Cloud data center load prediction method and system based on improved timesnet

Abstract

The application provides a cloud data center load prediction method and system based on improved TimesNet, multi-period feature extraction is performed on time series of resource utilization data, a plurality of most significant periods are determined, two-dimensional time series change modeling is performed, change data of the time series on different time scales is obtained, features of the most significant periods are extracted, and the features are adaptively fused into TimesBlock based on ResNet; a plurality of TimesBlock are combined and stacked to form an improved TimesNet network, so that the resource demand load of the cloud platform in the future time is predicted, the plurality of periods and fluctuation of the load can be more accurately captured, the accuracy of long-term prediction is improved, ResNet is used for two-dimensional convolution, the rapid change of the cloud data center load can be more quickly adapted, the resource scheduling lag or redundancy is reduced through accurate load prediction, the resource utilization efficiency is improved, the operation cost caused by default and waste is reduced, and the load prediction has better adaptability and generalization ability.

Description

Technical Field

[0001] This invention relates to the field of cloud computing, and more particularly to a cloud data center load prediction method and system based on an improved TimesNet. Background Technology

[0002] In cloud computing environments, cloud vendors widely use time-series load forecasting technology to predict resource changes over a period of time based on historical CPU or memory resource utilization. This allows for proactive elastic scaling, improved resource utilization efficiency, and reduced business bottlenecks. Traditional load forecasting techniques primarily rely on classic time series analysis methods, such as ARIMA and Seasonal Time Series Forecasting (STL). These methods perform well in handling simple periodic and linear trends, but often struggle to accurately predict the high volatility and complex patterns of cloud data center loads. Furthermore, while deep learning methods like LSTM and Informer offer higher accuracy, they lack the capability to process multi-periodic cloud load data and incur significant resource overhead. Therefore, existing load forecasting methods suffer from insufficient waveform processing, limited prediction accuracy, and low computational efficiency, making it impossible to accurately predict loads with minimal overhead when loads exhibit non-linear and multi-periodic characteristics. Summary of the Invention

[0003] The purpose of this invention is to provide a cloud data center load prediction method and system based on an improved TimesNet. This method extracts multi-period features from the time series of resource utilization data, identifies the most significant periods, and uses these to model two-dimensional time series changes, obtaining the changes in the time series at different time scales. It then extracts the features of each of the most significant periods and adaptively fuses these features into a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network, thereby predicting the resource demand load of the cloud platform in the future. This method can more accurately capture the multi-periodicity and volatility of the load, improving the accuracy of long-term predictions. Using ResNet for two-dimensional convolution allows for faster adaptation to rapid changes in cloud data center load. Accurate load prediction reduces resource scheduling lag or redundancy, improves resource utilization efficiency, and lowers operating costs caused by default and waste, giving load prediction better adaptability and generalization capabilities.

[0004] This invention is achieved through the following technical solution:

[0005] A cloud data center load prediction method based on improved TimesNet includes:

[0006] An improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data to identify the most significant periods in the time series. Based on the most significant periods, a two-dimensional time series variation model is performed to obtain the variation data of the time series at different time scales.

[0007] The time series data at different time scales are processed to extract the features of the most significant periods, and the features of the most significant periods are adaptively fused to obtain a ResNet-based TimesBlock.

[0008] Multiple TimesBlocks are combined and stacked to form an improved TimesNet network; then, the improved TimesNet network is used to perform load prediction on the resource utilization data of the cloud platform to obtain the resource demand load of the cloud platform in the future.

[0009] Optionally, an improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data to determine the most significant periods in the time series; based on the most significant periods, two-dimensional time series variation modeling is performed to obtain the variation data of the time series at different time scales, including:

[0010] Historical resource utilization data of the cloud platform is acquired. Based on the generation time of the historical resource utilization data, the historical resource utilization data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. The historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods in the time series are determined.

[0011] Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

[0012] Optionally, the time series variation data at different time scales are processed to extract features from the most significant periods, and the features from the most significant periods are adaptively fused to obtain a ResNet-based TimesBlock, including:

[0013] The time series data at different time scales is processed by two-dimensional convolution using ResNet to extract the features of the most significant periods.

[0014] Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

[0015] Optionally, an operational anomaly determination is performed on the process that identifies the most significant periods in the time series, including:

[0016] Extract the duration of the most significant periods in the time series;

[0017] The period acquisition duration is compared with a preset duration threshold;

[0018] When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration.

[0019] By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series.

[0020] The process evaluation coefficient is obtained using the following formula:

[0021] Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula:

[0022] Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively; 01i and T 02iT represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T zmin This represents the minimum duration obtained from a period determined n times from a set of periods.

[0023] The process evaluation coefficient is compared with the preset evaluation coefficient;

[0024] When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

[0025] Optionally, multiple TimesBlocks are combined and stacked to form an improved TimesNet network; then, the improved TimesNet network is used to perform load prediction on the resource utilization data of the cloud platform to obtain the resource demand load of the cloud platform in the future, including:

[0026] Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network.

[0027] The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different length periods, so as to obtain the resource demand load of the cloud platform for different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

[0028] A cloud data center load prediction system based on an improved TimesNet includes:

[0029] The multi-period feature extraction module is used to extract multi-period features from the time series of cloud platform resource utilization data using an improved TimesNet model, and to determine the most significant periods in the time series.

[0030] The two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods to obtain the variation data of the time series at different time scales;

[0031] A two-dimensional convolution processing module is used to process the change data of the time series at different time scales and extract the features of the most significant periods.

[0032] An adaptive fusion module is used to adaptively fuse the features of the most significant several periods to obtain a ResNet-based TimesBlock.

[0033] The combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network;

[0034] The load prediction module is used to perform load prediction on the resource utilization data of the cloud platform using the improved TimesNet network, so as to obtain the resource demand load of the cloud platform in the future.

[0035] Optionally, the multi-period feature extraction module is used to perform multi-period feature extraction on the time series of cloud platform resource utilization data using an improved TimesNet model, and to determine the most significant periods in the time series, including:

[0036] Historical resource utilization data of the cloud platform is acquired. Based on the generation time of the historical resource utilization data, the historical resource utilization data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. The historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods in the time series are determined.

[0037] The two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods, to obtain the variation data of the time series at different time scales, including:

[0038] Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

[0039] Optionally, an operational anomaly determination is performed on the process that identifies the most significant periods in the time series, including:

[0040] Extract the duration of the most significant periods in the time series;

[0041] The period acquisition duration is compared with a preset duration threshold;

[0042] When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration.

[0043] By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series.

[0044] The process evaluation coefficient is obtained using the following formula:

[0045] Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula:

[0046] Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively; 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T zmin This represents the minimum duration obtained from a period determined n times from a set of periods.

[0047] The process evaluation coefficient is compared with the preset evaluation coefficient;

[0048] When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

[0049] Optionally, the two-dimensional convolution processing module is used to process the change data of the time series at different time scales, and extract the features of the most significant periods, including:

[0050] The time series data at different time scales is processed by two-dimensional convolution using ResNet to extract the features of the most significant periods.

[0051] The adaptive fusion module is used to adaptively fuse the features of the most significant several periods to obtain a ResNet-based TimesBlock, including:

[0052] Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

[0053] Optionally, the combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network, including:

[0054] Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network.

[0055] The load forecasting module is used to perform load forecasting on the resource utilization data of the cloud platform using the improved TimesNet network, to obtain the resource demand load of the cloud platform in the future, including:

[0056] The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different length periods, so as to obtain the resource demand load of the cloud platform for different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The cloud data center load prediction method and system based on improved TimesNet provided in this application extracts multi-period features from the time series of resource utilization data, identifies the most significant periods, and uses these to perform two-dimensional time series variation modeling to obtain the variation data of the time series at different time scales. Features from the most significant periods are then extracted and adaptively fused into a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network, thereby predicting the resource demand load of the cloud platform in the future. This method can more accurately capture the multi-periodicity and volatility of the load, improving the accuracy of long-term predictions. Using ResNet for two-dimensional convolution allows for faster adaptation to rapid changes in cloud data center load. Accurate load prediction reduces resource scheduling lag or redundancy, improves resource utilization efficiency, and lowers operating costs caused by default and waste, giving load prediction better adaptability and generalization capabilities. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0060] Figure 1 This is a flowchart illustrating the cloud data center load prediction method based on the improved TimesNet provided by the present invention.

[0061] Figure 2 This is a schematic diagram of the structure of the cloud data center load prediction system based on the improved TimesNet provided by the present invention. Detailed Implementation

[0062] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, it should be noted that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, not the entire structure. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0063] The terms “comprising” and “having”, and any variations thereof, used in this application are intended to cover non-exclusive inclusion. For example, a process, method, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0064] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0065] Please see Figure 1 As shown, an embodiment of this application provides a cloud data center load prediction method based on an improved TimesNet. This cloud data center load prediction method based on an improved TimesNet includes:

[0066] An improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data to identify the most significant periods in the time series. Based on these most significant periods, a two-dimensional time series variation model is performed to obtain the variation data of the time series at different time scales.

[0067] The time series data at different time scales is processed to extract the features of the most significant periods, and the features of the most significant periods are adaptively fused to obtain the ResNet-based TimesBlock.

[0068] Multiple TimesBlocks are combined and stacked to form an improved TimesNet network; then, this improved TimesNet network is used to predict the resource utilization data of the cloud platform and obtain the resource demand load of the cloud platform in the future.

[0069] The beneficial effects of the above embodiments are as follows: This cloud data center load prediction method based on improved TimesNet extracts multi-period features from the time series of resource utilization data, identifies the most significant periods, and uses these to perform two-dimensional time series change modeling to obtain the change data of the time series at different time scales. It then extracts the features of each of the most significant periods and adaptively fuses these features into a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network, thereby predicting the resource demand load of the cloud platform in the future. This method can more accurately capture the multi-periodicity and volatility of the load, improving the accuracy of long-term predictions. Using ResNet for two-dimensional convolution allows for faster adaptation to rapid changes in cloud data center load. Accurate load prediction reduces resource scheduling lag or redundancy, improves resource utilization efficiency, reduces operating costs caused by default and waste, and gives load prediction better adaptability and generalization capabilities.

[0070] In another embodiment, an improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data to determine the most significant periods in the time series. Based on these most significant periods, a two-dimensional time series variation model is performed to obtain the variation data of the time series at different time scales, including:

[0071] Historical resource utilization data of the cloud platform is acquired. Based on the generation time of this historical resource utilization data, the data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. This historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods of the time series are determined.

[0072] Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

[0073] The beneficial effects of the above embodiments are that cloud platforms need to call different types of resources such as CPU and memory during the processing of different tasks. The number and speed of these resource calls directly affect the reliability of task processing on the cloud platform. However, it is not always better for the cloud platform to call more and faster CPU and memory resources during task processing. This depends on the actual amount of resources required during task processing. Calling resources too much or too fast will lead to resource redundancy and waste. Therefore, it is necessary to accurately predict the amount of resources that the cloud platform will call and utilize in the future based on the resource utilization data of the cloud platform in its historical operation. In order to ensure the matching between the historical resource utilization data of the cloud platform and the time series prediction, it is necessary to process the historical resource utilization data in advance. Based on the generation time of the historical resource utilization data, the historical resource utilization data is segmented according to the time sequence to obtain the corresponding time series of the cloud platform's resource utilization data, which facilitates subsequent direct processing of the time series. Furthermore, an improved TimesNet model is used to perform a Fast Fourier Transform (FFT) on the time series to obtain its frequency component intensities. These intensities are then used as multi-period features of the time series. Obtaining the frequency component intensities through FFT is a standard technique in this field and will not be described in detail here. The frequency component intensities characterize the importance of data within the corresponding periods of the time series; a higher frequency component intensity indicates greater importance of the data within the corresponding period. In this case, the periods corresponding to the k largest frequency component intensities are considered the most significant periods of the time series. This allows for the capture of the multi-periodicity of time series data, enabling multi-periodic feature extraction and allowing subsequent load forecasting to adapt to the high volatility and multi-periodicity of different scenarios. Then, based on the k most significant periods, the time series is divided according to the length of each period. (where i = 1, 2, ..., k) are folded to transform the one-dimensional time series into a series with a width of ... The two-dimensional tensor is used to capture the temporal variation data of the periodic components within the period using a two-dimensional convolutional network. Through the above transformation, the variation patterns of the time series at different time scales can be analyzed and learned simultaneously.

[0074] In another embodiment, the process of determining the most significant periods in the time series includes:

[0075] Extract the duration of the most significant periods in the time series;

[0076] The period acquisition duration is compared with a preset duration threshold;

[0077] When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration.

[0078] By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series.

[0079] The process evaluation coefficient is obtained using the following formula:

[0080] Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula:

[0081] Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively;01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T zmin This represents the minimum duration obtained from a period determined n times from a set of periods.

[0082] The process evaluation coefficient is compared with the preset evaluation coefficient;

[0083] When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

[0084] The beneficial effects of the above embodiments are that the technical solution can monitor the determination process of several most significant periods in a time series in real time. By extracting and comparing the period acquisition duration with a preset duration threshold, it can promptly detect potential anomalies. When the period acquisition duration exceeds the preset threshold, the solution can retrieve relevant data from historical records for auxiliary judgment. This includes the processing duration of historical resource utilization data and the acquisition duration of frequency component intensity, which provides rich reference information for subsequent anomaly judgment. By introducing a process evaluation coefficient S, the technical solution can quantitatively evaluate whether there are anomalies in the process of determining several most significant periods in a time series. The calculation of the process evaluation coefficient S considers the determination duration of multiple periods in historical records, as well as the weights of the first coefficient S01 and the second coefficient S02, making the evaluation more comprehensive and accurate. The calculation method of the first coefficient S01 and the second coefficient S02 considers the maximum and minimum values ​​of the period acquisition duration, which allows the process evaluation coefficient S to be dynamically adjusted according to different historical data, enhancing the adaptability and flexibility of the technical solution. When the process evaluation coefficient is not lower than the preset evaluation coefficient, the technical solution can determine that there is an anomaly in the process of determining several most significant periods in a time series and issue an anomaly alarm. This helps to identify problems in a timely manner and take appropriate countermeasures, reducing potential risks and losses.

[0085] In summary, this technical solution effectively improves the accuracy and efficiency of anomaly detection in identifying the most significant periods in a time series by employing real-time monitoring, historical data-assisted judgment, quantitative evaluation of anomaly severity, dynamic adjustment and adaptability, and timely alarms and responses. This is of great significance for ensuring the stable operation of the system and responding promptly to potential problems.

[0086] In another embodiment, the time series data at different time scales is processed to extract features from the most significant periods, and the features from these most significant periods are adaptively fused to obtain a ResNet-based TimesBlock, including:

[0087] By performing two-dimensional convolution on the time series data at different time scales using ResNet, the features of the most significant periods are extracted.

[0088] Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

[0089] The beneficial effects of the above embodiments are that by using ResNet to perform two-dimensional convolution processing on the changing data of the time series at different time scales, the powerful feature extraction capabilities and convergence speed of ResNet can be leveraged to improve the prediction accuracy and training efficiency of the model. Furthermore, based on the frequency component intensities of all periods within the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock; where the greater the frequency component intensity of a certain period, the greater its weighting proportion in the weighted fusion process. By employing an adaptive fusion strategy of weighted fusion, features from different periods are weighted and fused, enabling subsequent load prediction for different period lengths, thus exhibiting better adaptability and generalization ability.

[0090] In another embodiment, multiple TimesBlocks are combined and stacked to form an improved TimesNet network; this improved TimesNet network is then used to perform load prediction on the resource utilization data of the cloud platform to obtain the resource demand load of the cloud platform in the future, including:

[0091] Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network.

[0092] The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different time periods, so as to obtain the resource demand load of the cloud platform for different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

[0093] The beneficial effects of the above embodiments are that each TimesBlock can perform load forecasting within a corresponding period. To accurately predict the multi-periodic and volatile nature of load demand, multiple TimesBlocks are analyzed to determine the residuals of all TimesBlocks. These residuals are then linked to form an improved TimesNet network. This improved TimesNet network can achieve globally accurate forecasting of loads with multi-periodic and volatile characteristics. Furthermore, the resource utilization data of the cloud platform is input into this improved TimesNet network for load forecasting over different time periods, obtaining the resource demand load of the cloud platform for different time intervals in the future. This effectively reduces scheduling lag and redundancy in load forecasting, enabling faster adaptation to rapid changes in cloud data center load and improving the accuracy of long-term load forecasting.

[0094] Please see Figure 2 As shown, one embodiment of this application provides a cloud data center load prediction system based on an improved TimesNet. This cloud data center load prediction system based on an improved TimesNet includes:

[0095] The multi-period feature extraction module is used to extract multi-period features from the time series of cloud platform resource utilization data using an improved TimesNet model, and to determine the most significant periods in the time series.

[0096] The two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods, and obtain the variation data of the time series at different time scales;

[0097] The two-dimensional convolution processing module is used to process the variation data of the time series at different time scales and extract the features of the most significant periods.

[0098] The adaptive fusion module is used to adaptively fuse the features of the most significant periods to obtain a ResNet-based TimesBlock.

[0099] The combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network;

[0100] The load forecasting module is used to perform load forecasting on the cloud platform's resource utilization data using the improved TimesNet network, thereby obtaining the cloud platform's resource demand load in the future.

[0101] The beneficial effects of the above embodiments are as follows: This cloud data center load prediction system based on improved TimesNet extracts multi-period features from the time series of resource utilization data, identifies the most significant periods, and uses these to perform two-dimensional time series change modeling to obtain the change data of the time series at different time scales. It then extracts the features of each of the most significant periods and adaptively fuses these features into a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network, thereby predicting the resource demand load of the cloud platform in the future. This system can more accurately capture the multi-periodicity and volatility of the load, improving the accuracy of long-term predictions. Using ResNet for two-dimensional convolution allows for faster adaptation to rapid changes in cloud data center load. Accurate load prediction reduces resource scheduling lag or redundancy, improves resource utilization efficiency, and lowers operating costs caused by default and waste, giving load prediction better adaptability and generalization capabilities.

[0102] In another embodiment, the multi-period feature extraction module is used to perform multi-period feature extraction on the time series corresponding to the resource utilization data of the cloud platform using an improved TimesNet model, and to determine the most significant periods in the time series, including:

[0103] Historical resource utilization data of the cloud platform is acquired. Based on the generation time of this historical resource utilization data, the data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. This historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods of the time series are determined.

[0104] This two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods, obtaining the variation data of the time series at different time scales, including:

[0105] Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

[0106] The beneficial effects of the above embodiments are that cloud platforms need to call different types of resources such as CPU and memory during the processing of different tasks. The number and speed of these resource calls directly affect the reliability of task processing on the cloud platform. However, it is not always better for the cloud platform to call more and faster CPU and memory resources during task processing. This depends on the actual amount of resources required during task processing. Calling resources too much or too fast will lead to resource redundancy and waste. Therefore, it is necessary to accurately predict the amount of resources that the cloud platform will call and utilize in the future based on the resource utilization data of the cloud platform in its historical operation. In order to ensure the matching between the historical resource utilization data of the cloud platform and the time series prediction, it is necessary to process the historical resource utilization data in advance. Based on the generation time of the historical resource utilization data, the historical resource utilization data is segmented according to the time sequence to obtain the corresponding time series of the cloud platform's resource utilization data, which facilitates subsequent direct processing of the time series. Furthermore, an improved TimesNet model is used to perform a Fast Fourier Transform (FFT) on the time series to obtain its frequency component intensities. These intensities are then used as multi-period features of the time series. Obtaining the frequency component intensities through FFT is a standard technique in this field and will not be described in detail here. The frequency component intensities characterize the importance of data within the corresponding periods of the time series; a higher frequency component intensity indicates greater importance of the data within the corresponding period. In this case, the periods corresponding to the k largest frequency component intensities are considered the most significant periods of the time series. This allows for the capture of the multi-periodicity of time series data, enabling multi-periodic feature extraction and allowing subsequent load forecasting to adapt to the high volatility and multi-periodicity of different scenarios. Then, based on the k most significant periods, the time series is divided according to the length of each period. (where i = 1, 2, ..., k) are folded to transform the one-dimensional time series into a series with a width of ... The two-dimensional tensor is used to capture the temporal variation data of the periodic components within the period using a two-dimensional convolutional network. Through the above transformation, the variation patterns of the time series at different time scales can be analyzed and learned simultaneously.

[0107] In another embodiment, the process of determining the most significant periods in the time series includes:

[0108] Extract the duration of the most significant periods in the time series;

[0109] The period acquisition duration is compared with a preset duration threshold;

[0110] When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration.

[0111] By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series.

[0112] The process evaluation coefficient is obtained using the following formula:

[0113] Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula:

[0114] Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively; 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T zmin This represents the minimum duration obtained from a period determined n times from a set of periods.

[0115] The process evaluation coefficient is compared with the preset evaluation coefficient;

[0116] When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

[0117] The beneficial effects of the above embodiments are that the technical solution can monitor the determination process of several most significant periods in a time series in real time. By extracting and comparing the period acquisition duration with a preset duration threshold, it can promptly detect potential anomalies. When the period acquisition duration exceeds the preset threshold, the solution can retrieve relevant data from historical records for auxiliary judgment. This includes the processing time of historical resource utilization data and the acquisition duration of frequency component intensity, which provides rich reference information for subsequent anomaly judgment. By introducing a process evaluation coefficient S, the technical solution can quantitatively evaluate whether there are anomalies in the process of determining several most significant periods in a time series. The calculation of the process evaluation coefficient S considers the determination duration of multiple periods in historical records, as well as the weights of the first coefficient S01 and the second coefficient S02, making the evaluation more comprehensive and accurate. The calculation method of the first coefficient S01 and the second coefficient S02 considers the maximum and minimum values ​​of the period acquisition duration, which allows the process evaluation coefficient S to be dynamically adjusted according to different historical data, enhancing the adaptability and flexibility of the technical solution. When the process evaluation coefficient is not lower than the preset evaluation coefficient, the technical solution can determine that there is an anomaly in the process of determining several most significant periods in a time series and issue an anomaly alarm. This helps to identify problems in a timely manner and take appropriate countermeasures, reducing potential risks and losses.

[0118] In summary, this technical solution effectively improves the accuracy and efficiency of anomaly detection in identifying the most significant periods in a time series by employing real-time monitoring, historical data-assisted judgment, quantitative evaluation of anomaly severity, dynamic adjustment and adaptability, and timely alarms and responses. This is of great significance for ensuring the stable operation of the system and responding promptly to potential problems.

[0119] In another embodiment, the two-dimensional convolution processing module is used to process the variation data of the time series at different time scales, and extract the features of the most significant periods, including:

[0120] By performing two-dimensional convolution on the time series data at different time scales using ResNet, the features of the most significant periods are extracted.

[0121] This adaptive fusion module is used to adaptively fuse features from the most salient periods to obtain a ResNet-based TimesBlock, including:

[0122] Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

[0123] The beneficial effects of the above embodiments are that by using ResNet to perform two-dimensional convolution processing on the changing data of the time series at different time scales, the powerful feature extraction capabilities and convergence speed of ResNet can be leveraged to improve the prediction accuracy and training efficiency of the model. Furthermore, based on the frequency component intensities of all periods within the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock; where the greater the frequency component intensity of a certain period, the greater its weighting proportion in the weighted fusion process. By employing an adaptive fusion strategy of weighted fusion, features from different periods are weighted and fused, enabling subsequent load prediction for different period lengths, thus exhibiting better adaptability and generalization ability.

[0124] In another embodiment, the combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network, including:

[0125] Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network.

[0126] This load forecasting module utilizes the improved TimesNet network to perform load forecasting on the cloud platform's resource utilization data, obtaining the cloud platform's resource demand load in the future, including:

[0127] The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different time periods, so as to obtain the resource demand load of the cloud platform in different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

[0128] The beneficial effects of the above embodiments are that each TimesBlock can perform load forecasting within a corresponding period. To accurately predict the multi-periodic and volatile nature of load demand, multiple TimesBlocks are analyzed to determine the residuals of all TimesBlocks. These residuals are then linked to form an improved TimesNet network. This improved TimesNet network can achieve globally accurate forecasting of loads with multi-periodic and volatile characteristics. Furthermore, the resource utilization data of the cloud platform is input into this improved TimesNet network for load forecasting over different time periods, obtaining the resource demand load of the cloud platform for different time intervals in the future. This effectively reduces scheduling lag and redundancy in load forecasting, enabling faster adaptation to rapid changes in cloud data center load and improving the accuracy of long-term load forecasting.

[0129] In summary, this cloud data center load forecasting method and system based on improved TimesNet extracts multi-period features from the time series of resource utilization data, identifies the most significant periods, and uses these to model two-dimensional time series changes, obtaining the changes in the time series at different time scales. It then extracts the features of each of the most significant periods and adaptively fuses these features into a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network, thereby predicting the resource demand load of the cloud platform in the future. This method can more accurately capture the multi-periodicity and volatility of the load, improving the accuracy of long-term forecasts. Using ResNet for two-dimensional convolution allows for faster adaptation to rapid changes in cloud data center load. Accurate load forecasting reduces resource scheduling lag or redundancy, improves resource utilization efficiency, and lowers operating costs caused by default and waste, giving load forecasting better adaptability and generalization capabilities.

[0130] The above is only one specific embodiment of the present invention, and any improvements made based on the concept of the present invention shall be considered within the scope of protection of the present invention.

Claims

1. A cloud data center load prediction method based on an improved TimesNet, characterized in that, include: An improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data to identify the most significant periods in the time series. Based on the most significant periods, a two-dimensional time series variation model is performed to obtain the variation data of the time series at different time scales. The time series data at different time scales are processed to extract the features of the most significant periods, and the features of the most significant periods are adaptively fused to obtain a ResNet-based TimesBlock. Multiple TimesBlocks are combined and stacked to form an improved TimesNet network; then, the improved TimesNet network is used to perform load prediction on the resource utilization data of the cloud platform to obtain the resource demand load of the cloud platform in the future. The process of determining the most significant periods in the time series includes the following: Extract the duration of the most significant periods in the time series; The period acquisition duration is compared with a preset duration threshold; When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration. By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series. The process evaluation coefficient is obtained using the following formula: Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula: Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively; zmin This represents the minimum duration obtained from a period determined n times from a set of periods. The process evaluation coefficient is compared with the preset evaluation coefficient; When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

2. The cloud data center load prediction method based on improved TimesNet as described in claim 1, characterized in that: An improved TimesNet model is used to extract multi-period features from the time series of cloud platform resource utilization data, identifying the most significant periods in the time series. Based on these most significant periods, a two-dimensional time series variation model is performed to obtain the variation data of the time series at different time scales, including: Historical resource utilization data of the cloud platform is acquired. Based on the generation time of the historical resource utilization data, the historical resource utilization data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. The historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods in the time series are determined. Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

3. The cloud data center load prediction method based on improved TimesNet as described in claim 1, characterized in that: The time series data at different time scales is processed to extract features from the most significant periods. These features are then adaptively fused to obtain a ResNet-based TimesBlock, including: The time series data at different time scales is processed by two-dimensional convolution using ResNet to extract the features of the most significant periods. Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

4. The cloud data center load prediction method based on improved TimesNet as described in claim 1, characterized in that: Multiple TimesBlocks are combined and stacked to form an improved TimesNet network; then, the improved TimesNet network is used to perform load prediction on the resource utilization data of the cloud platform to obtain the resource demand load of the cloud platform in the future, including: Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network. The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different length periods, so as to obtain the resource demand load of the cloud platform for different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

5. A cloud data center load prediction system based on an improved TimesNet, characterized in that: The multi-period feature extraction module is used to extract multi-period features from the time series of cloud platform resource utilization data using an improved TimesNet model, and to determine the most significant periods in the time series. The two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods to obtain the variation data of the time series at different time scales; A two-dimensional convolution processing module is used to process the change data of the time series at different time scales and extract the features of the most significant periods. An adaptive fusion module is used to adaptively fuse the features of the most significant several periods to obtain a ResNet-based TimesBlock. The combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network; The load prediction module is used to perform load prediction on the resource utilization data of the cloud platform using the improved TimesNet network, so as to obtain the resource demand load of the cloud platform in the future. The process of determining the most significant periods in the time series includes the following: Extract the duration of the most significant periods in the time series; The period acquisition duration is compared with a preset duration threshold; When the period acquisition duration exceeds the preset duration threshold, the processing duration and frequency component intensity of historical resource utilization data in the historical record are retrieved to acquire the duration. By utilizing historical resources in the historical records, the processing time and frequency component intensity of the data are used to determine the process evaluation coefficients of the most significant periods in the time series. The process evaluation coefficient is obtained using the following formula: Where S represents the process evaluation coefficient; n represents the number of times the most significant periods of the time series have been determined in the historical records, excluding the current determination; T 01i and T 02i T represents the processing time and frequency component intensity acquisition time of historical resource utilization data corresponding to the determination of several cycles in the i-th time, respectively; zmax This represents the maximum duration obtained from a given number of cycles, determined n times; T z This indicates the periodic acquisition duration exceeding the preset duration threshold; S 01 and S 02 Let represent the first coefficient and the second coefficient, respectively, and the first coefficient and the second coefficient are obtained by the following formula: Among them, S 01 and S 02 T represents the first coefficient and the second coefficient, respectively; zmin This represents the minimum duration obtained from a period determined n times from a set of periods. The process evaluation coefficient is compared with the preset evaluation coefficient; When the process evaluation coefficient is not lower than the preset evaluation coefficient, it is determined that the process of the most significant period in the time series is abnormal, and an abnormality alarm is triggered.

6. The cloud data center load prediction system based on improved TimesNet as described in claim 5, characterized in that: The multi-period feature extraction module is used to extract multi-period features from the time series of resource utilization data of the cloud platform using an improved TimesNet model, and to determine the most significant periods in the time series, including: Historical resource utilization data of the cloud platform is acquired. Based on the generation time of the historical resource utilization data, the historical resource utilization data is processed to obtain the corresponding time series of resource utilization data of the cloud platform. The historical resource utilization data includes historical CPU and / or memory utilization data. The improved TimesNet model is used to perform a fast Fourier transform on the time series to obtain the frequency component intensity of the time series, which is used as the multi-period feature of the time series. Then, based on the frequency component intensity, the most significant periods in the time series are determined. The two-dimensional time series variation modeling module is used to perform two-dimensional time series variation modeling based on the most significant several periods, to obtain the variation data of the time series at different time scales, including: Based on the most significant periods, the time series is folded according to the length corresponding to each period, thereby converting the one-dimensional time series into a two-dimensional tensor; wherein the width of the two-dimensional tensor is equal to the length of the corresponding period; then a two-dimensional convolutional network is used to capture the temporal change data of the two-dimensional tensor within and between periods, which is used as the change data of the time series at different time scales.

7. The cloud data center load prediction system based on improved TimesNet as described in claim 5, characterized in that: The two-dimensional convolution processing module is used to process the variation data of the time series at different time scales, and extract the features of the most significant periods, including: The time series data at different time scales is processed by two-dimensional convolution using ResNet to extract the features of the most significant periods. The adaptive fusion module is used to adaptively fuse the features of the most significant several periods to obtain a ResNet-based TimesBlock, including: Based on the frequency component intensity of each period in the time series, the features of the most significant periods are weighted and fused to obtain a ResNet-based TimesBlock.

8. The cloud data center load prediction system based on improved TimesNet as described in claim 5, characterized in that: The combined stacking module is used to combine and stack multiple TimesBlocks to form an improved TimesNet network, including: Analyze multiple TimesBlocks to determine the residuals of all TimesBlocks, and then link the residuals of multiple TimesBlocks to form an improved TimesNet network. The load prediction module is used to perform load prediction on the resource utilization data of the cloud platform using the improved TimesNet network, to obtain the resource demand load of the cloud platform in the future, including: The resource utilization data of the cloud platform is input into the improved TimesNet network to perform load prediction for different length periods, so as to obtain the resource demand load of the cloud platform for different time intervals in the future; wherein the resource demand load includes the data demand load of CPU and / or memory.

Images