Cyclone similarity recognition method and storm surge prediction method based on multi-algorithm fusion of weight optimization

Abstract

The application discloses a kind of multi-algorithm fusion cyclone similarity identification method and storm surge forecasting method based on weight optimization, it is related to cyclone similarity identification field, it is by constructing multiple key factors multidimensional time series, combined with weighted dynamic time warping algorithm, similarity calculation function is established.Such realization multidimensional comprehensive representation to cyclone space-time evolution process, overcome the identification deviation caused by traditional method only relying on single feature.In this regard, genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are used to iteratively optimize weight vector, with the minimum dispersion coefficient of candidate similar cyclone set as the target, to search for the global optimal weight vector corresponding to each algorithm;Again based on preset fusion strategy, determine the target optimal weight vector from multiple global optimal solutions.The closed-loop identification process effectively integrates multi-dimensional feature modeling and scientific weight optimization, significantly improves the objectivity and accuracy of cyclone similarity identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cyclone similarity identification, and in particular to a cyclone similarity identification method based on weight optimization and a storm surge prediction method based on multi-algorithm fusion. BACKGROUND

[0002] China is located in the west coast of the northwest Pacific and is one of the regions with the most frequent tropical cyclone activities in the world. With the intensification of climate change, the generation frequency and intensity of tropical cyclones and extratropical cyclones are on the rise, which poses a serious threat to coastal areas. The analogy analysis method based on the evolution characteristics of historical similar cyclones has important application value in business warning because it can reflect the overall behavior pattern of the cyclone system. However, the existing cyclone similarity identification technology still has obvious shortcomings in feature expression and weight determination, which restricts its application in high-precision disaster deduction.

[0003] First, the existing method has the problem of single feature dimension in similarity criterion. Traditional methods mostly rely on single indicators such as cyclone path trajectory or central intensity for matching, and fail to fully utilize the multi-dimensional dynamic information contained in the path points, such as center position, moving speed, center maximum wind speed and center minimum pressure, etc. Such single-dimensional or low-dimensional matching strategy is difficult to fully describe the spatio-temporal evolution characteristics of cyclones, resulting in insufficient recognition ability for complex path turning or intensity mutation events, and the similarity cyclone screening result is prone to deviation.

[0004] Second, the weight determination of key factors relies on subjective experience and lacks scientific optimization mechanism. In the current business, expert weighting or equal weight averaging is commonly used to set the weight of each feature parameter, and the weight value is fixed and lacks data-driven support. Since the contribution of each factor to the similarity judgment is different in different cyclone events, the artificially set weight is difficult to approach the optimal discriminant function. In addition, the existing method mostly uses Euclidean distance or simple dynamic time warping (DTW) for similarity calculation, without considering the relative importance difference of each factor in time series matching, which affects the accuracy and robustness of similarity ranking.

[0005] Under this background, it is of great significance to develop a cyclone similarity identification method that can fuse multi-dimensional path features and automatically optimize the weight based on historical matching results, in order to improve the objectivity and precision of similar cyclone screening. Further, the complete path information of the identified optimal similar cyclone (including the wind field, pressure field and measured open boundary tidal level of the subsequent path points) is input into the refined storm surge numerical model as driving and boundary conditions, which can realize the rapid analog prediction of the storm surge process that may be caused by the current cyclone, making up for the deficiency of traditional physical models in real-time, and providing efficient and reliable disaster warning support for coastal areas. SUMMARY

[0006] In order to solve the above technical problems, the present application provides a multi-algorithm fusion cyclone similarity identification method based on weight optimization, comprising:

[0007] Obtaining historical cyclone data, the historical cyclone data including path records of a plurality of historical cyclones, each path record being composed of a plurality of path points distributed at equal time intervals, each path point containing a plurality of key factors; performing global standardization processing on the values of each key factor in all path records; for each path record of a historical cyclone, based on the corresponding standardized key factors of its successive path points, constructing a multi-dimensional time series representing the spatio-temporal evolution process of the cyclone;

[0008] Constructing a weight vector containing the weight coefficients corresponding to each key factor as the parameter structure of weight optimization; based on the weight vector and the multi-dimensional time series of the cyclone, constructing a similarity calculation function using a weighted dynamic time warping algorithm, which is used to calculate the similarity value between two cyclones;

[0009] Respectively using genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm, initializing the population based on the parameter structure, and iteratively optimizing the weight vectors in the population; each algorithm constructs a candidate similar cyclone set based on the current weight vector and the similarity calculation function during the iteration process, and calculates the dispersion coefficient corresponding to the set, taking the dispersion coefficient as the optimization objective, searching for the optimal solution by minimizing the dispersion coefficient, and finally obtaining the global optimal weight vector corresponding to each algorithm; based on a preset fusion strategy, determining the target optimal weight vector therefrom;

[0010] Substituting the target optimal weight vector into the similarity calculation function, recalculating the similarity values of the current measured cyclone and each historical cyclone, generating a similarity ranking list, and selecting the historical cyclone with the highest similarity value as the optimal similar cyclone.

[0011] Further, the key factors include: the central longitude of the cyclone, the central latitude, the central maximum wind speed, the central minimum pressure and the horizontal moving speed.

[0012] Further, the genetic algorithm is used to initialize the population based on the parameter structure, and iteratively optimize the weight vectors in the population, specifically including:

[0013] Initializing the population: randomly generating a plurality of weight vector individuals based on the parameter structure to form an initial population;

[0014] Calculating fitness value: for each weight vector individual in the population, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate its similarity value with all historical cyclones, form a similarity ranking list; select the top pre-set number of historical cyclones from the ranking list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the individual;

[0015] Selecting parent individuals: sort the individuals in the population in descending order according to the fitness values, the higher the fitness value, the better the consistency of the candidate similar cyclone set corresponding to the weight vector; select several individuals as parents using roulette selection method;

[0016] Performing crossover operation: randomly pairing the parent individuals two by two, performing arithmetic crossover on the weight vectors of each pair of parent individuals to generate two new offspring individuals; wherein, the acquisition method of each offspring individual is: taking a random number between 0 and 1 as a combination coefficient, linearly weighting and combining the two parent weight vectors;

[0017] Performing mutation operation: applying random disturbance to the weight components of a pre-set proportion of offspring individuals, the disturbance amplitude obeys a Gaussian distribution with a mean of zero; after the disturbance, performing non-negative correction and normalization processing on the weight vector to ensure that each component is non-negative and the sum is 1;

[0018] Iterative updating: taking the newly generated offspring individuals as the next generation population, repeating the calculation of fitness value, the selection of parent individuals, the crossover and mutation operations; in each iteration, update and record the global optimal weight vector, which is the weight vector with the highest fitness value among all individuals up to the current generation; when the dispersion coefficient corresponding to the optimal weight vector changes by less than a pre-set threshold within a certain number of consecutive generations, or reaches the maximum number of iterations, terminate the optimization and output the global optimal weight vector as the optimal weight vector obtained by the genetic algorithm.

[0019] Further, a particle swarm optimization algorithm is used to initialize the population based on the parameter structure, and the weight vectors in the population are iteratively optimized, specifically including:

[0020] Initializing the particle swarm: randomly generating a plurality of particles based on the parameter structure to form an initial population, each particle corresponding to a weight vector, and initializing a velocity vector for each particle;

[0021] Calculating fitness value: for each weight vector corresponding to a particle, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate its similarity value with all historical cyclones, form a similarity ranking list; select the top pre-set number of historical cyclones from the ranking list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the particle;

[0022] Update individual and global optimal positions: For each particle, record the weight vector with the highest fitness value obtained in all iterations since initialization as the individual's historical optimal position; at the same time, maintain a global optimal position in the entire population, which corresponds to the weight vector with the highest fitness achieved by all particles since initialization.

[0023] Update velocity and position: For each particle, calculate the updated velocity vector based on its current velocity, the deviation of its current weight vector from its individual historical best position, and the deviation from its global best position; update its corresponding weight vector based on this velocity vector.

[0024] Constraint processing: The updated weight vector is corrected for nonnegativity and normalized to ensure that each component is nonnegative and the sum is 1;

[0025] Iterative update: Repeatedly perform fitness value calculation, optimal position update, velocity and position update and constraint processing operations; in each iteration, update and record the global optimal weight vector; when the deviation coefficient corresponding to the optimal weight vector changes less than a preset threshold within several consecutive generations, or when the maximum number of iterations is reached, the optimization is terminated, and the global optimal weight vector is output as the optimal weight vector obtained by particle swarm optimization.

[0026] Furthermore, the population is initialized based on the parameter structure using the differential evolution algorithm, and the weight vector in the population is iteratively optimized, specifically including:

[0027] Population initialization: Based on the parameter structure, a preset number of weight vectors, i.e. individuals, are randomly generated to form the initial population;

[0028] Calculate fitness value: For each weight vector in the population, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate its similarity value with all historical cyclones, and form a similarity ranking list; select the first preset number of historical cyclones according to the ranking list to form a candidate similar cyclone set, calculate the deviation coefficient corresponding to the set, and take the reciprocal of the deviation coefficient as the fitness value of the individual.

[0029] Generate a test vector: For each weight vector in the population, randomly select three other weight vectors that are different from each other as the parent, and generate a mutation vector accordingly; then perform a crossover operation between the mutation vector and the current weight vector to generate a test vector.

[0030] Boundary constraints and normalization: The generated test vectors are corrected for non-negativity and normalized to ensure that each component is non-negative and the sum is 1;

[0031] Selection operation: comparing the test vector with the current weight vector used to generate the test vector in terms of fitness value, selecting the vector with higher fitness value into the next generation population;

[0032] Iterative update: repeatedly performing the calculation of fitness value, the generation of test vector, the boundary constraint and normalization processing and the selection operation; updating and recording the weight vector with the highest fitness value among all individuals up to the current generation as the global optimal weight vector in each generation; when the dispersion coefficient corresponding to the global optimal weight vector changes by less than a preset threshold value for a plurality of consecutive generations, or reaches the maximum number of iterations, the optimization is terminated, and the global optimal weight vector is output as the optimal weight vector obtained by the differential evolution algorithm.

[0033] Further, the dispersion coefficient corresponding to the set is calculated, specifically:

[0034] The average value of the similarity values of each historical cyclone in the candidate similar cyclone set and the current measured cyclone is calculated; the mean value of the square of the difference between each similarity value and the average value is calculated to obtain the variance; the standard deviation is obtained by taking the square root of the variance; and the dispersion coefficient corresponding to the set is obtained by dividing the standard deviation by the average value of the similarity values.

[0035] Further, the target optimal weight vector is determined based on a preset fusion strategy; specifically: the global optimal weight vectors obtained by the genetic algorithm, the particle swarm optimization algorithm and the differential evolution algorithm are substituted into the similarity calculation function, respectively, to calculate the dispersion coefficients of the candidate similar cyclone sets corresponding thereto; the global optimal weight vector with the smallest dispersion coefficient is selected as the target optimal weight vector.

[0036] The embodiment of the application also provides a storm surge forecasting method, comprising:

[0037] The optimal similar cyclone most similar to the current measured cyclone is screened out according to the multi-algorithm fusion cyclone similarity identification method described above;

[0038] The path record of the current measured cyclone is obtained, and the complete path record of the optimal similar cyclone is obtained;

[0039] The latest observation time of the current measured cyclone is aligned with the corresponding relative development time point in the optimal similar cyclone path;

[0040] Each path point of the optimal similar cyclone after the alignment time is taken as a subsequent path point, and the key factors corresponding to each subsequent path point are extracted respectively; the key factors include: cyclone center position, center maximum wind speed, maximum wind speed radius and horizontal movement speed;

[0041] Adopt Jelesnianski wind field model and Holland air pressure model, utilize key factor calculation to wind field distribution and air pressure field distribution corresponding subsequent path point;

[0042] The wind field distribution, air pressure field distribution and optimal similar cyclone measured open boundary tidal level at corresponding path point are taken as driving and boundary conditions, input into ADCIRC refined storm surge prediction model, numerical simulation is carried out, and the storm surge water level field corresponding to each subsequent path point is output;

[0043] The storm surge water level field corresponding to each subsequent path point is taken as the storm surge prediction result corresponding to the current measured cyclone.

[0044] Further, the Jelesnianski wind field model is adopted, and the wind field distribution corresponding to the subsequent path point is calculated by using the key factor, specifically as follows:

[0045] The spatial distance between the center position of each grid point in the simulation area and the center position of the cyclone is calculated;

[0046] The wind speed and direction at each grid point are calculated based on the maximum wind speed radius, the horizontal moving speed and the wind field inflow angle;

[0047] Based on the spatial distance between the center position of each grid point and the center position of the cyclone and the wind speed and direction at each grid point, the asymmetric wind field distribution evolving with the movement of the cyclone center is generated.

[0048] Further, the Holland air pressure model is adopted, and the air pressure field distribution corresponding to the subsequent path point is calculated by using the key factor, specifically as follows:

[0049] The key factor corresponding to the subsequent path point is obtained, including the center position of the cyclone, the central minimum air pressure, the environmental background air pressure and the maximum wind speed radius;

[0050] The spatial distance between the center position of each grid point in the simulation area and the center position of the cyclone is calculated by the key factor;

[0051] The air pressure value at each grid point is calculated based on the maximum wind speed radius as the structural core, the pressure difference between the central minimum air pressure and the environmental background air pressure, and the spatial distance between the center position of each grid point and the center position of the cyclone, according to the nonlinear law that the air pressure decreases with distance;

[0052] The dynamic air pressure field with the center position of the cyclone as the low pressure core and the axisymmetric distribution is constructed based on the air pressure value at each grid point.

[0053] Compared with the prior art, the present application has at least the following beneficial effects:

[0054] (1) The present application realizes multi-dimensional comprehensive representation of the spatio-temporal evolution process of cyclones by constructing a multi-dimensional time series containing multiple key factors, combining a weighted dynamic time warping algorithm, and establishing a similarity calculation function. This overcomes the recognition bias caused by the dependence of traditional methods on a single feature, significantly improving the ability to describe complex path turning and intensity mutation events. On this basis, genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are used to iteratively optimize the weight vector, with the minimum dispersion coefficient of the candidate cyclone set as the target, to search for the global optimal weight vector corresponding to each algorithm; and based on the preset fusion strategy, the target optimal weight vector is determined from multiple global optimal solutions. This multi-algorithm collaborative optimization mechanism avoids the problems of single optimization method falling into local optimum or unstable convergence. Finally, based on the target optimal weight vector, the historical cyclone with the highest similarity to the current measured cyclone is selected as the optimal similar cyclone. This closed-loop identification process effectively integrates multi-dimensional feature modeling and scientific weight optimization, significantly improving the objectivity, accuracy and business applicability of cyclone similarity identification.

[0055] (2) In the weight optimization process, for each candidate cyclone set, the average value, variance, and further the ratio of standard deviation to mean of the similarity values of the candidate cyclone set to the current measured cyclone are calculated to obtain the dispersion coefficient as the optimization target. This realizes quantitative evaluation of the consistency within the similar cyclone set, overcomes the problem of neglecting dispersion caused by the dependence of traditional methods on the mean similarity, and significantly improves the statistical reasonableness and robustness of the optimization target. On this basis, the global optimal weight vectors obtained by genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are substituted into the similarity function, and the dispersion coefficients of the corresponding candidate sets are compared to select the minimum one as the target optimal weight vector. This fusion strategy effectively avoids the convergence bias or local optimum risk that may exist in a single optimization algorithm, significantly enhancing the stability and global optimality of the weight determination process.

[0056] (3) After the optimal similar cyclone is identified, the complete path record thereof is obtained, and the latest observation time of the current measured cyclone is aligned with the corresponding relative development time point in the path of the similar cyclone. In this way, dynamic matching of the cyclone life cycle is realized, the path deduction deviation caused by inconsistent time starting points in the traditional analogy method is overcome, and the timing consistency of subsequent path prediction is significantly improved. On this basis, the key factors corresponding to each path point of the optimal similar cyclone after the alignment time are extracted as the basis for deducing the future path of the current cyclone, and the Jelesnianski wind field model and the Holland pressure model are combined to calculate the wind field and pressure field distribution of each subsequent path point. This analogy deduction mechanism does not need to rely on the numerical weather prediction model to provide future wind field and pressure field, avoids the dependence on atmospheric initial field assimilation, and significantly improves the rapid response ability and business operability of storm surge prediction.

[0057] (4) The present application adopts the Jelesnianski wind field model, calculates the spatial distance between the center of each grid point in the simulation area and the center of the cyclone based on the key factors of the subsequent path points, and combines the maximum wind speed radius, horizontal moving speed and wind field inflow angle to construct an asymmetric wind field distribution. In this way, the high-fidelity reconstruction of the typical wind field structure of the tropical cyclone is realized, and the physical authenticity of the wind field forcing is significantly improved. On this basis, the generated wind field distribution is input as a driving condition into the ADCIRC refined model, effectively enhancing the accuracy of storm surge simulation, especially in the nearshore complex terrain area.

[0058] (5) The present application adopts the Holland pressure model, calculates the pressure value of each grid point based on the center position of the subsequent path point of the cyclone, the center minimum pressure, the environmental background pressure and the maximum wind speed radius, and constructs an axisymmetric pressure field with the center of the cyclone as the low pressure core. In this way, the reasonable modeling of the cyclone pressure structure is realized, the distortion problem of the pressure gradient force caused by the uniform or linear pressure field assumption is overcome, and the accuracy of the pressure field driving is significantly improved. Further, the wind field distribution, the pressure field distribution and the measured open boundary tidal level of the optimal similar cyclone at the corresponding path point are collectively input as driving and boundary conditions into the ADCIRC model to output the storm surge water level field. This multi-source data fusion driving mechanism fully combines the advantages of analogy deduction and physical model, realizes high-precision prediction under minute-level rapid response, and significantly improves the intelligentization and business level of the marine disaster warning system. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 It is a weight optimization-based multi-algorithm fusion cyclone similarity identification method flow chart.

[0060] Figure 2 It is a storm surge prediction method flow chart. DETAILED DESCRIPTION

[0061] The following is a specific embodiment of the present application and further describes the technical solutions of the present application in conjunction with the drawings, but the present application is not limited to these embodiments.

[0062] The present application is to solve the problem of low matching accuracy caused by single feature dimension and subjective weight setting in the existing cyclone similarity identification method, such as Figure 1 As shown in the figure, the embodiment of the present application proposes a multi-algorithm fusion cyclone similarity identification method based on weight optimization, which includes:

[0063] Obtain historical cyclone data, the historical cyclone data includes path records of a plurality of historical cyclones, each path record is composed of a plurality of path points distributed at equal time intervals, and each path point contains a plurality of key factors; the values of each key factor in all path records are globally standardized; for each path record of a historical cyclone, a multi-dimensional time series representing the spatio-temporal evolution process of the cyclone is constructed based on the standardized key factors corresponding to the continuous path points contained therein.

[0064] The key factors include: central longitude of the cyclone, central latitude of the cyclone, central maximum wind speed, central minimum pressure, and horizontal moving speed.

[0065] In this embodiment, based on the tropical cyclone and extratropical cyclone path record data collected by the China Meteorological Administration from 1945 to 2024, path records of a plurality of historical cyclones are obtained. Each path record is sampled at a fixed time interval of 6 hours, and each path point contains measured data of five key factors of central longitude, central latitude, central maximum wind speed, central minimum pressure, and horizontal moving speed, forming a cyclone multi-dimensional feature database with uniform spatio-temporal resolution. Among them, the "central maximum wind speed" refers to the maximum sustained wind speed (usually 1 minute or 10 minute average wind speed, depending on the data source standard) appearing in the eye wall area of the cyclone, located on the annular region at a certain distance from the center of the cyclone, used to represent the intensity of the cyclone; "maximum wind speed radius" is the horizontal distance from the center of the cyclone to the position of the maximum wind speed, reflecting the compactness of the wind circle structure. The above parameters are derived from the China Meteorological Administration Typhoon Yearbook and international general data sets (such as IBTrACS, CMA-STI), provided in the form of discrete path points, which can be directly used to construct a multi-dimensional time series representing the spatio-temporal evolution process of the cyclone, and support subsequent similarity identification and storm surge prediction calculation.

[0066] In this embodiment, the values of each key factor (including central longitude, central latitude, central maximum wind speed, central minimum pressure and horizontal movement speed) in all path records are globally standardized. Specifically, for each key factor, the overall mean and overall standard deviation of all historical path points are used to normalize all observed values of the factor: the original observed value is subtracted from the overall mean of the factor and then divided by the overall standard deviation of the factor, thereby eliminating the influence of different factors due to the differences in dimension and numerical range. After standardization, the values of each factor are converted into dimensionless relative variables, forming a standardized multi-dimensional feature vector, and arranged in chronological order, finally generating a high-resolution multi-dimensional time series representing the spatio-temporal evolution process of the cyclone, which is used for subsequent similarity calculation. Among them:

[0067] The formula expression of the multi-dimensional feature vector is: ;

[0068] In the formula, , , , , represent the longitude of the cyclone center position corresponding to the path point, the latitude of the center position, the maximum wind speed of the center, the minimum pressure of the center and the horizontal movement speed, respectively.

[0069] The formula expression of the multi-dimensional time series is:

[0070] ;

[0071] In the formula, represents the multi-dimensional time series corresponding to the path record of any cyclone, represents the number of path points in the path record, represents the multi-dimensional feature vector corresponding to the i-th path point.

[0072] A weight vector containing the weight coefficients corresponding to each key factor is constructed as the parameter structure of weight optimization; based on the weight vector and the multi-dimensional time series of the cyclone, a similarity calculation function using the weighted dynamic time warping algorithm is constructed, which is used to calculate the similarity value between two cyclones; specifically:

[0073] Let the multi-dimensional time series of the current measured cyclone to be compared be:

[0074] ;

[0075] Let the multi-dimensional time series of a certain historical cyclone be: ;

[0076] Among them: , are the number of path points corresponding to the cyclone, respectively.

[0077] and respectively represent the multi-dimensional feature vector of the current measured cyclone's th path point and the multi-dimensional feature vector of the historical cyclone's th path point.

[0078] The local distance between two path points is calculated by weighted Euclidean distance:

[0079] ;

[0080] wherein: to are the weight coefficients of each key factor (cyclone center position longitude, center position latitude, center maximum wind speed, center minimum pressure and horizontal moving speed) respectively, which constitute the weight vector: , and satisfy ;

[0081] Based on the weighted dynamic time warping algorithm, the cumulative distance matrix M is constructed, and its recursive relationship is as follows:

[0082] ; and ;

[0083] The boundary condition is:

[0084] , which means that the cumulative distance is zero when the starting path points of the two sequences are aligned;

[0085] ; wherein, , represents the cumulative distance when the first path points of the current measured cyclone are sequentially aligned with the first path point of the historical cyclone;

[0086] ; wherein, , represents the cumulative distance when the first j path points of the historical cyclone are sequentially aligned with the first path point of the current measured cyclone.

[0087] Finally, the last element of the cumulative distance matrix is taken as the minimum cumulative distance between the two sequences:

[0088] ;

[0089] The matching path corresponding to the minimum cumulative distance is the optimal nonlinear alignment relationship between the path points of the two cyclones, reflecting the structural similarity in the evolution process of the two cyclones.

[0090] A similarity calculation function is constructed based on the above-mentioned weighted dynamic time warping algorithm, and the accumulated distance is converted into a normalized similarity value: the formula expression of the similarity calculation function is

[0091]

[0092] wherein, d is a similarity value, and d is a normalized similarity value. represents an empirical scale parameter. The function is a similarity calculation function using the weighted dynamic time warping algorithm, and the output value represents the similarity degree of the two cyclones, and the greater the value, the higher the similarity, represents a similarity value.

[0093] The genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are respectively used to initialize the population based on the parameter structure, and the weight vectors in the population are iteratively optimized; each algorithm constructs a candidate similar cyclone set based on the current weight vector and the similarity calculation function during the iteration process, and calculates the dispersion coefficient corresponding to the set, taking the dispersion coefficient as the optimization target, searching for the optimal solution by minimizing the dispersion coefficient, and finally obtaining the global optimal weight vector corresponding to each algorithm; based on a preset fusion strategy, a target optimal weight vector is determined therefrom;

[0094] The genetic algorithm is used to initialize the population based on the parameter structure, and the weight vectors in the population are iteratively optimized, specifically including:

[0095] Initializing the population: a plurality of weight vector individuals are randomly generated based on the parameter structure to form an initial population;

[0096] Calculating the fitness value: for each weight vector individual in the population, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate the similarity values of all historical cyclones, form a similarity ranking list; according to the ranking list, select the first pre-set historical cyclones to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the individual;

[0097] Selecting parent individuals: the individuals in the population are sorted in descending order according to the fitness value, and the higher the fitness value, the better the consistency of the candidate similar cyclone set corresponding to the weight vector; a plurality of individuals are selected as parents by using the roulette wheel selection method;

[0098] Performing a crossover operation: randomly pairing the parent individuals, performing arithmetic crossover on the weight vectors of each pair of parent individuals to generate two new offspring individuals; wherein, each offspring individual is obtained by: taking a random number between 0 and 1 as a combination coefficient, and linearly weighting and combining the two parent weight vectors;

[0099] For this purpose, the following example is given:​

[0100] Suppose two parent individuals (weight vectors) are:

[0101] Parent 1: ;

[0102] Parent 2: ;

[0103] Randomly generate a combination coefficient , such as .

[0104] Perform linear combination to generate two offspring:

[0105] Offspring 1: ;

[0106] Offspring 2: ;

[0107] This is equivalent to taking two symmetric points on the parent vector line to generate two new weight vectors.

[0108] Perform mutation operation: apply random disturbance to the weight components of a preset proportion of offspring individuals, and the disturbance amplitude follows a Gaussian distribution with a mean of zero; After disturbance, the weight vector is non-negatively corrected and normalized to ensure that each component is non-negative and the sum is 1;

[0109] Iterative update: take the newly generated offspring individuals as the next generation population, and repeat the calculation of fitness value, selection of parent individuals, crossover and mutation operation; In each iteration, update and record the global optimal weight vector, which is the weight vector with the highest fitness value among all individuals up to the current generation; When the dispersion coefficient corresponding to the optimal weight vector changes by less than a preset threshold in continuous generations, or reaches the maximum iteration number, terminate the optimization, and output the global optimal weight vector as the optimal weight vector obtained by the genetic algorithm.

[0110] Initialize the population based on the parameter structure using the particle swarm optimization algorithm, and iteratively optimize the weight vectors in the population, specifically including:

[0111] Initialize the particle swarm: randomly generate multiple particles based on the parameter structure, and form an initial population, each particle corresponds to a weight vector, and initialize a velocity vector for each particle;

[0112] Calculating fitness value: for each weight vector corresponding to a particle, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate its similarity value with all historical cyclones, form a similarity ranking list; select the top pre-set historical cyclones according to the ranking list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the particle;

[0113] Updating individual and global optimal position: for each particle, record the weight vector with the highest fitness value obtained in all iterations since initialization as the individual historical optimal position; at the same time, maintain a global optimal position in the entire population, which corresponds to the weight vector with the highest fitness value reached by all particles since initialization;

[0114] Updating speed and position: for each particle, calculate the updated speed vector according to the deviation of its current speed, current weight vector and individual historical optimal position, and the deviation from the global optimal position; update its corresponding weight vector based on the speed vector;

[0115] Constraint processing: non-negative correction and normalization processing is performed on the updated weight vector to ensure that each component is non-negative and the sum is 1;

[0116] Iterative update: repeat the calculation of fitness value, update of optimal position, update of speed and position, and constraint processing operation; in each iteration, update and record the global optimal weight vector; when the dispersion coefficient corresponding to the optimal weight vector changes by less than a pre-set threshold in consecutive generations, or reaches the maximum number of iterations, terminate the optimization, and output the global optimal weight vector as the optimal weight vector obtained by particle swarm optimization.

[0117] The differential evolution algorithm is used to initialize the population based on the parameter structure, and the weight vectors in the population are iteratively optimized, specifically including:

[0118] Initializing the population: randomly generating pre-set weight vectors, i.e. individuals, based on the parameter structure, and forming an initial population;

[0119] Calculating fitness value: for each weight vector in the population, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate its similarity value with all historical cyclones, form a similarity ranking list; select the top pre-set historical cyclones according to the ranking list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the individual;

[0120] Generating a trial vector: for each weight vector in the population, randomly select three other weight vectors that are different from each other as parents, and generate a mutation vector therefrom; then cross the mutation vector with the current weight vector to generate a trial vector;

[0121] Specifically, the process of generating a trial vector is as follows:

[0122] For each weight vector in the population (referred to as the "target vector"), randomly select three other weight vectors that are different from each other as parent individuals, denoted as and generate a mutation vector according to the difference mutation strategy, the generation formula is:

[0123] ;

[0124] In the formula, denotes a scaling factor, usually in the range of , used to control the perturbation amplitude of the difference vector.

[0125] Subsequently, cross the mutation vector with the current target vector (assuming ) to generate a trial vector, the cross operation determines whether the component of the trial vector comes from the mutation vector or the target vector with a preset cross rate CR (such as 0.8) as the probability. For example, as follows (to simplify the explanation, assume is a three-dimensional vector here):

[0126] Assume:

[0127] Target vector ;

[0128] Mutation vector , after truncating the negative value to 0:

[0129] ;

[0130] Set the cross rate CR = 0.8, randomly generate the cross decision of each dimension (for example: the first dimension "yes", the second dimension "no", and the third dimension "yes"), then the trial vector is constructed as follows:

[0131] First dimension: inherit from mutation vector → 0.7;

[0132] Second dimension: inherit from target vector → 0.5;

[0133] Third dimension: inherit from mutation vector → 0.44;

[0134] Get the trial vector: .

[0135] Boundary constraint and normalization processing: the generated test vector is non-negative and normalized to ensure that each component is non-negative and the sum is 1;

[0136] Selection operation: comparing the test vector with the current weight vector used to generate the test vector in terms of fitness value, selecting the vector with higher fitness value into the next generation population;

[0137] Iterative update: repeatedly performing the calculation of the fitness value, the generation of the test vector, the boundary constraint and the normalization processing, and the selection operation; updating and recording the weight vector with the highest fitness value among all individuals up to the current generation as the global optimal weight vector in each generation; when the variation of the dispersion coefficient corresponding to the global optimal weight vector is less than a preset threshold value for a plurality of consecutive generations, or the maximum number of iterations is reached, the optimization is terminated, and the global optimal weight vector is output as the optimal weight vector obtained by the differential evolution algorithm.

[0138] The dispersion coefficient corresponding to the set is calculated, specifically:

[0139] The average value of the similarity values of each historical cyclone in the candidate similar cyclone set and the current measured cyclone is calculated (specifically: the similarity value of each historical cyclone in the set with the current measured cyclone is calculated, and then the average value of these similarity values is calculated); the mean of the square of the difference between each similarity value and the average value is calculated to obtain the variance; the standard deviation is obtained by taking the square root of the variance; the standard deviation is divided by the average value of the similarity value to obtain the dispersion coefficient corresponding to the set.

[0140] To overcome the limitations of traditional subjective weighting method in weight distribution, the present application introduces the dispersion coefficient (Coefficient of Variation, ) as the optimization objective function of genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm. The optimization objective function is used to measure the relative dispersion degree of the similarity values of each historical cyclone in the candidate similar cyclone set and the current measured cyclone, and the smaller the value is, the higher the consistency of the similarity in the set is, and the stronger the system coordination is. In the optimization process, the algorithm adjusts the weight distribution of each key factor through iteration to minimize the dispersion coefficient as the target, so as to select the candidate similar cyclone set which is highly consistent with the current measured cyclone in each key factor and has stable internal similarity distribution, and realize the precise optimization of weight.

[0141] The target optimal weight vector is determined based on the preset fusion strategy; specifically: the global optimal weight vectors obtained by genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are substituted into the similarity calculation function respectively, and the dispersion coefficients of the corresponding candidate similar cyclone sets are calculated; the global optimal weight vector with the smallest dispersion coefficient is selected as the target optimal weight vector.

[0142] In the weight optimization process, for each candidate similar cyclone set, the average value, variance of the similarity value between the current measured cyclone and the candidate similar cyclone set are calculated, and the ratio of the standard deviation to the average value is further calculated to obtain a dispersion coefficient as an optimization target. In this way, the consistency of the similar cyclone set is quantitatively evaluated, the problem of neglecting dispersion caused by relying only on the average value of similarity in the traditional method is overcome, and the statistical rationality and robustness of the optimization target are significantly improved. On this basis, the global optimal weight vector obtained by the genetic algorithm, the particle swarm optimization algorithm and the differential evolution algorithm is substituted into the similarity function, the dispersion coefficients of the corresponding candidate sets are compared, and the minimum one is selected as the target optimal weight vector. The fusion strategy effectively avoids the convergence deviation or local optimization risk that may exist in a single optimization algorithm, and significantly enhances the stability and global optimality of the weight determination process.

[0143] The target optimal weight vector is substituted into the similarity calculation function, the similarity values between the current measured cyclone and each historical cyclone are recalculated, a similarity ranking list is generated, and the historical cyclone with the highest similarity value is selected as the optimal similar cyclone.

[0144] The present application establishes a similarity calculation function by constructing a multi-dimensional time series containing multiple key factors and combining a weighted dynamic time warping algorithm. In this way, the multi-dimensional comprehensive representation of the spatio-temporal evolution process of the cyclone is realized, the recognition bias caused by relying only on a single feature in the traditional method is overcome, and the ability to describe complex path turning and intensity mutation events is significantly improved. On this basis, genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are used to iteratively optimize the weight vector, and the global optimal weight vector corresponding to each algorithm is searched by minimizing the dispersion coefficient of the candidate similar cyclone set. Based on the preset fusion strategy, the target optimal weight vector is determined from multiple global optimal solutions. The multi-algorithm cooperative optimization mechanism avoids the problem that a single optimization method is prone to local optimization or unstable convergence. Finally, based on the target optimal weight vector, the historical cyclone with the highest similarity to the current measured cyclone is selected as the optimal similar cyclone. The closed-loop identification process effectively integrates multi-dimensional feature modeling and scientific weight optimization, significantly improving the objectivity, accuracy and business applicability of cyclone similarity identification.

[0145] Embodiment two

[0146] In order to overcome the defects of the traditional storm surge prediction model, such as the dependence on numerical weather prediction and the lack of real-time performance, as shown in Figure 2 The present application also provides a storm surge prediction method, which comprises the following steps:

[0147] According to the multi-algorithm fusion cyclone similarity identification method described above, the optimal similar cyclone most similar to the current measured cyclone is screened out;

[0148] obtaining a path record of a current measured cyclone and a complete path record of the optimal similar cyclone;

[0149] aligning a latest observation time of the current measured cyclone with a corresponding relative development time point in the optimal similar cyclone path;

[0150] In the embodiment, aligning the latest observation time of the current measured cyclone with the corresponding relative development time point in the optimal similar cyclone path means that, based on the similarity matching result generated by the process of “substituting the target optimal weight vector into the similarity calculation function, recalculating the similarity values of the current measured cyclone and each historical cyclone, generating a similarity sorting list, and selecting the historical cyclone with the highest similarity value as the optimal similar cyclone” in the multi-algorithm fusion cyclone similarity identification method, a nonlinear spatiotemporal correspondence between the current measured cyclone and the optimal similar cyclone is obtained (wherein the similarity value is determined by the minimum cumulative distance of the weighted dynamic time warping algorithm, and the optimal matching path corresponding to the similarity value is the nonlinear spatiotemporal correspondence between the current measured cyclone and the optimal similar cyclone); according to the correspondence, the path point position of the latest observation time (such as T time) of the current measured cyclone in the optimal similar cyclone path is determined, that is, the “relative development time point”, so as to realize the alignment of the two in the life cycle stage (such as generation, enhancement, maturity or decay). The alignment provides an accurate time reference for subsequent calling of the measured storm surge data of the optimal similar cyclone, and supports analog prediction.

[0151] taking each path point of the optimal similar cyclone after the alignment time as a subsequent path point, and extracting a key factor corresponding to each subsequent path point; the key factor includes a cyclone center position, a center maximum wind speed, a maximum wind speed radius and a horizontal moving speed;

[0152] using the Jelesnianski wind field model and the Holland pressure model, calculating the wind field distribution and the pressure field distribution of the corresponding subsequent path point by using the key factor;

[0153] using the Jelesnianski wind field model, calculating the wind field distribution of the corresponding subsequent path point by using the key factor, specifically:

[0154] calculating the spatial distance between the center position of each grid point in the simulation area and the cyclone center position;

[0155] taking the maximum wind speed radius as the structural core, combining the horizontal moving speed and the wind field inflow angle, and calculating the wind speed and direction at each grid point;

[0156] Based on the spatial distance between the center position of each grid point and the center position of the cyclone, the wind speed and direction on each grid point, an asymmetric wind field distribution evolving with the movement of the cyclone center is generated.

[0157] The application adopts the Jelesnianski wind field model, calculates the spatial distance between the center position of each grid point in the simulation area and the cyclone center based on the key factors of the subsequent path point, and combines the maximum wind speed radius, horizontal movement speed and wind field inflow angle to construct the asymmetric wind field distribution. In this way, the typical wind field structure of the tropical cyclone is reconstructed with high fidelity, and the physical reality of the wind field forcing is significantly improved. On this basis, the generated wind field distribution is input as the driving condition into the ADCIRC refined model, effectively enhancing the accuracy of the storm surge simulation, especially in the nearshore complex terrain area, showing stronger adaptability and prediction ability.

[0158] The Holland pressure model is adopted, and the pressure field distribution corresponding to the subsequent path point is calculated by using the key factors, specifically:

[0159] The key factors corresponding to the subsequent path point are obtained, including the center position of the cyclone, the center minimum pressure, the environmental background pressure and the maximum wind speed radius;

[0160] The spatial distance between the center position of each grid point in the simulation area and the center position of the cyclone is calculated by using the key factors;

[0161] Taking the maximum wind speed radius as the core structure, combining the pressure difference of the center minimum pressure and the environmental background pressure, and the spatial distance between the center position of each grid point and the center position of the cyclone, the pressure values on each grid point are calculated according to the nonlinear law that the pressure decreases with distance;

[0162] Based on the pressure values on each grid point, a dynamic pressure field is constructed with the center position of the cyclone as the core of low pressure and in an axisymmetric distribution.

[0163] The application adopts the Holland pressure model, calculates the pressure values on each grid point based on the center position of the cyclone, the center minimum pressure, the environmental background pressure and the maximum wind speed radius of the subsequent path point, and constructs an axisymmetric pressure field with the center of the cyclone as the core of low pressure. In this way, the cyclone pressure structure is reasonably modeled, the distortion problem of the pressure gradient force caused by the uniform or linear pressure field assumption is overcome, and the accuracy of the pressure field driving is significantly improved. Further, the wind field distribution, the pressure field distribution and the measured open boundary tidal level of the optimal similar cyclone at the corresponding path point are collectively input as the driving and boundary conditions into the ADCIRC model to output the storm surge water level field. The multi-source data fusion driving mechanism fully combines the advantages of analogy and physical model, realizes high-precision prediction under the condition of minute-level rapid response, and significantly improves the intelligentization and business level of the marine disaster warning system.

[0164] The wind field distribution, the atmospheric pressure field distribution and the measured open boundary tidal level of the optimal similar cyclone at the corresponding path point are taken as driving and boundary conditions, input into the ADCIRC refined storm surge prediction model, numerical simulation is carried out, and the storm surge water level field corresponding to each subsequent path point is output;

[0165] In the embodiment, the measured open boundary tidal level of the optimal similar cyclone at the corresponding path point refers to the actual tidal level data at the open boundary section provided by the marine observation station or reanalysis data corresponding to each path point time of the historical observation path of the optimal similar cyclone, which has deducted the astronomical tide component and retained the storm surge water level part caused by the cyclone, as the open boundary water level driving condition of the ADCIRC refined storm surge prediction model.

[0166] The storm surge water level field corresponding to each subsequent path point output is taken as the storm surge prediction result corresponding to the current measured cyclone.

[0167] The ADCIRC model adopts unstructured triangular mesh and Galerkin finite element method to solve shallow water dynamic equations, and can simulate the storm surge water level field with high precision. The present application realizes the analog prediction of the storm surge possibly caused by the current measured cyclone by using the mature physical mechanism.

[0168] After the optimal similar cyclone is identified, the complete path record thereof is obtained, and the latest observation time of the current measured cyclone is aligned with the corresponding relative development time point in the similar cyclone path. In this way, the dynamic matching of the cyclone life cycle is realized, the path deduction deviation caused by the inconsistent time starting point in the traditional analog method is overcome, and the time sequence consistency of the subsequent path prediction is significantly improved. On this basis, the key factors corresponding to each path point of the optimal similar cyclone after the alignment time are extracted as the basis for deducing the future path of the current cyclone, and the Jelesnianski wind field model and the Holland atmospheric pressure model are combined to calculate the wind field and atmospheric pressure field distribution of each subsequent path point. The analog deduction mechanism does not need to rely on the numerical weather prediction model to provide the future wind field and atmospheric pressure field, avoids the dependence on the assimilation of atmospheric initial field, and significantly improves the rapid response ability and business operability of the storm surge prediction.

[0169] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, motion condition, etc. between components in a certain posture (as shown in the drawings), if the certain posture changes, the directional indications also change accordingly.

[0170] In addition, in the present application, the description such as "first", "second", "one" and the like are only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the technical features or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise explicitly specified and limited.

[0171] In the present application, unless otherwise explicitly specified and limited, the terms "connection", "fixing" and the like should be understood broadly, for example, "fixing" can be fixed connection, or detachable connection, or integral; can be mechanical connection, or electrical connection; can be directly connected, or indirectly connected through intermediate medium, can be the internal communication of two elements or the interaction relationship between two elements, unless otherwise explicitly limited. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0172] In addition, the technical solutions of various embodiments of the present application can be combined with each other, but it must be based on the realization of the ordinary skilled in the art, when the combination of technical solutions appears contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist, nor within the scope of protection required by the present application.

Claims

1. A multi-algorithm fusion cyclone similarity identification method based on weight optimization, characterized in that, The application comprises the following steps: acquiring historical cyclone data, which comprises path records of a plurality of historical cyclones, each path record being composed of a plurality of path points distributed at equal time intervals, each path point containing a plurality of key factors; and performing global standardization processing on the values of each key factor in all path records; for each path record of a historical cyclone, based on the corresponding standardized key factors of the successive path points contained therein, a multi-dimensional time series representing the spatio-temporal evolution process of the cyclone is constructed; a weight vector containing the corresponding weight coefficients of each key factor is constructed as a parameter structure for weight optimization; based on the weight vector and the multi-dimensional time series of the cyclone, a similarity calculation function using a weighted dynamic time warping algorithm is constructed, which is used to calculate the similarity value between two cyclones; genetic algorithm, particle swarm optimization algorithm and differential evolution algorithm are respectively used to initialize a population based on the parameter structure and to iteratively optimize the weight vectors in the population; in the iteration process, each algorithm constructs a candidate similar cyclone set based on the current weight vector and the similarity calculation function, and calculates the dispersion coefficient corresponding to the set; the dispersion coefficient is taken as the optimization objective, and the optimal solution is searched by minimizing the dispersion coefficient, so as to finally obtain the global optimal weight vector corresponding to each algorithm; based on a preset fusion strategy, a target optimal weight vector is determined from the global optimal weight vectors obtained by the three algorithms; the preset fusion strategy is specifically as follows: the global optimal weight vectors obtained by the genetic algorithm, the particle swarm optimization algorithm and the differential evolution algorithm are substituted into the similarity calculation function respectively to calculate the dispersion coefficients of the candidate similar cyclone sets corresponding to the global optimal weight vectors; the global optimal weight vector with the smallest dispersion coefficient is selected as the target optimal weight vector; the target optimal weight vector is substituted into the similarity calculation function to recalculate the similarity values of the current measured cyclone and each historical cyclone, a similarity ranking list is generated, and the historical cyclone with the highest similarity value in the list is selected as the optimal similar cyclone.

2. The method according to claim 1, wherein, The key factors include the central longitude, the central latitude, the central maximum wind speed, the central minimum pressure and the horizontal moving speed of the cyclone.

3. The method according to claim 2, wherein, The genetic algorithm is used to initialize a population based on the parameter structure and to iteratively optimize the weight vectors in the population, which specifically comprises the following steps: initializing the population: a plurality of weight vector individuals are randomly generated based on the parameter structure to form an initial population; calculating the fitness value: for each weight vector individual in the population, the individual is substituted into the similarity calculation function to calculate the similarity values of the individual and all historical cyclones based on the current measured cyclone, so as to form a similarity ranking list; a candidate similar cyclone set is formed by selecting the first preset historical cyclones in the ranking list, the dispersion coefficient corresponding to the set is calculated, and the reciprocal of the dispersion coefficient is taken as the fitness value of the individual; selecting parent individuals: the individuals in the population are sorted in descending order according to the fitness values, and the higher the fitness value, the better the consistency of the candidate similar cyclone set corresponding to the weight vector; a plurality of individuals are selected as parents by using the roulette wheel selection method. Performing crossover operation: randomly pairing the parent individuals two by two, and performing arithmetic crossover on the weight vectors of each pair of parent individuals to generate two new offspring individuals; wherein the obtaining manner of each offspring individual is: taking a random number between 0 and 1 as a combination coefficient, and linearly weighting and combining the two parent weight vectors; Performing mutation operation: applying random disturbance to the weight components of a preset proportion of offspring individuals, and the disturbance amplitude obeys Gaussian distribution with a mean of zero; after the disturbance, performing non-negative correction and normalization processing on the weight vector to ensure that each component is non-negative and the sum is 1; Iterative updating: taking the newly generated offspring individuals as the next generation population, and repeatedly performing the calculation of the fitness value, the selection of the parent individual, the crossover and mutation operation; in each iteration, update and record the global optimal weight vector, which is the weight vector with the highest fitness value among all individuals up to the current generation; when the variation of the dispersion coefficient corresponding to the optimal weight vector is less than a preset threshold value for a continuous number of generations, or reaches the maximum number of iterations, terminate the optimization, and output the global optimal weight vector as the optimal weight vector obtained by the genetic algorithm.

4. The method of claim 2, wherein the method is characterized by, The particle swarm optimization algorithm is used to initialize the population based on the parameter structure, and the weight vectors in the population are iteratively optimized, specifically including: Initializing the particle swarm: generating a plurality of particles based on the parameter structure to form an initial population, each particle corresponding to a weight vector, and initializing a velocity vector for each particle; Calculating the fitness value: for each weight vector corresponding to a particle, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate the similarity values of all historical cyclones, and form a similarity sorting list; select the top pre-set historical cyclones from the sorting list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the particle; Updating the individual and the global optimal position: for each particle, record the weight vector with the highest fitness value obtained in all iterations since initialization as the individual historical optimal position; at the same time, maintain a global optimal position in the entire population, which corresponds to the weight vector with the highest fitness value reached by all particles since initialization; Updating the speed and position: for each particle, calculate the updated velocity vector according to the deviation of its current velocity, current weight vector and individual historical optimal position, and the deviation of the global optimal position; update the weight vector corresponding to the particle based on the velocity vector; Constraint processing: performing non-negative correction and normalization processing on the updated weight vector to ensure that each component is non-negative and the sum is 1; Iterative updating: repeatedly performing the calculation of the fitness value, the updating of the optimal position, the updating of the speed and position, and the constraint processing operation; in each iteration, update and record the global optimal weight vector; when the variation of the dispersion coefficient corresponding to the optimal weight vector is less than a preset threshold value for a continuous number of generations, or reaches the maximum number of iterations, terminate the optimization, and output the global optimal weight vector as the optimal weight vector obtained by the particle swarm optimization.

5. The method of claim 2, wherein the method is characterized by, The differential evolution algorithm is used to initialize a population based on the parameter structure, and the weight vectors in the population are iteratively optimized, specifically including: Initializing the population: generating a preset number of weight vectors, i.e., individuals, based on the parameter structure, and forming an initial population; Calculating the fitness value: for each weight vector in the population, substitute it into the similarity calculation function, take the current measured cyclone as the benchmark, calculate the similarity values of all historical cyclones, and form a similarity ranking list; select the top preset number of historical cyclones from the ranking list to form a candidate similar cyclone set, calculate the dispersion coefficient corresponding to the set, and take the reciprocal of the dispersion coefficient as the fitness value of the individual; Generating a test vector: for each weight vector in the population, randomly select three other weight vectors that are not the same as the current weight vector as parents, and generate a mutation vector based on them; then cross the mutation vector with the current weight vector to generate a test vector; Boundary constraint and normalization processing: non-negative correction and normalization processing are performed on the generated test vector to ensure that each component is non-negative and the sum is 1; Selection operation: compare the fitness values of the test vector and the current weight vector used to generate the test vector, and select the vector with higher fitness value to enter the next generation population; Iterative update: repeatedly perform fitness value calculation, test vector generation, boundary constraint and normalization processing, and selection operation; update and record the weight vector with the highest fitness value among all individuals up to the current generation in each generation as the global optimal weight vector; when the dispersion coefficient corresponding to the global optimal weight vector changes by less than a preset threshold in consecutive generations, or reaches the maximum number of iterations, terminate the optimization, and output the global optimal weight vector as the optimal weight vector obtained by the differential evolution algorithm.

6. The method according to any one of claims 3 to 5, wherein, The calculation of the dispersion coefficient corresponding to the set is specifically: Calculate the average of the similarity values of each historical cyclone in the candidate similar cyclone set and the current measured cyclone; calculate the average of the squares of the differences between each similarity value and the average, to obtain the variance; take the square root of the variance to obtain the standard deviation; divide the standard deviation by the average of the similarity values to obtain the dispersion coefficient corresponding to the set.

7. A storm surge forecasting method, characterized by, Including: The multi-algorithm fusion cyclone similarity identification method of claim 1 selects the optimal similar cyclone most similar to the current measured cyclone; Obtain the path record of the current measured cyclone, and the complete path record of the optimal similar cyclone; Align the latest observation time of the current measured cyclone with its corresponding relative development time point in the optimal similar cyclone path; Take each path point of the optimal similar cyclone after the alignment time as a subsequent path point, and extract the key factors corresponding to each subsequent path point; The key factors include: cyclone center position, center maximum wind speed, maximum wind speed radius, and horizontal movement speed; Use the Jelesnianski wind field model and the Holland pressure model to calculate the wind field distribution and pressure field distribution of the corresponding subsequent path point using the key factors; The wind field distribution, the atmospheric pressure field distribution and the measured open boundary tidal level of the optimal similar cyclone at the corresponding path point are taken as driving and boundary conditions, input into the ADCIRC refined storm surge prediction model, numerical simulation is carried out, and the storm surge water level field corresponding to each subsequent path point is output; All the output storm surge water level fields corresponding to each subsequent path point are taken as the storm surge prediction results corresponding to the current measured cyclone.

8. A storm surge forecasting method according to claim 7, characterised in that, The Jelesnianski wind field model is adopted, and the wind field distribution of the corresponding subsequent path point is calculated by using the key factor, specifically as follows: The spatial distance between the center position of each grid point in the simulation area and the center position of the cyclone is calculated; The wind speed and direction at each grid point are calculated based on the maximum wind speed radius, the horizontal moving speed and the wind field inflow angle; Based on the spatial distance between the center position of each grid point and the center position of the cyclone and the wind speed and direction at each grid point, the asymmetric wind field distribution evolving with the movement of the cyclone center is generated.

9. A storm surge forecasting method according to claim 8, characterised in that, The Holland atmospheric pressure model is adopted, and the atmospheric pressure field distribution of the corresponding subsequent path point is calculated by using the key factor, specifically as follows: The key factor corresponding to the subsequent path point is obtained, including: the center position of the cyclone, the central minimum atmospheric pressure, the environmental background atmospheric pressure and the maximum wind speed radius; The spatial distance between the center position of each grid point in the simulation area and the center position of the cyclone is calculated by the key factor; The atmospheric pressure value at each grid point is calculated based on the maximum wind speed radius as the structural core, the pressure difference between the central minimum atmospheric pressure and the environmental background atmospheric pressure, and the spatial distance between the center position of each grid point and the center position of the cyclone, according to the nonlinear law that the atmospheric pressure decreases with the distance; The dynamic atmospheric pressure field with the center position of the cyclone as the low-pressure core and the axisymmetric distribution is constructed based on the atmospheric pressure value at each grid point.

Images