Digital bay hydrological condition early warning method and system

Abstract

The invention relates to the technical field of marine disaster early warning, and discloses a digital bay hydrological condition early warning method and system, and the method comprises the steps: collecting historical meteorological data and real-time meteorological data of a to-be-predicted bay region; according to the historical meteorological data, establishing a wind and wave relation model by using a deep reinforcement learning and genetic algorithm, and generating wave prediction data based on the relation model and the real-time wind data of the to-be-predicted bay area; based on the theoretical model wind field and the background wind field, synthesizing a synthesized wind field of the to-be-predicted bay area; performing storm surge simulation by using a hydrodynamic mathematical model and a tidal current mathematical model according to the synthesized wind field and the real-time meteorological data; and on the basis of the wave prediction data and the storm surge simulation result, hydrological condition early warning information is generated in combination with an early warning index threshold value, the prediction precision and early warning timeliness of waves and storm surge in the bay area are improved, and accurate grading response of complex meteorological disasters is realized.

Description

Technical Field

[0001] The present invention relates to the field of marine disaster early warning technology, and in particular to a digital bay hydrological condition early warning method and system. Background Art

[0002] The hydrological conditions in the Gulf region are affected by a variety of meteorological factors, especially wind, waves, and storm surges. In order to effectively warn of possible extreme weather phenomena such as storm surges, traditional hydrological forecasting methods usually rely on a single meteorological data source, which is difficult to fully reflect the interaction between the complex meteorological field and the Gulf environment. With the advancement of computing technology and algorithms, composite warning methods based on multi-source data have gradually received attention in recent years.

[0003] This composite warning method of multi-source data fusion integrates different meteorological and ocean data sources to more comprehensively describe the complex bay environment, especially the interactive effects of wind fields on waves and storm surges. However, the prior art often uses theoretical typhoon models or background wind fields alone to simulate wind fields, resulting in overestimation of wind speed in the core area of ​​the typhoon or accumulation of errors in the peripheral wind fields. For example, patent CN109583456A proposes to superimpose theoretical wind fields and background wind fields with fixed weights, but the fixed weights cannot adapt to the dynamic changes in typhoon intensity and distance, resulting in significant errors in the synthetic wind field in the peripheral area of ​​the typhoon. In addition, the robustness of the wind-wave relationship model in the prior art is poor, and traditional wave prediction relies more on physical models (such as SWAN) or statistical models (such as ARIMA), which makes it difficult to capture the complex spatiotemporal correlation between wind and waves. For example, patent CN112950083A uses LSTM neural networks to predict waves, but the model is prone to fall into local optimality and is not adaptable enough to sudden changes in wind direction. Existing machine learning methods have not effectively combined time series feature extraction (such as wind field gradient) with global parameter optimization, resulting in limited generalization capabilities of the model in complex waters (such as multi-island bays).

[0004] Therefore, there is an urgent need for a new digital bay hydrological condition early warning method and system that can simulate wind field changes at different levels and scales, so as to more accurately predict ocean hydrological conditions. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a digital bay hydrological condition early warning method and system, which can improve the prediction accuracy and early warning time of waves and storm surges in the bay area, and realize accurate graded response to complex meteorological disasters.

[0006] The present invention provides a digital bay hydrological condition early warning method, comprising the following steps: S1. Collect historical and real-time meteorological data of the bay area to be predicted; The meteorological data includes: wind data, wave data and tide data; S2. Based on historical meteorological data, deep reinforcement learning and genetic algorithms are used to establish a relationship model between wind and waves, and wave prediction data is generated based on the relationship model and real-time wind data of the bay area to be predicted; S3. Based on the theoretical model wind field and the background wind field, synthesize the synthetic wind field of the bay area to be predicted; S4, performing storm surge simulation using a hydrodynamic mathematical model and a tidal mathematical model according to the synthetic wind field and real-time meteorological data; S5. Generate hydrological condition warning information based on the wave prediction data and storm surge simulation results in combination with warning indicator thresholds.

[0007] Furthermore, the S2 specifically includes: S21. Based on the wind data and wave data in the historical meteorological data, pre-train the deep reinforcement learning agent to extract the correlation features between wind and waves; S22, forming an initial population based on the initial individuals recommended by the trained deep reinforcement learning agent and the randomly generated individuals; S23, calculating the fitness of individuals in the population according to the wind propagation delay parameter and the wave response time; S24, perform genetic operations at a preset ratio according to fitness, and generate offspring populations through crossover and mutation operations on new offspring individuals; S25, repeating steps S22 to S24 until the maximum number of iterations is reached, then stopping the iteration and outputting the relationship model between wind and waves; S26, inputting the real-time wind data of the bay area to be predicted into the relationship model between wind and waves to generate wave prediction data for the bay area to be predicted.

[0008] Furthermore, the calculation formula of the fitness in S23 is: Take n points in the bay area to be predicted and divide the impact time into m uniform time periods; then the calculation formula of fitness F is: ; In the formula, u i,j is the wind speed of the wind data to be predicted; v i,j is the wind direction of the wind data to be predicted; U i,j is the wind speed of historical wind data; V i,j is the wind direction of the historical wind data; i=1,…n; j=1,…m; R is the reward value output by the machine learning agent; λ is the adaptive balance coefficient.

[0009] Furthermore, the reward value R is calculated as follows: S231, define the state space as the gradient matrix of the current wind data with the wave data vector [Hs,T]; S232, the machine learning agent outputs a reward value through a policy network: ; in, w 1, w 2 is the network learnable parameter.

[0010] Furthermore, the S3 specifically includes: ; Where: V c represents the synthetic wind field; V represents the theoretical model wind field; V ERA represents the background wind field; E represents the weight coefficient; Furthermore, the weight coefficient is calculated as follows: ; Where d represents the distance from the typhoon center to the target location; d0 represents the reference distance, which is used to standardize the distance d; and k represents the attenuation rate coefficient.

[0011] Furthermore, the attenuation rate coefficient k is determined by the following steps: S31, calculating the maximum impact radius R of historical typhoons in the target sea area max ; S32, set k = ln2 / (R max / d0)2, when d=R max When E=0.5.

[0012] Furthermore, the S4 specifically includes: Set the open boundary conditions, closed boundary conditions and moving boundary conditions of the hydrodynamic model according to the tide forecast data and real-time tide data; Based on the synthetic wind field and real-time meteorological data, the hydrodynamic mathematical model is used to calculate the change of output water level over time, the velocity field of water flow, surface wind stress and bed shear stress; The data output by the hydrodynamic model is used to simulate the storm surge through the tidal mathematical model.

[0013] Furthermore, the S5 specifically includes: Determine hydrological condition warning indicators and set warning thresholds based on wave forecast data and storm surge simulation results; According to the threshold of hydrological condition warning indicators, the hydrological condition warning is divided into 4 warning levels and corresponding response measures are formulated; Based on wave prediction data and storm surge simulation results, the warning level for the bay area to be predicted is determined and corresponding measures are implemented.

[0014] The present invention also provides a digital bay hydrological condition early warning system, which is used to implement the digital bay hydrological condition early warning method mentioned above, and is characterized in that the system includes the following modules: Data collection module: used to collect historical meteorological data and real-time meteorological data of the bay area to be predicted; the meteorological data includes: wind data, wave data and tidal data; Wave prediction module: connected to the data collection module, used to establish a relationship model between wind and waves based on historical meteorological data using deep reinforcement learning and genetic algorithms, and generate wave prediction data based on the relationship model and real-time wind data of the bay area to be predicted; Storm surge prediction module: connected to the data collection module, used to synthesize the synthetic wind field of the bay area to be predicted by using the theoretical model wind field and the background wind field; based on the synthetic wind field, combined with real-time meteorological data, storm surge simulation is performed using a hydrodynamic mathematical model and a tidal mathematical model; The early warning module is connected to the wave prediction module and the storm surge prediction module, and generates hydrological condition early warning information based on the wave prediction data and the storm surge simulation result in combination with the early warning index threshold.

[0015] The embodiments of the present invention have the following technical effects: 1. By integrating historical meteorological data, real-time meteorological data (such as wind speed, wave data, tidal data) and the output data of the hydrodynamic model, it is possible to more accurately predict the hydrological conditions in the bay area and improve the prediction accuracy of natural disasters such as storm surges and waves.

[0016] 2. By combining the theoretical model wind field and the background wind field, a more accurate synthetic wind field is synthesized. The two are fused through dynamic weights, and the weight decays in a Gaussian manner with the distance between the typhoon center and the target location, which is more in line with the actual physical law that the typhoon wind speed decays with distance. The attenuation rate coefficient is determined by historical typhoon statistics rather than manually preset, which enhances the model's adaptability to different typhoon intensities, retains the high-precision simulation of the typhoon core, and uses the background field to correct the peripheral wind field error, thereby providing a more reliable driving wind field for storm surge simulation and reducing the impact of the error of a single wind field model on the simulation results.

[0017] 3. Using deep reinforcement learning (DRL) combined with genetic algorithms, wave data can be automatically learned and predicted from a large amount of historical data, which greatly shortens the time required for prediction and improves the accuracy of the solution. DRL dynamically captures the nonlinear time-series relationship between wind and waves (such as the lagged effect of sudden changes in wind speed on wave height) through reward mechanisms (such as policy networks based on wind field gradient matrices and wave data), solving the problem of insufficient modeling of time-series dependence in traditional models (such as LSTM and ARIMA); genetic algorithms select the optimal parameter combination through fitness functions (integrating wind speed / wind direction errors and DRL reward values) to avoid the model from falling into local optimality and improve the generalization ability in complex sea areas (such as multi-island areas); through deep reinforcement learning (DRL) agent pre-training, the time-series correlation characteristics of wind field gradient matrices and wave parameters are extracted, and the initial population individuals are generated and input into the genetic algorithm, replacing completely random initialization, thereby providing a guiding initial position for the genetic algorithm, which can guide the search process to approach the global optimal solution faster. Compared with completely random initialization, initialization based on the predicted value of existing knowledge can guide the search direction and reduce unnecessary blind exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 The present invention provides a flowchart of a digital bay hydrological condition early warning method.

[0020] Figure 2 It is the relative error of the wind data forecast within 18 hours of a digital bay hydrological condition early warning method provided by an embodiment of the present invention.

[0021] Figure 3 The invention provides a digital bay hydrological condition early warning method and a relative error of wind data in 30 hours of forecast. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0023] Figure 1is a flow chart of a digital bay hydrological condition early warning method provided by an embodiment of the present invention. Figure 1 , specifically including: S1. Collect historical and real-time meteorological data of the bay area to be predicted; Meteorological data include: wind data, wave data and tide data.

[0024] Collecting historical and real-time meteorological data of the bay area to be predicted is the data cornerstone for building a hydrological early warning system. Historical meteorological data covers continuous observation records of wind, wave and tidal fields over the past few decades, including wind elements such as wind speed and direction, wave parameters such as effective wave height, wave period and propagation direction, as well as astronomical tide level, storm surge process, tidal phase and other tidal information. Real-time meteorological data can be dynamically obtained through meteorological satellite remote sensing, coastal automatic weather stations, buoy arrays and radar detection equipment.

[0025] S2. Based on historical meteorological data, deep reinforcement learning and genetic algorithms are used to establish a relationship model between wind and waves, and wave prediction data is generated based on the relationship model and real-time wind data of the bay area to be predicted; In the digital bay hydrological early warning method, establishing a relationship model between wind and waves is the core link, and the key lies in capturing the nonlinear correlation between meteorological elements through intelligent algorithms. The implementation process of step S2 combines the advantages of deep reinforcement learning and genetic algorithms to form a hybrid modeling framework, which not only utilizes the feature extraction ability of reinforcement learning, but also gives play to the global optimization characteristics of genetic algorithms, and finally constructs a high-precision wind and wave response model.

[0026] In some embodiments, S2 specifically includes: S21. Based on the wind data and wave data in the historical meteorological data, a deep reinforcement learning agent is pre-trained to extract the correlation features between wind and waves.

[0027] First, the deep reinforcement learning agent is pre-trained using wind data (including wind speed and direction) and wave data (such as significant wave height Hs and period T) from historical meteorological data. The goal of this stage is to enable the agent to extract the correlation features between wind and waves. In this way, it is possible to capture how changes in the wind field affect wave parameters and how this effect varies over time and space.

[0028] S22. Form an initial population based on the initial individuals recommended by the trained deep reinforcement learning agent and the randomly generated individuals.

[0029] The initial population is formed based on the initial individuals recommended by the above trained deep reinforcement learning agent and the randomly generated individuals. This step combines the results of reinforcement learning with the search ability of genetic algorithms to provide a diverse starting point for the subsequent optimization process.

[0030] For example, based on the results of pre-training, the agent may recommend a set of parameters as a starting point. For example, it may recommend a set of model parameters that it believes can better simulate the relationship between wind and waves based on past experience. Suppose this set of parameters is A ={a1, a2, ..., an}; At the same time, we also randomly generate a set of parameters B = {b1, b2, ..., bn}. These parameters represent different model configurations, which may be completely randomly chosen to explore different parts of the search space.

[0031] These two sets of parameters are combined, along with a few additional random individuals, to form the initial population. For example, the initial population may contain 5 individuals: A, B, and 3 other randomly generated individuals C, D, and E. The purpose of this is to ensure that the initial population contains both "experienced" solutions provided by the deep reinforcement learning agent and completely random new attempts, thereby ensuring diversity and extensive exploration.

[0032] S23, calculating the fitness of individuals in the population according to the wind propagation delay parameter and the wave response time; In some embodiments, the fitness is calculated as: Take n points in the bay area to be predicted and divide the impact time into m uniform time periods; then the calculation formula of fitness F is: ; In the formula, u i,j is the wind speed of the wind data to be predicted; v i,j is the wind direction of the wind data to be predicted; U i,j is the wind speed of historical wind data; V i,j is the wind direction of the historical wind data; i=1,…n; j=1,…m; R is the reward value output by the machine learning agent; λ is the adaptive balance coefficient.

[0033] The fitness function is designed to take into account multiple points (n) and time intervals (m uniform time periods) in the area to be predicted. The fitness of each individual is calculated by comparing the predicted wind data (u i,j , v i,j ) and historical wind data (U i,j , V i,j ) and the reward value R output by the machine learning agent, where λ is the adaptive balance coefficient.

[0034] Furthermore, the reward value R is calculated as follows: S231, define the state space as the gradient matrix of the current wind data with the wave data vector [Hs,T]; S232, the machine learning agent outputs the reward value through the policy network: ; in, w 1, w 2 is the network learnable parameter.

[0035] Optionally, λ is used as an adaptive balancing factor to adjust the relative importance of different factors (such as the impact of historical data and real-time data). It can be determined by cross-validation, grid search, or Bayesian optimization.

[0036] Learning parameters w 1, w 2 are learnable parameters in the policy network, which determine how to output the reward value based on the current state (the gradient matrix of wind data and the wave data vector). They can be determined by back propagation algorithm, initialization and regularization, hyperparameter tuning and other methods.

[0037] S24. Genetic operations are performed at preset ratios according to fitness, and crossover and mutation operations are performed on new offspring individuals to generate a sub-generation population.

[0038] Specifically, individuals are sorted according to the fitness function, and excellent individuals are selected to enter the next generation according to a preset ratio (for example, the top 20% of individuals). This process can be done by roulette selection, tournament selection, and other methods. The selected parent individuals are paired, and part of the gene information is exchanged at a certain crossover rate to produce new offspring. Common crossover methods include single-point crossover, multi-point crossover, and uniform crossover. In order to increase population diversity and avoid premature convergence, some gene values ​​are randomly changed at a certain mutation rate for the newly generated offspring individuals. Mutation can be achieved by replacing a gene value or perturbing it. Through the above genetic operations (selection, crossover, mutation), the offspring population is generated and prepared for the next round of iteration.

[0039] S25. Repeat steps S22 to S24 until the maximum number of iterations is reached, then stop the iteration and output the relationship model between wind and waves.

[0040] Specifically, the process of steps S22 to S24 needs to be repeated continuously, that is, the fitness of each individual in the offspring population is recalculated, and then the selection, crossover and mutation operations are performed again to generate a new generation of population. A maximum number of iterations is set as the stopping criterion. When the number of iterations reaches the preset maximum value, the algorithm stops running. The best individual in the population output at this time represents the optimized wind-wave relationship model.

[0041] S26, inputting the real-time wind data of the bay area to be predicted into the relationship model between wind and waves to generate wave prediction data for the bay area to be predicted.

[0042] The real-time wind data of the bay area to be predicted (such as wind speed u i,j and wind direction i,j ) is input into the final wind and wave relationship model. Based on the input real-time wind data, the model will output corresponding wave prediction data (such as significant wave height Hs and period T). These prediction data can be used to evaluate the hydrological conditions in the Gulf region and support decision-making of the early warning system.

[0043] The above step S2 describes a method for combining deep reinforcement learning with genetic algorithms to establish a wind and wave relationship model. The core of this method is to use the complementary advantages of the two intelligent algorithms to improve prediction accuracy. The deep reinforcement learning agent learns the wind speed, wind direction and corresponding wave data (such as effective wave height Hs and period T) in the historical meteorological data through the pre-training stage, so that it can identify how wind field changes affect wave parameters and capture the complex nonlinear relationship between the two. On this basis, the genetic algorithm introduces an evolutionary mechanism, evaluates the rationality of different wind propagation delay assumptions through fitness functions, and explores the optimal solution space using crossover and mutation operations. The two work together to form a closed loop of "feature extraction-global optimization": reinforcement learning mines local features, and genetic algorithms implement parameter tuning, which jointly improves the generalization ability of the model.

[0044] For example, based on the wind and wave data in Bohai Bay over the past 30 years, deep reinforcement learning and genetic algorithms are used to compare the predicted current value of a single point with the measured current value. Figure 2 and Figure 3 They are the relative forecast errors of wind data at 18 hours and 30 hours respectively.

[0045] S3. Based on the theoretical model wind field and the background wind field, synthesize the synthetic wind field of the bay area to be predicted.

[0046] Step S3 describes how to synthesize the synthetic wind field of the bay area to be predicted based on the theoretical model wind field and the background wind field. This process introduces a weight coefficient E to balance the theoretical wind field (V) and the background wind field (V ERA ) to generate a synthetic wind field (V c ), the specific implementation details are as follows: In some embodiments, S3 specifically includes: ; Where: V c represents the synthetic wind field; V represents the theoretical model wind field, which represents the wind field predicted based on physical or mathematical models; V ERArepresents the background wind field, which usually refers to the climate average state or background wind field conditions constructed from long-term observation data; E represents the weight coefficient, which is used to adjust the relative contribution between the theoretical model wind field and the background wind field.

[0047] Optionally, the theoretical model wind field is a Holland gradient wind field; and the background wind field is an ECMWF background wind field.

[0048] In some embodiments, the weight coefficient is calculated as follows: ; Where d represents the distance from the typhoon center to the target location; d0 represents the reference distance, which is used to standardize the distance d; and k represents the attenuation rate coefficient.

[0049] In some embodiments, the attenuation rate coefficient k is determined by the following steps: S31, statistically analyzing the maximum impact radius R of historical typhoons in the target sea area. max This step involves analyzing past typhoon events in the region, recording the impact range of each event, and calculating an average maximum impact radius as R max S32, set k = ln2 / (R max / d0)2, when d=R max The purpose of this step is to ensure that when d=R max When the weight coefficient E = 0.5, it means that within the maximum impact range of the typhoon, the contribution of the theoretical wind field and the background wind field to the synthetic wind field is equal. max , the influence of the theoretical wind field gradually decreases, while the effect of the background wind field increases accordingly.

[0050] Through the above steps, the advantages of the theoretical model wind field and the background wind field can be effectively combined to generate a synthetic wind field model that can reflect the dynamic changes of the current weather system and maintain long-term meteorological characteristics. This method is particularly suitable for complex and changeable marine environments, which helps to improve the accuracy of wind and wave forecasts and support more effective hydrological warning decisions. In addition, by adjusting parameters such as d0 and k, customized adjustments can be made according to the characteristics of different geographical regions, further improving the applicability and reliability of the model.

[0051] S4. Based on the synthetic wind field and real-time meteorological data, storm surge simulation is performed using a hydrodynamic mathematical model and a tidal mathematical model; In some embodiments, S4 specifically includes: Set the open boundary conditions, closed boundary conditions and moving boundary conditions of the hydrodynamic model according to the tide forecast data and real-time tide data; Open boundary conditions: Set the open boundary conditions of the hydrodynamic model based on tidal forecast data and real-time tidal data. Open boundaries are usually located at the outer boundaries of the study area and are used to describe the interaction between the ocean or large lake and the external water body. These conditions may include parameters such as water level and flow rate. Closed boundary conditions: Define the boundaries within the study area that are not directly connected to the external water body, such as coastlines, dams, etc. On these boundaries, the water flow velocity is usually set to zero (no penetration condition), or other physical constraints are set according to the specific situation. Moving boundary conditions: For some special areas, such as estuaries and wetlands, there may be periodic dry and wet alternations. Moving boundary conditions need to be set to dynamically adjust the boundary position and properties to more accurately simulate the actual environment.

[0052] Based on the synthetic wind field and real-time meteorological data, the hydrodynamic mathematical model is used to calculate the change of output water level over time, the velocity field of water flow, surface wind stress and bed shear stress; Specifically, synthetic wind fields and real-time meteorological data are used as input, including but not limited to wind speed, wind direction, air pressure, etc. Using a hydrodynamic mathematical model, based on the above boundary conditions and input data, the output water level changes over time, the velocity field of the water flow, the surface wind stress, and the bottom bed shear stress are calculated. These calculation results are an important basis for understanding the impact of storm surges. Water level changes: reflect the changes in sea level during the storm. Water flow velocity field: shows the direction and intensity of water flow at different times and locations. Surface wind stress and bottom bed shear stress: help analyze the force of wind on the water surface and the possibility of transporting seabed sediments.

[0053] The data output by the hydrodynamic model is used to simulate the storm surge through the tidal mathematical model.

[0054] The data obtained from the hydrodynamic model, especially the water level change and water velocity field, are used as input into the tidal mathematical model. The tidal mathematical model is used to further simulate the specific impact of storm surges, including but not limited to the inundation range of coastal areas, flood depth and duration, etc. This step helps to assess the potential threat of storm surges to human activities and the natural environment, and supports the formulation of effective disaster prevention and mitigation strategies.

[0055] S5. Generate hydrological condition warning information based on wave prediction data and storm surge simulation results combined with warning indicator thresholds.

[0056] In some embodiments, S5 specifically includes: Determine hydrological condition warning indicators and set warning thresholds based on wave forecast data and storm surge simulation results; First, based on wave prediction data (such as significant wave height Hs, period T) and storm surge simulation results (such as maximum inundation depth, flood duration), a series of key hydrological condition warning indicators are determined. These indicators include but are not limited to maximum wave height Hs, average wave period T, highest water level caused by storm surge, flood duration, etc.

[0057] According to the threshold of hydrological condition warning indicators, the hydrological condition warning is divided into 4 warning levels and corresponding response measures are formulated; For example, a mild warning is issued when the maximum wave height Hs is between 2 meters and 3 meters; a moderate warning is issued when the maximum wave height Hs is between 3 meters and 4 meters or the highest water level caused by the storm surge exceeds the warning water level by less than 0.5 meters; a high warning is issued when the maximum wave height Hs is greater than 4 meters or the highest water level caused by the storm surge exceeds the warning water level by more than 0.5 meters; an extreme warning is issued when the maximum wave height Hs exceeds 6 meters or serious flood disasters are expected.

[0058] Develop detailed response measures for each warning level, for example: Mild warning: Strengthen monitoring and notify relevant departments to prepare emergency supplies. Moderate warning: Activate emergency plans and evacuate residents in low-lying areas. High warning: Fully mobilize rescue forces to ensure the safe evacuation of all personnel. Extreme warning: Emergency closure of coastal facilities to ensure the safety of life and property.

[0059] Based on wave prediction data and storm surge simulation results, the warning level for the bay area to be predicted is determined and corresponding measures are implemented.

[0060] Based on the wave prediction data and storm surge simulation results, the individual warning index values ​​are determined and compared with the pre-set thresholds to determine which warning level the current bay area is in. According to the determined warning level, corresponding measures of the corresponding level are implemented.

[0061] The present invention also provides a digital bay hydrological condition early warning system, which is used to implement the digital bay hydrological condition early warning method mentioned above, and is characterized in that the system includes the following modules: Data collection module: used to collect historical and real-time meteorological data of the bay area to be predicted; meteorological data include: wind data, wave data and tide data; Wave prediction module: connected to the data collection module, used to establish a wind-wave relationship model based on historical meteorological data using deep reinforcement learning and genetic algorithms, and generate wave prediction data based on the relationship model and real-time wind data of the bay area to be predicted; Storm surge prediction module: connected to the data collection module, used to synthesize the synthetic wind field of the bay area to be predicted by using the theoretical model wind field and the background wind field; based on the synthetic wind field, combined with real-time meteorological data, the storm surge simulation is carried out using the hydrodynamic mathematical model and the tidal mathematical model; The early warning module is connected with the wave prediction module and the storm surge prediction module, and generates hydrological condition early warning information based on the wave prediction data and storm surge simulation results combined with the early warning index threshold.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A digital bay hydrological condition early warning method, characterized in that: The steps include: S1. Collect historical and real-time meteorological data of the bay area to be predicted; The meteorological data includes: wind data, wave data and tide data; S2. Based on historical meteorological data, deep reinforcement learning and genetic algorithms are used to establish a relationship model between wind and waves, and wave prediction data is generated based on the relationship model and real-time wind data of the bay area to be predicted; S3. Based on the theoretical model wind field and the background wind field, synthesize the synthetic wind field of the bay area to be predicted; S4, performing storm surge simulation using a hydrodynamic mathematical model and a tidal mathematical model according to the synthetic wind field and real-time meteorological data; S5. Generate hydrological condition warning information based on the wave prediction data and storm surge simulation results in combination with warning indicator thresholds.

2. A digital bay hydrological condition early warning method according to claim 1, characterized in that: The S2 specifically includes: S21. Based on the wind data and wave data in the historical meteorological data, pre-train the deep reinforcement learning agent to extract the correlation features between wind and waves; S22, forming an initial population based on the initial individuals recommended by the trained deep reinforcement learning agent and the randomly generated individuals; S23, calculating the fitness of individuals in the population according to the wind propagation delay parameter and the wave response time; S24, perform genetic operations at a preset ratio according to fitness, and generate offspring populations through crossover and mutation operations on new offspring individuals; S25, repeating steps S22 to S24 until the maximum number of iterations is reached, then stopping the iteration and outputting the relationship model between wind and waves; S26, inputting the real-time wind data of the bay area to be predicted into the relationship model between wind and waves to generate wave prediction data for the bay area to be predicted.

3. A digital bay hydrological condition early warning method according to claim 2, characterized in that: The calculation formula of fitness in S23 is: Take n points in the bay area to be predicted and divide the impact time into m uniform time periods; then the calculation formula of fitness F is: ； In the formula, u i,j is the wind speed of the wind data to be predicted; v i,j is the wind direction of the wind data to be predicted; U i,j is the wind speed of historical wind data; V i,j is the wind direction of the historical wind data; i=1,…n; j=1,…m; R is the reward value output by the machine learning agent; λ is the adaptive balance coefficient.

4. A digital bay hydrological condition early warning method according to claim 3, characterized in that: The reward value R is calculated as follows: S231, define the state space as the gradient matrix of the current wind data with the wave data vector [Hs,T]; S232, the machine learning agent outputs a reward value through a policy network: ； in, w 1, w 2 is the network learnable parameter.

5. A digital bay hydrological condition early warning method according to claim 1, characterized in that: The S3 specifically includes: ； Where: V c represents the synthetic wind field; V represents the theoretical model wind field; V ERA represents the background wind field; E represents the weight coefficient.

6. A digital bay hydrological condition early warning method according to claim 5, characterized in that: The weight coefficient is calculated as follows: ； Where d represents the distance from the typhoon center to the target location; d0 represents the reference distance, which is used to standardize the distance d; and k represents the attenuation rate coefficient.

7. A digital bay hydrological condition early warning method according to claim 6, characterized in that: The attenuation rate coefficient k is determined by the following steps: S31, statistically analyzing the maximum impact radius R of historical typhoons in the target sea area. max ; S32, set k = ln2 / (R max / d0)2, when d=R max When E=0.

5.

8. A digital bay hydrological condition early warning method according to claim 1, characterized in that: The S4 specifically includes: Set the open boundary conditions, closed boundary conditions and moving boundary conditions of the hydrodynamic model according to the tide forecast data and real-time tide data; Based on the synthetic wind field and real-time meteorological data, the hydrodynamic mathematical model is used to calculate the change of output water level over time, the velocity field of water flow, surface wind stress and bed shear stress; The data output by the hydrodynamic model is used to simulate the storm surge through the tidal mathematical model.

9. A digital bay hydrological condition early warning method according to claim 1, characterized in that: The S5 specifically includes: Determine hydrological condition warning indicators and set warning thresholds based on wave forecast data and storm surge simulation results; According to the threshold of hydrological condition warning indicators, the hydrological condition warning is divided into 4 warning levels and corresponding response measures are formulated; Based on wave prediction data and storm surge simulation results, the warning level for the bay area to be predicted is determined and corresponding measures are implemented.

10. A digital bay hydrological condition early warning system, used to implement a digital bay hydrological condition early warning method as described in any one of claims 1 to 9, characterized in that: The system includes the following modules: Data collection module: used to collect historical meteorological data and real-time meteorological data of the bay area to be predicted; the meteorological data includes: wind data, wave data and tidal data; Wave prediction module: connected to the data collection module, used to establish a relationship model between wind and waves based on historical meteorological data using deep reinforcement learning and genetic algorithms, and generate wave prediction data based on the relationship model and real-time wind data of the bay area to be predicted; Storm surge prediction module: connected to the data collection module, used to synthesize the synthetic wind field of the bay area to be predicted by using the theoretical model wind field and the background wind field; based on the synthetic wind field, combined with real-time meteorological data, storm surge simulation is performed using a hydrodynamic mathematical model and a tidal mathematical model; The early warning module is connected to the wave prediction module and the storm surge prediction module, and generates hydrological condition early warning information based on the wave prediction data and the storm surge simulation result in combination with the early warning index threshold.

Images