Weather forecast large model rapid forecasting method, system and device under low computing resources, medium and terminal

Abstract

The invention belongs to the technical field of weather forecast, and discloses a method, a system, equipment, a medium and a terminal for quickly forecasting a weather forecast large model under low computing resources. Based on a fine tuning technology of a neural network, by utilizing an integration thought, aiming at the problems that large parameter training of an existing large model has high requirements on data and computing resources and is difficult to train, a large model rapid forecasting method under low computing resources is designed for the meteorological field, and the technical barrier of existing large model training is broken through. And an effective large model can be quickly forecasted from zero under the condition of low computing resources. According to the method, large model rapid forecasting under low computing resources can be realized, and the high computing resource requirement of existing large model training is avoided. According to the method, the large model is formed by integrating various small models, so that the problem of difficulty in large-parameter training of the existing large model is solved. The method focuses on a forecasting model in the meteorological field and is also applicable to other fields.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological forecasting, and particularly relates to a rapid forecasting method, system, device, medium and terminal of a large meteorological forecasting model with low computing resources. Background Art

[0002] In recent years, large artificial intelligence models have demonstrated astonishing levels of intelligence. However, the current training of large artificial models requires long-term and sufficient computing resource support. The allocation of a large amount of computing resources has greatly hindered the research on large models by a large number of ordinary research institutions. Secondly, in terms of model structure, the basic architecture of these large models is transformer, and the model architecture is single; in terms of training difficulty, for these large models, a large amount of data and a huge number of parameters mean that the training difficulty of the model is extremely high. In addition, the training of these large models also faces complex engineering problems, such as model training under a distributed framework, how to split the model and data, and how to effectively communicate between different nodes. Solving these problems requires the close cooperation of professionals in multiple industries, which further raises the industry barriers of large models.

[0003] Through the above analysis, the problems and defects existing in the prior art are as follows: (1) The existing large models are all integrated large model architectures, which have a huge number of parameters, require a large amount of data for training, and have a high training difficulty.

[0004] (2) The existing large model training has high requirements for computing resources.

[0005] (3) The data requires a large amount of manual participation in annotation, and the model training needs to be fine-tuned by combining expert experience and reinforcement learning techniques during the training process.

[0006] (4) The model training requires the participation of professionals in multiple fields, involving the close cooperation of professionals in multiple professional fields such as artificial intelligence, communication maintenance, and computing resource management. Summary of the Invention

[0007] Meteorology is closely related to human activities. Accurate meteorological forecasting is of great significance to fields such as daily travel and urban planning. The current meteorological forecasting is mainly numerical forecasting or AI forecasting, which has a series of problems such as short forecasting time, low prediction accuracy, and strong model dependence. The powerful potential demonstrated by large artificial intelligence models provides a good solution for improving meteorological forecasting capabilities.

[0008] Aiming at the problems existing in the prior art, the present invention provides a rapid forecasting method, system, device, medium and terminal of a large meteorological forecasting model with low computing resources.

[0009] The present invention is implemented as follows. A rapid forecasting method of a large meteorological forecasting model with low computing resources includes: Step 1: Construct a large meteorological forecasting model dataset; Step 2: Split the large meteorological forecasting model task into subtasks 1, 2, …, N, and determine the training sets for subtasks 1 - N; Step 3: Decompose the output of each subtask, and determine the subtask output branches 1, 2, …, Q for each subtask; Step 4: Collect multiple small models that can perform the same task for each subtask type, and construct a small model alternative set; Step 5: Select subtask type N, and initialize N as 1; Step 6: For the subtask N selected in Step 5, train each small model in the small model alternative set for the subtask output branch 1 of this subtask, and determine the small model with the best performance as the base model; Step 7: Fine - tune the small models on the remaining output branches of subtask N: On the base model in Step 6, the remaining output branches of subtask N adopt the same small model structure as the base model, initialize the parameters of the small models on the remaining output branches of subtask N with the parameters of the base model trained in Step 6, and perform parameter fine - tuning; Step 8: Construct an integrated small model for subtask N: Based on the small models trained on all output branches of subtask N in Step 6 and Step 7, use the same small model structure to construct an integrated small model for subtask N, and input the outputs of all output branches under subtask N into the integrated small model for subtask N; Initialize the integrated small model for subtask N with the parameters of the base model in Step 6, and perform parameter fine - tuning; Step 9: Update subtask type N = N + 1, repeat Steps 6 - 8, and train the current integrated module for subtask N; Step 10: Integrate the outputs of the integrated modules for subtasks N and N - 1, select the same network structure as the base model in Step 6, initialize the integrated small models for subtasks N and N - 1 with the parameters of the base model in Step 6, and fine - tune the parameters of the integrated small models for subtasks N and N - 1; Step 11: Repeat Steps 9 - 10 until all subtasks are integrated.

[0010] Furthermore, Step 1 includes the following steps: S11. According to the task type, determine the form of training data, collect a large amount of data in the determined form, and form an initial dataset; S12. Pre - process the initial dataset collected in S11, such as screening, enhancement, denoising, annotation, etc., to form a pre - processed dataset; S13. Divide the pre - processed dataset in S12 into a large model training set and a large model test set at a certain ratio.

[0011] Furthermore, Step 2 includes the following steps: S21. Analyze the large model tasks and determine the output form of the large model; S22. Evenly split the output of the large model in S21 into multiple sub - outputs, and regard each sub - output as a sub - task; S23. Output the sub - task types, sub - task 1, sub - task 2,..., sub - task N; S24. Construct a sub - task training set. For sub - tasks 1 - N, each sub - task randomly samples a certain proportion from the large model training set determined in S13 of step 1 to form the training set of this sub - task; Furthermore, step 3 includes the following steps: S31. Based on the sub - tasks output in S23 of step 2, evenly split the output of each sub - task; S32. Determine the output branches of each sub - task after splitting: sub - task output branch 1, sub - task output branch 2,..., sub - task output branch Q; S33. Construct the training sets corresponding to different output branches under this sub - task: According to sub - task output branches 1 - Q in S32, each output branch randomly samples a certain proportion from the sub - task training set determined in S24 of step 2, so as to construct the training sets corresponding to different output branches of the sub - task: sub - task training set 1, sub - task training set 2,..., sub - task training set Q; Furthermore, step 4 includes the following steps: S41. For the sub - task types determined in S22 of step 2, determine the input and output data forms of the model; S42. Construct a set of alternative small models. Referring to the input and output data forms of the model determined in S41, collect a variety of small models with similar input and output data forms; Furthermore, step 5 includes the following steps: S51. Select sub - task type N and initialize N as 1; Step 6 includes the following steps: S61. Select sub - task N output branch 1 and select a small model from the set of alternative small models; S62. Configure the training of the small model, select the optimizer of the small model, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S63. Determine the loss function of sub - task N according to the specific task form of the selected sub - task N; S64. According to S33 in step 3, select the sub - task N training set 1 corresponding to sub - task N output branch 1; S65. Input the training set in S64 into the small model of sub - task N output branch 1, and calculate the training loss with the loss function determined in S63; S66. According to the training loss in S65, execute the backpropagation algorithm to obtain the parameter update gradient of the small model of the output branch 1 of subtask N; S67. Using the optimizer configured in S62, update the parameters of the small model of the output branch 1 of subtask N until the loss calculated in S65 converges, and save the small model of the output branch 1 of subtask N after convergence during the training process; S68. Repeat steps S61 - S67, select the small model after training convergence, and use this small model as the base model of subtask N; Step 7 includes the following steps: S71. Sequentially select the output q branches of subtask N, where q = 2,..., Q; S72. Select the base model determined in S68 of step 6 as the fine-tuning small model of the output branch q of subtask N; S73. Fine-tune the training configuration of the small model of the output branch q of subtask N, select the optimizer for the small model of the output branch q of subtask N, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S74. According to the task type of the selected output branch q of subtask N, determine the loss function of the output branch q of subtask N; S75. According to S33 in step 3, select the training set q of subtask N corresponding to the output branch q of subtask N; S76. Initialize the small model of the output branch q of subtask N in S72 with the base model parameters determined in S68 of step 6; S77. Input the training set in S75 into the small model of the output branch q of subtask N, and calculate the training loss with the loss function determined in S74; S78. According to the training loss in S77, execute the backpropagation algorithm to obtain the parameter update gradient of the small model on the output branch q of subtask N; S79. Using the optimizer configured in S73, update the parameters of the small model on the output branch q of subtask N until the loss calculated in S77 converges, and save the small model after convergence during the training process; S710. Repeat steps S71 - S79 until the training of the small models of all remaining output branches on subtask N is completed; Step 8 includes the following steps: S81. Respectively obtain the base model of subtask N in S68 of step 6 and the small models of all remaining output branches on subtask N in S710 of step 7; S82. Set the structure of the ensemble small model of subtask N as the base model; S83. Select all the training data of subtask N as the training set of the ensemble small model of subtask N in S82; S84. Select the optimizer for the sub-task N integrated small model, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer. S85. Select the loss function for the sub-task N integrated small model according to the task type of sub-task N. S86. Initialize the sub-task N integrated small model in S82 with the basic model parameters determined in S68 of step 6. S87. Input the training set in S83 into the output branch 1-Q of sub-task N, and input the output of the small model on the output branch 1-Q into the sub-task N integrated small model, and calculate the training loss with the loss function determined in S85. S88. According to the training loss in S87, execute the backpropagation algorithm to obtain the parameter update gradient of the sub-task N integrated small model. S89. Use the optimizer configured in S84 to update the parameters of the sub-task N integrated small model until the loss change calculated in S87 is stable, and save the minimum loss during the training process and the corresponding sub-task N integrated small model. Step 9 includes the following steps: S91. Update the sub-task N type, and set N = N + 1. S92. Repeat steps 6-8 until the integration module training of all sub-tasks is completed. Step 10 includes the following steps: S1001. Obtain the integration modules of sub-task N and N-1. S1002. Select the optimizer for the integrated small model of sub-task N and N-1, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer. S1003. Select the loss function for the integrated small model of sub-task N and N-1 according to the task types of sub-task N and N-1. S1004. Initialize the integrated small model of sub-task N and N-1 in S1002 with the basic model parameters determined in S68 of step 6. S1005. Input the training set in S1002 into the integration module of sub-task N and N-1, input the output of the integration module of sub-task N and N-1 into the integrated small model of sub-task N and N-1, and calculate the training loss with the loss function determined in S1003. S1006. According to the training loss in S1005, execute the backpropagation algorithm to obtain the parameter update gradient of the integrated small model of sub-task N and N-1. S1007. Use the optimizer configured in S1002 to update the parameters of the integrated small model of sub-task N and N-1 until the loss calculated in S1005 converges, and save the integrated small model of sub-task N and N-1 corresponding after the training converges. Step 11 includes the following steps: S1101. Repeat step 9 and save the integration modules for all remaining subtasks. S1102. Update the subtask N type, N = N + 1. S1103. Repeat step 10 to complete the training of the integrated small models for subtasks N and N - 1. S1104. Repeat the execution of S1102 - S1103 until the training of all subtask integrated small models is completed.

[0012] Another object of the present invention is to provide a fast weather forecasting method for a large weather forecasting model with low computing resources, including: A dataset construction module for constructing a dataset for the large weather forecasting model. A splitting module for splitting the large weather forecasting model task into subtask 1, subtask 2,..., subtask N, and determining the training sets for subtasks 1 - N. A decomposition module for decomposing the output of each subtask to determine the subtask output branches 1, subtask output branches 2,..., subtask output branches Q for each subtask. An alternative set construction module for collecting multiple small models that can perform the same task for a subtask type and constructing a small model alternative set. An initialization module for selecting subtask type N and initializing N to 1. A training module for, for the selected subtask N, training each small model in the small model alternative set for the output branch 1 of this subtask, and determining the small model with the optimal performance as the base model. A fine - tuning module for fine - tuning the small models on the remaining output branches of subtask N: on the base model, the remaining output branches of subtask N adopt the same small model structure as the base model, initialize the parameters of the small models on the remaining output branches of subtask N with the trained base model parameters, and perform parameter fine - tuning; construct the integrated small model for subtask N: based on the small models trained on all output branches of subtask N, use the same small model structure for the integrated small model of subtask N, input the outputs of all output branches under subtask N into the integrated small model of subtask N; initialize the integrated small model of subtask N with the base model parameters and perform parameter fine - tuning. An update module for updating subtask type N = N + 1 and training the integrated module for the current subtask N. An integration module for integrating the outputs of the integrated modules of subtasks N and N - 1, selecting the same network structure as the base model, initializing the integrated small models of subtasks N and N - 1 with the base model parameters, and fine - tuning the parameters of the integrated small models of subtasks N and N - 1; until all subtasks are integrated.

[0013] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor is caused to execute the steps of the method for rapidly predicting a large meteorological prediction model under low computing resources.

[0014] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is caused to execute the steps of the method for rapidly predicting a large meteorological prediction model under low computing resources.

[0015] Another object of the present invention is to provide an information data processing terminal for implementing the rapid construction of a large meteorological prediction model under low computing resources.

[0016] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows: First, there are many technical barriers in the construction of current large models, such as the difficulty of training large models with a large number of parameters, the high requirements of large models for data and computing resources, and the need for cooperation of experts in multiple fields. To break through these technical barriers of large models, this solution focuses on the meteorological field and proposes a method for rapidly predicting a large meteorological prediction model under low computing resources. Based on the fine-tuning technology of neural networks and the use of the integration idea, it is possible to rapidly construct an effective meteorological prediction large model from scratch under low computing resources.

[0017] The present invention can realize the rapid construction of a large model under low computing resources, avoiding the high computing resource requirements for training existing large models.

[0018] The present invention solves the problem of the difficulty of training large models with a large number of parameters by integrating multiple small models to form a large model.

[0019] The present invention focuses on the prediction model in the meteorological field and is also applicable to other fields.

[0020] Second, the expected benefits and commercial values after the transformation of the technical solution of the present invention are as follows: Greatly reduce the computing resource requirements for training large models, save training costs, and reduce the consumption of material resources.

[0021] Reduce the training difficulty of large models and reduce the consumption of human resources for training large models.

[0022] Lower the industry barriers for large model research, enabling research institutions in computing-intensive special industries such as meteorology, manufacturing, and healthcare, which have a large amount of data but limited computing resources, to also develop effective industry large models and empower the intelligent development of the industry.

[0023] The technical solution of the present invention fills the technical gaps at home and abroad in the industry: Existing large model training usually requires a large amount of computing resources, has a long training time, and requires the participation and cooperation of experts in multiple fields. This solution can quickly predict an effective large model with low computing resources, filling the technical gap that there is no large model that can be quickly predicted under existing low computing resources.

[0024] The technical solution of the present invention solves the technical problems that people have been eager to solve but have never been successful in: The solution of the present invention fits the actual situation of most domestic research institutions, realizes the rapid and effective training of large models with low resources, and reduces the industry barriers of large models.

[0025] Third, the existing technical problems solved by the technical solution of the present invention in industrial applications: 1. The problem of high computing resource requirements Existing large meteorological forecasting models usually rely on large-scale computing resources, such as high-performance computers and GPU clusters, which are difficult for ordinary institutions or enterprises to afford. The present invention greatly reduces the dependence on computing resources through task decomposition and modular small model construction, enabling efficient model training and inference in low computing power environments.

[0026] 2. The balance problem between model complexity and performance optimization Due to its holistic design, traditional meteorological models are difficult to achieve a balance between performance and model complexity. The present invention simplifies the model structure and improves the overall performance by splitting the task into multiple subtasks and training small models for each subtask.

[0027] 3. The problem of being difficult to optimize for specific scenarios Existing meteorological forecasting models are usually general models and are difficult to adapt to specific regional or task requirements. The present invention realizes efficient optimization for specific meteorological variables or scenarios through the decomposition of subtasks and a diverse set of alternative small models, improving the pertinence and adaptability of the forecast.

[0028] 4. The problem of low efficiency in constructing integrated models In traditional methods, the construction and optimization of multi-subtask integrated models usually require multiple independent trainings, resulting in low efficiency. The present invention greatly shortens the model training time by using fine-tuning and parameter inheritance techniques to quickly transfer the training results of the basic model to the subtask output branches and integration modules.

[0029] The significant technical progress of the present invention: 1. A significant improvement in resource utilization efficiency Through modular task decomposition and small model selection, the present invention achieves forecasting performance close to that of traditional large-scale meteorological models under low computing power conditions, significantly improves resource utilization efficiency, and reduces the threshold for industrial deployment.

[0030] 2. Improvement of Model Accuracy and Adaptability Through the refinement of subtasks and the optimization of branch outputs, each subtask is supported by a specially optimized small model, and finally an overall model is formed through the subtask integration module, greatly improving the accuracy and scenario adaptability of meteorological forecasting.

[0031] 3. Improvement of Training Efficiency and Scalability The present invention adopts the basic model parameter inheritance and fine-tuning technology in model construction, reduces the computational amount of repeated training, and improves training efficiency. At the same time, the task decomposition method makes the model expansion more flexible and can quickly adapt to new variables or tasks.

[0032] 4. Applicable to Multiple Industrial Scenarios This technical solution is not only applicable to large-scale meteorological forecasting systems, but also can be extended to small-scale and regional forecasting requirements, such as agricultural meteorological services, urban meteorological disaster warnings and other scenarios, and has broad industrial application value.

[0033] 5. Reduction of R & D and Deployment Costs Compared with the traditional large meteorological models that require expensive equipment and long-term development, the present invention greatly reduces the R & D cycle and deployment costs through the selection of small models and modular training methods, providing feasibility for more enterprises and institutions.

[0034] By solving the key problems of the existing technology and achieving significant technological progress, the present invention has important application value and competitive advantages in the meteorological forecasting industry. Description of the Drawings

[0035] Figure 1 It is a flowchart of a fast forecasting method for a large meteorological forecasting model under low computing resources provided by an embodiment of the present invention.

[0036] Figure 2 It is a structural block diagram of a fast forecasting system for a large meteorological forecasting model under low computing resources provided by an embodiment of the present invention.

[0037] Figure 3 It is a schematic diagram of the principle of a fast forecasting method for a large model under low computing resources provided by an embodiment of the present invention. Detailed Embodiments

[0038] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] As Figure 1 shown, a fast weather forecasting method for a large weather forecasting model with low computing resources provided by an embodiment of the present invention includes the following steps: S101. Construct a large weather forecasting model dataset; S102. Split the large weather forecasting model task into subtasks 1, subtasks 2,..., subtasks N, and determine the training sets of subtasks 1-N; S103. Decompose the output of each subtask, and determine the subtask output branches 1, subtask output branches 2,..., subtask output branches Q of each subtask; S104. Collect multiple small models that perform the same task for the subtask type, and construct a small model alternative set; S105. Select subtask type N and initialize N to 1; S106. For the subtask type N selected in S105, train each small model in the small model alternative set for the subtask output branch 1 of this subtask, and determine the small model with the best performance as the basic model; S107. Fine-tune the small models on the remaining output branches of subtask N: On the basic model in S106, the other output branches of subtask N adopt the same small model structure as the basic model, initialize the parameters of the small models on the other output branches of subtask N with the parameters of the basic model trained in S106, and perform parameter fine-tuning; S108. Construct an integrated small model for subtask N: Based on the small models trained on all output branches of subtask N in S106 and S107, use the same small model structure to construct an integrated small model for subtask N, and input the outputs of all output branches under subtask N into the integrated small model for subtask N; Initialize the parameters of the integrated small model for subtask N with the parameters of the basic model in S106, and perform parameter fine-tuning; S109. Update subtask type N = N + 1, repeat S106 - S108, and train the integrated model of the current subtask N; S110. Integrate the output of the integrated models of subtasks N and N - 1, select the same network structure as the basic model in S106, initialize the parameters of the integrated small models of subtasks N and N - 1 with the parameters of the basic model in S106, and fine-tune the parameters of the integrated small models of subtasks N and N - 1; S111. Repeat S109 - S110 until all subtasks are integrated.

[0040] An embodiment of the present invention proposes a method for quickly predicting a large meteorological prediction model with low computing resources. Through modular task decomposition and small model optimization, the bottleneck of traditional meteorological prediction models in computing resource consumption is solved. The working principle includes four main steps: dataset construction, task decomposition, model selection, parameter optimization, and module integration.

[0041] First, in the dataset construction stage (S101), according to the requirements of the meteorological prediction task, multiple meteorological data sources are integrated to form a large-scale dataset covering multiple variables, laying a foundation for subsequent training. Subsequently, the complex meteorological prediction task is decomposed into multiple independent subtasks (S102), and each subtask makes predictions for specific meteorological variables or regions, simplifying the overall problem into subproblems that can be solved independently. The output of each subtask is further decomposed into multiple output branches (S103) to ensure that the model has high adaptability to fine-grained prediction targets.

[0042] Secondly, in the model training stage (S104 - S107), for each subtask, first, multiple small models are collected and evaluated, and the optimal small model is selected as the base model according to performance (S106). Subsequently, through the fine-tuning mechanism (S107), using the parameters of the base model as the initialization point, the small models of other output branches of the subtask are further optimized, thereby achieving efficient training and improving the model performance. This modular small model training method reduces the computational complexity while ensuring high accuracy of each subtask model.

[0043] Thirdly, in the model integration stage (S108), all output branches of the subtasks are integrated into a subtask-level integrated small model through the same small model structure, and the parameters of the integrated small model are fine-tuned to ensure the consistency and coordination of the internal outputs of the subtasks. In this way, each subtask independently constructs an integration module, providing an extensible basic module for the construction of the final large model. Subsequently, the integration between subtasks is carried out step by step in a progressive manner (S109 - S111). The output of the integration modules of subtask N and N - 1 is further integrated through the same small model structure, and the parameters are optimized using the fine-tuning mechanism to achieve efficient cooperation between subtasks.

[0044] Finally, the whole process takes modular integration as the core until the integration modules of all subtasks are constructed. Through distributed subtask processing, optimized model selection, a progressive strategy of step-by-step integration, and fine-tuning technology, the present invention achieves the goal of constructing an efficient large meteorological prediction model in a low computing resource environment, not only significantly reducing resource consumption but also ensuring the accuracy and robustness of the model.

[0045] S101 provided by the embodiment of the present invention includes the following steps: S11. Determine the form of the training data according to the task type, collect a large amount of data in the determined form, and form an initial data set. S12. Preprocess the initial data set collected in S11, such as screening, enhancement, denoising, annotation, etc., to form a preprocessed data set. S13. Divide the preprocessed data set in S12 into a large model training set and a large model test set in a certain proportion.

[0046] Step S102 provided by the embodiment of the present invention includes the following steps: S21. Analyze the large model task and determine the output form of the large model. S22. Evenly split the output of the large model in S21 into multiple sub-outputs, and regard each sub-output as a sub-task. S23. Output the sub-task types, sub-task 1, sub-task 2,..., sub-task N. S24. Construct a sub-task training set. For sub-tasks 1 - N, a certain proportion of random sampling is performed on each sub-task from the large model training set determined in S13 of S101 to form the training set for this sub-task. Step S103 provided by the embodiment of the present invention includes the following steps: S31. Based on the sub-tasks output in S23 of S102, evenly split the output of each sub-task. S32. Determine the output branches of each sub-task after splitting: sub-task output branch 1, sub-task output branch 2,..., sub-task output branch Q. S33. Construct the training sets corresponding to different output branches under this sub-task: According to sub-task output branches 1 - Q in S32, a certain proportion of random sampling is performed on each output branch from the sub-task training set determined in S24 of S102, so as to construct the training sets corresponding to different output branches of the sub-task: sub-task training set 1, sub-task training set 2,..., sub-task training set Q. Step S104 provided by the embodiment of the present invention includes the following steps: S41. For the sub-task types determined in S22 of S102, determine the input and output data forms of the model. S42. Construct a set of alternative small models. Referring to the model input and output data forms determined in S41, collect a variety of small models with similar input and output data forms. Step S105 provided by the embodiment of the present invention includes the following steps: S51. Select sub-task type N and initialize N to 1. Step S106 includes the following steps: S61. Select sub-task N output branch 1 and select a small model from the set of alternative small models. S62, Small model training configuration, select the optimizer for the small model, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S63, According to the specific task form of the selected sub-task N, determine the loss function of sub-task N; S64, According to S33 in S103, select the training set 1 of sub-task N corresponding to output branch 1 of sub-task N; S65, Input the training set in S64 into the small model of output branch 1 of sub-task N, and calculate the training loss with the loss function determined in S63; S66, According to the training loss in S65, execute the backpropagation algorithm to obtain the parameter update gradient of the small model of output branch 1 of sub-task N; S67, Use the optimizer configured in S62 to update the parameters of the small model of output branch 1 of sub-task N until the loss change calculated in S65 is stable, save the minimum loss during the training process and the corresponding small model of output branch 1 of sub-task N; S68, Repeat steps S61 - S67, select the small model with the minimum loss, and use this small model as the base model of sub-task N; S107 includes the following steps: S71, Sequentially select the output q branches of sub-task N, where q = 2,..., Q; S72, Select the base model determined in S68 of S106 as the fine-tuning small model of output branch q of sub-task N; S73, Fine-tune the training configuration of the small model of output branch q of sub-task N, select the optimizer for the small model of output branch q of sub-task N, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S74, According to the task type of the selected output branch q of sub-task N, determine the loss function of output branch q of sub-task N; S75, According to S33 in S103, select the training set q of sub-task N corresponding to output branch q of sub-task N; S76, Initialize the small model of output branch q of sub-task N in S72 with the base model parameters determined in S68 of S106; S77, Input the training set in S75 into the small model of output branch q of sub-task N, and calculate the training loss with the loss function determined in S74; S78, According to the training loss in S77, execute the backpropagation algorithm to obtain the parameter update gradient of the small model on output branch q of sub-task N; S79, Use the optimizer configured in S73 to update the parameters of the small model on output branch q of sub-task N until the loss change calculated in S77 is stable, save the minimum loss during the training process and the corresponding small model under the minimum loss; S710. Repeat steps S71 - S79 until the training of all the small models on the remaining output branches of subtask N is completed; S108 includes the following steps: S81. Obtain the base model of subtask N in S68 of S106 and the small models of all the remaining output branches of subtask N in S710 of S107 respectively; S82. Set the integrated small model structure of subtask N as the base model; S83. Select all the training data of subtask N as the training set for the integrated small model of subtask N in S82; S84. Select the optimizer for the integrated small model of subtask N and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S85. Select the loss function for the integrated small model of subtask N according to the task type of subtask N; S86. Initialize the integrated small model of subtask N in S82 with the base model parameters determined in S68 of S106; S87. Input the training set in S83 into output branches 1 - Q of subtask N, and input the outputs of the small models on output branches 1 - Q into the integrated small model of subtask N, and calculate the training loss with the loss function determined in S85; S88. According to the training loss in S87, execute the backpropagation algorithm to obtain the parameter update gradient of the integrated small model of subtask N; S89. Use the optimizer configured in S84 to update the parameters of the integrated small model of subtask N until the change in the loss calculated in S87 is stable, and save the minimum loss during the training process and the corresponding integrated small model of subtask N; S109 includes the following steps: S91. Update the type of subtask N, set N = N + 1; S92. Repeat S106 - S108 until the training of the integrated modules for all subtasks is completed; S110 includes the following steps: S1001. Obtain the integrated modules of subtask N and N - 1; S1002. Select the optimizer for the integrated small models of subtask N and N - 1 and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S1003. Select the loss functions for the integrated small models of subtask N and N - 1 according to the task types of subtask N and N - 1; S1004. Initialize the integrated small models of subtask N and N - 1 in S1002 with the base model parameters determined in S68 of S106; S1005. Input the training set in S1002 into the integrated module of subtasks N and N - 1. The output of the integrated module of subtasks N and N - 1 is input into the integrated small model of subtasks N and N - 1, and calculate the training loss with the loss function determined in S1003. S1006. According to the training loss in S1005, execute the backpropagation algorithm to obtain the parameter update gradients of the integrated small model of subtasks N and N - 1. S1007. Use the optimizer configured in S1002 to update the parameters of the integrated small model of subtasks N and N - 1 until the loss calculated in S1005 converges, and save the integrated small model corresponding to subtasks N and N - 1 after training convergence. S111 includes the following steps: S1101. Repeat S109 and save the integrated modules of all remaining subtasks. S1102. Update the subtask N type, N = N + 1. S1103. Repeat S110 to complete the training of the integrated small model of subtasks N and N - 1. S1104. Repeat the execution of S1102 - S1103 until the training of all integrated small models of subtasks is completed.

[0047] As Figure 2 shown, a rapid weather forecasting large model system with low computing resources provided by an embodiment of the present invention includes: A dataset construction module for constructing a weather forecasting large model dataset; A splitting module for splitting the weather forecasting large model task into subtasks 1, subtasks 2,..., subtasks N, and determining the training sets of subtasks 1 - N; A decomposition module for decomposing the output of each subtask to determine the subtask output branches 1, subtask output branches 2,..., subtask output branches Q of each subtask; An alternative set construction module for collecting multiple small models that can perform the same task for a subtask type and constructing a small model alternative set; An initialization module for selecting the subtask type N and initializing N to 1; A training module for, for the selected subtask N, training each small model in the small model alternative set for the subtask output branch 1 of this subtask, and determining the small model with the best performance as the base model. A fine-tuning module for fine-tuning the small model on the remaining output branches of sub-task N: On the basis model, the same small model structure as the basis model is adopted for the remaining output branches of sub-task N. The parameters of the small model on the remaining output branches of sub-task N are initialized with the trained basis model parameters and then fine-tuned; Constructing the integrated small model for sub-task N: Based on the trained small models on all output branches of sub-task N, an integrated small model for sub-task N is constructed with the same small model structure, and the outputs of all output branches under sub-task N are input into the integrated small model for sub-task N; The parameters of the integrated small model for sub-task N are initialized with the basis model parameters and then fine-tuned; An update module for updating the sub-task type N = N + 1 and training the current sub-task N integration module; An integration module for integrating the output of the integration modules of sub-task N and N - 1, selecting the same network structure as the basis model, initializing the integrated small models of sub-task N and N - 1 with the basis model parameters, and fine-tuning the parameters of the integrated small models of sub-task N and N - 1; until all sub-task integrations are completed.

[0048] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the method for quickly predicting a large meteorological forecasting model under low computing resources.

[0049] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method for quickly predicting a large meteorological forecasting model under low computing resources.

[0050] Another object of the present invention is to provide an information data processing terminal for implementing the rapid construction of a large meteorological forecasting model under low computing resources.

[0051] Specific implementation of the present invention: The technical principle diagram of the present invention is as Figure 3 shown, and the technical solutions adopted are as follows: 1. Construct a large meteorological forecasting model dataset; 2. Split the large meteorological forecasting model task into sub-task 1, sub-task 2,..., sub-task N, and determine the training sets of sub-tasks 1 - N.

[0052] 3. Decompose the output of each sub-task and determine the sub-task output branches 1, sub-task output branches 2,..., sub-task output branches Q of each sub-task.

[0053] 4. Collect multiple small models that can perform the same task for the sub-task type and construct a small model alternative set.

[0054] 5. Select the sub - task type N and initialize N to 1.

[0055] 6. For the selected sub - task N in step 5, output branch 1 for this sub - task, train each small model in the small model candidate set, and determine the small model with the best performance as the base model.

[0056] 7. Fine - tune the small models on the remaining output branches of sub - task N: On the base model in step 6, for the other output branches of sub - task N, use the same small model structure as the base model, initialize the parameters of the small models on the other output branches of sub - task N with the parameters of the base model trained in step 6, and perform parameter fine - tuning.

[0057] 8. Construct the ensemble small model for sub - task N: Based on the small models trained on all output branches of sub - task N in steps 6 and 7, use the same small model structure to construct the ensemble small model for sub - task N, and input the outputs of all output branches under sub - task N into the ensemble small model for sub - task N. Initialize the parameters of the ensemble small model for sub - task N with the parameters of the base model in step 6, and perform parameter fine - tuning.

[0058] 9. Update the sub - task type N = N + 1, repeat steps 5 - 8 to train the current sub - task N ensemble module.

[0059] 10. Integrate the outputs of the ensemble modules of sub - task N and N - 1, select the same network structure as the base model in step 6, initialize the parameters of the ensemble small models of sub - task N and N - 1 with the parameters of the base model in step 6, and fine - tune the parameters of the ensemble small models of sub - task N and N - 1.

[0060] 11. Repeat steps 9 - 10 until all sub - task integrations are completed.

[0061] Figure 3 Schematic diagram of the large - model fast prediction method under low computing resources Step 1 includes the following steps: S11. According to the task type, determine the form of the training data, collect a large amount of data in the determined form, and form an initial data set.

[0062] S12. Pre - process the initial data set collected in S11, such as screening, enhancement, denoising, annotation, etc., to form a pre - processed data set.

[0063] S13. Divide the pre - processed data set in S12 into a large - model training set and a large - model test set at a certain ratio.

[0064] Step 2 includes the following steps: S21. Analyze the large - model task and determine the output form of the large model.

[0065] S22. Evenly split the output of the large model in S21 into multiple sub-outputs, and consider each sub-output as a sub-task.

[0066] S23. Output the sub-task types, sub-task 1, sub-task 2, …, sub-task N.

[0067] S24. Construct a sub-task training set. For each of the sub-tasks 1 - N, randomly sample a certain proportion from the large model training set determined in S13 of step 1 to form the training set for this sub-task.

[0068] Step 3 includes the following steps: S31. Based on the sub-tasks output in S23 of step 2, evenly split the output of each sub-task.

[0069] S32. Determine the output branches for each of the split sub-tasks: sub-task output branch 1, sub-task output branch 2, …, sub-task output branch Q.

[0070] S33. Construct the training sets corresponding to different output branches for this sub-task: According to the sub-task output branches 1 - Q in S32, for each output branch, randomly sample a certain proportion from the sub-task training set determined in S24 of step 2, so as to construct the training sets corresponding to different output branches of the sub-task: sub-task training set 1, sub-task training set 2, …, sub-task training set Q.

[0071] Step 4 includes the following steps: S41. For the sub-task types determined in S22 of step 2, determine the input and output data forms of the model.

[0072] S42. Construct a set of alternative small models. Referring to the model input and output data forms determined in S41, collect multiple small models with similar input and output data forms.

[0073] Step 5 includes the following steps: S51. Select sub-task type N and initialize N to 1.

[0074] Step 6 includes the following steps: S61. Select sub-task N output branch 1 and select a small model from the set of alternative small models.

[0075] S62. Configure the training of the small model. Select the optimizer for the small model and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer.

[0076] S63. According to the specific task form of the selected sub-task N, determine the loss function of sub-task N.

[0077] S64. According to S33 in step 3, select the subtask N training set 1 corresponding to the output branch 1 of subtask N.

[0078] S65. Input the training set in S64 into the small model of the output branch 1 of subtask N, and calculate the training loss using the loss function determined in S63.

[0079] S66. According to the training loss in S65, execute the backpropagation algorithm to obtain the parameter update gradient of the small model of the output branch 1 of subtask N.

[0080] S67. Using the optimizer configured in S62, update the parameters of the small model of the output branch 1 of subtask N until the change in the loss calculated in S65 is stable, and save the minimum loss during the training process and the corresponding small model of the output branch 1 of subtask N.

[0081] S68. Repeat steps S61 - S67, select the small model with the minimum loss, and use this small model as the base model of subtask N.

[0082] Step 7 includes the following steps: S71. Sequentially select the output q branches of subtask N, where q = 2, …, Q.

[0083] S72. Select the base model determined in S68 in step 6 as the fine-tuning small model of the output branch q of subtask N.

[0084] S73. Fine-tune the training configuration of the small model of the output branch q of subtask N, select the optimizer of the small model of the output branch q of subtask N, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer.

[0085] S74. According to the task type of the selected output branch q of subtask N, determine the loss function of the output branch q of subtask N.

[0086] S75. According to S33 in step 3, select the subtask N training set q corresponding to the output branch q of subtask N.

[0087] S76. Initialize the small model of the output branch q of subtask N in S72 with the base model parameters determined in S68 in step 6.

[0088] S77. Input the training set in S75 into the small model of the output branch q of subtask N, and calculate the training loss using the loss function determined in S74.

[0089] S78. According to the training loss in S77, execute the backpropagation algorithm to obtain the parameter update gradient of the small model on the output branch q of subtask N.

[0090] S79. Update the small model parameters on the output branch q of subtask N with the optimizer configured in S73 until the loss change calculated in S77 is stable, and save the minimum loss during the training process and the corresponding small model under the minimum loss.

[0091] S710. Repeat steps S71 - S79 until the training of the small models on all remaining output branches of subtask N is completed.

[0092] Step 8 includes the following steps: S81. Obtain the base model of subtask N in step 6 (S68) and the small models on all remaining output branches of subtask N in S710 of step 7 respectively.

[0093] S82. Set the integrated small model structure of subtask N as the base model.

[0094] S83. Select all the training data of subtask N as the training set for the integrated small model of subtask N in S82.

[0095] S84. Select the optimizer for the integrated small model of subtask N, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer.

[0096] S85. Select the loss function for the integrated small model of subtask N according to the task type of subtask N.

[0097] S86. Initialize the integrated small model of subtask N in S82 with the base model parameters determined in S68 of step 6.

[0098] S87. Input the training set in S83 into output branches 1 - Q of subtask N, input the outputs of the small models on output branches 1 - Q into the integrated small model of subtask N, and calculate the training loss with the loss function determined in S85.

[0099] S88. According to the training loss in S87, execute the backpropagation algorithm to obtain the parameter update gradient of the integrated small model of subtask N.

[0100] S89. Update the parameters of the integrated small model of subtask N with the optimizer configured in S84 until the loss change calculated in S87 is stable, and save the minimum loss during the training process and the corresponding integrated small model of subtask N.

[0101] Step 9 includes the following steps: S91. Update the type of subtask N, set N = N + 1.

[0102] S92. Repeat steps 5 - 8 until the training of the integrated modules for all subtasks is completed.

[0103] Step 10 includes the following steps: S1001, Obtain the integration modules for subtasks N and N - 1.

[0104] S1002, Select the integration small model optimizer for subtasks N and N - 1, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer.

[0105] S1003, According to the task types of subtasks N and N - 1, select the integration small model loss function for subtasks N and N - 1.

[0106] S1004, Initialize the integration small models for subtasks N and N - 1 in S1002 with the basic model parameters determined in step 6.

[0107] S1005, Input the training set in S1002 into the integration module for subtasks N and N - 1. The output of the integration module for subtasks N and N - 1 is input into the integration small models for subtasks N and N - 1, and calculate the training loss with the loss function determined in S1003.

[0108] S1006, According to the training loss in S1005, perform the backpropagation algorithm to obtain the parameter update gradients of the integration small models for subtasks N and N - 1.

[0109] S1007, Use the optimizer configured in S1002 to update the parameters of the integration small models for subtasks N and N - 1 until the loss change calculated in S1005 is stable, and save the minimum loss during the training process and the corresponding integration small models for subtasks N and N - 1.

[0110] Step 11 includes the following steps: S1101, Repeat step 9 and save the integration modules for all the remaining subtasks.

[0111] S1102, Update the type of subtask N, N = N + 1.

[0112] S1103, Repeat step 10 to complete the training of the integration small models for subtasks N and N - 1.

[0113] S1104, Repeat the execution of S1102 - S1103 until the training of all the integration small models is completed.

[0114] This solution includes: 1. Construct a meteorological large model dataset: A11, Collect data on different meteorological elements globally from the ERA5 platform for 10 years (2013 - 2023); A12, Process the data in A11 into a tensor format of H*W*C*T, where H, W, C, and T represent the height, width, number of meteorological elements, and time of the data respectively; A13. Divide the meteorological data in step A12 in a ratio of 7:3 from the time dimension to construct a training set and a test set.

[0115] 2. Construction and training of the meteorological large model.

[0116] A21. Determine the subtasks of the meteorological forecast large model and construct a subtask training set.

[0117] 1.1 Determine that the task of the meteorological forecast large model is time series modeling, which inputs all meteorological elements at the previous t0 moments and predicts all meteorological elements at the subsequent t1 moment. The input data is in the form of a tensor of H*W*C*t0, and the output is H*W*C*(t0 + t1).

[0118] 1.2 Evenly split the output H*W*C*(t0 + t1) in the time dimension in A21 into N sub-outputs, and the data format of each subtask output is: H*W*C*((t0 + t1) / N).

[0119] 1.3. Output the subtask types, subtask 1, subtask 2,..., subtask N. The input and output tensor sizes of subtasks 1 - N satisfy: input tensor size: H*W*C*t0, output tensor size: H*W*C*((t0 + t1) / N).

[0120] 1.4. Construct a subtask training set. For subtasks 1 - N, each subtask randomly samples 70% from the large model training set determined in 1.3) to form the training set of this subtask.

[0121] A22. Split the subtask output and construct a subtask output branch training set.

[0122] 2.1. Based on the subtasks in 1.3, evenly split the output of each subtask. The input and output tensor shapes of the subtask output branch are: H*W*C*t0, H*W*C*((t0 + t1) / N / Q).

[0123] 2.2 According to the output f shape of the subtask output branch in 2.1, determine each subtask output branch in turn: subtask output branch 1, subtask output branch 2,..., subtask output branch Q.

[0124] 2.3. According to subtask output branches 1 - Q in 2.2, each output branch randomly samples 70% of the data from the subtask training set determined in 1)4 to construct the training sets on different output branches of the subtask: subtask training set 1, subtask training set 2,..., subtask training set Q.

[0125] A23. Determine the small model candidate set 3.1. For the task type in 1.1 and the input and output tensor formats, select alternative small models that can perform the same task and have the same input data format.

[0126] 3.2. Determine multiple alternative small models and construct a set of alternative small models.

[0127] A24, Training of the basic model for subtask N 4.1. Select subtask type N and initialize N to 1.

[0128] 4.2. Select output branch 1 of subtask N and choose a small model from the set of alternative small models.

[0129] 4.3. Configure the training of the small model. Select Adam as the optimizer for the small model, and set the learning rate of the optimizer to 0.0001, the weight decay coefficient to 0.001, and the adaptive moment estimation parameters to 0.9 and 0.99.

[0130] 4.4. Determine that the loss function for output branch 1 of subtask N is MSE.

[0131] 4.5. Select the corresponding subtask training set 1 for output branch 1 of subtask N from 2.3.

[0132] 4.6. Input the training set in 4.5 into the small model and calculate the training loss using the loss function determined in 4.4.

[0133] 4.7. According to the training loss in 4.6, execute the backpropagation algorithm to obtain the parameter update gradient of the small model.

[0134] 4.8. Using the Adam optimizer configured in 4.3, update the parameters of the small model until the loss calculated in 4.6 converges, and save the small model corresponding to the converged training.

[0135] 4.9. Repeat steps 4.2 - 4.8, select the small model with converged training, and use this small model as the basic model.

[0136] A25, Fine-tuning the small models of the remaining output branches of subtask N 5.1. Sequentially select output branch q of subtask N, where q = 2, …, Q, and initialize q = 2.

[0137] 5.2. Select the basic model in 4.9 as the small model on output branch q of subtask N.

[0138] 5.3. Configure the training of the fine-tuning small model for branch q of subtask N. Select Adam as the optimizer for the small model, and set the learning rate of the optimizer to 0.0001, the weight decay coefficient to 0.001, and the adaptive moment estimation parameters to 0.9 and 0.99.

[0139] 5.4. Select MSE as the loss function for the output branch q of subtask N.

[0140] 5.5. Select the subtask training set q corresponding to the output branch q of subtask N from 2.3.

[0141] 5.6. Initialize the small model on the output branch q of subtask N in 5.2 with the basic model parameters in 4.9.

[0142] 5.7. Input the training set in 5.5 into the small model on the output branch q of subtask N, and calculate the training loss with the loss function determined in 5.4.

[0143] 5.8. According to the training loss in 5.7, execute the backpropagation algorithm to obtain the updated gradient of the parameters of the small model on the output branch q of subtask N.

[0144] 5.9. Use the optimizer configured in 5.3 to update the parameters of the small model on the output branch q of subtask N until the change in the loss calculated in 5.7 is stable, save the minimum loss during training and the corresponding small model on the output branch q of subtask N.

[0145] 5.10. Update the output branch q of subtask N, q = q + 1, and repeat steps 5.1 - 5.9 until the training of the small models on all remaining output branches of subtask N is completed.

[0146] A26, Fine-tune the integrated small model of subtask N 6.1. Obtain the basic model on the output branch 1 of subtask N and the small models on the remaining output branches of subtask N from 4.9 and 5.10 respectively.

[0147] 6.2. Set the structure of the integrated small model of subtask N as the basic model.

[0148] 6.3. Select all the training data of subtask N from 1.4 as the training set of the integrated small model of subtask N in 6.2.

[0149] 6.4. Select Adam as the optimizer for the integrated small model of subtask N, and set the learning rate of the optimizer to 0.0001, the weight decay coefficient to 0.001, and the adaptive moment estimation parameters to 0.9 and 0.99.

[0150] 6.5. Select the loss function of the integrated small model of subtask N as MSE.

[0151] 6.6. Initialize the integrated small model of subtask N in 6.2 with the basic model parameters in 4.9.

[0152] 6.7. Input the training set of subtask N in 6.3 into the small model on the output branch 1-Q of subtask N, and input the output of the small model on the output branch 1-Q of subtask N into the integrated small model of subtask N, and calculate the training loss with the loss function determined in 6.5.

[0153] 6.8. According to the training loss in 6.7, execute the backpropagation algorithm to obtain the parameter update gradient of the integrated small model of subtask N.

[0154] 6.9. Use the optimizer configured in 6.4 to update the parameters of the integrated small model of subtask N until the change of the loss calculated in 6.7 is stable, and save the minimum loss during the training process and the corresponding integrated small model of subtask N.

[0155] A27. Train other subtask integration modules 7.1. Update the type of subtask N, and set N = N + 1.

[0156] 7.2. Repeat steps A22 - A26 until all remaining subtask integration modules are completed.

[0157] A28. Integration of subtask N and N - 1 8.1. Obtain the integration modules of subtask N and N - 1.

[0158] 8.2. Set the structure of the integrated small model on subtask N and N - 1 as the basic small model.

[0159] 8.3. Select all the training data on subtask N and N - 1 from 1.4 as the training set of the integrated small model of subtask N and N - 1 in 8.2 8.4. Set the optimizer of the integrated small model of subtask N and N - 1 in 8.2 to Adam, and set the learning rate of the optimizer to 0.0001, the weight decay coefficient to 0.001, and the adaptive moment estimation parameters to 0.9 and 0.99.

[0160] 8.5. Select MSE as the loss function of the integrated small model on subtask N and N - 1.

[0161] 8.6. Initialize the integrated small model of subtask N and N - 1 in 8.2 with the basic model parameters in 4.9.

[0162] 8.7. Input the training set in 8.3 into the integration module of subtask N and N - 1, and input the output of the integration module of subtask N and N - 1 into the integrated small model of subtask N and N - 1, and calculate the training loss with the loss function determined in 8.5.

[0163] 8.8. According to the training loss in 8.7, execute the backpropagation algorithm to obtain the parameter update gradient of the integrated small model of subtask N and N - 1.

[0164] 8.9. Update the integrated small model parameters of subtasks N and N - 1 using the optimizer configured in 8.4 until the loss change calculated in 8.7 becomes stable, and save the minimum loss during the training process and the corresponding integrated small models of subtasks N and N - 1.

[0165] A29, Integration of remaining subtasks 9.1. Repeat A27 to complete the training of the integration module for all remaining subtasks.

[0166] 9.2. Update the type of subtask N, N = N + 1.

[0167] 9.3. Repeat A28 to complete the training of the integrated small models of subtasks N and N - 1.

[0168] 9.4. Repeat the execution of 9.2 - 9.3 until the training of all integrated small models of subtasks is completed.

[0169] 3. Testing of the large meteorological forecasting model A31, Input the test data into subtask 1, subtask 2, …, subtask N in sequence.

[0170] A32, Obtain the output results of the integrated small models of subtasks 1, 2, …, N.

[0171] The present invention provides a fast forecasting method for large models with low computing resources, which successfully overcomes the problem of high computing resource requirements in the training process of existing large models. This method avoids the bottleneck of traditional large models in terms of hardware resources and provides the possibility for developing large - scale meteorological forecasting models in scenarios with limited computing resources. Secondly, the present invention constructs a large model by integrating multiple small models, significantly reducing the training difficulty brought by a large number of parameters. It decomposes complex problems into subtask modules and trains and integrates them step by step, improving the efficiency and scalability of model training. In addition, although the present invention focuses on the forecasting model in the meteorological field, its methodology is also applicable to the construction of large models in other fields, with strong generality and cross - domain adaptability.

[0172] The technical solutions in the present invention have a certain degree of flexibility, and similar inventive purposes can be achieved through the following alternative solutions: 1. Replace the base model: This solution uses the TAU model as the base small model. However, for different tasks, it can be flexibly replaced with other lightweight small models, such as network structures like LSTM, Transformer, or CNN, and adjust the model structure and configuration according to specific task requirements.

[0173] 2. Retrain the sub-task model: In the current solution, the training of each sub-task branch small model and the sub-task integration model adopts the fine-tuning training method. In the alternative solution, a retraining method from scratch can be adopted. Although it requires a longer training time, it can perform a more refined model fitting for new data.

[0174] 3. Optimize the training configuration: During the training process of the base model, sub-task branch models, and integration models, other types of optimizers (such as AdamW, SGD, etc.) can be selected, and the learning rate or weight decay parameters can be adjusted to improve the convergence speed and prediction accuracy of the model.

[0175] The technical key points of the present invention lie in its large model integration design method under low computing resources and the integration training method adapted to large parameter quantities. It mainly includes the following points: 1. Large model integration design method: By decomposing the prediction task of future meteorological variables into sub-task modules and sub-task branches, the problem of computing resource limitations caused by the high parameter quantity of large models is solved. Each sub-task branch model focuses on the prediction of specific time periods and variable dimensions, greatly improving the training efficiency and model performance.

[0176] 2. Integration training method: Adopting a modular design, by independently training sub-task models and integrating sub-task outputs in the integration stage, the decoupling between sub-tasks is ensured, the training difficulty of complex models is reduced, and at the same time, efficient prediction performance is achieved.

[0177] Based on ERA5 meteorological data, the present invention proposes a large model construction method, using the TAU model as the base small model, inputting 20 variables at 6 times and 4 pressure levels, and outputting 20 meteorological elements corresponding to future 4, 8, 12, and 16 times. The specific model design includes: 1. Sub-task decomposition: The variable output at future 16 times is decomposed into 4 sub-tasks, and each sub-task is responsible for predicting meteorological variables at 1-4, 4-8, 8-12, and 12-16 times respectively.

[0178] 2. Sub-task branch design: Each sub-task is further refined into 4 branches, which are responsible for outputting meteorological elements at positions 1-5, 5-10, 10-15, and 15-20 respectively, forming a finer-grained task division.

[0179] 3. Sub-task integration: Gradually integrate the outputs of each sub-task into the complete 20-variable output at future 1-4, 1-8, 1-12, and 1-16 times, realizing the collaborative integration between tasks and ensuring the consistency and integrity of the model output.

[0180] With this design, the present invention decomposes complex weather forecasting tasks into multiple subtasks in a modular manner for independent training and collaborative integration, greatly improving the training efficiency and prediction accuracy. It is applicable to resource-constrained scenarios and also has strong scalability and application value.

[0181] Evidence related to the technical effects obtained in the embodiments of the present invention.

[0182] The experimental results are as follows: MSE test results of the large weather forecasting model It should be noted that the embodiments of the present invention can be implemented through hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, such as provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software, such as firmware.

[0183] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention by those skilled in the art within the disclosed technical scope of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A method for rapid forecasting of a large meteorological forecast model under low computing resources, characterized in that: The following steps are involved: Step 1: Construct a large weather forecast model dataset; Step 2: split the weather forecast model task into subtask 1, subtask 2, ..., subtask N, and determine the training sets of subtasks 1-N; Step 3, decompose each subtask output, and determine subtask output branch 1, subtask output branch 2, ..., subtask output branch Q of each subtask; Step 4: Collect multiple small models that perform the same task according to the subtask type and build a small model candidate set; Step 5: Select subtask type N and initialize N to 1; Step 6: Select subtask type N for step 5, output branch 1 for this subtask, train each small model in the small model candidate set, and determine the small model with the best performance as the basic model; Step 7, fine-tune the small model on the remaining output branch of subtask N: On the basic model in step 6, the remaining output branch of subtask N adopts the same small model structure as the basic model, initializes the parameters of the small model on the remaining output branch of subtask N with the parameters of the basic model trained in step 6, and performs parameter fine-tuning; Step 8. Construct a subtask N integrated small model: Based on the small models trained on all output branches of subtask N in steps 6 and 7, use the same small model structure as the subtask N integrated small model, and input the outputs of all output branches under subtask N into the subtask N integrated small model; Initialize the subtask N integrated model with the parameters of the basic model in step 6 and fine-tune the parameters; Step 9: Update subtask type N=N+1, repeat steps 5-8, and train the current subtask N integrated model; Step 10: Integrate the integrated model outputs of subtasks N and N-1, select the same network structure as the basic model in step 6, initialize the integrated small model of subtasks N and N-1 with the parameters of the basic model in step 6, and fine-tune the parameters of the integrated small model of subtasks N and N-1; Step 11: Repeat steps 9-10 until all subtasks are integrated.

2. The method for rapid forecasting of a large weather forecast model with low computing resources as claimed in claim 1, characterized in that: The step 1 comprises the following steps: S11, according to the subtask type, collect and determine the form of data to form an initial data set; S12, preprocessing the initial data set collected in S11, including screening, enhancement, denoising, and labeling, to form a preprocessed data set; S13, dividing the preprocessed data set in S12 into a weather forecast large model training set and a weather forecast large model test set.

3. The method for rapid weather forecasting using a large model with low computing resources as claimed in claim 1, characterized in that: The step 2 comprises the following steps: S21, analyzing the large model task and determining the output form of the large model; S22, split the large model output in S21 equally into multiple sub-outputs, and each sub-output is regarded as a subtask; S23, output subtask types, subtask 1, subtask 2, ..., subtask N; Decompose each subtask output and determine subtask output branch 1, subtask output branch 2, ..., subtask output branch Q of each subtask; S24, constructing a subtask training set. For subtasks 1-N, each subtask is randomly sampled from the large model training set determined in S13 in step 1 to form the training set of the subtask.

4. The method for rapid weather forecasting using a large model with low computing resources as claimed in claim 1, characterized in that: The step 3 comprises the following steps: S31, based on the subtasks output by S23 in step 2, the output of each subtask is equally split; S32, determining the output branch of each subtask after the splitting: subtask output branch 1, subtask output branch 2, ..., subtask output branch Q; S33, construct the training set corresponding to the output branch of each subtask after the split: according to the subtask output branches 1-Q in S32, each output branch is randomly sampled from the subtask training set determined in S24 in step 2, and the training set corresponding to the output branch of each subtask after the split is constructed: subtask training set 1, subtask training set 2,..., subtask training set Q.

5. The method for rapid weather forecasting using a large model with low computing resources as claimed in claim 1, characterized in that: The step 4 comprises the following steps: S41, determining the input and output data forms of the model for the subtask type determined in S22 in step 2; S42, constructing a selection set of small models, referring to the model input and output data forms determined in S41, and collecting small models with various input and output data forms.

6. The method for rapid weather forecasting using a large model with low computing resources as claimed in claim 1, characterized in that: The step 5 comprises the following steps: S51, select subtask type N, and initialize N to 1; Step 6 contains the following steps: S61, select subtask N output branch 1, and select a small model from the candidate small model set; S62, small model training configuration, select the optimizer of the small model, set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S63, determining a loss function of subtask N according to a specific task form of the selected subtask N; S64, according to S33 in step 3, select subtask N training set 1 corresponding to subtask N output branch 1; S65, input the training set in S64 into the small model of the output branch 1 of subtask N, and calculate the training loss using the loss function determined in S63; S66, executing a back propagation algorithm to obtain a parameter update gradient of a small model of output branch 1 of subtask N according to the training loss in S65; S67, using the optimizer configured in S62, updating the small model parameters of the output branch 1 of the subtask N until the loss calculated in S65 changes steadily, and saving the minimum loss in the training process and the small model corresponding to the output branch 1 of the subtask N; S68, repeating steps S61-S67, selecting a small model with the minimum loss, and using the small model as the basic model for subtask N; Step 7 includes the following steps: S71, select the output q branch of subtask N in sequence, q=2, ...Q; S72, selecting the basic model determined in S68 in step 6 as the fine-tuned small model of the output branch q of subtask N; S73, fine-tune the small model training configuration of the subtask N output branch q, select the small model optimizer of the subtask N output branch q, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S74, determining a loss function of the output branch q of the subtask N according to the task type of the output branch q of the subtask N; S75, according to S33 in step 3, select the subtask N training set q corresponding to the subtask N output branch q; S76, initializing the small model of the output branch q of subtask N in S72 with the basic model parameters determined in S68 in step 6; S77, input the training set in S75 into the small model of the output branch q of subtask N, and calculate the training loss using the loss function determined in S74; S78, executing the back propagation algorithm to obtain the parameter update gradient of the small model on the output branch q of subtask N according to the training loss in S77; S79, using the optimizer configured in S73, updating the small model parameters on the output branch q of subtask N until the loss calculated in S77 changes steadily, and saving the minimum loss in the training process and the corresponding small model under the minimum loss; S710, repeating steps S71-S79 until the training of the small models of all remaining output branches on subtask N is completed; Step 8 contains the following steps: S81, respectively obtain the base model of subtask N in S68 of step 6 and the small models of all remaining output branches of subtask N in S710 of step 7; S82, setting the integrated small model structure of subtask N as the basic model; S83, selecting all training data of subtask N as the training set of the integrated small model of subtask N in S82; S84, select the optimizer of the integrated small model of subtask N, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S85, selecting a subtask N integrated small model loss function according to the task type of the subtask N; S86, initializing the subtask N integrated small model in S82 with the basic model parameters determined in S68 in step 6; S87, input the training set in S83 into the output branch 1-Q of subtask N, input the output of the small model on the output branch 1-Q into the integrated small model of subtask N, and calculate the training loss using the loss function determined in S85; S88, according to the training loss in S87, execute the back propagation algorithm to obtain the parameter update gradient of the subtask N integrated small model; S89, using the optimizer configured in S84, updating the parameters of the subtask N integrated small model until the loss calculated in S87 changes steadily, and saving the minimum loss in the training process and the corresponding subtask N integrated small model; Step 9 contains the following steps: S91, update the type of subtask N, set N=N+1; S92, repeating steps 5-8 until the integrated module training of all subtasks is completed; Step 10 includes the following steps: S1001, obtaining an integrated module of subtasks N and N-1; S1002, select the integrated small model optimizer for subtasks N and N-1, and set the learning rate, weight decay coefficient, and adaptive moment estimation parameters of the optimizer; S1003, selecting an integrated small model loss function of subtasks N and N-1 according to the task types of subtasks N and N-1; S1004, initializing the integrated small model of subtasks N and N-1 in S1002 with the basic model parameters determined in S68 in step 6; S1005, input the training set in S1002 into the subtask N and N-1 integration module, input the output of the subtask N and N-1 integration module into the subtask N and N-1 integration small model, and calculate the training loss using the loss function determined in S1003; S1006, executing the back propagation algorithm to obtain the parameter update gradient of the integrated small model of subtasks N and N-1 according to the training loss in S1005; S1007, using the optimizer configured in S1002, update the integrated small model parameters of subtasks N and N-1 until the loss calculated in S1005 changes steadily, and save the minimum loss in the training process and the integrated small model corresponding to subtasks N and N-1; Step 11 includes the following steps: S1101, repeat step 9 to save the integration modules of all remaining subtasks; S1102, update subtask N type, N=N+1; S1103, repeat step 10 to complete the integrated small model training of subtasks N and N-1; S1104, repeat S1102-S1103 until the training of all subtask integrated small models is completed.

7. A system for rapid forecasting of a large meteorological forecast model with low computing resources, which implements the method for rapid forecasting of a large meteorological forecast model with low computing resources as claimed in any one of claims 1 to 6, characterized in that: The meteorological forecast large model rapid forecast system under low computing resources includes: Dataset construction module, used to construct a large weather forecast model dataset; A splitting module is used to split the large weather forecast model task into subtask 1, subtask 2, ..., subtask N, and determine the training sets of subtasks 1-N; A decomposition module is used to decompose each subtask output and determine subtask output branch 1, subtask output branch 2, ..., subtask output branch Q of each subtask; The candidate set building module is used to collect multiple small models that perform the same task according to the subtask type and build a small model candidate set; Initialization module, used to select subtask type N, and initialize N to 1; A training module is used to select a subtask N, output branch 1 for the subtask, train each small model in the small model candidate set, and determine the small model with the best performance as the basic model; A fine-tuning module is used to fine-tune the small models on the remaining output branches of subtask N: on the basic model, the remaining output branches of subtask N adopt the same small model structure as the basic model, initialize the small model parameters on the remaining output branches of subtask N with the trained basic model parameters, and perform parameter fine-tuning; construct a subtask N integrated small model: based on the small models trained on all output branches of subtask N, use the same small model structure as the subtask N integrated small model, input the outputs of all output branches under subtask N into the subtask N integrated small model; initialize the subtask N integrated small model with the basic model parameters, and perform parameter fine-tuning; Update module, used to update subtask type N=N+1 and train the current subtask N integration module; The integration module is used to integrate the integration module outputs of subtasks N and N-1, select the same network structure as the basic model, initialize the integrated small model of subtasks N and N-1 with the basic model parameters, and fine-tune the integrated small model parameters of subtasks N and N-1; until all subtasks are integrated.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for rapid forecasting of a large meteorological forecast model under low computing resources as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the steps of the method for rapid forecasting of a large meteorological forecast model under low computing resources as described in any one of claims 1-6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the large-scale rapid forecasting system for meteorological forecasting with low computing resources as described in claim 7.

Images