Optimization method and device for time series data processing and electronic equipment

Abstract

The invention relates to the technical field of time series analysis, and provides an optimization method and device for time series data processing and electronic equipment, and the optimization method for time series data processing comprises the steps: obtaining a first time series data subset; an iterative search step of the following transformation strategy is executed through a pre-selected optimization algorithm until a preset iteration termination condition is reached, a target transformation strategy for the time series data is obtained, and the transformation strategy comprises an operator and a corresponding hyper-parameter: the hyper-parameter of the transformation strategy is adjusted based on an evaluation result of last iteration; applying the adjusted hyper-parameter of the transformation strategy to the first time sequence data subset to obtain a transformed data set; and evaluating the zero sample prediction performance of the large time sequence model on the transformed data set to obtain an evaluation result. According to the method, the optimal transformation strategy can be automatically searched and determined in a short time, the requirement of manual intervention is reduced, and the efficiency of a time sequence analysis system can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of time series analysis, and in particular to an optimization method, device and electronic equipment for time series data processing. Background Art

[0002] Time series analysis is an important branch of data analysis, which focuses on modeling and predicting data that changes over time. Time series big models (such as Timer, Moiria, Chronos) have demonstrated excellent capabilities in processing complex time series data. These models can effectively capture long-term dependencies and complex patterns in the data, especially in zero-shot prediction tasks. By training on large-scale and diverse data sets, these big models can still provide high-precision prediction results without explicit annotations or a small number of samples. Therefore, time series big models have broad application potential in many fields such as financial forecasting, energy management, and traffic flow forecasting, and provide strong support for efficient decision-making.

[0003] In order for these advanced models to perform at their best, high-quality input data is essential, which requires the selection of appropriate data transformation strategies to process the data. However, the existing techniques for manually selecting appropriate transformation strategies have significant deficiencies in time series data preprocessing. First, manually selecting and adjusting preprocessing strategies is very time-consuming, especially when faced with large-scale data sets. Developers need to repeatedly experiment with different parameter combinations, which not only consumes a lot of time and computing resources, but also in practical applications, data sets may contain millions or even billions of data points, making manual adjustment of preprocessing strategies almost an impossible task. Second, the lack of a standardized method to guide the entire optimization process leads to poor consistency and repeatability of the results. Different developers may adopt different strategies, making it difficult to compare and verify the results. The lack of a systematic approach also means that you need to start from scratch every time, and you cannot effectively use previous experience and knowledge. In addition, these methods are highly dependent on the developer's experience and intuition, and it is difficult to guarantee that the optimal configuration is found every time. Even experienced developers may miss the best solution due to personal bias or knowledge limitations. For novices or team members, this method has a high learning curve. Once the data characteristics change, the original preprocessing strategy may no longer be applicable and needs to be readjusted. This means that every time the data is updated or changed, the preprocessing strategy needs to be readjusted, increasing maintenance costs. Data changes may be caused by external factors, such as seasonal changes, emergencies, etc. These changes require the preprocessing strategy to be highly flexible and adaptive.

[0004] In addition, replacing a model may require redesigning the data processing process. Especially when using large models, data quality is more sensitive to model performance. High-quality data can significantly improve the accuracy and robustness of the model, while low-quality data may lead to poor model performance. Therefore, as the model is continuously iterated and improved, the preprocessing steps also need to be updated to ensure that the data can meet the needs of the new model. Such frequent adjustments not only increase the workload, but may also introduce new errors and inconsistencies, further increasing the complexity and risk of the project.

[0005] In summary, existing manual or semi-automatic preprocessing methods have obvious shortcomings in efficiency, consistency and flexibility, and more automated and intelligent solutions are urgently needed to meet these challenges. Summary of the invention

[0006] The present invention provides an optimization method, device and electronic device for time series data processing, which are used to solve the defect of low efficiency in manually selecting appropriate transformation strategies in the prior art. The method can automatically search and determine the best transformation strategy in a short time, reduce the need for manual intervention, help improve development efficiency, speed up project progress, and ultimately improve the performance of the overall system.

[0007] The present invention provides an optimization method for time series data processing, comprising: obtaining a first time series data subset; executing the following iterative search steps of a transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, and obtaining a target transformation strategy for the time series data, wherein the transformation strategy includes an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on an evaluation result of a previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; and evaluating the zero-sample prediction performance of a large time series model on the transformed data set to obtain an evaluation result.

[0008] According to the optimization method for time series data processing provided by the present invention, obtaining the first time series data subset includes: sampling the time series data subset to be enhanced from the original time series data set; applying the time series data enhancement method to perform data enhancement on the time series data subset to be enhanced, so as to obtain the first time series data subset.

[0009] According to the optimization method for time series data processing provided by the present invention, the transformation strategy includes context transformation, normalization transformation and outlier transformation. The context transformation constructs the reasoning context, the normalization transformation adjusts the range of input and output values, and the outlier transformation is used to reduce the impact of outliers on model performance.

[0010] According to the optimization method for time series data processing provided by the present invention, the optimization algorithm includes a Bayesian optimization algorithm.

[0011] According to the optimization method for time series data processing provided by the present invention, the evaluation indicators in the evaluation results include at least one of the following: mean square error, mean absolute error, root mean square error, mean absolute percentage error, mean square percentage error and statistics of mean square error, mean absolute error, root mean square error, mean absolute percentage error and mean square percentage error, and the statistics include at least one of the following: mean, median, standard deviation and interquartile range, and the iterative search step of the following transformation strategy is performed by a pre-selected optimization algorithm until a pre-set iteration termination condition is reached to obtain a target transformation strategy for time series data, including: determining the target transformation strategy according to the Pareto ranking of the evaluation indicators.

[0012] According to the optimization method for time series data processing provided by the present invention, the target transformation strategy is determined according to the Pareto ranking of the evaluation indicators, including: based on the evaluation results obtained in the iterative search step, obtaining the hyperparameters of each group of transformation strategies that rank in the top preset proportion on any preset indicator combination; sampling a second time series data subset from the original time series data set; obtaining the weighted ranking results of the evaluation indicators of the hyperparameters of each group of transformation strategies on the second time series data subset; and determining the target transformation strategy according to the weighted ranking results.

[0013] The present invention also provides an optimization device for time series data processing, comprising: an acquisition module, configured to acquire a first time series data subset; an optimization module, configured to execute the iterative search steps of the following transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, thereby obtaining a target transformation strategy for the time series data, wherein the transformation strategy includes an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on the evaluation results of the previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; and evaluating the zero-sample prediction performance of a large time series model on the transformed data set to obtain an evaluation result.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the optimization method for time series data processing as described in any one of the above is implemented.

[0015] 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 any of the above-described optimization methods for time series data processing.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the optimization method for processing time series data as described above is implemented.

[0017] The optimization method, device and electronic device for time series data processing provided by the present invention execute the iterative search steps of the transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, thereby obtaining a target transformation strategy for the time series data. The method, device and electronic device can automatically search and determine the best transformation strategy in a short time, reducing the need for manual intervention and helping to improve the efficiency of the time series analysis system. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0019] Figure 1 It is a flow chart of the optimization method for time series data processing provided by the present invention.

[0020] Figure 2 It is a schematic diagram of the optimization process of time series data processing provided by the present invention.

[0021] Figure 3 It is a structural schematic diagram of the optimization device for time series data processing provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one", "one" or "the" do not indicate a quantitative limitation, but indicate the existence of at least one. Words such as "include" or "comprise" mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0025] The following is a brief explanation of the terms involved in the present invention.

[0026] Combine the following Figure 1-Figure 4 The invention describes an optimization method, device and electronic device for time series data processing.

[0027] Figure 1 is a flow chart of the optimization method for time series data processing provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 101: Obtain a first time series data subset.

[0028] In this embodiment, the first time series data subset can be obtained by sampling the original time series data set, and in order to generate diversified data samples, it can also be enhanced by a data enhancement method. In addition, the first time series data subset is used to optimize the transformation strategy, and another part of the data can be extracted from the original time series data set as the second time series data subset to verify the effect of the optimized transformation strategy.

[0029] Step 102: Execute the following iterative search steps of the transformation strategy through the pre-selected optimization algorithm until the pre-set iteration termination condition is reached to obtain the target transformation strategy for the time series data: adjust the hyperparameters of the transformation strategy based on the evaluation result of the previous iteration; apply the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain the transformed data set; evaluate the zero-sample prediction performance of the large time series model on the transformed data set to obtain the evaluation result.

[0030] In this embodiment, the transformation strategy includes an operator and corresponding hyperparameters. The transformation strategy can be divided into multiple categories, such as context transformation, normalization transformation, and outlier transformation. Each type of transformation strategy can include multiple specific operators, and each operator has corresponding hyperparameters. The iteration termination condition is set according to the specific optimization algorithm, and may include reaching a preset number of iterations or convergence conditions. In addition, before adjusting the hyperparameters of the transformation strategy for the first time, an initialization step can be performed to initialize the parameters of the transformation strategy and obtain an evaluation result.

[0031] The optimization algorithm can use Bayesian optimization algorithm, hyperparameter optimization method based on grid search or random search. Grid search finds the optimal configuration by exhaustively enumerating all possible hyperparameter combinations, while random search explores the hyperparameter space through random sampling. Although these two methods are simple and direct, they may be very time-consuming when faced with high-dimensional and complex hyperparameter spaces. Another alternative is to use genetic algorithms or other evolutionary algorithms for hyperparameter optimization. These methods simulate natural selection and genetic mechanisms to gradually iterate the optimal hyperparameter combination. Although genetic algorithms have strong global search capabilities, their convergence speed and computing resource requirements may be high.

[0032] In addition, reinforcement learning-based methods can also be considered to dynamically adjust the transformation strategy. By training an agent to learn how to select the best preprocessing steps based on the current data characteristics, this method can adaptively adjust the strategy to cope with data changes. However, reinforcement learning methods require a large amount of training data and computing resources, and the design and tuning of the model are relatively complex.

[0033] The optimization method for time series data processing provided by the present invention automatically optimizes the preprocessing strategy of time series data by executing iterative search steps through a pre-selected optimization algorithm. Compared with the traditional method of manually selecting and adjusting the preprocessing strategy, it can adapt to the characteristics of different data sets by systematically and standardizedly selecting and adjusting the optimal data preprocessing configuration, thereby improving the zero-sample prediction performance of large time series models (LTSMs).

[0034] In some optional implementations, obtaining a first time series data subset includes: sampling a time series data subset to be enhanced from an original time series data set; applying a time series data enhancement method to enhance the time series data subset to be enhanced to obtain a first time series data subset. In order to enhance data diversity and reduce potential data offset, a variety of time series data enhancement methods can be used, including but not limited to amplitude flipping, time flipping, amplitude distortion, time distortion, noise injection and translation transformation. These enhancement methods can generate diverse data samples while retaining the intrinsic characteristics of the original data. Generating rich and diverse training samples through diverse data enhancement methods can effectively reduce the impact of data offset on model performance and enhance the generalization ability of the model. In addition, generative adversarial networks (GANs) can also be used to generate more diverse time series data. GANs can generate new samples similar to the original data distribution through adversarial training of generators and discriminators, thereby increasing data diversity and reducing the impact of data offset.

[0035] In some optional implementations, the transformation strategies include context transformation, normalization transformation, and outlier transformation. Context transformation builds an inference context, such as using a sliding window to determine the length of an input sequence. Normalization transformation adjusts the range of input and output values. Outlier transformation is used to mitigate the impact of outliers on model performance.

[0036] In addition, normalization transformations can include minimum-maximum normalization, Z-score standardization, robust standardization, P-norm normalization, and decimal scaling, which improve the stability and convergence speed of model training by converting data to a uniform range. For the processing of missing data, commonly used methods include reserved value interpolation, mean or median interpolation, piecewise linear interpolation, piecewise cubic spline interpolation, hot and cold deck interpolation, expectation maximization algorithm, support vector machine interpolation, and multiple interpolation. In addition, for outlier detection, commonly used techniques include median absolute deviation (MAD), Grubbs test, generalized extreme studentized deviation (GESD), interquartile range (IQR), local outlier factor (LOF), density-based spatial clustering with applied noise (DBSCAN), and isolation forest (iForest), which help identify and correct outliers in the data to avoid their negative impact on the model.

[0037] In some optional implementations, the optimization algorithm includes a Bayesian optimization algorithm. Advanced Bayesian optimization algorithms, such as the Tree-structured Parzen Estimator (TPE), can be used to search for the optimal transformation configuration. In each iteration or trial, a set of transformation configurations is selected and applied to the input data, and the predictive performance of the model is evaluated. After multiple iterations, the optimal transformation strategy can be determined based on the Pareto ranking or other ranking results of multiple indicators.

[0038] In some optional implementations, the evaluation indicators in the evaluation results may include at least one of the following: mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), (MAPE) and mean square percentage error (MSPE) and statistics of mean square error, mean absolute error, root mean square error, mean absolute percentage error and mean square percentage error, the statistics including at least one of the following: mean, median, standard deviation (STD) and interquartile range (IQR), and the following iterative search steps of the transformation strategy are performed by a pre-selected optimization algorithm until a pre-set iteration termination condition is reached to obtain a target transformation strategy for time series data, including: determining the target transformation strategy according to the Pareto ranking of the evaluation indicators.

[0039] In some optional implementations, a target transformation strategy is determined based on the Pareto ordering of evaluation indicators, including: based on the evaluation results obtained in the iterative search step, obtaining hyperparameters of each group of transformation strategies that rank in the top preset proportion on any preset indicator combination; sampling a second time series data subset from the original time series data set; obtaining a weighted ranking result of the evaluation indicators of the hyperparameters of each group of transformation strategies on the second time series data subset; and determining the target transformation strategy based on the weighted ranking result.

[0040] This implementation method determines the optimal target transformation strategy through the Pareto sorting method of multiple indicators (such as MSE, MAE, RMSE, MAPE and MSPE), ensuring that the selected strategy performs well on multiple evaluation indicators, further enhancing the robustness of the model. In addition, the two-stage Pareto sorting method not only selects the experiments that rank in the top preset proportion, such as 30%, on any combination of indicators, but also performs the final sorting based on the weighted sum of all indicators in the second stage, ensuring the comprehensiveness and reliability of the selected strategy.

[0041] See also Figure 2 , Figure 2It is a schematic diagram of the optimization process of time series data processing provided by the present invention, which is divided into two main stages: data preparation and time series conversion optimization. The data preparation stage starts with the collection of time series samples, and then a variety of data enhancement methods are applied, including amplitude deformation (such as smoothing, window sliding), time deformation (such as translation, jitter, spikes), slope adjustment (such as flipping), etc., to generate an enhanced data set. By comparing the window paired samples and the true values, the effect of the enhanced data can be evaluated. The transformation strategies included in the time series conversion optimization stage involve context transformations, such as trimmers, samplers, and aligners; abnormal transformations, such as denoisers, interpolators, and clippers; normalization transformations, such as scalers, warpers, and differencers. The settings of the transformation configuration can include pruning, such as fixed blocks, 4x blocks, and 8x blocks; denoising, such as None and exponentially weighted moving average (EWMA); and scaling, such as Min-max and Standard. Through black-box optimization, Bayesian optimization algorithms such as TPE are used to search for the optimal transformation configuration. In zero-sample prediction and evaluation, the optimized transformation strategy is used to perform zero-sample prediction on the time series model. The prediction results are post-processed and compared with the true values, and evaluation indicators such as MSE, MAE, RMSE and MAPE are used to measure the model performance. A two-stage Pareto sorting method is used to determine the optimal configuration.

[0042] The present invention searches for the optimal time series data preprocessing configuration in an automated manner, thereby improving the prediction performance of the model and reducing the workload of developers. Specifically, an advanced optimization algorithm is used to automatically search and determine the best combination of preprocessing parameters, including context window size, normalization method and anomaly detection strategy, so that the optimal configuration can be found in a short time, greatly improving work efficiency. The present invention provides a standardized method to ensure that each optimization process follows the same process, improving the consistency and repeatability of the results. At the same time, it can dynamically adapt to changes in data characteristics, automatically adjust the preprocessing strategy, and ensure that the model is always in the best state. In addition, the present invention can be easily integrated into existing workflows, supports multiple programming languages ​​and frameworks, and is easy for developers to use. This method not only improves the adaptability and accuracy of the model, but also simplifies the data preparation process, allowing developers to focus more on the research and development and improvement of core algorithms. By reducing the need for manual intervention, the present invention helps to improve development efficiency, accelerate the progress of the project, and ultimately improve the performance of the overall system.

[0043] The following is a description of the time series data processing optimization device provided by the present invention. The time series data processing optimization device described below and the time series data processing optimization method described above can be referenced to each other.

[0044] Figure 4 A schematic diagram of the structure of the optimization device for time series data processing provided in the embodiment of the present application, such as Figure 4 As shown, it specifically includes: an acquisition module 401, configured to acquire a first time series data subset; an optimization module 402, configured to execute the iterative search steps of the following transformation strategy through a pre-selected optimization algorithm, until a pre-set iteration termination condition is reached, and a target transformation strategy for the time series data is obtained, the transformation strategy includes an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on the evaluation result of the previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; evaluating the zero-sample prediction performance of the large time series model on the transformed data set to obtain an evaluation result.

[0045] In some optional implementations, the acquisition module 401 is further configured to: sample a time series data subset to be enhanced from the original time series data set; and perform data enhancement on the time series data subset to be enhanced using a time series data enhancement method to obtain a first time series data subset.

[0046] In some optional implementations, the transformation strategy includes context transformation, normalization transformation and outlier transformation. The context transformation constructs the reasoning context, the normalization transformation adjusts the range of input and output values, and the outlier transformation is used to reduce the impact of outliers on model performance.

[0047] In some optional implementations, the optimization algorithm includes a Bayesian optimization algorithm.

[0048] In some optional implementations, the evaluation indicators in the evaluation results include at least one of the following: mean square error, mean absolute error, root mean square error, mean absolute percentage error, mean square percentage error, and statistics of the mean square error, mean absolute error, root mean square error, mean absolute percentage error, and mean square percentage error, and the statistics include at least one of the following: mean, median, standard deviation, and interquartile range, and the optimization module 402 is further configured to determine the target transformation strategy according to the Pareto ranking of the evaluation indicators.

[0049] In some optional implementations, the optimization module 402 is further configured to: obtain hyperparameters of each group of transformation strategies that rank in the top preset proportion on any preset indicator combination based on the evaluation results obtained in the iterative search step; sample a second time series data subset from the original time series data set; obtain weighted ranking results of the evaluation indicators of the hyperparameters of each group of transformation strategies on the second time series data subset; and determine the target transformation strategy based on the weighted ranking results.

[0050] The optimization device for time series data processing provided by the present invention executes the iterative search steps of the transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, thereby obtaining a target transformation strategy for the time series data. It can automatically search and determine the best transformation strategy in a short time, reducing the need for manual intervention and helping to improve the efficiency of the time series analysis system.

[0051] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the optimization method for time series data processing, the method comprising: obtaining a first time series data subset; performing the following iterative search steps of the transformation strategy by a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, and obtaining a target transformation strategy for the time series data, the transformation strategy including an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on the evaluation result of the previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; evaluating the zero-sample prediction performance of the large time series model on the transformed data set to obtain an evaluation result.

[0052] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0053] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the optimization method for time series data processing provided by the above methods, the method including: obtaining a first subset of time series data; executing the following iterative search steps of the transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, and obtaining a target transformation strategy for the time series data, the transformation strategy including an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on the evaluation results of the previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; evaluating the zero-sample prediction performance of the large time series model on the transformed data set to obtain an evaluation result.

[0054] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the optimization method for time series data processing provided by the above-mentioned methods, the method comprising: obtaining a first subset of time series data; executing the following iterative search steps of the transformation strategy through a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, to obtain a target transformation strategy for the time series data, the transformation strategy comprising an operator and corresponding hyperparameters: adjusting the hyperparameters of the transformation strategy based on the evaluation results of the previous iteration; applying the adjusted hyperparameters of the transformation strategy to the first subset of time series data to obtain a transformed data set; evaluating the zero-sample prediction performance of the large time series model on the transformed data set to obtain an evaluation result.

[0055] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0056] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optimization method for time series data processing, characterized in that: include: Get the first time series data subset; The following transformation strategy iterative search steps are performed by a pre-selected optimization algorithm until a pre-set iteration termination condition is reached, and a target transformation strategy for time series data is obtained, wherein the transformation strategy includes an operator and corresponding hyperparameters: Adjust the hyperparameters of the transformation strategy based on the evaluation results of the previous iteration; Applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; Evaluate the zero-shot forecasting performance of large time series models on the transformed dataset and obtain the evaluation results.

2. The optimization method for time series data processing according to claim 1, characterized in that: The obtaining of the first time series data subset includes: Sample a subset of the time series data to be enhanced from the original time series dataset; A time series data enhancement method is applied to perform data enhancement on the time series data subset to be enhanced to obtain the first time series data subset.

3. The optimization method for time series data processing according to claim 1, characterized in that: The transformation strategy includes context transformation, normalization transformation and outlier transformation. The context transformation constructs the reasoning context, the normalization transformation adjusts the range of input and output values, and the outlier transformation is used to reduce the impact of outliers on model performance.

4. The optimization method for time series data processing according to claim 1, characterized in that: The optimization algorithm includes a Bayesian optimization algorithm.

5. The optimization method for time series data processing according to claim 1, characterized in that: The evaluation index in the evaluation result includes at least one of the following: mean square error, mean absolute error, root mean square error, mean absolute percentage error, mean square percentage error and statistics of mean square error, mean absolute error, root mean square error, mean absolute percentage error and mean square percentage error, and the statistics include at least one of the following: mean, median, standard deviation and interquartile range, and the iterative search step of the following transformation strategy is performed by the pre-selected optimization algorithm until a pre-set iteration termination condition is reached to obtain a target transformation strategy for time series data, including: The target transformation strategy is determined according to the Pareto ranking of the evaluation indicators.

6. The optimization method for time series data processing according to claim 5, characterized in that: The step of determining the target transformation strategy according to the Pareto ranking of the evaluation indicators includes: Based on the evaluation results obtained in the iterative search step, obtaining the hyperparameters of each group of transformation strategies that rank in the top preset proportion on any preset indicator combination; Sampling a second time series data subset from the original time series data set; Obtaining weighted ranking results of evaluation indicators of the hyperparameters of each group of transformation strategies on the second time series data subset; A target change strategy is determined according to the weighted ranking result.

7. An optimization device for time series data processing, characterized in that: include: An acquisition module, configured to acquire a first time series data subset; The optimization module is configured to perform the following iterative search steps of the transformation strategy by a pre-selected optimization algorithm until a pre-set iteration termination condition is reached to obtain a target transformation strategy for the time series data, wherein the transformation strategy includes an operator and corresponding hyperparameters: Adjust the hyperparameters of the transformation strategy based on the evaluation results of the previous iteration; Applying the adjusted hyperparameters of the transformation strategy to the first time series data subset to obtain a transformed data set; Evaluate the zero-shot forecasting performance of large time series models on the transformed dataset and obtain the evaluation results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the optimization method for time series data processing according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the optimization method for time series data processing according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the optimization method for time series data processing according to any one of claims 1 to 6 is implemented.

Images