A data processing method, device and equipment applied to model training and a medium

Abstract

Embodiments of the present application disclose a data processing method and device applied to model training, equipment, and medium. A smooth signal sequence is extracted from original equipment data. In order to expand the equipment data, ensure that the expanded equipment data has difference with the original equipment data, and also conforms to the characteristics of the business scenario, the jump information of the original equipment data can be adjusted according to the set jump ratio to obtain jump feature data. The jump information can include jump frequency and / or jump amplitude. According to the jump feature data and the smooth signal sequence, enhanced equipment data can be obtained. The enhanced equipment data has difference with the original equipment data and conforms to the characteristics of the business scenario. The enhanced equipment data meets the high-quality demand of the data set, so that the initial equipment prediction model is trained by using the original equipment data and the enhanced equipment data, and a high-precision equipment prediction model can be obtained.

Description

Technical Field

[0001] This application relates to the field of model training technology, and in particular to a data processing method, apparatus, device, and computer-readable storage medium for model training. Background Technology

[0002] Equipment data is data collected at different times, reflecting the state or extent of change of a thing or phenomenon over time. Time series analysis is a statistical method for processing equipment data. Based on stochastic process theory and mathematical statistics, this method studies the statistical laws followed by equipment data in order to solve practical problems.

[0003] Device data can be used for model training. For example, to predict the changing trend of system resource usage, the resource usage of each device over a past period can be collected as a dataset. This dataset can then be used to train a model, enabling the trained model to analyze current resource usage and predict future trends. To improve the accuracy of model analysis, a sufficient dataset must be collected for training. This requires a large number of samples with sufficient diversity to cover the distribution characteristics of the scenario. However, in real-world environments, collected datasets are often insufficient in quantity, have limited data features, and lack strong coverage of business scenarios. Low-quality collected datasets result in poor predictive performance of device prediction models trained on such datasets.

[0004] It is evident that improving the accuracy of equipment prediction models is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a data processing method, apparatus, device, and computer-readable storage medium for model training, which can improve the accuracy of the device's predictive model.

[0006] To address the aforementioned technical problems, embodiments of this application provide a data processing method for model training, comprising:

[0007] Extract smoothed signal sequences from raw device data;

[0008] The transition information of the original device data is adjusted according to the set transition ratio to obtain transition characteristic data; wherein, the transition information includes transition frequency and / or transition amplitude;

[0009] Based on the jump characteristic data and the smoothed signal sequence, the enhanced device data is obtained;

[0010] The initial equipment prediction model is trained using the original equipment data and the enhanced equipment data to obtain an equipment prediction model that meets the accuracy requirements.

[0011] Optionally, adjusting the transition information of the original device data according to a set transition ratio to obtain transition feature data includes:

[0012] The switching frequency of the original device data is adjusted according to the set switching ratio to obtain the switching position sequence;

[0013] The amplitude of the original device data is adjusted according to the set amplitude jump ratio to obtain the jump amplitude sequence.

[0014] Optionally, adjusting the transition frequency of the original device data according to a set frequency transition ratio to obtain a transition position sequence includes:

[0015] Perform a difference operation on the original device data to obtain a difference sequence;

[0016] Based on the difference sequence and the set standard deviation principle, the initial jump position sequence of the original device data is determined;

[0017] Based on the frequency jump ratio and the jump frequency of the original device data, a new jump frequency is randomly generated;

[0018] The initial jump position sequence is adjusted using the new jump frequency to obtain the jump position sequence.

[0019] Optionally, before randomly generating a new switching frequency based on the switching frequency of the frequency switching ratio and the original device data, the method further includes:

[0020] Determine whether the intervals of the initial jump position sequence conform to an exponential distribution;

[0021] If the intervals of the initial jump position sequence conform to an exponential distribution, the step of randomly generating a new jump frequency based on the frequency jump ratio and the jump frequency of the original device data is performed.

[0022] Optionally, adjusting the jump amplitude of the original device data according to a set jump ratio to obtain a jump amplitude sequence includes:

[0023] The mean and variance of the original equipment data are used as initial normal distribution parameters;

[0024] The initial normal distribution parameters are adjusted according to the set amplitude jump ratio to obtain new initial normal distribution parameters;

[0025] Based on the new normal distribution parameters, the original equipment data is fitted with a normal distribution to generate random variables;

[0026] Select N random variables that satisfy the standard deviation principle as the jump amplitude sequence; wherein the value of N is the same as the number of jump positions contained in the initial jump position sequence.

[0027] Optionally, the transition feature data includes a transition position sequence and a transition amplitude sequence; obtaining the enhanced device data based on the transition feature data and the smoothed signal sequence includes:

[0028] The jump position sequence and the jump amplitude sequence are combined to obtain the jump sequence;

[0029] The jump sequence and the smooth signal sequence are superimposed to obtain the enhanced device data.

[0030] Optionally, the extraction of the smoothed signal sequence from the raw device data includes:

[0031] A smoothed signal sequence is obtained by performing a moving average operation on the original device data.

[0032] This application also provides a data processing apparatus for model training, including an extraction unit, an adjustment unit, an enhancement unit, and a training unit;

[0033] The extraction unit is used to extract a smoothed signal sequence from the raw device data;

[0034] The adjustment unit is used to adjust the jump information of the original device data according to a set jump ratio to obtain jump characteristic data; wherein, the jump information includes jump frequency and / or jump amplitude;

[0035] The enhancement unit is used to obtain enhanced device data based on the jump feature data and the smooth signal sequence;

[0036] The training unit is used to train the initial device prediction model using the original device data and the enhanced device data to obtain a device prediction model that meets the accuracy requirements.

[0037] Optionally, the adjustment unit includes a frequency adjustment subunit and an amplitude adjustment subunit;

[0038] The frequency adjustment subunit is used to adjust the switching frequency of the original device data according to a set frequency switching ratio to obtain a switching position sequence.

[0039] The amplitude adjustment subunit is used to adjust the jump amplitude of the original device data according to a set amplitude jump ratio to obtain a jump amplitude sequence.

[0040] Optionally, the frequency adjustment subunit is used to perform differential operations on the original device data to obtain a differential sequence; determine the initial jump position sequence of the original device data according to the differential sequence and the set standard deviation principle; randomly generate a new jump frequency based on the frequency jump ratio and the jump frequency of the original device data; and adjust the initial jump position sequence using the new jump frequency to obtain a jump position sequence.

[0041] Optionally, it also includes a judgment unit;

[0042] The judgment unit is used to determine whether the interval of the initial jump position sequence conforms to an exponential distribution; if the interval of the initial jump position sequence conforms to an exponential distribution, the frequency adjustment subunit is triggered to execute the step of randomly generating a new jump frequency based on the jump frequency of the frequency jump ratio and the original device data.

[0043] Optionally, the amplitude adjustment subunit is used to take the mean and variance of the original equipment data as initial normal distribution parameters; adjust the initial normal distribution parameters according to a set amplitude jump ratio to obtain new initial normal distribution parameters; perform normal distribution fitting on the original equipment data based on the new normal distribution parameters to generate random variables; select N random variables that satisfy the standard deviation principle as the jump amplitude sequence; wherein, the value of N is the same as the number of jump positions contained in the initial jump position sequence.

[0044] Optionally, the jump feature data includes a jump position sequence and a jump amplitude sequence; the enhancement unit includes a synthesis subunit and a superposition subunit;

[0045] The synthesis subunit is used to synthesize the jump position sequence and the jump amplitude sequence to obtain a jump sequence;

[0046] The superposition subunit is used to superimpose the transition sequence and the smoothed signal sequence to obtain enhanced device data.

[0047] Optionally, the extraction unit is used to perform a moving average operation on the original device data to obtain a smooth signal sequence.

[0048] This application also provides a data processing device for model training, including:

[0049] Memory, used to store computer programs;

[0050] A processor for executing the computer program to implement the steps of the data processing method applied to model training as described in any of the preceding methods.

[0051] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data processing method for model training as described in any of the above embodiments.

[0052] As can be seen from the above technical solution, a smoothed signal sequence is extracted from the original equipment data. This smoothed signal sequence can represent the overall trend of the original equipment data, excluding jump features. Jump information characterizes the degree of change in the original equipment data, and can include jump frequency and / or jump amplitude. To expand the equipment data, ensure that the expanded equipment data differs from the original equipment data, and also conforms to the characteristics of the business scenario, the jump information corresponding to the original equipment data can be appropriately adjusted to obtain new jump feature data. In this technical solution, the jump information of the original equipment data can be adjusted according to a set jump ratio to obtain jump feature data. Based on the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The initial equipment prediction model is trained using the original equipment data and the enhanced equipment data to obtain an equipment prediction model that meets the accuracy requirements. In this technical solution, adjusting the jump information of the original equipment data yields expanded jump feature data. By combining the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The enhanced equipment data differs from the original equipment data and conforms to the characteristics of the business scenario. The enhanced device data meets the requirements of high-quality datasets. Therefore, by training the initial device prediction model using both the original device data and the enhanced device data, a device prediction model with higher accuracy can be obtained. Attached Figure Description

[0053] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A schematic diagram of a data augmentation scenario provided for an embodiment of this application;

[0055] Figure 2 A flowchart illustrating a data processing method for model training provided in this application embodiment;

[0056] Figure 3This is a schematic diagram of the structure of a data processing device for model training provided in an embodiment of this application;

[0057] Figure 4 This is a schematic diagram of the structure of a data processing device for model training provided in an embodiment of this application. Detailed Implementation

[0058] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0059] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0060] In traditional methods, to enable models to predict results more accurately, a sufficient dataset must be collected for training. However, currently collected datasets often suffer from insufficient quantity, limited data features, and weak coverage of business scenarios, resulting in poor prediction performance of models trained on such datasets.

[0061] Therefore, embodiments of this application provide a data processing method, apparatus, device, and computer-readable storage medium for model training. A smoothed signal sequence is extracted from raw device data. See also... Figure 1 The diagram illustrates a data augmentation scenario. In practical applications, a smoothed signal sequence can be extracted from the original device data using a moving average method. This smoothed signal sequence represents the overall trend of the original device data, excluding jump features. Jump information characterizes the degree of change in the original device data, and can include jump frequency and / or jump amplitude. To augment the device data, ensuring the augmented data differs from the original data and conforms to the characteristics of the business scenario, the jump information of the original device data can be adjusted according to a set jump ratio to obtain jump feature data. The augmented device data is obtained by superimposing the smoothed signal sequence and the jump feature data. The augmented device data differs from the original data and conforms to the characteristics of the business scenario. The augmented device data meets the high-quality requirements of the dataset; therefore, training the initial device prediction model using both the original and augmented device data yields a device prediction model with high accuracy.

[0062] Next, we will describe in detail a data processing method for model training provided by an embodiment of this application. Figure 2 A flowchart of a data processing method for model training provided in this application embodiment, the method including:

[0063] S201: Extract a smoothed signal sequence from the raw device data.

[0064] Raw equipment data is time-series data that can be used to reflect the state or degree of change of a thing or phenomenon over time.

[0065] Smoothed signal sequences can be used to represent the overall trend of raw device data, excluding abrupt changes. In practice, a smoothed signal sequence can be obtained by performing a moving average operation on the raw device data.

[0066] Raw device data can take many forms, and the specific form of raw device data to be acquired can be determined based on the actual business scenario. For example, when it is necessary to predict the system resource usage, the raw device data can be the data on the system resource usage of each device within a preset time period; when it is necessary to classify the operating status of devices, the raw device data can be the corresponding operating status data of the devices within a preset time period.

[0067] In this embodiment of the application, the specific format of the raw device data is not limited, and can be determined according to the actual business scenario.

[0068] S202: Adjust the transition information of the original device data according to the set transition ratio to obtain transition characteristic data.

[0069] Transition information can characterize the degree of change in raw device data. Transition information can include transition frequency and / or transition amplitude.

[0070] The diversity of transition feature data can improve the coverage of the dataset for business scenarios and enhance the data quality of the dataset. Therefore, in this embodiment, transition feature data can be enhanced to expand the dataset.

[0071] To ensure that the obtained transition feature data reflects the characteristics of the business scenario, it is necessary to ensure that the obtained transition feature data differs from the original transition feature data, but without significant deviation. Therefore, in this embodiment, a transition ratio can be set for the transition information of the original device data. The transition ratios corresponding to different types of transition information can be the same or different, and this is not limited here.

[0072] S203: Obtain the enhanced device data based on the jump characteristic data and the smoothed signal sequence.

[0073] In this embodiment of the application, the transition feature data and the smooth signal sequence can be superimposed to obtain enhanced device data.

[0074] S204: Train the initial equipment prediction model using the original equipment data and the enhanced equipment data to obtain an equipment prediction model that meets the accuracy requirements.

[0075] After training the initial device prediction model using the original device data and the enhanced device data, when new device data is received subsequently, the trained device prediction model can be directly used to analyze the new device data, thereby obtaining the analysis results.

[0076] The type of raw device data determines the corresponding device prediction model. For example, when the raw device data represents the system resource usage of each device within a preset time period, the corresponding device prediction model can be a resource usage prediction model. When the raw device data represents the operating status data of devices within a preset time period, the corresponding device prediction model can be a device status prediction model. Device status can include normal load status, overload status, and idle status.

[0077] As can be seen from the above technical solution, a smoothed signal sequence is extracted from the original equipment data. This smoothed signal sequence can represent the overall trend of the original equipment data, excluding jump features. Jump information characterizes the degree of change in the original equipment data, and can include jump frequency and / or jump amplitude. To expand the equipment data, ensure that the expanded equipment data differs from the original equipment data, and also conforms to the characteristics of the business scenario, the jump information corresponding to the original equipment data can be appropriately adjusted to obtain new jump feature data. In this technical solution, the jump information of the original equipment data can be adjusted according to a set jump ratio to obtain jump feature data. Based on the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The initial equipment prediction model is trained using the original equipment data and the enhanced equipment data to obtain an equipment prediction model that meets the accuracy requirements. In this technical solution, adjusting the jump information of the original equipment data yields expanded jump feature data. By combining the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The enhanced equipment data differs from the original equipment data and conforms to the characteristics of the business scenario. The enhanced device data meets the requirements of high-quality datasets. Therefore, by training the initial device prediction model using both the original device data and the enhanced device data, a device prediction model with higher accuracy can be obtained.

[0078] The following section will use transition information, including transition frequency and transition amplitude, as an example to elaborate on the topic.

[0079] Taking the adjustment of the switching frequency as an example, in practical applications, the switching frequency of the original device data can be adjusted according to the set frequency switching ratio to obtain the switching position sequence; the switching amplitude of the original device data can be adjusted according to the set amplitude switching ratio to obtain the switching amplitude sequence.

[0080] The transition frequency can be determined based on the location where the transition occurs. In a practical implementation, differential operations can be performed on the raw device data to obtain a differential sequence.

[0081] The first-order difference of the original equipment data yields the difference data, which is then summarized as the difference sequence. The difference data refers to the change data between two adjacent time periods. The change Δ = y(x+1) - y(x) between consecutive adjacent terms in the discrete function constructed from the original equipment data is taken as the first-order difference of the function y(x) at point x.

[0082] Based on the difference sequence and the established standard deviation principle, the initial jump position sequence of the original device data is determined.

[0083] In this embodiment, the initial jump position sequence can be determined according to the principle of three times the standard deviation. In a specific implementation, the distribution of the first-order difference can be statistically analyzed to obtain its mean μ and standard deviation σ. According to the principle of three times the standard deviation, the jump point position and jump amplitude of the original device data can be identified based on the first-order difference value belonging to the interval (-∞, μ-3*σ] or [μ-3*σ, +∞).

[0084] Based on the frequency jump ratio and the jump frequency of the original equipment data, a new jump frequency is randomly generated; the initial jump position sequence is adjusted using the new jump frequency to obtain the jump position sequence.

[0085] The initial jump position sequence often conforms to a Poisson distribution, and the parameter λ of the Poisson distribution can be used to characterize the jump frequency. Therefore, in this embodiment, the parameter λ of the Poisson distribution can be scaled according to a set frequency jump ratio. By scaling λ, the jump intensity of the initial jump position sequence is controlled, thus generating a new jump frequency. By adjusting the initial jump position sequence using the new jump frequency, a new jump position sequence can be obtained.

[0086] After determining the location of the transition point, the difference between two adjacent transition point locations can be calculated to obtain the transition point location interval sequence.

[0087] If the sequence of transition point intervals conforms to an exponential distribution, then the initial transition point sequence truly conforms to a Poisson distribution. Therefore, in this embodiment, to verify that the initial transition point sequence truly conforms to a Poisson distribution, it can be determined whether the intervals of the initial transition point sequence conform to an exponential distribution. If the intervals of the initial transition point sequence conform to an exponential distribution, then the initial transition point sequence conforms to a Poisson distribution. In this case, the step of randomly generating a new transition frequency based on the frequency transition ratio and the transition frequency of the original device data can be performed.

[0088] Taking the adjustment of jump amplitude as an example, in practical applications, the mean and variance of the original equipment data can be used as the initial normal distribution parameters; the initial normal distribution parameters are adjusted according to the set amplitude jump ratio to obtain new initial normal distribution parameters; based on the new normal distribution parameters, the original equipment data is fitted with a normal distribution to generate random variables; N random variables that satisfy the standard deviation principle are selected as the jump amplitude sequence; where the value of N is the same as the number of jump positions contained in the initial jump position sequence.

[0089] In practical implementation, in order to ensure that the new initial normal distribution parameters are different from the original initial normal distribution parameters and that the deviation is not too large, the amplitude jump ratio can be set to a smaller range. For example, the amplitude jump ratio can be set to [0.9, 1.1].

[0090] In practical applications, N random variables can be selected from the random variables as the jump amplitude sequence according to the principle of three standard deviations.

[0091] In this embodiment, the number of data points contained in the generated jump position sequence and jump amplitude sequence are corresponding, so the jump position sequence and jump amplitude sequence can be directly synthesized to obtain the jump sequence.

[0092] After obtaining the transition sequence, the transition sequence and the smoothed signal sequence can be superimposed to obtain the enhanced device data.

[0093] As can be seen from the above technical solutions, the jump frequency and jump amplitude of the original equipment data can characterize the degree of change in the original equipment data; the smoothing signal sequence can be used to represent the overall trend of the original equipment data excluding jump characteristics. To expand the equipment data, the expanded equipment data should differ from the original equipment data and also conform to the characteristics of the business scenario. Therefore, the jump frequency and jump amplitude corresponding to the original equipment data can be appropriately adjusted to obtain new equipment data. In this technical solution, the jump frequency of the original equipment data can be adjusted according to a set frequency jump ratio to obtain a jump position sequence; the jump amplitude of the original equipment data can be adjusted according to a set amplitude jump ratio to obtain a jump amplitude sequence. Based on the jump position sequence, jump amplitude sequence, and smoothing signal sequence, the enhanced equipment data can be obtained. In this technical solution, the jump characteristics of the original equipment data are analyzed. By combining the jump position sequence, jump amplitude sequence, and smoothing signal sequence, enhanced equipment data can be obtained. The enhanced equipment data differs from the original equipment data and conforms to the characteristics of the business scenario. The enhanced equipment data meets the high-quality requirements of the dataset.

[0094] Figure 3 A schematic diagram of a data processing device for model training provided in an embodiment of this application includes an extraction unit 31, an adjustment unit 32, an enhancement unit 33, and a training unit 34;

[0095] Extraction unit 31 is used to extract a smoothed signal sequence from the raw device data;

[0096] The adjustment unit 32 is used to adjust the jump information of the original equipment data according to the set jump ratio to obtain jump characteristic data; wherein, the jump information includes jump frequency and / or jump amplitude;

[0097] Enhancement unit 33 is used to obtain enhanced device data based on transition characteristic data and smoothed signal sequence;

[0098] Training unit 34 is used to train the initial device prediction model using the original device data and the enhanced device data to obtain a device prediction model that meets the accuracy requirements.

[0099] Optionally, the adjustment unit includes a frequency adjustment subunit and an amplitude adjustment subunit;

[0100] The frequency adjustment subunit is used to adjust the switching frequency of the original device data according to the set frequency switching ratio to obtain the switching position sequence;

[0101] The amplitude adjustment subunit is used to adjust the jump amplitude of the original device data according to the set amplitude jump ratio to obtain the jump amplitude sequence.

[0102] Optionally, the frequency adjustment subunit is used to perform differential operations on the original equipment data to obtain a differential sequence; determine the initial jump position sequence of the original equipment data according to the differential sequence and the set standard deviation principle; randomly generate a new jump frequency based on the frequency jump ratio and the jump frequency of the original equipment data; and adjust the initial jump position sequence using the new jump frequency to obtain the jump position sequence.

[0103] Optionally, it also includes a judgment unit;

[0104] The judgment unit is used to determine whether the interval of the initial jump position sequence conforms to an exponential distribution; if the interval of the initial jump position sequence conforms to an exponential distribution, the frequency adjustment subunit is triggered to execute the step of randomly generating a new jump frequency based on the frequency jump ratio and the original device data.

[0105] Optionally, the amplitude adjustment subunit is used to take the mean and variance of the original equipment data as initial normal distribution parameters; adjust the initial normal distribution parameters according to the set amplitude jump ratio to obtain new initial normal distribution parameters; based on the new normal distribution parameters, fit the original equipment data to a normal distribution to generate random variables; select N random variables that satisfy the standard deviation principle as the jump amplitude sequence; wherein, the value of N is the same as the number of jump positions contained in the initial jump position sequence.

[0106] Optionally, the jump feature data includes a jump position sequence and a jump amplitude sequence; the enhancement unit includes a synthesis subunit and a superposition subunit;

[0107] Synthesis subunit, used to synthesize jump position sequence and jump amplitude sequence to obtain jump sequence;

[0108] The superposition subunit is used to superimpose the transition sequence and the smoothed signal sequence to obtain enhanced device data.

[0109] Optionally, the extraction unit is used to perform a moving average operation on the original device data to obtain a smooth signal sequence.

[0110] Figure 3 The description of the features in the corresponding embodiments can be found in [reference needed]. Figure 2 The relevant descriptions of the corresponding embodiments will not be repeated here.

[0111] As can be seen from the above technical solution, a smoothed signal sequence is extracted from the original equipment data. This smoothed signal sequence can represent the overall trend of the original equipment data, excluding jump features. Jump information characterizes the degree of change in the original equipment data, and can include jump frequency and / or jump amplitude. To expand the equipment data, ensure that the expanded equipment data differs from the original equipment data, and also conforms to the characteristics of the business scenario, the jump information corresponding to the original equipment data can be appropriately adjusted to obtain new jump feature data. In this technical solution, the jump information of the original equipment data can be adjusted according to a set jump ratio to obtain jump feature data. Based on the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The initial equipment prediction model is trained using the original equipment data and the enhanced equipment data to obtain an equipment prediction model that meets the accuracy requirements. In this technical solution, adjusting the jump information of the original equipment data yields expanded jump feature data. By combining the jump feature data and the smoothed signal sequence, enhanced equipment data can be obtained. The enhanced equipment data differs from the original equipment data and conforms to the characteristics of the business scenario. The enhanced device data meets the requirements of high-quality datasets. Therefore, by training the initial device prediction model using both the original device data and the enhanced device data, a device prediction model with higher accuracy can be obtained.

[0112] Figure 4 A schematic diagram of the structure of a data processing device 40 for model training provided in this application embodiment includes:

[0113] Memory 41 is used to store computer programs;

[0114] Processor 42 is used to execute computer programs to implement the steps of any of the above-described data processing methods applied to model training.

[0115] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described data processing methods applied to model training.

[0116] The foregoing has provided a detailed description of a data processing method, apparatus, device, and computer-readable storage medium for model training, as provided in the embodiments of this application. The various embodiments are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0117] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0118] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

Claims

1. A data processing method applied to model training, characterized in that, The method comprises the following steps: extracting a smooth signal sequence from original equipment data; the smooth signal sequence is used to represent the overall trend of the original equipment data excluding the jump feature; adjusting the jump information of the original equipment data according to a set jump proportion to obtain jump feature data; wherein the jump information comprises a jump frequency and / or a jump amplitude; the jump feature data reflects the characteristics of a service scenario; obtaining enhanced equipment data according to the jump feature data and the smooth signal sequence; training an initial equipment prediction model using the original equipment data and the enhanced equipment data to obtain an equipment prediction model meeting the accuracy requirement. 2.The data processing method for model training of claim 1, wherein, The step of adjusting the jump information of the original equipment data according to a set jump proportion to obtain jump feature data comprises: adjusting the jump frequency of the original equipment data according to a set frequency jump proportion to obtain a jump position sequence; adjusting the jump amplitude of the original equipment data according to a set amplitude jump proportion to obtain a jump amplitude sequence. 3.The data processing method for model training of claim 2, wherein, The step of adjusting the jump frequency of the original equipment data according to a set frequency jump proportion to obtain a jump position sequence comprises: performing a difference operation on the original equipment data to obtain a difference sequence; determining an initial jump position sequence of the original equipment data according to the difference sequence and a set standard deviation principle; randomly generating a new jump frequency based on the frequency jump proportion and the jump frequency of the original equipment data; adjusting the initial jump position sequence using the new jump frequency to obtain a jump position sequence. 4.The data processing method for model training of claim 3, wherein, Before the step of randomly generating a new jump frequency based on the frequency jump proportion and the jump frequency of the original equipment data, the method further comprises: judging whether the interval of the initial jump position sequence conforms to an exponential distribution; in the case that the interval of the initial jump position sequence conforms to an exponential distribution, performing the step of randomly generating a new jump frequency based on the frequency jump proportion and the jump frequency of the original equipment data. 5.The data processing method for model training of claim 2, wherein, The step of adjusting the jump amplitude of the original equipment data according to a set amplitude jump proportion to obtain a jump amplitude sequence comprises: taking the mean and variance of the original equipment data as initial normal distribution parameters; adjusting the initial normal distribution parameters according to a set amplitude jump proportion to obtain new initial normal distribution parameters; performing normal distribution fitting on the original equipment data based on the new normal distribution parameters to generate random variables; selecting N random variables meeting a standard deviation principle as a jump amplitude sequence; wherein the value of N is the same as the number of jump positions contained in the initial jump position sequence. 6.The data processing method for model training of claim 1, wherein, The jump feature data comprises a jump position sequence and a jump amplitude sequence; and the step of obtaining enhanced equipment data according to the jump feature data and the smooth signal sequence comprises: combining the jump position sequence and the jump amplitude sequence to obtain a jump sequence; superimposing the jump sequence and the smooth signal sequence to obtain enhanced equipment data.

7. The data processing method for model training according to any one of claims 1 to 6, characterized in that, The step of extracting a smooth signal sequence from original equipment data comprises: performing a moving average operation on the original equipment data to obtain a smooth signal sequence.

8. A data processing apparatus applied to model training, characterized in that, The method comprises an extracting unit, an adjusting unit, an enhancing unit and a training unit. The extracting unit is configured to extract a smooth signal sequence from the original equipment data, wherein the smooth signal sequence is used to represent the overall trend of the original equipment data excluding the jump feature; The adjusting unit is configured to adjust jump information of the original equipment data according to a set jump ratio to obtain jump feature data, wherein the jump information comprises a jump frequency and / or a jump amplitude, and the jump feature data reflects the characteristics of the service scenario; The enhancing unit is configured to obtain enhanced equipment data according to the jump feature data and the smooth signal sequence; The training unit is configured to train an initial equipment prediction model by using the original equipment data and the enhanced equipment data to obtain an equipment prediction model meeting the accuracy requirement. 9.A data processing device applied to model training, characterized in that, The method comprises: a memory configured to store a computer program; a processor configured to execute the computer program to implement the steps of the data processing method for model training according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored on the computer readable storage medium and is executed by the processor to implement the steps of the data processing method for model training according to any one of claims 1 to 7.

Images