A Parallel Multi-Objective Stochastic Parameter Optimization Method and Apparatus

Abstract

This invention provides a parallel multi-objective stochastic parameter optimization method and apparatus. The method includes: configuring the number of nodes in a parallel computing cluster and generating a random seed sequence; loading and compiling the model to be optimized and obtaining its global configuration; sending the global configuration to all processes, with each process generating an initial parameter population satisfying the global configuration constraints based on the random seeds, and serially evaluating the objective function value; each process iteratively executing a shuffling complex evolution until convergence is met; and each process outputting its optimal parameter vector and corresponding objective function value to the root process, which then determines the global optimal solution. This invention significantly improves the accuracy and efficiency of parameter optimization, solving the problem of poor model optimization performance in existing related technologies.

Description

Technical Field

[0001] This invention relates to the field of computer modeling and optimization technology, and in particular to a parallel multi-objective stochastic parameter optimization method and apparatus. Background Technology

[0002] With the increasing complexity of models in fields such as ecology, climate, and hydrology, parameter optimization and calibration have become crucial steps in achieving accurate predictions and reliable simulations. Traditional single-objective parameter optimization methods struggle to simultaneously address diverse evaluation metrics for model performance, while existing multi-objective optimization techniques, such as genetic algorithms and particle swarm optimization, often suffer from slow convergence, susceptibility to local optima, and low computational efficiency. Furthermore, when dealing with high-dimensional parameter spaces, conventional methods incur heavy computational burdens, failing to fully utilize high-performance computing resources and significantly limiting their practical application in large-scale ecological environment simulation and numerical simulation studies.

[0003] In recent years, stochastic search methods, represented by Shuffled Complex Evolution (SCE), have made significant progress in the field of parameter calibration (e.g., Duan, QY, Sorooshian, S., & Gupta, V. 1992. Effective and efficient global optimization for conceptual rainfall-runoff models [J]. Water Resources Research, 28(4), 1015-1031.). However, how to effectively combine these methods to achieve efficient parallel execution, especially in optimizing multi-objective parameters in large-scale cluster environments, still faces technical bottlenecks. Therefore, developing a parallel multi-objective stochastic parameter optimization method applicable to fields such as ecology, climate, and hydrology to significantly improve parameter optimization efficiency and global search quality has important theoretical significance and application value.

[0004] There is currently no effective solution to the problem of poor model optimization performance in existing related technologies. Summary of the Invention

[0005] This invention provides a parallel multi-objective stochastic parameter optimization method and apparatus to address the shortcomings of poor model optimization performance in existing related technologies.

[0006] In a first aspect, the present invention provides a parallel multi-objective stochastic parameter optimization method, comprising:

[0007] Configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds;

[0008] The structure of the model to be optimized is loaded and compiled, and the global configuration of the model to be optimized is obtained;

[0009] The global configuration is sent to all processes, and each process generates an initial parameter population that satisfies the global configuration constraints based on the random seed, and the objective function value is obtained by serial evaluation.

[0010] Each process iteratively executes the shuffle complex evolution separately until the convergence condition is met;

[0011] Each process outputs its optimal parameter vector and corresponding objective function value to the root process, which then determines the global optimal solution.

[0012] According to the present invention, a parallel multi-objective stochastic parameter optimization method is provided, which configures the number of nodes in a parallel computing cluster and generates a random seed sequence, including:

[0013] Configure the number of nodes in the parallel computing cluster;

[0014] Initialize the communication environment based on the messaging interface or an equivalent messaging standard;

[0015] Externally driven data and corresponding observation data are loaded into a shared storage medium to generate a random seed sequence of a specific length.

[0016] According to a parallel multi-objective stochastic parameter optimization method provided by the present invention, the method performs structural loading and compilation on the model to be optimized, and obtains the global configuration of the model to be optimized, including:

[0017] The model to be optimized is loaded and compiled, and the time step, output options and error processing parameters are set so that the model to be optimized is in a runnable state.

[0018] Set upper and lower bounds and explicit constraints for each parameter of the model to be optimized, and determine the calibration variable, objective function and corresponding weights of the model to be optimized.

[0019] According to a parallel multi-objective stochastic parameter optimization method provided by the present invention, the global configuration is sent to all processes, each process generates an initial parameter population that satisfies the global configuration constraints based on the stochastic seed, and the objective function value is obtained through serial evaluation, including:

[0020] The global configuration is sent to all processes via a broadcast operation using a message passing interface;

[0021] Each process is assigned a unique random seed;

[0022] Each process generates an initial parameter population that satisfies the global configuration constraints based on the random seed, and sequentially evaluates the objective function value.

[0023] According to the parallel multi-objective stochastic parameter optimization method provided by the present invention, each process iteratively performs shuffle complex evolution, including:

[0024] The current parameter population is sorted according to the objective function value to obtain an ordered population;

[0025] The ordered population is divided into several complexes using an alternating interval allocation rule; each complex contains several parameter vectors.

[0026] Each process sequentially performs competitive complex evolution within the complex to obtain an improved solution for the complex.

[0027] Each process merges all complexes to form the next generation population.

[0028] According to a parallel multi-objective stochastic parameter optimization method provided by the present invention, each process sequentially performs competitive complex evolution within the complex to obtain an improved solution to the complex, including:

[0029] The parameter vectors within the composite are sorted in ascending order according to the objective function values;

[0030] Based on the current probability distribution, several parameter vectors are randomly selected without replacement to form a sub-complex;

[0031] The reflection, contraction, and random perturbation operations are performed sequentially to generate a new parameter vector. The corresponding objective function value is evaluated, and if it is better than the worst parameter vector, it is replaced.

[0032] According to the parallel multi-objective stochastic parameter optimization method provided by the present invention, in each evolution of the sub-complex, only one candidate parameter vector is maintained. The candidate parameter vector is reassigned and used to replace the worst parameter vector according to the following rules:

[0033] Perform a reflection operation to reassign values ​​to the candidate parameter vector;

[0034] If the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints, a shrinkage operation is performed to reassign the candidate parameter vector.

[0035] If the contraction operation fails to improve the objective function of the candidate parameter vector, then the candidate parameter vector is regenerated randomly using a normal distribution with the mean of the optimal parameter vector of the subcomplex as the mean and the standard deviation of the current parameter vector as the scale.

[0036] According to the parallel multi-objective stochastic parameter optimization method provided by the present invention, the competitive complex evolution operation is performed iteratively for a preset number of steps on each complex, including:

[0037] Reorder all parameter vectors within the composite in ascending order according to the objective function value;

[0038] A specific number of parameter vectors are randomly extracted without replacement according to the weights corresponding to the objective function, and then formed into sub-complexes;

[0039] The centroid of the sub-complex is determined, the candidate parameter vectors are reassigned to generate new parameter vectors, and the objective function value is evaluated. If the objective function value is better than the worst parameter vector, it is replaced.

[0040] If the number of times the candidate parameter vector is reassigned reaches a certain number or the number of times the target function is called reaches a threshold, then the current competitive complex evolution operation ends.

[0041] According to the parallel multi-objective stochastic parameter optimization method provided by the present invention, the process stops iterative shuffling complex evolution when any of the following convergence conditions are met:

[0042] The cumulative number of times the target function has been called exceeds the threshold.

[0043] Within a specific number of shuffle cycles, the rate of change of the optimal objective function did not reach the preset standard;

[0044] The range of the population geometric mean is less than the range threshold.

[0045] Secondly, the present invention also provides a parallel multi-objective stochastic parameter optimization device, comprising:

[0046] The configuration module is used to configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds.

[0047] The loading module is used to load and compile the structure of the model to be optimized, and to obtain the global configuration of the model to be optimized.

[0048] An evaluation module is used to send the global configuration to all processes, each process generates an initial parameter population that satisfies the global configuration constraints based on the random seed, and evaluates the objective function value serially.

[0049] An evolution module is used to iteratively perform the shuffle complex evolution for each process individually until the convergence condition is met.

[0050] The determination module is used to enable each process to output its own optimal parameter vector and corresponding objective function value to the root process, and to determine the global optimal solution through the root process.

[0051] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the parallel multi-objective stochastic parameter optimization method as described in the first aspect above.

[0052] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the parallel multi-objective random parameter optimization method as described in the first aspect above.

[0053] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the parallel multi-objective random parameter optimization method as described in the first aspect above.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The parallel multi-objective stochastic parameter optimization method provided by this invention assigns a unique random seed to each process, and allows multiple processes to run independently and in parallel. Multiple random trajectories maintain population diversity, significantly reducing the risk of getting trapped in local optima. Combined with the shuffle complex evolution framework, it can significantly improve the accuracy and efficiency of parameter optimization, solving the problem of poor model optimization performance in existing related technologies. Furthermore, the algorithm and processes of this method do not share state, and it can be directly scaled to large-scale clusters of hundreds of nodes. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the parallel multi-objective random parameter optimization method provided by the present invention;

[0058] Figure 2 This is a schematic diagram illustrating the process of optimizing model parameters in an embodiment of the present invention;

[0059] Figure 3 This is a flowchart of the microbial-enzyme-mediated carbon-nitrogen-water coupling model in an embodiment of the present invention;

[0060] Figure 4 This is a schematic diagram of the simulation results of heterotrophic respiratory flux in an embodiment of the present invention;

[0061] Figure 5 This is a schematic diagram illustrating the simulation results of microbial biomass in an embodiment of the present invention;

[0062] Figure 6 This is a schematic diagram illustrating the simulation results of the carbon-nitrogen ratio of microbial biomass in an embodiment of the present invention;

[0063] Figure 7 This is a schematic diagram of the simulation results of soil ammonia concentration in an embodiment of the present invention;

[0064] Figure 8 This is a structural block diagram of the parallel multi-objective random parameter optimization device provided by the present invention;

[0065] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0067] This invention provides a parallel multi-objective stochastic parameter (ParMOSPO) optimization method. Figure 1 This is a flowchart of the parallel multi-objective stochastic parameter optimization method provided by the present invention, as follows: Figure 1 As shown, the method includes the following steps:

[0068] Step S101: Configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds.

[0069] Step S102: Load and compile the structure of the model to be optimized, and obtain the global configuration of the model to be optimized;

[0070] Step S103: The global configuration is sent to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on a random seed, and the objective function value is obtained by serial evaluation.

[0071] Step S104: Each process iteratively executes the shuffle complex evolution separately until the convergence condition is met;

[0072] In step S105, each process outputs its own optimal parameter vector and corresponding objective function value to the root process, and the root process determines the global optimal solution.

[0073] In this method, firstly, the number of nodes in the parallel computing cluster is configured, and a random seed sequence is generated. Then, the model to be optimized is structurally loaded and compiled, and its global configuration is obtained, which characterizes the model's state performance. The global configuration is then sent to all processes, and each process generates an initial parameter population that satisfies the global configuration constraints based on the random seed, and the objective function value is obtained through serial evaluation. Next, each process iteratively executes Shuffled Complex Evolution (SCE) until the convergence condition is met. Through the parallel Shuffled Complex Evolution framework, the cluster's computing power is fully utilized, making the high-dimensional parameter search speed increase nearly linearly with the number of nodes. Finally, each process outputs its optimal parameter vector and corresponding objective function value to the root process, which determines the global optimum. In the above process, by assigning a unique random seed to each process and ensuring that multiple processes run independently and in parallel, multiple random trajectories maintain population diversity, significantly reducing the risk of getting trapped in local optima. Combined with the Shuffled Complex Evolution framework, this significantly improves the accuracy and efficiency of parameter optimization, solving the problem of poor model optimization performance in existing related technologies. Furthermore, the algorithm and process of this method do not share state, and can be directly scaled to large-scale clusters of more than a hundred nodes.

[0074] Figure 2 This is a schematic diagram illustrating the process of optimizing model parameters in an embodiment of the present invention, such as... Figure 2 As shown, in some embodiments, step S101, configuring the number of nodes in the parallel computing cluster and generating a random seed sequence, includes: configuring the number of nodes in the parallel computing cluster; initializing a communication environment based on a Message Passing Interface (MPI) or an equivalent message passing standard; loading externally forced driving data and corresponding observation data into a shared storage medium to generate a random seed sequence of a specific length.

[0075] For example, the number of nodes in the parallel computing cluster is configured, a communication environment based on the Message Passing Interface (MPI) or an equivalent message passing standard is initialized, and externally forced driving data and corresponding observation data are loaded into a shared storage medium to generate a random seed sequence of length N for subsequent distribution.

[0076] In some embodiments, step S102 involves loading and compiling the model to be optimized and obtaining its global configuration, including: loading and compiling the model to be optimized, setting the time step, output options, and error processing parameters to make the model to be optimized runnable; setting upper and lower bounds and explicit constraints for each parameter to be optimized in the model to be optimized, and determining the calibration variables, objective function, and corresponding weights of the model to be optimized.

[0077] In this embodiment, the microbial-enzyme-mediated carbon-nitrogen-water coupling model (MEND) is applied for parameter optimization, and the flowchart is as follows. Figure 3 As shown, Figure 3 This is a flowchart of the microbial-enzyme-mediated carbon-nitrogen-water coupling model in this embodiment of the invention. The MEND model is a microbial-enzyme-mediated carbon-nitrogen-water coupling model (i.e., the carbon-nitrogen coupling version), which includes 23 soil carbon-nitrogen pools:

[0078] 1. Five soil organic matter (SOM) pools: Oxidized particulate organic matter (POM) O Particulate organic matter (POM) decomposed by hydrolytic enzymes H Mineral-associated organic matter (MOM) and soluble organic matter (DOM), as well as quinone-like organic matter (QOM) that interacts with soluble organic matter through adsorption and desorption.

[0079] 2. Seven inorganic nitrogen reservoirs: adsorbed ammonium (NH4) + Adsorb), non-adsorbed ammonium (NH4) + ), nitrates (NO3) - ), nitrite (NO2) - ), nitric oxide (NO), nitrous oxide (N2O) and nitrogen (N2);

[0080] 3. Two functional groups of microbes: activated and dormant microbes (MBA and MBD);

[0081] 4. Three enzyme functional groups that break down SOM: oxidase (EPO), hydrolase (EPH), and comprehensive decomposition enzyme (EM);

[0082] 5. Six functional groups of enzymes controlling inorganic nitrogen conversion: nitrogenase (ENH4), nitrase (ENO3), and denitrase (ENO2, ENO, EN2O, EN2).

[0083] A biosphere reserve ( The evergreen monsoon broad-leaved forest is used as an example. The simulation of the model to be optimized is driven by measured climate and environmental data, mainly including soil temperature, humidity, pH, litter biomass, and dry and wet nitrogen deposition rates. The field experimental data period is from August 2009 to December 2012. In the MEND model calibration, the time step is one hour, and the number of parallel computing nodes is 30.

[0084] In this embodiment, each parameter of the model is... x j Set upper and lower bounds [b l,j b u,j ] and explicit constraints, and determine the model calibration variables, objective function, and their weights. The initial population contains at least one parameter vector specified by the user, and the remaining parameter vectors are obtained through [b l, j b u, j The parameters to be optimized for MEND model calibration and their ranges are shown in Table 1, which contains a total of 12 parameters: [The parameters are generated by uniform random sampling within the specified range].

[0085] Table 1. Parameters to be optimized in the MEND model

[0086]

[0087] In Table 1, POM o POM represents oxidized particulate organic matter. H This indicates hydrolyzable particulate organic matter; MOM indicates mineral-bound organic matter (non-protected mode); DOM indicates soluble organic matter; EP indicates POM-degrading enzymes, including oxidases (EP...). o ) and hydrolase (EP) H ), EM represents a complex decomposing enzyme.

[0088] The calibration variables and objective function of the MEND model are shown in Table 2:

[0089] Table 2 Objective Function Table for MEND Model Parameterization

[0090]

[0091] Based on the number of observation points for the response variable, heterotrophic respiratory flux (R) h Microbial biomass carbon (MBC) and microbial biomass carbon-to-nitrogen ratio (MBN) CN ), ammonium concentration (NH4) + The weights of the objectives are 3 / 7, 2 / 7, 3 / 14, and 1 / 14, respectively, and the objective function is quantified by the following two indicators: 1-coefficient of determination (R²). 2 ); Mean Absolute Relative Error (MARE).

[0092] In some embodiments, step S103, sending the global configuration to all processes, where each process generates an initial parameter population satisfying the global configuration constraints based on a random seed, and sequentially evaluates the objective function value, includes: sending the global configuration to all processes using a broadcast operation of a message passing interface; and distributing a unique random seed to each process. rank Each process generates an initial parameter population that satisfies global configuration constraints based on a random seed, and sequentially evaluates the objective function value. f obj .

[0093] Based on this embodiment, in step S104, each process iteratively executes the shuffle complex evolution, including: adjusting the current parameter population according to the objective function value. f obj Sort the data to obtain an ordered population D; divide the ordered population D into several (k) complexes using an alternating interval allocation rule; each complex contains several (m) parameter vectors; each process sequentially performs competitive complex evolution (CCE) within a complex to obtain an improved solution for the complex; each process merges all complexes to form the next generation population.

[0094] In the above embodiments, each process sequentially performs competitive complex evolution within the complex to obtain an improved solution for the complex, including: adjusting the parameter vectors within the complex according to the objective function value. f obj Sort in ascending order; randomly select several without replacement according to the current probability distribution. nps The subcomplex is composed of (1) parameter vectors; reflection, contraction, and random perturbation operations are performed sequentially to generate new parameter vectors, and the corresponding objective function values ​​are evaluated. f obj If the parameter vector is better than the worst one, then it is replaced. In this embodiment, the three-stage update of reflection-contraction-perturbation combined with dynamic complex adjustment effectively escapes local extrema and accelerates convergence.

[0095] Specifically, in each evolution of the sub-complex, only one candidate parameter vector is maintained. s new The candidate parameter vectors are reassigned and used to replace the worst parameter vector, i.e., the worst point, according to the following rules. : Perform a reflection operation to reassign values ​​to the candidate parameter vector; if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints, then perform a contraction operation to reassign values ​​to the candidate parameter vector; if the contraction operation fails to improve the objective function of the candidate parameter vector, then randomly generate a new candidate parameter vector using a normal distribution with the mean of the optimal parameter vector of the subcomplex as the mean and the standard deviation of the current parameter vector as the scale.

[0096] For example, a reflection operation is first performed to reassign values ​​to the candidate parameter vector, as shown in the following formula:

[0097]

[0098] in, Represents a candidate parameter vector. This indicates that the subcomplex, excluding the worst point, is otherwise... nps -1 point of center of gravity, Represents the reflection coefficient. This represents the worst-case scenario. If the reflection operation fails to improve the objective function or violates explicit constraints, a contraction operation is performed, calculated as follows:

[0099]

[0100] in, Represents a candidate parameter vector. This indicates that the subcomplex, excluding the worst point, is otherwise... nps -1 point of center of gravity, Indicates the shrinkage coefficient. This represents the worst-case scenario. If the contraction operation still fails to improve the objective function, then a new candidate parameter vector is randomly generated using a normal distribution with the mean of the subcomplex's worst point as the mean and the current parameter standard deviation as the scale. s new .

[0101] More specifically, the competitive complex evolution operation iterates through each complex for a preset number of steps, including: re-sorting all parameter vectors within the complex in ascending order according to the objective function value; randomly extracting a specific number of parameter vectors without replacement according to the weights corresponding to the objective function, and forming sub-complexes; determining the centroid of the sub-complexes, reassigning values ​​to candidate parameter vectors to generate new parameter vectors, and evaluating the objective function value; if it is better than the worst parameter vector, it is replaced; if the number of reassignment attempts of candidate parameter vectors reaches a specific number or the number of calls to the objective function reaches a threshold, the current competitive complex evolution operation ends.

[0102] For example, recalculate based on the objective function value f objSort all m parameter vectors in the complex in ascending order, then randomly select a specific number of parameter vectors without replacement according to the weights corresponding to the objective function, and form sub-complexes. The weights are calculated using the following formula:

[0103]

[0104] in, Indicates weight, Indicates the number of parameter vectors. This represents the parameter vector number. Then, the centroid of the sub-complex is calculated using the following formula:

[0105]

[0106] in, Indicates the first The centroid of each parameter, Indicates the number of points other than the worst point. The vector in the th... The values ​​that can be taken on each parameter This indicates the number of parameter vectors selected. Candidate parameter vectors are reassigned sequentially to generate new parameter vectors, and the objective function value is evaluated. If a new parameter vector is better than the worst one, it is replaced. If the number of reassignment attempts for candidate parameter vectors reaches a certain threshold... The number of times the target function is called or the number of calls reaches the threshold. maxn If the current competitive complex evolution operation ends, the operation will continue until the preset number of steps has not been reached. nspl .

[0107] During the iteration process, when the number of complexes k is greater than the minimum number of complexes... mings Furthermore, when the convergence progress meets the preset conditions, the number of complexes is dynamically reduced and the population is redistributed in an alternating manner. The formula for calculating the geometric mean range is as follows:

[0108]

[0109] in, Indicates the range of the geometric mean. Indicates the number of parameter vectors. Indicates the first The maximum value of each parameter vector in the current population. Indicates the first The minimum value of the parameter vector in the current population. Indicates the first The upper bound of the parameter vector. Indicates the first The lower bound of the parameter vector. This indicates that the value of zero should be avoided for extremely small positive numbers.

[0110] In this embodiment, the number of complexes is 2, the number of points for each complex is 13, the initial population number of points is 26, the number of points for each sub-complex is 7, and the number of evolutionary steps for each complex is 13.

[0111] The process stops iterative shuffling complex evolution when any of the following convergence conditions are met: the cumulative number of calls to the objective function exceeds a threshold; the rate of change of the optimal objective function does not reach a preset standard within a specific number of shuffling cycles; or the range of the population geometric mean is less than a range threshold.

[0112] For example, convergence conditions include:

[0113] 1. The cumulative number of times the target function is called reaches or exceeds the threshold. maxn = 10000;

[0114] 2. Continuous kstop = The rate of change of the optimal objective function within 10 shuffle cycles is less than the preset standard pcento = 0.0001;

[0115] 3. Current population geometric mean range gnrng Less than the range threshold = 0.001.

[0116] Each process outputs its optimal parameter vector and objective function value, and sends the results to the root process, which then determines the global optimal solution.

[0117] For example, each process atomically appends a fixed-length record to the same log file using the MPI_File_write_shared interface of MPI. The root process writes the file header, while the other processes only append data records. After receiving the optimal results from all processes, the root process selects the minimum overall objective function value through comparison or the MPI_Reduce operation and outputs the corresponding parameter vector as the global optimal solution. The optimization process record includes at least: process ID (rank), number of iterations, optimal objective function value, worst objective function value, and so on. gnrng And the current optimal parameter vector.

[0118] In this example, 30 parallel processes were used, with a total of 69,810 iterations. The optimal objective function value obtained by this method is 0.33; the average of the optimal values ​​of each process is 0.34; and the overall average of the objective function across all processes in all iterations is 0.49. The optimal parameters obtained after calibration of the model to be optimized are shown in Table 1, and the calibration results are as follows. Figure 4-7 As shown. Among them, Figure 4 This is a schematic diagram illustrating the simulation results of heterotrophic respiratory flux in an embodiment of the present invention. Figure 4 In the middle, heterotrophic respiratory flux (R h The simulated values ​​and measured values ​​are in high agreement, and the coefficient of determination (R²) is high.2 The value of 0.70 indicates that the model has high reliability in simulating the soil carbon release process. Figure 5 This is a schematic diagram illustrating the simulation results of microbial biomass in an embodiment of the present invention. Figure 5 In the study, the simulation of microbial biomass carbon (MBC) showed good results, and R... 2 The value of 0.67 indicates that the model can accurately reflect the growth dynamics of the microbial community. Figure 6 This is a schematic diagram illustrating the simulation results of the carbon-nitrogen ratio of microbial biomass in an embodiment of the present invention. Figure 6 In China, the carbon-to-nitrogen ratio of microbial biomass (MB) CN The simulation accuracy is high, with a mean absolute relative error (MARE) of only 0.33, which verifies the advantages of model optimization in characterizing microbial stoichiometry. Figure 7 This is a schematic diagram illustrating the simulation results of soil ammonia concentration in an embodiment of the present invention. Figure 7 In the middle, ammonium concentration (NH4) + The MARE of 0.49 was slightly higher than that of MB. CN However, these results remain within an acceptable range, indicating the stability of the optimization algorithm in simulating nitrogen transformation processes. These results fully demonstrate that the parallel multi-objective stochastic parameter optimization method proposed in this invention significantly improves the overall performance of the model in simulating complex soil carbon and nitrogen cycles by collaboratively optimizing multiple key ecological process parameters, achieving excellent prediction accuracy (R²). 2 >0.65, MARE<0.5) provides an innovative technical means for modeling ecosystem processes.

[0119] The present invention also provides a parallel multi-objective random parameter optimization device. The parallel multi-objective random parameter optimization device provided by the present invention will be described below. The parallel multi-objective random parameter optimization device described below and the parallel multi-objective random parameter optimization method described above can be referred to in correspondence. Figure 8 This is a structural block diagram of the parallel multi-objective stochastic parameter optimization device provided by the present invention, as shown below. Figure 8 As shown, the device includes:

[0120] Configuration module 801 is used to configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds.

[0121] Loading module 802 is used to load and compile the structure of the model to be optimized, and to obtain the global configuration of the model to be optimized;

[0122] Evaluation module 803 is used to send the global configuration to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on a random seed, and evaluates the objective function value serially.

[0123] Evolution module 804 is used to iteratively perform the shuffle complex evolution for each process individually until the convergence condition is met.

[0124] The determination module 805 is used to enable each process to output its own optimal parameter vector and corresponding objective function value to the root process, and the root process determines the global optimal solution.

[0125] In operation, this device first configures the number of nodes in the parallel computing cluster and generates a random seed sequence. Then, the loading module 802 loads and compiles the model to be optimized and obtains its global configuration, which characterizes the model's state performance. The evaluation module 803 then sends the global configuration to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on the random seed and evaluates the objective function value serially. Next, the evolution module 804 iteratively executes Shuffled Complex Evolution (SCE) for each process individually until convergence is achieved. This parallel shuffled complex evolution framework fully utilizes the cluster's computing power, resulting in a near-linear increase in high-dimensional parameter search speed with the number of nodes. Finally, the determination module 805 outputs the optimal parameter vector and corresponding objective function value from each process to the root process, which then determines the global optimal solution. In the aforementioned process, by assigning a unique random seed to each process and ensuring that multiple processes run independently and in parallel, the diversity of the population is maintained through multiple random trajectories, significantly reducing the risk of getting trapped in local optima. Combined with the shuffle complex evolution framework, this significantly improves the accuracy and efficiency of parameter optimization, solving the problem of poor model optimization performance in existing related technologies. Furthermore, the algorithm and processes in this device do not share state, allowing for direct scaling to large-scale clusters of over a hundred nodes.

[0126] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include: a processor 901, a communication interface 902, a memory 903, and a communication bus 904. The processor 901, communication interface 902, and memory 903 communicate with each other via the communication bus 904. The processor 901 can call logical instructions from the memory 903 to execute a parallel multi-objective stochastic parameter optimization method, which includes:

[0127] Configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds;

[0128] The structure of the model to be optimized is loaded and compiled, and the global configuration of the model to be optimized is obtained.

[0129] The global configuration is sent to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on a random seed, and the objective function value is obtained by serial evaluation.

[0130] Each process iterates through the evolution of the shuffle complex separately until the convergence condition is met.

[0131] Each process outputs its optimal parameter vector and corresponding objective function value to the root process, which then determines the global optimal solution.

[0132] Furthermore, the logical instructions in the aforementioned memory 903 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0133] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the parallel multi-objective stochastic parameter optimization method provided by the above methods, the method comprising:

[0134] Configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds;

[0135] The structure of the model to be optimized is loaded and compiled, and the global configuration of the model to be optimized is obtained.

[0136] The global configuration is sent to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on a random seed, and the objective function value is obtained by serial evaluation.

[0137] Each process iterates through the evolution of the shuffle complex separately until the convergence condition is met.

[0138] Each process outputs its optimal parameter vector and corresponding objective function value to the root process, which then determines the global optimal solution.

[0139] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the parallel multi-objective stochastic parameter optimization method provided by the methods described above, the method comprising:

[0140] Configure the number of nodes in the parallel computing cluster and generate a random seed sequence; the random seed sequence contains several random seeds;

[0141] The structure of the model to be optimized is loaded and compiled, and the global configuration of the model to be optimized is obtained.

[0142] The global configuration is sent to all processes. Each process generates an initial parameter population that satisfies the global configuration constraints based on a random seed, and the objective function value is obtained by serial evaluation.

[0143] Each process iterates through the evolution of the shuffle complex separately until the convergence condition is met.

[0144] Each process outputs its optimal parameter vector and corresponding objective function value to the root process, which then determines the global optimal solution.

[0145] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0146] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A parallel multi-objective stochastic parameter optimization method, characterized in that, The method comprises the following steps: configuring the number of nodes of a parallel computing cluster, generating a random seed sequence; the random seed sequence contains a plurality of random seeds; performing structural loading and compiling on a to-be-optimized model, and obtaining the global configuration of the to-be-optimized model; sending the global configuration to all processes, and each process generates an initial parameter population meeting the global configuration constraint based on the random seed, and obtains a target function value through serial evaluation; each process independently iteratively performs a shuffle complex evolution until a convergence condition is met; each process outputs a respective optimal parameter vector and a corresponding target function value to a root process, and determines a global optimal solution through the root process; sending the global configuration to all processes, and each process generates an initial parameter population meeting the global configuration constraint based on the random seed, and obtains a target function value through serial evaluation, comprising: sending the global configuration to all processes using the broadcast operation of the message passing interface; distributing a unique random seed to each process; each process generates an initial parameter population meeting the global configuration constraint based on the random seed, and obtains a target function value through serial evaluation; each process independently iteratively performs a shuffle complex evolution, comprising: sorting the current parameter population according to the target function value to obtain an ordered population; dividing the ordered population into a plurality of complexes using an interleaved interval allocation rule; each complex contains a plurality of parameter vectors; each process performs a competitive complex evolution in the complex to obtain an improved solution of the complex; each process combines all complexes to form a next-generation population.

2. The parallel multi-objective random parameter optimization method of claim 1, wherein, configuring the number of nodes of a parallel computing cluster, generating a random seed sequence, comprising: configuring the number of nodes of a parallel computing cluster; initializing a communication environment based on the message passing interface or an equivalent message passing standard; loading external forced driving data and corresponding observation data into a shared storage medium, and generating a random seed sequence of a specific length.

3. The parallel multi-objective random parameter optimization method of claim 1, wherein, performing structural loading and compiling on a to-be-optimized model, and obtaining the global configuration of the to-be-optimized model, comprising: performing structural loading and compiling on the to-be-optimized model, setting the time step, output options, and error handling parameters, so that the to-be-optimized model is in a runnable state; setting the upper and lower bounds and display constraints for each to-be-optimized parameter of the to-be-optimized model, and determining the calibration variables, target function, and corresponding weights of the to-be-optimized model.

4. The parallel multi-objective stochastic parameter optimization method of claim 1, wherein, each process performs a competitive complex evolution in the complex to obtain an improved solution of the complex, comprising: sorting the parameter vectors in the complex in ascending order according to the target function value; randomly selecting a plurality of parameter vectors to form a sub-complex according to the current probability distribution without replacement; sequentially performing reflection, contraction, and random disturbance operations to generate new parameter vectors and evaluate the corresponding target function values, and if better than the worst parameter vector, replacing it.

5. The parallel multi-objective stochastic parameter optimization method of claim 1, wherein, In each sub-complex evolution process, only one candidate parameter vector is maintained, which is re-assigned and used to replace the worst parameter vector according to the following rules: performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; 6. The parallel multi-objective random parameter optimization method of claim 5, wherein, performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; 7. The parallel multi-objective random parameter optimization method of claim 1, wherein, performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; 8. A parallel multi-objective stochastic parameter optimization apparatus for implementing the parallel multi-objective stochastic parameter optimization method of any one of claims 1 to 7, characterized in that, performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter vector if the reflection operation fails to improve the objective function of the candidate parameter vector or violates the explicit constraints; performing a reflection operation to reassign the candidate parameter vector; performing a contraction operation to reassign the candidate parameter

Images