Cloud resource demand prediction method and system based on period division

Abstract

The invention discloses a cloud resource demand prediction method and system based on period division, and the method comprises the following steps: obtaining cloud resource state data of a cloud resource pool, and carrying out the preprocessing of the data, and obtaining historical time series data; extracting a long-term change trend through a smoothing technology; performing period difference labeling, dividing each period of the period difference labeling by using a capacity span-based heuristic method, determining a cloud resource delivery period of the cloud resource pool, and obtaining cloud resource state data corresponding to the cloud resource delivery period; performing time sequence prediction model training, and predicting to obtain virtual cloud resource demand information of the cloud resource pool in a cloud resource delivery period; and determining the occurrence time when the cloud resource demand exceeds the warning capacity of the cloud resource pool and the amount of cloud resources needing to be expanded when the cloud resource pool maintains a stable operation state. According to the invention, operation and maintenance personnel can be assisted to expand the capacity of the cloud resource pool which is about to be exhausted in advance, and the utilization efficiency of cloud resources is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of time series prediction, and in particular to a method and system for predicting cloud resource demand based on period division. Background Art

[0002] The capacity change of the resource pool shows obvious phased characteristics, which can usually be divided into two stages: resource delivery period and resource pool maintenance period. When the resource pool is in the resource delivery period, the resource utilization rate rises rapidly, and the curve of this stage can reflect the user's demand for resources. When the resource pool is in the resource maintenance period, the management personnel will manually control the application and release of resources to prevent resource exhaustion. For cloud vendors, the capacity changes dramatically during the resource delivery period, and forecasting technology is needed to assist resource procurement.

[0003] Current demand forecasting technologies often use the entire life cycle data of a resource pool as samples and use machine learning algorithms (such as LSTM, Informer, etc.) to predict resource changes in the resource pool. This will cause the following two problems:

[0004] 1. The procurement of cloud resources is discontinuous and sudden. Directly using raw data for prediction will affect the training speed and generalization ability of the model.

[0005] 2. The traditional prediction model learns all the data during the resource delivery period and the maintenance period. These two periods are affected by human factors to different degrees, so the data distribution is very different, which will greatly affect the prediction accuracy. Summary of the invention

[0006] The purpose of the present invention is to provide a cloud resource demand prediction method and system based on period division in order to solve the above problems.

[0007] To achieve the above purpose, the technical solution provided by the present invention is:

[0008] The present application provides a method for predicting cloud resource demand based on period division, comprising the following steps:

[0009] Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation state; pre-process the cloud resource status data to obtain historical time series data;

[0010] The long-term trend of historical time series data is extracted through smoothing technology; the trend of the smoothed historical time series data is marked with period differences;

[0011] The periods marked by period difference are divided by using a heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool and obtain the cloud resource status data corresponding to the cloud resource delivery period;

[0012] Based on the cloud resource status data corresponding to the cloud resource delivery period, a time series prediction model is trained, and the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model;

[0013] The virtual cloud resource demand information is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool.

[0014] To optimize the above technical solutions, the specific measures taken also include:

[0015] Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation status; pre-process the cloud resource status data to obtain historical time series data, including:

[0016] Extracting cloud resource capacity change records within the cloud resource pool from the operation log of the cloud resource pool, analyzing the cloud resource capacity change records, and determining cloud resource capacity change rate status information within the cloud resource pool; based on the cloud resource capacity change rate status information, determining a first time point and a second time point corresponding to when the cloud resource capacity change rate within the cloud resource pool first exceeded a preset change rate threshold and when the cloud resource capacity change rate last exceeded the preset change rate threshold, respectively, and determining the time interval between the first time point and the second time point as a historical global time interval in which the cloud resource pool was in an effective operation state;

[0017] Based on all time points covered by the historical global time interval, the operation logs are screened and processed to obtain the cloud resource status data of the cloud resource pool in the historical global time interval; wherein the cloud resource status data includes the type and quantity of cloud resources released externally; based on the generation time sequence of the cloud resource status data, the cloud resource status data is converted into time series cloud resource status data as historical time series data.

[0018] Furthermore, the historical time series data is smoothed to extract the long-term trend. Specifically, the weighted moving average is used as the smoothing method to smooth the historical time series data to extract its long-term trend. The weighting method is:

[0019] The following formula is used to obtain the weight value of the cloud resource status data corresponding to each time according to the time series cloud resource status data:

[0020]

[0021] Among them, G(b_a) represents the weight value of the cloud resource status data corresponding to the bth time under the condition of the cloud resource status data corresponding to the ath time; t(a) represents the ath time moment; t(b) represents the bth time moment; T represents the total duration of the time series; n represents the total number of cloud resource status data in the time series; || represents the absolute value;

[0022] The following formula is used to perform weighted moving average processing on the time series cloud resource status data according to the weight value of the cloud resource status data corresponding to each time, thereby obtaining smoothed time series cloud resource status data:

[0023]

[0024] Where D′2(a) represents the binary form of the cloud resource status data corresponding to the ath time after weighted moving average processing; D2(b) represents the original cloud resource status data corresponding to the bth time; [] 10 Indicates converting the value in the brackets to a decimal number; {}2 means converting the value in the brackets to a binary number;

[0025] Using the following formula, based on the obtained smoothed time series cloud resource status data and the original time series cloud resource status data, a difference blue bar is generated to reflect the degree of difference in weighted moving average processing:

[0026]

[0027] Among them, μ represents the difference generated by the control (the blue bar shows the percentage); e represents the natural constant.

[0028] Furthermore, the changing trend of the smoothed historical time series data is labeled with period differences. Specifically, the changing trend of the historical time series data is labeled by difference: the smoothed historical time series data is differentiated, and the thresholds Tup and Tdown are set according to the distribution of the differential historical time series data. When the differential value is greater than Tup, it is marked as an upward period, and when it is less than Tdown, it is marked as a downward period, and other periods are marked as stable periods.

[0029] Furthermore, the periods marked by the period difference are divided using a heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool, and obtain the cloud resource status data corresponding to the cloud resource delivery period, specifically:

[0030] Perform threshold comparison on the smoothed time series cloud resource status data to determine the distribution time range information of all rising intervals, all falling intervals and all stable intervals corresponding to the smoothed time series cloud resource status data, and use this as the characteristic information of the change trend;

[0031] Based on the distribution time range information of all rising intervals corresponding to the smoothed time series cloud resource status data, the start and end times and capacity spans of all rising intervals are determined. According to the capacity span, the start and end times are compared with the historical global time intervals to determine the cloud resource delivery period of the cloud resource pool; and the cloud resource types and cloud resource quantities virtualized by the cloud resource pool during the cloud resource delivery period are obtained as cloud resource status data.

[0032] Furthermore, the resource delivery period data set calculation method for the cloud resource delivery period is:

[0033] Record the start and end time of all rising periods (start i ,end i ), and calculate the capacity utilization span Span i , the capacity span is greater than the maximum span max(Span i ) multiplied by the threshold T is divided into the resource delivery period D of the resource pool, and the corresponding capacity data constitute the resource delivery period data set:

[0034] D={(start i ,end i )|Span i ≥T·max(Span i )},0 <T<1。

[0035] Furthermore, a time series prediction model is trained based on the cloud resource status data corresponding to the cloud resource delivery period, and the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model; the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool are determined by the virtual cloud resource demand information, specifically:

[0036] The cloud resource status data corresponding to the cloud resource delivery period is input into the time series prediction model to train the time series prediction model; based on the trained time series prediction model, the type and quantity of cloud resources that the outside world expects to obtain from the cloud resource pool per unit time during the cloud resource delivery period are predicted as virtual cloud resource demand information;

[0037] Compare the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information with the warning capacity of the cloud resource pool to determine the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool; and based on the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information, determine the CPU, memory storage and hard disk storage that need to be expanded to maintain a stable operating state of the cloud resource pool.

[0038] A second aspect of the present application provides a cloud resource demand forecasting system based on period division, including:

[0039] The cloud resource status data acquisition and analysis module is used to obtain the cloud resource status data of the cloud resource pool in the effective operation status historical global time interval; pre-process the cloud resource status data to obtain historical time series data;

[0040] The delivery period calibration module is used to extract the long-term change trend of the historical time series data through smoothing technology; to mark the change trend of the smoothed historical time series data by period difference; to divide each period of the period difference mark by using the heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool, and obtain the cloud resource status data corresponding to the cloud resource delivery period;

[0041] The model training and prediction module is used to train the time series prediction model based on the cloud resource status data corresponding to the cloud resource delivery period, and obtain the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period through the time series prediction model prediction;

[0042] The resource adjustment module is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool based on the virtual cloud resource demand information.

[0043] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned cloud resource demand prediction method based on period division is implemented.

[0044] A fourth aspect of the present application provides a computer-readable storage medium storing a computer program, which enables a computer to execute the above-mentioned cloud resource demand prediction method based on period division.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The cloud resource demand prediction method based on period division of the present invention obtains the cloud resource status data of the historical global time interval in which the cloud resource pool is in an effective operation state and performs preprocessing, and then performs differential identification processing on the preprocessed cloud resource status data to obtain the change trend characteristic information of the cloud resource status data, so as to divide the historical global time interval into periods, so as to determine the cloud resource delivery period of the cloud resource pool, so as to facilitate the subsequent accurate extraction of the cloud resource status data corresponding to the cloud resource delivery period, provide a reliable data source for training the time series prediction model, accurately predict and obtain the virtual cloud resource demand information of the cloud resource pool in the cloud resource delivery period, thereby determining the occurrence time when the cloud resource demand exceeds the warning capacity of the cloud resource pool and the amount of cloud resources that need to be expanded to maintain a stable operation state of the cloud resource pool, assisting operation and maintenance personnel to expand the capacity of the cloud resource pool that is about to be exhausted in advance, and improving the utilization efficiency of cloud resources.

[0047] Aiming at the capacity variation law of private cloud resource pool, the present invention introduces the concepts of resource delivery period and maintenance period, and points out the data feature gap caused by human factor control.

[0048] The present invention divides the resource pool into different periods using a heuristic method based on capacity span. For the resource pool in the resource delivery period, only the extracted resource delivery period data set is used to train the time series prediction model, thereby avoiding the negative impact of inconsistency in data distribution in different periods on the prediction accuracy of the model.

[0049] The present invention extracts long-term change trends from historical time series through smoothing technology, thereby reducing the impact of discontinuity and mutation in cloud resource procurement on the model convergence speed and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 : Schematic diagram of the flow of the cloud resource demand prediction method based on period division of the present invention.

[0051] Figure 2 : A schematic diagram of the structure of the cloud resource demand forecasting system based on period division of the present invention.

[0052] Figure 3 : Use smoothing technology to extract long-term change trend charts from historical time series data.

[0053] Figure 4 : Annotate the changing trend of the smoothed historical time series data into period difference charts.

[0054] Figure 5 : Computation graph of resource delivery period dataset during cloud resource delivery period. DETAILED DESCRIPTION

[0055] The above contents of the present invention are further described in detail below in the form of embodiments, but this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the above contents of the present invention belong to the scope of the present invention.

[0056] Some of the terms involved in the solution of the present invention are explained as follows:

[0057] Time Series Forecasting: A time series is a sequence of data points arranged in chronological order, such as stock prices, temperature records, sales data, etc. Time series forecasting refers to the process of predicting future values ​​based on historical time series data through methods such as machine learning.

[0058] This application provides a cloud resource demand prediction method based on period division, such as Figure 1 As shown, the following steps are included:

[0059] Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation state; pre-process the cloud resource status data to obtain historical time series data;

[0060] The long-term trend of historical time series data is extracted through smoothing technology; the trend of the smoothed historical time series data is marked with period differences;

[0061] The periods marked by period difference are divided by using a heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool and obtain the cloud resource status data corresponding to the cloud resource delivery period;

[0062] Based on the cloud resource status data corresponding to the cloud resource delivery period, a time series prediction model is trained, and the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model;

[0063] The virtual cloud resource demand information is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool.

[0064] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods:

[0065] Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation status; pre-process the cloud resource status data to obtain historical time series data, including:

[0066] Extracting cloud resource capacity change records within the cloud resource pool from the operation log of the cloud resource pool, analyzing the cloud resource capacity change records, and determining cloud resource capacity change rate status information within the cloud resource pool; based on the cloud resource capacity change rate status information, determining a first time point and a second time point corresponding to when the cloud resource capacity change rate within the cloud resource pool first exceeded a preset change rate threshold and when the cloud resource capacity change rate last exceeded the preset change rate threshold, respectively, and determining the time interval between the first time point and the second time point as a historical global time interval in which the cloud resource pool was in an effective operation state;

[0067] Based on all time points covered by the historical global time interval, the operation logs are screened and processed to obtain the cloud resource status data of the cloud resource pool in the historical global time interval; wherein the cloud resource status data includes the type and quantity of cloud resources released externally; based on the generation time sequence of the cloud resource status data, the cloud resource status data is converted into time series cloud resource status data as historical time series data.

[0068] A large number of different types of cloud resources are stored inside the cloud resource pool. When the cloud resource pool releases and configures cloud resources or changes cloud resources, a cloud resource capacity change record is generated accordingly. The cloud resource capacity change record inside the cloud resource pool is extracted from the operation log of the cloud resource pool, and the cloud resource capacity change log is analyzed to determine the cloud resource capacity change rate state information inside the cloud resource pool. The cloud resource capacity change rate state information may be, but is not limited to, the distribution of the change rate of each type of cloud resource capacity size inside the cloud resource pool in the corresponding time period. Then, based on the cloud resource capacity change rate state information, the first time point and the second time point corresponding to the first time the cloud resource capacity change rate inside the cloud resource pool exceeds the preset change rate threshold and the last time the cloud resource capacity change rate threshold are determined. When the cloud resource capacity change rate inside the cloud resource pool exceeds the preset change rate threshold, it indicates that the cloud resource pool is currently in an active cloud resource operation state, so that the historical global time interval in which the cloud resource pool is in an effective operation state can be accurately determined. Since the historical global time interval in which the cloud resource pool is in an effective operation state is correlated with the cloud resource delivery period of the cloud resource pool, the operation log is screened based on all time points covered by the historical global time interval to obtain the cloud resource status data of the cloud resource pool in the historical global time interval, and based on the generation time sequence of the cloud resource status data, the cloud resource status data is converted into time series cloud resource status data, and the time series cloud resource status data is weighted moving averaged to obtain smoothed time series cloud resource status data, which can reduce the impact of the discontinuity and burstiness of the cloud resource status data on the convergence speed and generalization ability of the time series prediction model.

[0069] In some implementations, the historical time series data is smoothed to extract the long-term trend, specifically: weighted moving average is used as a smoothing method to smooth the historical time series data to extract its long-term trend. The weighting method is:

[0070] The following formula is used to obtain the weight value of the cloud resource status data corresponding to each time according to the time series cloud resource status data:

[0071]

[0072] Among them, G(b_a) represents the weight value of the cloud resource status data corresponding to the bth time under the condition of the cloud resource status data corresponding to the ath time; t(a) represents the ath time moment; t(b) represents the bth time moment; T represents the total duration of the time series; n represents the total number of cloud resource status data in the time series; || represents the absolute value;

[0073] The following formula is used to perform weighted moving average processing on the time series cloud resource status data according to the weight value of the cloud resource status data corresponding to each time, thereby obtaining smoothed time series cloud resource status data:

[0074]

[0075] Where D′2(a) represents the binary form of the cloud resource status data corresponding to the ath time after weighted moving average processing; D2(b) represents the original cloud resource status data corresponding to the bth time; [] 10 Indicates converting the value in the brackets to a decimal number; {}2 means converting the value in the brackets to a binary number;

[0076] Using the following formula, based on the obtained smoothed time series cloud resource status data and the original time series cloud resource status data, a difference blue bar is generated to reflect the degree of difference in weighted moving average processing:

[0077]

[0078] Among them, μ represents the difference generated by the control (the blue bar shows the percentage); e represents the natural constant.

[0079] By using the above formula (1), the weight value of the cloud resource status data corresponding to each time is obtained according to the time series cloud resource status data, so as to obtain the weight of the cloud resource status data corresponding to each time to ensure the reliability of the algorithm of weighted moving average processing; then by using the above formula (2), according to the weight value of the cloud resource status data corresponding to each time, the time series cloud resource status data is subjected to weighted moving average processing, so as to obtain smoothed time series cloud resource status data. The weighted moving average processing is completely and automatically performed through the algorithm, which reflects the intelligence of the system; then by using the above formula (3), according to the obtained smoothed time series cloud resource status data and the original time series cloud resource status data, a difference blue bar is generated to reflect the difference degree of the weighted moving average processing, so that the difference degree of the weighted moving average processing can be known according to the difference blue bar, which is convenient for subsequent tracing and review of the data.

[0080] In some implementations, the trend of the smoothed historical time series data is differentially labeled, specifically: the trend of the historical time series data is differentially labeled, for example Figure 3 As shown in the figure: The smoothed historical time series data is differentiated, and the thresholds Tup and Tdown are set according to the distribution of the differential historical time series data. When the differential value is greater than Tup, it is marked as an upward period, and when it is less than Tdown, it is marked as a downward period. Other periods are marked as stable periods. For example Figure 4 shown.

[0081] In some implementations, a heuristic method based on capacity span is used to divide each period marked by the period difference, so as to determine the cloud resource delivery period of the cloud resource pool, and obtain the cloud resource status data corresponding to the cloud resource delivery period, specifically:

[0082] Perform threshold comparison on the smoothed time series cloud resource status data to determine the distribution time range information of all rising intervals, all falling intervals and all stable intervals corresponding to the smoothed time series cloud resource status data, and use this as the characteristic information of the change trend;

[0083] Based on the distribution time range information of all rising intervals corresponding to the smoothed time series cloud resource status data, the start and end times and capacity spans of all rising intervals are determined. According to the capacity span, the start and end times are compared with the historical global time intervals to determine the cloud resource delivery period of the cloud resource pool; and the cloud resource types and cloud resource quantities virtualized by the cloud resource pool during the cloud resource delivery period are obtained as cloud resource status data.

[0084] The smoothed time series cloud resource status data is compared with the threshold value to determine the distribution time range information of all rising intervals, all falling intervals and all stable intervals corresponding to the smoothed time series cloud resource status data, which is used as the characteristic information of the change trend. Specifically, the smoothed time series cloud resource status data is differentiated and the threshold T is set. up and T down , and T up >T down , when the difference value is greater than T up , it is marked as an ascending interval; when the difference value is less than T down , it is marked as a descending interval; other intervals are marked as stable intervals. Based on the distribution time range information of all rising intervals corresponding to the smoothed time series cloud resource status data, the start and end times of all rising intervals are determined, and the start and end times are compared with the time points of the historical global time intervals. The union of the start and end times of all rising intervals corresponding to the interval range of the historical global time interval is determined as the cloud resource delivery period of the cloud resource pool. The cloud resource types and cloud resource quantities released by the cloud resource pool during the cloud resource delivery period are obtained as cloud resource status data, which can ensure the comprehensiveness and accuracy of cloud resource status data.

[0085] In some implementations, the resource delivery period data set calculation method for the cloud resource delivery period is:

[0086] Record the start and end time of all rising periods (start i ,end i ), and calculate the capacity utilization span Span i , the capacity span is greater than the maximum span max(Span i ) multiplied by the threshold T is divided into a number of rising periods as the resource delivery period D of the resource pool. For example Figure 5 As shown, the corresponding capacity data constitutes the resource delivery period data set:

[0087] D={(start i ,end i )|Span i ≥T·max(Span i )},0 <T<1。

[0088] In some implementations, a time series prediction model is trained based on cloud resource status data corresponding to the cloud resource delivery period, and virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model; the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool are determined by the virtual cloud resource demand information, specifically:

[0089] The cloud resource status data corresponding to the cloud resource delivery period is input into the time series prediction model to train the time series prediction model; based on the trained time series prediction model, the type and quantity of cloud resources that the outside world expects to obtain from the cloud resource pool per unit time during the cloud resource delivery period are predicted as virtual cloud resource demand information;

[0090] Compare the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information with the warning capacity of the cloud resource pool to determine the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool; and based on the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information, determine the CPU, memory storage and hard disk storage that need to be expanded to maintain a stable operating state of the cloud resource pool.

[0091] The cloud resource status data corresponding to the cloud resource delivery period is input into the time series prediction model to train the time series prediction model; based on the trained time series prediction model, the type and quantity of cloud resources that the outside world expects to obtain from the cloud resource pool per unit time during the cloud resource delivery period are predicted, and this is used as the virtual cloud resource demand information, which can improve the accuracy and credibility of the prediction of the outside world's cloud resource demand. In addition, the information on the change of virtual cloud resource demand over time included in the virtual cloud resource demand information is compared with the warning capacity of the cloud resource pool to determine the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool, so as to determine the CPU, memory storage and hard disk storage that need to be expanded to maintain a stable operation state of the cloud resource pool. For example, it is determined that the cloud resource pool's demand for cloud resources is expected to exceed the warning capacity of the cloud resource pool (such as 80% of the total cloud resources in the cloud resource pool) after 35 days. At this time, if the cloud resource pool is to maintain a stable operation state, it is necessary to expand the CPU by a preset number of cores, the memory storage by a preset number of GB and the hard disk storage by a preset number of TB. This can accurately assist the operation and maintenance personnel to reasonably purchase cloud resources for the cloud resource pool in advance to avoid waste caused by purchasing too many cloud resources or resource depletion caused by purchasing too few cloud resources.

[0092] The second aspect of the present application provides a cloud resource demand forecasting system based on period division, such as Figure 2 As shown, including:

[0093] The cloud resource status data acquisition and analysis module is used to obtain the cloud resource status data of the cloud resource pool in the effective operation status historical global time interval; pre-process the cloud resource status data to obtain historical time series data;

[0094] The delivery period calibration module is used to extract the long-term change trend of the historical time series data through smoothing technology; to mark the change trend of the smoothed historical time series data by period difference; to divide each period of the period difference mark by using the heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool, and obtain the cloud resource status data corresponding to the cloud resource delivery period;

[0095] The model training and prediction module is used to train the time series prediction model based on the cloud resource status data corresponding to the cloud resource delivery period, and obtain the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period through the time series prediction model prediction;

[0096] The resource adjustment module is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool based on the virtual cloud resource demand information.

[0097] It also includes an operation log analysis module for analyzing the operation log of the cloud resource pool to determine the historical global time interval in which the cloud resource pool is in a valid operation state.

[0098] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned cloud resource demand prediction method based on period division is implemented.

[0099] A fourth aspect of the present application provides a computer-readable storage medium storing a computer program, which enables a computer to execute the above-mentioned cloud resource demand prediction method based on period division.

[0100] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0101] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any technician familiar with the profession, without departing from the scope of the technical solution of the present invention and based on the technical essence of the present invention, may make any simple modifications, equivalent substitutions and improvements to the above embodiments, which are still within the protection scope of the technical solution of the present invention.

Claims

1. A cloud resource demand prediction method based on period division, characterized in that: The following steps are involved: Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation state; pre-process the cloud resource status data to obtain historical time series data; Extract long-term trend of historical time series data through smoothing technology; Mark the changing trend of the smoothed historical time series data by period difference; The periods marked by period difference are divided by using a heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool and obtain the cloud resource status data corresponding to the cloud resource delivery period; Based on the cloud resource status data corresponding to the cloud resource delivery period, a time series prediction model is trained, and the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model; The virtual cloud resource demand information is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool.

2. The cloud resource demand forecasting method based on period division according to claim 1 is characterized by: Obtain cloud resource status data of the cloud resource pool in the historical global time interval of the effective operation status; pre-process the cloud resource status data to obtain historical time series data, including: Extracting cloud resource capacity change records within the cloud resource pool from the operation log of the cloud resource pool, analyzing the cloud resource capacity change records, and determining cloud resource capacity change rate status information within the cloud resource pool; based on the cloud resource capacity change rate status information, determining a first time point and a second time point corresponding to when the cloud resource capacity change rate within the cloud resource pool first exceeded a preset change rate threshold and when the cloud resource capacity change rate last exceeded the preset change rate threshold, respectively, and determining the time interval between the first time point and the second time point as a historical global time interval in which the cloud resource pool was in an effective operation state; Based on all time points covered by the historical global time interval, the operation logs are screened and processed to obtain the cloud resource status data of the cloud resource pool in the historical global time interval; wherein the cloud resource status data includes the type and quantity of cloud resources released externally; based on the generation time sequence of the cloud resource status data, the cloud resource status data is converted into time series cloud resource status data as historical time series data.

3. The cloud resource demand forecasting method based on period division according to claim 1 is characterized by: The historical time series data is smoothed to extract the long-term trend. Specifically, the weighted moving average is used as the smoothing method to smooth the historical time series data to extract its long-term trend. The weighting method is: The following formula is used to obtain the weight value of the cloud resource status data corresponding to each time according to the time series cloud resource status data: Among them, G(b_a) represents the weight value of the cloud resource status data corresponding to the bth time under the condition of the cloud resource status data corresponding to the ath time; t(a) represents the ath time moment; t(b) represents the bth time moment; T represents the total duration of the time series; n represents the total number of cloud resource status data in the time series; | | represents the absolute value; The following formula is used to perform weighted moving average processing on the time series cloud resource status data according to the weight value of the cloud resource status data corresponding to each time, thereby obtaining smoothed time series cloud resource status data: Where D′2(a) represents the binary form of the cloud resource status data corresponding to the ath time after weighted moving average processing; D2(b) represents the original cloud resource status data corresponding to the bth time; [] 10 Indicates that the value in the brackets is converted to a decimal number; {}2 means that the value in the brackets is converted to a binary number; Using the following formula, based on the obtained smoothed time series cloud resource status data and the original time series cloud resource status data, a difference blue bar is generated to reflect the degree of difference in weighted moving average processing: Among them, μ represents the difference generated by the control (the blue bar shows the percentage); e represents the natural constant.

4. The cloud resource demand forecasting method based on period division according to claim 1 is characterized by: The changing trend of the smoothed historical time series data is marked by period differences. Specifically, the changing trend of the historical time series data is marked by differences: the smoothed historical time series data is differentiated, and the thresholds Tup and Tdown are set according to the distribution of the differential historical time series data. When the differential value is greater than Tup, it is marked as an upward period, and when it is less than Tdown, it is marked as a downward period, and other periods are marked as stable periods.

5. The cloud resource demand forecasting method based on period division according to claim 1 is characterized by: The periods marked by period difference are divided by using a heuristic method based on capacity span to determine the cloud resource delivery period of the cloud resource pool and obtain the cloud resource status data corresponding to the cloud resource delivery period, specifically: Perform threshold comparison on the smoothed time series cloud resource status data to determine the distribution time range information of all rising intervals, all falling intervals and all stable intervals corresponding to the smoothed time series cloud resource status data, and use this as the characteristic information of the change trend; Based on the distribution time range information of all the rising intervals corresponding to the smoothed time series cloud resource status data, the start and end times and capacity spans of all the rising intervals are determined. According to the capacity span, the start and end times are compared with the historical global time intervals to determine the cloud resource delivery period of the cloud resource pool; And obtain the cloud resource type and cloud resource quantity virtualized by the cloud resource pool during the cloud resource delivery period as cloud resource status data.

6. The method for predicting cloud resource demand based on period division according to claim 5 is characterized in that: The calculation method of the resource delivery period dataset during the cloud resource delivery period is: Record the start and end time of all rising periods (start i ,end i ), and calculate the capacity utilization span Span i , the capacity span is greater than the maximum span max(Span i ) multiplied by the threshold T is divided into the resource delivery period D of the resource pool, and the corresponding capacity data constitute the resource delivery period data set: D={(start i ,end i )|Span i ≥T·max(Span i )},0<T<1。 7. The cloud resource demand forecasting method based on period division according to claim 1 is characterized by: Based on the cloud resource status data corresponding to the cloud resource delivery period, the time series prediction model is trained, and the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period is predicted by the time series prediction model; the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool are determined by the virtual cloud resource demand information, specifically: the cloud resource status data corresponding to the cloud resource delivery period is input into the time series prediction model to train the time series prediction model; based on the trained time series prediction model, the type and quantity of cloud resources that the outside world expects to obtain from the cloud resource pool per unit time during the cloud resource delivery period are predicted as the virtual cloud resource demand information; Compare the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information with the warning capacity of the cloud resource pool to determine the time when the cloud resource demand exceeds the warning capacity of the cloud resource pool; and based on the information on changes in virtual cloud resource demand over time included in the virtual cloud resource demand information, determine the CPU, memory storage and hard disk storage that need to be expanded to maintain a stable operating state of the cloud resource pool.

8. A cloud resource demand forecasting system based on period division, characterized in that: include: A cloud resource status data acquisition and analysis module is used to obtain cloud resource status data of the cloud resource pool in the effective operation status historical global time interval; Preprocess the cloud resource status data to obtain historical time series data; The delivery period calibration module is used to extract long-term change trends from historical time series data through smoothing technology; Mark the changing trend of the smoothed historical time series data by period difference; The periods marked by period difference are divided by using a heuristic method based on capacity span, so as to determine the cloud resource delivery period of the cloud resource pool and obtain the cloud resource status data corresponding to the cloud resource delivery period; The model training and prediction module is used to train the time series prediction model based on the cloud resource status data corresponding to the cloud resource delivery period, and obtain the virtual cloud resource demand information of the cloud resource pool during the cloud resource delivery period through the time series prediction model prediction; The resource adjustment module is used to determine the time when the cloud resource demand exceeds the cloud resource pool warning capacity and the amount of cloud resources that need to be expanded to maintain a stable operating state of the cloud resource pool based on the virtual cloud resource demand information. The system is used to execute the cloud resource demand forecasting method based on period division as claimed in any one of claims 1 to 7.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a cloud resource demand forecasting method based on period division as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the cloud resource demand forecasting method based on period division as claimed in any one of claims 1 to 7.

Images