dirty data. Considering that there were not many missing values ​​in the original data, linear interpolation was used to handle outliers and missing values. Data from January 2019 to December 2022 were used as the training set, and data from January to September 2024 were used as the test set. Data from June to September of each year were used as flood season data, and the rest of the time were used as non-flood season data, which were used to train the DRSN-LSTM model. Finally, the root mean square error (RMSE), relative standard deviation (RSD), and average accuracy within ±0.30m are used to measure the prediction errors of various models.

[0089] The NS_TIDE model was used to verify and analyze the hourly tide levels at the confluence of the Qinhuai New River and the Yangtze River. Figure 5 Comparing the predicted and measured values ​​of the T_TIDE and NS_TIDE models, the NS_TIDE model shows a significantly smaller error than the T_TIDE model. In the area where this invention is performed, 185 km upstream, the tidal flow is significantly smaller than the runoff, and the river width variation is not significant, meeting the applicable scenario requirements of the NS_TIDE model. During the validation period, the overall RMSE of the T_TIDE and NS_TIDE models were 0.44 m and 0.21 m, respectively. Because the influence of runoff on tides is considered, the NS_TIDE model has a significantly smaller error than the T_TIDE model, and can better simulate the periodicity and regularity of tidal level changes. Both models show a seasonal variation trend in their errors. The NS_TIDE model still exhibits a large error when the runoff effect is strong, such as during Typhoon Gemi from July 25th to 28th, 2024. Figure 6As shown, its local RMSE is 0.27m. The main reason is that the NS_TIDE model only considers the influence of runoff and tides on the tide level under the average conditions of the past years. During floods or typhoons, meteorological, engineering forecasts and river hydrology factors change in a complex and drastic manner, which have a significant short-term impact on the magnitude of tide level changes. In order to further improve the forecast accuracy, it is necessary to consider the relationship between multiple influencing factors and tide level changes.

[0090] To improve the prediction accuracy of the DRSN-LSTM model, during the training and validation phases, it is necessary to make full use of data such as upstream measured runoff, downstream measured tide levels, river forecast water levels, meteorological and engineering forecast plans within the prediction time range. For example, when making a 24-hour tide level forecast at the current time t, the input should include the tide level data of Datong Station from t-19 to t, the water level data of Datong Station from t-33 to t-10, as well as the 24-hour forecast water level of Tiexinqiao, the Qinhuai New River Hub forecast plan, and meteorological forecast data. Figure 7 The root mean square error (RMSE) of high tide, low tide, and hourly tide forecasts for non-flood season (a) and flood season (b) with forecast periods of 3h, 6h, 12h, 24h, and 36h was statistically analyzed. The evaluation indicators show that the DRSN-LSTM model performs well with forecast periods shorter than 12h, effectively simulating short-term tide level trends. However, the prediction error increases with longer forecast periods.

[0091] During the non-flood season ( Figure 7 a) When influenced less by meteorological and artificial forecasting factors, the main factors affecting tidal level changes are the periodic and regular trends of runoff and tides. Short-term forecast accuracy is slightly better than the NS_TIDE model. However, limited by the performance of the LSTM module, the DRSN-LSTM model struggles to learn long-term nonlinear and non-stationary characteristics, resulting in lower simulation accuracy over long lead times compared to the NS_TIDE model. During the flood season ( Figure 7 b) Under the influence of multiple factors, the DRSN-LSTM model can effectively capture complex nonlinear relationships and simulate the short-term impact of wind direction, wind speed, rainfall, artificial prediction and the inflow of water from the upper reaches of the Qinhuai New River on the tide level. Its short-term forecast accuracy is better than that of the NS_TIDE model.

[0092] like Figure 8The hourly tide level forecasts during Typhoon Gemi, with RMSE values ​​of 0.19m, 0.21m, 0.29m, 0.31m, and 0.33m for different forecast periods, are shown in the graphs. Analysis reveals that the DRSN-LSTM model can effectively simulate the magnitude of tide level changes under the influence of multiple factors; however, in the long-term forecast period, it deviates significantly from the baseline value of the observed tide level. To further improve its long-term forecast accuracy, a coupled model is constructed, leveraging the simulation advantages of the NS_TIDE model in long-term trends, to correct the long-term forecast errors of the DRSN-LSTM model.

[0093] Both the NS_TIDE and DRSN-LSTM models have their advantages in terms of forecast performance. The NS_TIDE model performs better in simulating the trend characteristics of tidal level changes, while the DRSN-LSTM model performs better in simulating the influence of multiple factors on the amplitude of tidal level changes. Leveraging the parameter optimization capabilities of an improved genetic algorithm, the coupled NS_TIDE and DRSN-LSTM model combines the advantages of both models, making the tidal level forecast curve closer to the observed values. Figure 9 The RMSE of high tide, low tide and hourly tide forecasts were statistically analyzed for different forecast periods during non-flood season (a) and flood season (b).

[0094] Compared to the mechanism-based NS_TIDE model and the data-based DRSN-LSTM model, the coupled model's forecast values ​​are closer to the observed values. Its 3-hour hourly forecasts for the non-flood season and flood season have RMSE values ​​of 0.11m and 0.13m, respectively. Accurate high and low tide forecasts play a crucial supporting role in flood control and disaster reduction, as well as water flow prediction. The forecast accuracy of high and low tide levels has also been significantly improved compared to single-model forecasts. The high tide forecast error is slightly higher, possibly because the hydrodynamic changes during high tide are more complex, making it difficult for the model to capture their precise patterns. Under a 24-hour long-term forecast, the hourly forecasts for the non-flood season and flood season have RMSE values ​​of 0.15m and 0.18m, respectively, providing a guarantee for developing more scientific and reasonable forecasting plans. Under normal weather conditions during the non-flood season (… Figure 10 a), its hourly forecast accuracy within ±0.30m reached an average of 90.4%, typhoon "Gemi" ( Figure 10 b) and "Bebinca" Figure 10(b) The average accuracy rates during the period were 82.7% and 86.6%, indicating that the coupled model can effectively simulate the variation range of tidal levels under multiple influencing factors. This model has been applied to the Jiangsu Province Yangtze River Forecasting and Prediction Integrated System. After its official launch in 2024, it provided decision-making assistance for the operation forecast of the Qinhuai New River Water Conservancy Project. During the "6.29 Flood" of the Qinhuai River in 2024, the accurate prediction of the low tide level of the Yangtze River was used to drain approximately 14.61 million m3 of floodwater from the Qinhuai River, controlling the water level at the Dongshan Station of the Qinhuai River between 6.8 and 7.5 m, improving the city's drainage conditions and ensuring the safety of Nanjing's main urban area.

[0095] In one embodiment, the method further includes: capturing the spatial correlation between each harmonic component and the study area based on the harmonic components decomposed by the nonsteady harmonic model, combined with spatial autocorrelation analysis, and obtaining the overall spatial distribution characteristics of the tide in the study area according to the spatial correlation corresponding to each harmonic component; obtaining the attenuation rate of the tide level according to the overall spatial distribution characteristics, and generating continuous target prediction results according to the attenuation rate.

[0096] This embodiment performs spatial autocorrelation analysis based on the harmonic components decomposed by the non-steady-state harmonic model. On the one hand, it considers the nonlinear effects of runoff and offshore tidal range, improving the accuracy of tidal level prediction. On the other hand, it is more interpretable than spatial autocorrelation analysis based on tidal level.

[0097] By performing spatial autocorrelation analysis on each harmonic component, a local spatial autocorrelation model is constructed to describe the spatial heterogeneity of the local region and capture the spatial correlation between each harmonic component and the study area, such as the propagation direction, energy difference, and tidal level difference of the tide. By synthesizing the spatial correlation between each harmonic component and the study area, tidal frequency waves at any location can be predicted, including nonlinear effects. By superimposing a series of harmonic components with fixed frequencies, the tidal level prediction at the corresponding location can be obtained; that is, the spatial correlation of the harmonic components reflects the overall spatial distribution characteristics of the tide. Smoothing the tidal level predictions at continuous locations yields the attenuation rate of the tidal level at different locations within the study area. Based on the attenuation rate and the target prediction results at already determined observation points, continuous target prediction results can be generated.

[0098] In one embodiment, after obtaining the tidal decay rate based on the overall spatial distribution characteristics, the method further includes: evaluating and correcting the decay rate using a DRSN-LSTM model, and generating continuous target prediction results based on the corrected decay rate.

[0099] Under different environmental conditions, such as the influence of meteorological factors like typhoons, the rate of tidal decay can vary. This embodiment utilizes the good performance of the DRSN-LSTM model in short-term predictions under severe, variable, and extreme weather conditions to evaluate and correct the decay rate, adapting it to different environmental conditions and thus generating more accurate and continuous target prediction results.

[0100] A specific correction method can be a threshold comparison approach. Based on the DRSN-LSTM model for predicting tidal trends in different scenarios, spatial interpolation is performed using inverse distance weighted interpolation to obtain a large amount of training data. This training data is then used to adjust the decay rate. If the root mean square error between the target prediction result based on the decay rate and the training data obtained through spatial interpolation is greater than a set threshold, the decay rate needs to be corrected by incorporating expert knowledge and experience. If the root mean square error is less than the set threshold, the target prediction result based on the decay rate and the training data obtained through spatial interpolation are combined using methods such as weighted summation, and the decay rate is then updated.

[0101] In summary, current tidal level forecasting for tidal river sections suffers from problems such as complex and variable influencing factors and low forecast accuracy under abnormal or extreme weather conditions, making it difficult to meet the technical requirements of "digitalized scenarios, intelligent simulation, and precise decision-making" proposed by digital twin water conservancy construction. The coupled model constructed in this invention uses the mechanistic model NS_TIDE to provide a baseline value and predict the periodic and regular trends of tidal level changes. The data-driven model DRSN-LSTM provides the amplitude values ​​of tidal level changes under the influence of multiple factors. Based on this, an improved genetic algorithm optimizes the coupling parameters to obtain the final tidal level forecast value. Compared with single methods, this effectively improves forecast accuracy, and the forecast results during floods and typhoons are also significantly improved.

[0102] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0103] Based on the same inventive concept, this application also provides an intelligent tidal level prediction device for tidal river sections to implement the intelligent tidal level prediction method for tidal river sections mentioned above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent tidal level prediction device for tidal river sections provided below can be found in the limitations of the intelligent tidal level prediction method for tidal river sections above, and will not be repeated here.

[0104] In one embodiment, a smart tidal level prediction device for tidal river sections is provided, comprising:

[0105] The first model building module is used to construct a nonsteady harmonic analysis model based on real-time tidal level data of the study area, upstream runoff, and offshore tides.

[0106] The second model building module is used to build DRSN-LSTM models for non-flood season and flood season based on tidal data, water level data, meteorological data and engineering data of the study area.

[0107] The prediction output module is used to couple the unsteady harmonic analysis model and the DRSN-LSTM model to predict the tidal trend of the study area and obtain the target prediction results of the study area.

[0108] Among them, the tide trend prediction includes the prediction of the current time node, the prediction of the previous time node, and the change prediction determined by the prediction of the current time node and the previous time node.

[0109] Obtaining target prediction results for the study area includes:

[0110] Based on the previous time point predictions of the unsteady harmonic analysis model and the DRSN-LSTM model, obtain the tidal baseline value; determine the tidal change trend based on the change prediction of the unsteady harmonic analysis model, and obtain the tidal change amplitude based on the change prediction of the DRSN-LSTM model; determine the target prediction result based on the tidal baseline value, tidal change trend, and tidal change amplitude.

[0111] Each module in the aforementioned intelligent tidal level prediction device for tidal river sections can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0112] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in all of the above method embodiments.

[0113] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in all of the above method embodiments.

[0114] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in all of the above method embodiments.

[0115] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0116] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0117] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for intelligent prediction of tidal levels in tidal river sections, characterized in that, The method includes: Based on real-time tidal level data of the study area, upstream runoff, and offshore tides, a nonsteady harmonic analysis model is constructed. Based on the tidal data, water level data, meteorological data, and engineering data of the study area, DRSN-LSTM models for non-flood season and flood season are constructed. The unsteady harmonic analysis model and the DRSN-LSTM model are coupled to predict the tidal trend of the study area, and the target prediction results of the study area are obtained. The tide trend prediction includes the current time node prediction, the previous time node prediction, and the change prediction determined by the current time node prediction and the previous time node prediction. The acquisition of the target prediction results for the study area includes: Based on the previous time node predictions of the unsteady harmonic analysis model and the DRSN-LSTM model, a tidal baseline value is obtained; the tidal change trend is determined based on the change predictions of the unsteady harmonic analysis model; the tidal change amplitude is obtained based on the change predictions of the DRSN-LSTM model; and the target prediction result is determined based on the tidal baseline value, the tidal change trend, and the tidal change amplitude. The method further includes: A baseline coefficient and an amplitude coefficient are introduced. The baseline coefficient is used to adjust the influence of the previous time node prediction of the unsteady harmonic analysis model and the DRSN-LSTM model on the tide level baseline value. The amplitude coefficient is used to adjust the influence of the tide level change amplitude on the target prediction result. The target prediction result is expressed as follows: ； Among them, W t This represents the target prediction result at time t. This represents the prediction for the current time point based on the non-steady harmonic analysis model. This indicates the prediction at the previous time point based on the non-steady-state harmonic analysis model. This represents the prediction for the current time point based on the DRSN-LSTM model. This indicates the prediction at the previous time point based on the DRSN-LSTM model, where α and β represent the baseline coefficient and the amplitude coefficient, respectively.

2. The method according to claim 1, characterized in that, The method further includes: An improved genetic algorithm is used to optimize the baseline coefficient and the amplitude coefficient to obtain the optimal baseline coefficient and the optimal amplitude coefficient for different periods.

3. The method according to claim 2, characterized in that, The improved genetic algorithm includes: Based on the genetic algorithm, an adaptive operator is introduced to optimize the crossover probability and mutation probability in the genetic algorithm. The adaptive operator is determined based on the number of iterations, population size, fitness value of individuals in the population, and average fitness value of individuals in the population.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Based on the harmonic components decomposed by the nonsteady harmonic analysis model, combined with spatial autocorrelation analysis, the spatial correlation between each harmonic component and the study area is captured, and the overall spatial distribution characteristics of tides in the study area are obtained according to the spatial correlation corresponding to each harmonic component. The decay rate of the tide level is obtained based on the overall spatial distribution characteristics, and continuous target prediction results are generated based on the decay rate.

5. The method according to claim 4, characterized in that, After obtaining the tidal level decay rate based on the overall spatial distribution characteristics, the method further includes: The decay rate is evaluated and corrected using the DRSN-LSTM model, and continuous target prediction results are generated based on the corrected decay rate.

6. A smart tidal level prediction device for tidal river sections, characterized in that, The device includes: The first model building module is used to construct a nonsteady harmonic analysis model based on real-time tidal level data of the study area, upstream runoff, and offshore tides. The second model building module is used to build DRSN-LSTM models for non-flood season and flood season based on the tidal data, water level data, meteorological data and engineering data of the study area. The prediction output module is used to couple the unsteady harmonic analysis model and the DRSN-LSTM model to predict the tidal trend of the study area, and obtain the target prediction result of the study area. The tide trend prediction includes the current time node prediction, the previous time node prediction, and the change prediction determined by the current time node prediction and the previous time node prediction. The acquisition of the target prediction results for the study area includes: Based on the previous time node predictions of the unsteady harmonic analysis model and the DRSN-LSTM model, a tidal baseline value is obtained; the tidal change trend is determined based on the change predictions of the unsteady harmonic analysis model; the tidal change amplitude is obtained based on the change predictions of the DRSN-LSTM model; and the target prediction result is determined based on the tidal baseline value, the tidal change trend, and the tidal change amplitude. A baseline coefficient and an amplitude coefficient are introduced. The baseline coefficient is used to adjust the influence of the previous time node prediction of the unsteady harmonic analysis model and the DRSN-LSTM model on the tide level baseline value. The amplitude coefficient is used to adjust the influence of the tide level change amplitude on the target prediction result. The target prediction result is expressed as follows: ； Among them, W t This represents the target prediction result at time t. This represents the prediction for the current time point based on the non-steady harmonic analysis model. This indicates the prediction at the previous time point based on the non-steady-state harmonic analysis model. This represents the prediction for the current time point based on the DRSN-LSTM model. This indicates the prediction at the previous time point based on the DRSN-LSTM model, where α and β represent the baseline coefficient and the amplitude coefficient, respectively.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images