Numerical mode physical parameterization scheme optimization method and device based on artificial intelligence

Abstract

The invention provides a numerical mode physical parameterization scheme optimization method and device based on artificial intelligence, and the method comprises the steps: determining a numerical mode physical parameterization scheme corresponding to a target weather type according to the target weather type; optimizing the numerical mode physical parameterization scheme by using a heuristic intelligent optimization algorithm and taking forecast precision as an optimization target; on the basis of a pre-constructed nestable optimization objective function, through iterative computation, optimizing sensitive parameters in the optimized numerical mode physical parameterization scheme; different constraint conditions are designed according to the physical significance of the to-be-optimized sensitive parameters, Pareto optimal parameters optimized by considering multiple meteorological targets are calculated, and evaluation and optimization of a multi-level numerical mode physical process parameterization scheme are achieved. The problems existing in the optimization process of a numerical mode physical parameterization scheme are solved, so that the scheme optimization effect is improved, and then the forecasting precision of numerical weather forecasting is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data processing, and in particular, to an optimization method and device for a physical parameterization scheme of a numerical model based on artificial intelligence. Background Art

[0002] Currently, the optimization of the physical process parameterization scheme of numerical models mainly adopts the sensitivity test method, and its technical solution can be summarized into the following steps:

[0003] First, parameterization scheme combination screening. Based on expert experience or historical test results, select combinations of several physical process parameterization schemes (such as boundary layer schemes, cloud microphysical schemes, convection schemes, etc.) to form a candidate scheme pool;

[0004] Second, sensitivity test design. By adjusting the key parameters in the parameterization scheme (such as turbulent diffusion coefficient, cloud droplet concentration, convection triggering threshold, etc.), design multiple groups of test schemes;

[0005] Third, model integration and error evaluation. Run the numerical model for forecast tests, and calculate the forecast errors (such as mean square error, correlation coefficient) of each test scheme based on the observed data;

[0006] Finally, iterative optimization. Select better schemes according to the error results, and repeat adjusting the parameters or replacing the parameterization scheme combination until the preset accuracy threshold is reached.

[0007] Furthermore, the typical implementation processes of the existing physical parameterization scheme optimization include manual parameter tuning, combined tests, and statistical optimization; among them, in the manual parameter tuning process, the parameter range is adjusted depending on expert experience (such as gradually adjusting the cloud droplet concentration from 50 cm -3 to 200 cm -3 ); the combined tests use the "full permutation" or "orthogonal test" method to test different parameterization scheme combinations (such as testing the influence of the YSU and MYJ boundary layer schemes on cold wave forecasts in the WRF model); during statistical optimization, a statistical relationship (such as a linear regression model) between parameters and forecast errors is established based on historical data to guide parameter adjustment.

[0008] Based on the above technical means, the following technical problems exist in the prior art:

[0009] First, it relies on subjective experience and lacks systematicness; parameter adjustment and scheme combination selection highly depend on expert experience, it is difficult to cover the high-dimensional parameter space, and it is easy to fall into local optima; for example, the adjustment range of the raindrop terminal velocity in the cloud microphysical scheme is usually limited to the literature recommended values, lacking global search ability;

[0010] Second, the calculation cost is high; the full permutation and combination test requires a large number of model integrations to be run (for example, if 10 parameters each take 5 values, 10^5 tests are required). Limited by supercomputer resources, only a limited number of combinations can actually be tested;

[0011] Third, the coupling effect of physical processes is ignored; traditional methods optimize the parameters of a single physical process in isolation and do not consider the interaction effects between parameterization schemes (such as the momentum coupling between boundary layer turbulence and convective processes);

[0012] Fourth, there is a lack of multi-objective collaborative optimization; the existing optimization objective is single (such as only minimizing the temperature error), resulting in parameter adjustments that may deteriorate the prediction accuracy of other elements (such as humidity or wind speed).

[0013] In view of this, a method and device for optimizing the physical parameterization scheme of a numerical model based on artificial intelligence are provided to solve the problems existing in the optimization process of the physical parameterization scheme of the numerical model, thereby improving the optimization effect of the scheme and further improving the prediction accuracy of numerical weather forecasting. Summary of the Invention

[0014] The present invention aims to provide a method and device for optimizing the physical parameterization scheme of a numerical model based on artificial intelligence to at least partially solve the above technical problems.

[0015] The present invention provides a method for optimizing the physical parameterization scheme of a numerical model based on artificial intelligence, the method comprising:

[0016] Determine the physical parameterization scheme of the numerical model corresponding to the target weather type according to the target weather type;

[0017] Use a heuristic intelligent optimization algorithm to optimize the physical parameterization scheme of the numerical model with the prediction accuracy as the optimization objective;

[0018] Based on a pre-constructed nestable optimization objective function, through iterative calculation, optimize the sensitive parameters in the optimized physical parameterization scheme of the numerical model;

[0019] Design different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculate the Pareto optimal parameters considering multi-meteorological objective optimization, and realize the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels.

[0020] In some embodiments, the physical parameterization scheme of the numerical model includes boundary layer processes, cloud microphysical processes, cumulus convection processes, long and short wave radiation processes, and surface processes;

[0021] The sensitive parameters include boundary layer parameters, cloud microphysical parameters, cumulus convection parameters, long and short wave radiation parameters, and surface parameters.

[0022] In some embodiments, the heuristic intelligent optimization algorithm is a genetic algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0023] S31: Randomly generate a set of parameter combinations as the initial population. The parameter combinations include multiple parameters corresponding to the physical processes of the target weather type.

[0024] S32: Calculate the forecast error of each parameter combination.

[0025] S33: Select excellent individuals from the initial population according to the fitness function values.

[0026] S34: Randomly select two individuals from the current population as parent individuals, and generate two offspring individuals through crossover operations to generate a new generation of population.

[0027] Use the new generation of population as the initial population, and repeat steps S32 - S34 until the preset convergence degree is reached or the preset number of iterations is reached.

[0028] Use the parameter combination in the current population when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

[0029] In some embodiments, the heuristic intelligent optimization algorithm is a particle swarm optimization algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0030] S41: Randomly generate a set of parameter combinations as the initial particle swarm. Each particle represents a parameter combination, and each particle has a randomly generated initial velocity and initial position. The parameter combinations include multiple parameters corresponding to the physical processes of the target weather type.

[0031] S42: Update the velocity and position of each particle according to the personal historical best solution and the global best solution of each particle.

[0032] S43: Calculate the forecast error of each particle.

[0033] S44: Use the forecast error as the input of the fitness function, and calculate the current fitness output by the fitness function.

[0034] S45: Update the personal best solution and the global best solution according to the current fitness and the historical fitness.

[0035] Repeat steps S42 - S45 until the preset iteration termination condition is reached.

[0036] Use the current parameter combination when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

[0037] In some embodiments, the nestable optimization objective function is constructed by integrating mode integral calculation and error index calculation based on the optimization objective function with sample set error feedback, and the sample set includes historical sensitive parameters and semi-empirical parameters.

[0038] In some embodiments, optimizing the sensitive parameters specifically includes:

[0039] Initializing the sensitive parameters and randomly selecting multiple sensitive parameters to form a parameter combination;

[0040] For each parameter combination, calculating the mode integral result and the error index;

[0041] Adjusting the sensitive parameters in the numerical model physical parameterization scheme according to the obtained integral result and error index;

[0042] Taking the updated sensitive parameters as the parameter combination and performing iterative calculation until the preset iteration condition is met and the iteration stops.

[0043] In some embodiments, the evaluation and optimization of the multi-level numerical model physical process parameterization scheme specifically includes:

[0044] The first level: optimizing the sensitive parameters for a single physical process;

[0045] The second level: based on the optimization results of the first level, jointly optimizing the parameters of multiple physical processes and their interactions;

[0046] The third level: comprehensively considering all physical processes, performing global optimization, and outputting the Pareto optimal solution.

[0047] The present invention also provides an optimization device for a numerical model physical parameterization scheme based on artificial intelligence, and the device includes:

[0048] A scheme determination unit, configured to determine a numerical model physical parameterization scheme corresponding to the target weather type according to the target weather type;

[0049] A scheme optimization unit, configured to use a heuristic intelligent optimization algorithm and take the forecast accuracy as the optimization objective to optimize the numerical model physical parameterization scheme;

[0050] A parameter optimization unit, configured to optimize the sensitive parameters in the optimized numerical model physical parameterization scheme through iterative calculation based on a pre-constructed nestable optimization objective function;

[0051] A multi-level optimization unit is used to design different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculate the Pareto optimal parameters considering the optimization of multiple meteorological objectives, and realize the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels.

[0052] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned method is implemented.

[0053] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned method is implemented.

[0054] The method and device for optimizing the physical parameterization scheme of a numerical model based on artificial intelligence provided by the present invention determine the physical parameterization scheme of the numerical model corresponding to the target weather type according to the target weather type; use a heuristic intelligent optimization algorithm to optimize the physical parameterization scheme of the numerical model with the forecast accuracy as the optimization objective; based on a pre-constructed nestable optimization objective function, optimize the sensitive parameters in the optimized physical parameterization scheme of the numerical model through iterative calculation; design different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculate the Pareto optimal parameters considering the optimization of multiple meteorological objectives, and realize the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels.

[0055] Based on the physical mechanism of the interaction and mutual influence of momentum, heat, and water vapor among the physical process parameterization schemes in the numerical weather prediction model, the method and device construct a heuristic intelligent optimization algorithm suitable for optimizing the physical parameters of the numerical model, and realize the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels through a nestable optimization objective. Solve the problems existing in the optimization process of the physical parameterization scheme of the numerical model, thereby improving the optimization effect of the scheme and further improving the forecast accuracy of the numerical weather prediction. Description of the Drawings

[0056] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0057] Figure 1 It is a flowchart of the method for optimizing the physical parameterization scheme of a numerical model based on artificial intelligence provided by the present invention;

[0058] Figure 2Structural block diagram of the optimization device for the physical parameterization scheme of the numerical model based on artificial intelligence provided by the present invention;

[0059] Figure 3 Schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0060] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] To solve the above technical problems, the present invention provides an optimization method for the physical parameterization scheme of the numerical model based on artificial intelligence. By using a global search method for the parameterization scheme based on an intelligent optimization algorithm to replace the traditional sensitivity test, the problem of low search efficiency in the parameter space is solved; a nested optimization objective function that combines physical mechanisms and data-driven is designed to achieve the collaborative optimization of parameters under the coupling effect of multiple processes; a multi-level optimization framework is constructed to balance the optimization of parameters of a single physical process and multiple objectives (such as temperature, humidity, wind speed, etc.), and improve the consistency of forecast accuracy.

[0062] The optimization method for the physical parameterization scheme of the numerical model based on artificial intelligence proposed by the present invention, on the basis of considering the physical mechanisms of momentum, heat and water vapor interaction and mutual influence among the physical parameterization schemes in the numerical weather prediction model, constructs a heuristic intelligent optimization algorithm suitable for the physical parameter optimization of the numerical model by using optimization algorithms such as genetic algorithms and particle swarm algorithms based on a heuristic artificial intelligence optimization algorithm framework. By constructing a nestable optimization objective that combines model integration calculation and forecast error index calculation, the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels are realized.

[0063] In a specific implementation manner, as Figure 1 shown, the optimization method for the physical parameterization scheme of the numerical model based on artificial intelligence provided by the present invention includes the following steps:

[0064] S110: According to the target weather type, determine the physical parameterization scheme of the numerical model corresponding to the target weather type; it should be understood that the target weather type refers to the weather that is focused on, such as low-temperature cold wave, freezing rain and thunderstorm, etc. According to the characteristics of different weather types, the physical parameterization scheme of the numerical model affecting the target weather type is determined by analysis.

[0065] S120: Using a heuristic intelligent optimization algorithm, with the prediction accuracy as the optimization goal, optimize the physical parameterization scheme of the numerical model; that is, for the parameterization scheme of the key physical processes of the model, adopt heuristic intelligent optimization algorithms such as genetic algorithms and particle swarm algorithms, with the prediction accuracy as the optimization goal, and carry out the optimization of the physical process parameterization scheme of the numerical model. Among them, the optimization goal is to make the prediction results output by the model as close as possible to the observed data by adjusting the parameters in the physical parameterization scheme of the numerical model. The optimization goal specifically includes prediction accuracy, physical consistency, and multi-objective optimization; among them, the prediction accuracy is based on the observed data, and the error (such as the mean square error MSE) between the model output and the observed data is minimized; physical consistency means ensuring that the optimized parameterization scheme conforms to the basic laws of atmospheric physical processes (such as energy conservation and mass conservation); multi-objective optimization means optimizing the prediction accuracy of multiple meteorological elements (such as temperature, humidity, wind speed, etc.) simultaneously.

[0066] S130: Based on the pre-constructed nestable optimization objective function, through iterative calculation, optimize the sensitive parameters in the optimized physical parameterization scheme of the numerical model; sensitive parameters refer to the parameters that have a greater impact on the model output results, and the sensitive parameters include boundary layer parameters, cloud microphysical parameters, cumulus convection parameters, short-wave and long-wave radiation parameters, and surface parameters; among them, the boundary layer parameters include turbulent diffusion coefficient, mixing length, friction velocity, etc., the cloud microphysical parameters include cloud droplet initial concentration, ice crystal generation rate, raindrop terminal velocity, etc., the cumulus convection parameters include convection triggering threshold, convective available potential energy (CAPE) release time scale, etc., the short-wave and long-wave radiation parameters include cloud optical thickness, aerosol optical properties, surface albedo, etc., and the surface parameters include soil moisture, vegetation coverage, surface roughness, etc.

[0067] S140: Design different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculate the Pareto optimal parameters considering multi-meteorological objective optimization, and realize the evaluation and optimization of the physical process parameterization scheme of the multi-level numerical model. The evaluation and optimization of the physical process parameterization scheme of the multi-level numerical model specifically include:

[0068] The first level: Optimize its sensitive parameters for a single physical process;

[0069] The second level: On the basis of the optimization results of the first level, jointly optimize the parameters of multiple physical processes and their interactions;

[0070] The third level: Comprehensively consider all physical processes, conduct global optimization, and output the Pareto optimal solution.

[0071] Among them, the constraint conditions specifically include physical constraints, parameter range constraints, and multi-objective balance constraints. Among them, physical constraints include energy conservation, mass conservation, momentum conservation, etc.; parameter range constraints mean that parameter values must be within a reasonable range (such as the turbulent diffusion coefficient being positive); multi-objective balance constraints refer to the trade-off between the optimization objectives of different meteorological elements (such as the balance between temperature and humidity).

[0072] Among them, the Pareto optimal parameters refer to the parameter combinations in multi-objective optimization where it is impossible to improve one objective without deteriorating other objectives. The specific content is the non-dominated solution set of the multi-objective optimization results, that is, multi-objective trade-off, and simultaneously optimize the prediction accuracy of elements such as temperature, humidity, and wind speed.

[0073] Specifically, the numerical model physical parameterization scheme is a mathematical expression or empirical formula used in numerical weather prediction models to describe sub-grid scale physical processes. The numerical model physical parameterization scheme includes boundary layer processes, cloud microphysical processes, cumulus convection processes, short and long wave radiation processes, and surface processes.

[0074] Among them, the boundary layer process is used to describe the process of turbulent exchange, heat, and momentum transport within the atmospheric boundary layer, and the boundary layer process can be realized through the K-ε turbulence closure model, the first-order closure scheme, etc.; the cloud microphysical process is used to describe the generation, growth, and dissipation processes of cloud droplets, raindrops, ice crystals, and other microphysical particles, and the cloud microphysical process can be carried out through single-moment schemes, double-moment schemes (such as the Thompson scheme), etc.; the cumulus convection process is used to describe the heating and water vapor transport of sub-grid scale convective activities (such as thunderstorms, cumulus convection), and can be realized through the Kain-Fritsch scheme, the Betts-Miller scheme; the short and long wave radiation process is used to describe the absorption, scattering, and transmission of solar short wave radiation and terrestrial long wave radiation, and can be realized through the RRTM (Rapid Radiative Transfer Model); the surface process is used to describe the surface energy balance, water transport, and vegetation effects, and can be realized through the Noah land surface model, the CLM (Community Land Model).

[0075] The present invention adopts a heuristic intelligent optimization algorithm, and the heuristic intelligent optimization algorithm includes a genetic algorithm and a particle swarm optimization algorithm.

[0076] In the above step S120, the heuristic intelligent optimization algorithm is a genetic algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0077] S31: Randomly generate a set of parameter combinations as the initial population, and the parameter combinations include multiple parameters corresponding to the physical processes of the target weather type.

[0078] S32: Calculate the prediction error of each parameter combination.

[0079] S33: Select excellent individuals from the initial population according to the fitness function values;

[0080] S34: Randomly select two individuals from the current population as parent individuals, and generate two offspring individuals through crossover operations to generate a new generation of population;

[0081] Use the new generation of population as the initial population, and repeat steps S32 - S34 until the preset convergence degree is reached or the preset number of iterations is reached;

[0082] Use the parameter combination in the current population when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

[0083] In a specific usage scenario, for the sake of easy understanding, the implementation process of the genetic algorithm is described exemplarily below in combination with the parameterization scheme of the key physical processes in weather forecasting:

[0084] 1. Initialize the population

[0085] According to the characteristics of the parameterization scheme of the key physical processes in the weather forecasting model, determine the parameter dimensions and ranges to be optimized; for example, assume that the parameters to be optimized include cloud microphysical process parameters, boundary layer parameters, cumulus convection parameters, etc., and each parameter has its reasonable value range.

[0086] Randomly generate a set of parameter combinations as the initial population. The population size can be determined according to the problem scale and computing resources, generally dozens to hundreds of individuals; for example, the population size is set to 100, and each individual represents a set of parameter combinations, and its value is randomly generated within the value range of the corresponding parameter.

[0087] 2. Fitness evaluation

[0088] For each parameter combination, substitute it into the weather forecasting model to run and obtain the forecast result of the model; then, compare the forecast result with the actual observed data to calculate the forecast error; commonly used error metrics include mean square error (MSE), mean absolute error (MAE), etc.; for example, assume that the actually observed temperature is T obs , and the temperature forecast by the model is T pred , then the mean square error can be expressed as:

[0089]

[0090] where n is the number of data points.

[0091] 3. Fitness function

[0092] Take the calculated prediction error as the input of the fitness function. The smaller the value of the fitness function, the better the performance of the parameter combination. For example, the MSE can be directly used as the fitness function, or the reciprocal of the MSE can be taken as the fitness function to convert it into a maximization problem.

[0093] 4. Selection

[0094] Select excellent individuals from the current population according to the fitness function values to generate the next generation population. Common selection methods include roulette wheel selection, tournament selection, ranking selection, etc. Taking roulette wheel selection as an example, first calculate the proportion of the fitness of each individual in the total fitness as the probability of its being selected. Then, randomly select individuals according to these probabilities. For example, assume there are 4 individuals in the population with fitness values of 0.1, 0.2, 0.3, and 0.4 respectively. Then their selection probabilities are 10%, 20%, 30%, and 40% respectively. To ensure that excellent individuals are not lost, an elite retention strategy can be adopted, that is, directly retain several individuals with the highest fitness into the next generation population. For example, retain 5% of the individuals with the highest fitness.

[0095] 5. Crossover and Mutation

[0096] Randomly select two individuals from the current population as parent individuals and generate two offspring individuals through the crossover operation. The methods of the crossover operation include single-point crossover, multi-point crossover, uniform crossover, etc. Taking single-point crossover as an example, randomly select a crossover point and exchange the gene segments of the two parent individuals at this point. For example, assume the parameter combinations of the two parent individuals are [a1, a2, a3] and [b1, b2, b3] respectively, and the crossover point is 2. Then the generated offspring individuals are [a1, b2, b3] and [b1, a2, a3].

[0097] Perform mutation operations on the newly generated offspring individuals to randomly change some gene values of the individuals with a certain probability. The mutation operation can increase the diversity of the population and avoid the algorithm falling into local optimality. The mutation probability is generally set to a relatively low value, such as 0.01 - 0.1. For example, assume the parameter combination of an individual is [c1, c2, c3] and the mutation probability is 0.05. Then each gene value has a 5% probability of mutating, and the mutated parameter values are randomly generated within the value range of the corresponding parameters.

[0098] 6. Iteration

[0099] Repeat the above steps of fitness evaluation, selection, crossover and mutation until the iteration termination condition is met. The iteration termination condition can be reaching the preset maximum number of iterations, the fitness function value converging to a certain extent, etc. For example, the maximum number of iterations is set to 1000 times, or when the change in the fitness function value is less than a certain threshold in 10 consecutive iterations, it is considered that the fitness function value has converged.

[0100] During the iteration process, record the individual with the highest fitness in each generation of the population and its fitness value. Finally, output the individual with the highest fitness as the optimized parameter combination, and the corresponding minimum prediction error.

[0101] In practical applications, the performance of the genetic algorithm is affected by various parameters, such as population size, crossover probability, mutation probability, etc. By adjusting these parameters, the performance of the algorithm can be optimized. For example, increasing the population size can improve the global search ability of the algorithm, but it will increase the computational cost; appropriately increasing the crossover probability can increase the diversity of the population, but too high a crossover probability may lead to a slower convergence rate of the algorithm; the setting of the mutation probability needs to be balanced between maintaining population diversity and avoiding premature convergence.

[0102] To verify the effectiveness of the genetic algorithm in optimizing the parameterization scheme of key physical processes in weather forecasting, experimental verification can be carried out. By substituting the optimized parameter combination into the weather forecasting model for operation and comparing with the forecasting results under the original parameter combination, evaluate the improvement of the optimized model in terms of forecasting accuracy, stability, etc.

[0103] In some other embodiments, the heuristic intelligent optimization algorithm can also be a particle swarm optimization algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0104] S41: Randomly generate a group of parameter combinations as the initial particle swarm. Each particle represents a parameter combination, and each particle has a randomly generated initial velocity and initial position. The parameter combination includes multiple parameters corresponding to the physical processes of the target weather type;

[0105] S42: Update the velocity and position of each particle according to the individual historical optimal solution and the global optimal solution of each particle;

[0106] S43: Calculate the prediction error of each particle;

[0107] S44: Use the prediction error as the input of the fitness function and calculate the current fitness output by the fitness function;

[0108] S45: Update the individual optimal solution and the global optimal solution according to the current fitness and the historical fitness;

[0109] Repeat steps S42 - S45 until the preset iteration termination condition is reached;

[0110] Use the current parameter combination at the time of stopping iteration as the model parameters of the numerical model physical parameterization scheme.

[0111] For the sake of easy understanding, the implementation process of the Particle Swarm Optimization (PSO) algorithm is described exemplarily below in the context of optimizing the parameterization scheme of key physical processes in weather forecasting:

[0112] 1. Initialize the particle swarm

[0113] According to the characteristics of the parameterization scheme of key physical processes in the weather forecasting model, determine the parameter dimensions and ranges to be optimized; for example, assume that the parameters to be optimized include cloud microphysical process parameters, boundary layer parameters, cumulus convection parameters, etc., and each parameter has its reasonable value range.

[0114] Randomly generate a set of parameter combinations as the initial particle swarm. Each particle represents a parameter combination, and its initial position is randomly distributed within the parameter space. For example, assume that the parameter space is three-dimensional, and the value ranges of each parameter are [a1, b1], [a2, b2], [a3, b3] respectively. Then the initial position of each particle can be expressed as [x1, x2, x3], where x i ∈[a i ,b i .

[0115] Initialize a velocity vector for each particle. The initial value of the velocity is usually randomly generated within a certain range, and the dimension of the velocity vector is the same as that of the parameter space; for example, the velocity vector can be expressed as [v1, v2, v3], where v i is a randomly generated value.

[0116] 2. Update velocity and position

[0117] The velocity update formula for the particle is:

[0118]

[0119] Where: is the velocity of particle i at the t-th iteration, w is the inertia weight, used to balance the global search ability and the local search ability, and its value range is usually [0.4, 0.9]; c1 and c2 are learning factors, usually with values [1.5, 2.5], and r1 and r2 are random numbers between [0, 1], used to increase the randomness of the algorithm; is the historical best position of particle i itself at the t-th iteration; g (t) is the global best position, that is, the best position of all particles at the t-th iteration; is the position of particle i at the t-th iteration.

[0120] The position update formula for the particle is:

[0121]

[0122] The updated position needs to be ensured within the value range of the parameter space. If it exceeds the range, it will be adjusted to the nearest boundary value.

[0123] 3. Fitness evaluation

[0124] For the position of each particle (i.e., the parameter combination), substitute it into the weather forecast model to run and obtain the forecast result of the model; then, compare the forecast result with the actual observed data to calculate the forecast error; common error metrics include mean square error (MSE), mean absolute error (MAE), etc.

[0125] Take the calculated forecast error as the input of the fitness function. The smaller the value of the fitness function, the better the performance of the position (parameter combination) of the particle. For example, MSE can be directly used as the fitness function, or the reciprocal of MSE can be taken as the fitness function to convert it into a maximization problem.

[0126] 4. Update individual and global optimal solutions

[0127] For each particle, compare its current fitness with its own historical best fitness. If the current fitness is better, update its individual optimal position p i .

[0128] Among all the particles, find the particle with the best fitness and update the global optimal position g.

[0129] 5. Iteration

[0130] Repeat the above steps of updating velocity, updating position, fitness evaluation, and updating the optimal solution until the iteration termination condition is met; the iteration termination condition can be reaching the preset maximum number of iterations, the fitness function value converging to a certain degree, etc.; for example, the maximum number of iterations is set to 1000 times, or when the change in the fitness function value is less than a certain threshold in 10 consecutive iterations, it is considered that the fitness function value has converged.

[0131] During the iteration process, record the global optimal position and its fitness value in each generation. Finally, output the global optimal position as the optimized parameter combination and the corresponding minimum forecast error.

[0132] The particle swarm optimization algorithm (PSO) simulates the foraging behavior of bird flocks and uses the velocity and position update mechanism of particles to gradually optimize the physical process parameterization scheme in the weather forecast model. Through fitness evaluation and iterative optimization, PSO can find the parameter combination that minimizes the model forecast error, thereby improving the accuracy of weather forecasting.

[0133] In some embodiments, the nestable optimization objective function is constructed by integrating the pattern integral calculation and the error index calculation on the basis of the optimization objective function with sample set error feedback. The sample set includes historical sensitive parameters and semi-empirical parameters. Based on the optimized key physical process parameterization scheme, for the sensitive parameters in the parameterization scheme and the semi-empirical parameters such as some sub-grid processes that are difficult to explicitly analyze and express, design the optimization objective function with sample set error feedback, construct the nestable optimization objective function that integrates the pattern integral calculation and the error index calculation, and through repeated iterative calculations, carry out the optimization of the sensitive parameters in the key physical process parameterization scheme.

[0134] Semi-empirical parameters refer to parameters that cannot be directly analyzed through explicit physical formulas and are usually obtained by fitting empirical formulas or observational data. Semi-empirical parameters include cloud microphysical parameters, cumulus convection parameters, and long- and short-wave radiation parameters. Among them, cloud microphysical parameters include cloud droplet collision efficiency, ice crystal melting rate, etc.; cumulus convection parameters include convection triggering threshold, convective precipitation efficiency, etc.; long- and short-wave radiation parameters include aerosol scattering coefficient, cloud droplet effective radius, etc.

[0135] The optimization objective function combines the pattern integral calculation and the error index calculation, and its specific form is:

[0136] J = α·MSE + β·RMSE + γ·CC + δ·Phy_Penalty

[0137] Where:

[0138] MSE is the mean square error between the model output and the observational data;

[0139] RMSE is the root mean square error between the model output and the observational data;

[0140] CC is the correlation coefficient between the model output and the observational data;

[0141] Phy_Penalty is the physical constraint penalty term (such as energy conservation deviation);

[0142] α, β, γ, δ are weight coefficients.

[0143] The optimization of sensitive parameters adopts a nested optimization algorithm, including:

[0144] 1. Outer layer optimization: Adjust the sensitive parameters in the key physical process parameterization scheme;

[0145] 2. Inner layer optimization: For each set of parameter combinations, calculate the pattern integral result and the error index;

[0146] 3. Iterative optimization: Through multiple iterations, gradually approach the optimal parameter combination.

[0147] In some embodiments, the sensitive parameters are optimized, specifically including:

[0148] Initialize the sensitive parameters, and randomly select multiple sensitive parameters to form a parameter combination;

[0149] For each parameter combination, calculate the pattern integral result and the error index;

[0150] According to the obtained integral result and error index, adjust the sensitive parameters in the numerical model physical parameterization scheme;

[0151] Use the updated sensitive parameters as the parameter combination for iterative calculation until the preset iteration condition is met and the iteration stops.

[0152] For the sake of easy understanding, the following is an exemplary description of the specific process of optimizing the sensitive parameters in the parameterization scheme of the key physical processes in weather forecasting by using the nested optimization algorithm:

[0153] 1. Determine the sensitive parameters

[0154] According to the parameterization scheme of the key physical processes of the weather forecasting model, identify the sensitive parameters that have a significant impact on the model output. These parameters include:

[0155] Cloud microphysical parameters: cloud droplet coalescence efficiency, ice crystal melting rate, etc.;

[0156] Cumulus convection parameters: convection triggering threshold, convective precipitation efficiency, etc.;

[0157] Short-wave and long-wave radiation parameters: aerosol scattering coefficient, cloud droplet effective radius, etc.

[0158] 2. Construct the optimization objective function

[0159] The optimization objective function combines pattern integral calculation and error index calculation.

[0160] 3. The specific process of the nested optimization algorithm includes:

[0161] Outer layer optimization: Set a reasonable value range for each sensitive parameter, and calculate the fitness of each group of parameter combinations according to the result of the inner layer optimization (error index) obtained in the previous iteration.

[0162] Inner layer optimization: For each group of sensitive parameter combinations in the outer layer optimization, run the weather forecasting model for pattern integral calculation; compare the pattern integral result with the actual observation data, and calculate each error index in the optimization objective function; feedback the error index to the outer layer optimization algorithm for evaluating the fitness of the current parameter combination.

[0163] Iterative optimization: Adjust the sensitive parameter combination according to the feedback results of the inner-layer optimization. For example, in the PSO algorithm, update the velocity and position of the particles; recalculate the pattern integral and error metrics for the updated parameter combination; stop the iteration when the preset maximum number of iterations is reached or the optimization objective function value converges to a certain extent.

[0164] Output the optimal sensitive parameter combination obtained after multiple iterations of optimization. Through the nested optimization algorithm, the sensitive parameters in the weather forecasting model can be effectively optimized, improving the model's simulation ability and forecasting accuracy for complex atmospheric processes.

[0165] In the above specific implementation manner, the optimization method for the numerical model physical parameterization scheme based on artificial intelligence provided by the present invention determines the numerical model physical parameterization scheme corresponding to the target weather type according to the target weather type; uses a heuristic intelligent optimization algorithm to optimize the numerical model physical parameterization scheme with the forecasting accuracy as the optimization objective; based on a pre-constructed nested optimization objective function, optimizes the sensitive parameters in the optimized numerical model physical parameterization scheme through iterative calculation; designs different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculates the Pareto optimal parameters considering multi-meteorological objectives optimization, and realizes the evaluation and optimization of the multi-level numerical model physical process parameterization scheme.

[0166] Based on considering the physical mechanism of momentum, heat, and water vapor interaction and mutual influence among the physical process parameterization schemes in the numerical weather forecasting model, a heuristic intelligent optimization algorithm suitable for optimizing the physical parameters of the numerical model is constructed. Through the nested optimization objective, the evaluation and optimization of the multi-level numerical model physical process parameterization scheme are realized. Solve the problems existing in the optimization process of the numerical model physical parameterization scheme, thereby improving the optimization effect of the scheme and further improving the forecasting accuracy of the numerical weather forecasting.

[0167] In addition to the above method, the present invention also provides an optimization device for the numerical model physical parameterization scheme based on artificial intelligence, as Figure 2 shown. The device includes:

[0168] A scheme determination unit 210, configured to determine the numerical model physical parameterization scheme corresponding to the target weather type according to the target weather type;

[0169] A scheme optimization unit 220, configured to use a heuristic intelligent optimization algorithm to optimize the numerical model physical parameterization scheme with the forecasting accuracy as the optimization objective;

[0170] A parameter optimization unit 230, configured to optimize sensitive parameters in the optimized numerical model physical parameterization scheme through iterative calculation based on a pre-constructed nestable optimization objective function;

[0171] A multi-level optimization unit 240, configured to design different constraint conditions according to the physical meanings of the sensitive parameters to be optimized, calculate Pareto optimal parameters considering multi-weather objective optimization, and implement the evaluation and optimization of the numerical model physical process parameterization scheme at multiple levels.

[0172] In some embodiments, the numerical model physical parameterization scheme includes a boundary layer process, a cloud microphysical process, a cumulus convection process, a long-wave and short-wave radiation process, and a surface process;

[0173] The sensitive parameters include boundary layer parameters, cloud microphysical parameters, cumulus convection parameters, long-wave and short-wave radiation parameters, and surface parameters.

[0174] In some embodiments, the heuristic intelligent optimization algorithm is a genetic algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0175] S31: Randomly generate a set of parameter combinations as an initial population, where the parameter combinations include multiple parameters corresponding to the physical processes of the target weather type;

[0176] S32: Calculate the forecast error of each parameter combination;

[0177] S33: Select excellent individuals from the initial population according to the fitness function values;

[0178] S34: Randomly select two individuals from the current population as parent individuals, and generate two offspring individuals through crossover operations to generate a new generation of population;

[0179] Use the new generation of population as the initial population, and repeat steps S32 - S34 until the preset convergence degree is reached or the preset number of iterations is reached;

[0180] Use the parameter combination in the current population when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

[0181] In some embodiments, the heuristic intelligent optimization algorithm is a particle swarm optimization algorithm. At this time, optimizing the numerical model physical parameterization scheme specifically includes:

[0182] S41: Randomly generate a set of parameter combinations as an initial particle swarm. Each particle represents a parameter combination, and each particle has a randomly generated initial velocity and initial position. The parameter combinations include multiple parameters corresponding to the physical processes of the target weather type;

[0183] S42: Update the velocity and position of each particle according to its own historical optimal solution and the global optimal solution.

[0184] S43: Calculate the prediction error of each particle.

[0185] S44: Use the prediction error as the input of the fitness function, and calculate the current fitness output by the fitness function.

[0186] S45: Update the individual optimal solution and the global optimal solution according to the current fitness and the historical fitness.

[0187] Repeat steps S42 - S45 until the preset iteration termination condition is reached.

[0188] Use the current parameter combination at the end of iteration as the model parameters of the numerical model physical parameterization scheme.

[0189] In some embodiments, the nestable optimization objective function is constructed by integrating the model integration calculation and the error index calculation based on the optimization objective function with sample set error feedback, and the sample set includes historical sensitive parameters and semi - empirical parameters.

[0190] In some embodiments, optimizing the sensitive parameters specifically includes:

[0191] Initialize the sensitive parameters, and randomly select multiple sensitive parameters to form a parameter combination.

[0192] For each parameter combination, calculate the model integration result and the error index.

[0193] Adjust the sensitive parameters in the numerical model physical parameterization scheme according to the obtained integration result and error index.

[0194] Use the updated sensitive parameters as the parameter combination for iterative calculation until the iteration stops when the preset iteration condition is met.

[0195] In some embodiments, the evaluation and optimization of the multi - level numerical model physical process parameterization scheme specifically includes:

[0196] First level: Optimize the sensitive parameters for a single physical process.

[0197] Second level: On the basis of the optimization results of the first level, jointly optimize the parameters of multiple physical processes and their interactions.

[0198] Third level: Considering all physical processes comprehensively, conduct global optimization and output the Pareto optimal solution.

[0199] In the above specific embodiments, the optimization device for the physical parameterization scheme of the numerical model provided by the present invention determines the physical parameterization scheme of the numerical model corresponding to the target weather type according to the target weather type; uses a heuristic intelligent optimization algorithm to optimize the physical parameterization scheme of the numerical model with the forecast accuracy as the optimization goal; based on the pre-constructed nestable optimization objective function, optimizes the sensitive parameters in the optimized physical parameterization scheme of the numerical model through iterative calculation; designs different constraint conditions according to the physical meanings of the sensitive parameters to be optimized, and calculates the Pareto optimal parameters considering the optimization of multiple meteorological objectives, so as to realize the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels.

[0200] Based on considering the physical mechanism of the interaction and mutual influence of momentum, heat, and water vapor among the physical process parameterization schemes in the numerical weather prediction model, the device constructs a heuristic intelligent optimization algorithm suitable for optimizing the physical parameters of the numerical model, and realizes the evaluation and optimization of the physical process parameterization scheme of the numerical model at multiple levels through the nestable optimization objective. It solves the problems existing in the optimization process of the physical parameterization scheme of the numerical model, thereby improving the optimization effect of the scheme and further improving the forecast accuracy of the numerical weather prediction.

[0201] Figure 3 An example of the physical structure diagram of an electronic device is shown as Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communication interface 320, and the memory 330 complete mutual communication through the communication bus 340. The processor 310 can call the logical instructions in the memory 330 to execute the above method.

[0202] In addition, when the logical instructions in the above-mentioned memory 330 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0203] On the other hand, the present invention also provides a computer program product, where the computer program product includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the above method.

[0204] In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the above method.

[0205] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0206] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical discs, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0207] 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 foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optimization method for physical parameterization schemes of numerical models based on artificial intelligence, characterized in that, The method includes: Determining a numerical model physical parameterization scheme corresponding to the target weather type according to the target weather type; Using a heuristic intelligent optimization algorithm, with the forecast accuracy as the optimization goal, optimizing the numerical model physical parameterization scheme; Based on a pre-constructed nested optimization objective function, through iterative calculation, optimizing the sensitive parameters in the optimized numerical model physical parameterization scheme; Designing different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculating the Pareto optimal parameters considering multi-meteorological objective optimization, and realizing the evaluation and optimization of the multi-level numerical model physical process parameterization scheme.

2. The method for optimizing the physical parameterization scheme of the numerical model based on artificial intelligence according to claim 1, wherein The numerical model physical parameterization scheme includes boundary layer processes, cloud microphysical processes, cumulus convection processes, long- and short-wave radiation processes, and surface processes; The sensitive parameters include boundary layer parameters, cloud microphysical parameters, cumulus convection parameters, long- and short-wave radiation parameters, and surface parameters.

3. The optimization method for physical parameterization schemes of numerical models based on artificial intelligence according to claim 1, characterized in that When the heuristic intelligent optimization algorithm is a genetic algorithm, optimizing the numerical model physical parameterization scheme specifically includes: S31: Randomly generating a set of parameter combinations as the initial population, where the parameter combinations include multiple physical process parameters corresponding to the target weather type; S32: Calculating the forecast error of each parameter combination; S33: Selecting excellent individuals from the initial population according to the fitness function value; S34: Randomly selecting two individuals from the current population as parent individuals, and generating two offspring individuals through crossover operations to generate a new generation of population; Taking the new generation of population as the initial population, repeating steps S32 - S34 until the preset convergence degree is reached or the preset number of iterations is reached; Taking the parameter combination in the current population when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

4. The optimization method for physical parameterization scheme of numerical model based on artificial intelligence according to claim 1, characterized in that When the heuristic intelligent optimization algorithm is a particle swarm optimization algorithm, optimizing the numerical model physical parameterization scheme specifically includes: S41: Randomly generating a set of parameter combinations as the initial particle swarm, each particle representing a parameter combination, and each particle having a randomly generated initial velocity and initial position, where the parameter combinations include multiple physical process parameters corresponding to the target weather type; S42: Updating the velocity and position of each particle according to the personal historical best solution and the global best solution of each particle; S43: Calculating the forecast error of each particle; S44: Using the forecast error as the input of the fitness function, and calculating the current fitness output by the fitness function; S45: Updating the personal best solution and the global best solution according to the current fitness and the historical fitness; Repeating steps S42 - S45 until the preset iteration termination condition is reached; Taking the current parameter combination when the iteration stops as the model parameters of the numerical model physical parameterization scheme.

5. The optimization method for physical parameterization schemes of numerical models based on artificial intelligence according to claim 1, characterized in that The nested optimization objective function is constructed by integrating the model integration calculation and the error index calculation based on the optimization objective function with sample set error feedback, and the sample set includes historical sensitive parameters and semi-empirical parameters.

6. The optimization method for physical parameterization schemes of numerical models based on artificial intelligence according to claim 5, characterized in that Optimizing the sensitive parameters specifically includes: Initialize the sensitive parameters and randomly select multiple sensitive parameters to form a parameter combination; For each parameter combination, calculate the pattern integral result and the error index; According to the obtained integral result and error index, adjust the sensitive parameters in the numerical model physical parameterization scheme; Use the updated sensitive parameters as the parameter combination and perform iterative calculations until the iteration stops when the preset iteration condition is met.

7. The optimization method for physical parameterization scheme of numerical models based on artificial intelligence according to claim 1, wherein Evaluation and optimization of the multi-level numerical model physical process parameterization scheme, specifically including: The first level: optimize the sensitive parameters for a single physical process; The second level: on the basis of the optimization results of the first level, jointly optimize the parameters of multiple physical processes and their interactions; The third level: comprehensively consider all physical processes, perform global optimization, and output the Pareto optimal solution.

8. An optimization device for physical parameterization schemes of numerical models based on artificial intelligence, characterized in that, The device includes: A scheme determination unit, configured to determine a numerical model physical parameterization scheme corresponding to the target weather type according to the target weather type; A scheme optimization unit, configured to use a heuristic intelligent optimization algorithm and take the forecast accuracy as the optimization target to optimize the numerical model physical parameterization scheme; A parameter optimization unit, configured to optimize the sensitive parameters in the optimized numerical model physical parameterization scheme through iterative calculations based on a pre-constructed nestable optimization objective function; A multi-level optimization unit, configured to design different constraint conditions according to the physical meaning of the sensitive parameters to be optimized, calculate the Pareto optimal parameters considering multi-meteorological objective optimization, and implement the evaluation and optimization of the multi-level numerical model physical process parameterization scheme.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images