Method, device and system for determining computing power state

Abstract

The invention provides a method, device and system for determining computing power state data, and relates to the technical field of computing power networks. The method for determining the computing power state data comprises the following steps: acquiring the computing power state data in a first time period; encoding the computing power state data in the first time period according to an auto-encoder to obtain an overall feature vector of the computing power state data in the first time period, the dimension of the overall feature vector being smaller than the dimension of the computing power state data in the first time period; and processing the overall feature vector according to a machine learning model to obtain computing power state data in a second time period, the second time period being a time period after the first time period. Through the method, the processing efficiency of computing power state data prediction can be improved while the prediction accuracy of the computing power state data is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computing power network technology, and in particular to a method, device and system for determining computing power status. Background Art

[0002] As digital transformation in various industries deepens, massive amounts of data will be generated. The massive amounts of data generated by digital transformation urgently require a lot of computing power to process. As a key system that provides computing power resources, computing power networks will become a key link in supporting the processing of massive amounts of data. Moreover, the processing of massive amounts of data also places higher demands on cloud-network integration and computing-network integration. The computing power network and the communication network need to work together to complete the allocation, scheduling, and management of computing tasks.

[0003] In the collaborative collaboration between computing power networks and communication networks, the perception and prediction of computing power status data is a key issue. Summary of the invention

[0004] The present disclosure provides a method, device and system for determining computing power status data.

[0005] According to a first aspect of the present disclosure, a method for determining computing power status data is proposed, including: obtaining computing power status data within a first time period; encoding the computing power status data within the first time period according to an autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; processing the overall feature vector according to a machine learning model to obtain computing power status data within a second time period, wherein the second time period is a time period after the first time period.

[0006] In some embodiments, the computing power status data within the first time period includes multiple data subsequences, and encoding the computing power status data within the first time period according to the autoencoder to obtain an overall feature vector includes: encoding each data subsequence in the multiple data subsequences according to the autoencoder to obtain a local feature vector of each data subsequence; and fusing the local feature vectors of all data subsequences in the multiple data subsequences to obtain the overall feature vector.

[0007] In some embodiments, the fusing the local feature vectors of all data subsequences in the multiple data subsequences to obtain the overall feature vector includes: determining the weight of each data subsequence in the multiple data subsequences according to the distribution characteristics of the multiple data subsequences; and performing weighted fusion of the local feature vectors of all data subsequences in the multiple data subsequences according to the weights of all data subsequences in the multiple data subsequences to obtain the overall feature vector.

[0008] In some embodiments, the distribution characteristics of the multiple data subsequences are variance distribution characteristics of the multiple data subsequences.

[0009] In some embodiments, the variance distribution characteristics of the multiple data subsequences are determined according to the following method: determining the variance of each data subsequence in the multiple data subsequences; determining multiple variance value intervals based on the variances of all data subsequences in the multiple data subsequences; determining the number of data subsequences in each variance value interval in the multiple variance value intervals; and using the number of data subsequences in the variance value interval in which each data subsequence in the multiple data subsequences is located as the variance distribution characteristics of each data subsequence.

[0010] In some embodiments, the weight of each data subsequence is negatively correlated with the variance distribution characteristics of each data subsequence.

[0011] In some embodiments, the weight of each data subsequence is exponentially related to the variance distribution characteristic of each data subsequence.

[0012] In some embodiments, each data subsequence in the multiple data subsequences contains an equal number of data.

[0013] In some embodiments, the computing power status data includes available computing power resources of one or more computing power resources or used computing power resources of one or more computing power resources.

[0014] In some embodiments, the machine learning model includes a deep neural network, a support vector machine, or a gradient boosting decision tree.

[0015] In some embodiments, the method for determining computing power status data further includes: before encoding the computing power status data within the first time period according to the autoencoder, training the autoencoder and the machine learning model based on historical computing power status sample data.

[0016] According to a second aspect of the present disclosure, a device for determining computing power status data is proposed, including: an acquisition module, configured to acquire computing power status data within a first time period; an encoding module, configured to encode the computing power status data within the first time period according to an autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; a determination module, configured to process the overall feature vector according to a machine learning model to obtain computing power status data within a second time period, wherein the second time period is a time period after the first time period.

[0017] According to a third aspect of the present disclosure, a system for determining computing power status data is proposed, comprising: a communication network orchestrator, configured to send computing power status data within a first time period to a device for determining computing power status data; the device for determining computing power status data is configured to: encode the computing power status data within the first time period according to an autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; and process the overall feature vector according to a machine learning model to obtain computing power status data within a second time period, wherein the second time period is a time period after the first time period.

[0018] In some embodiments, the communication network orchestrator is further configured to: send a subscription request to a computing power network orchestrator, wherein the subscription request is used to indicate a subscription to computing power status data; establish a long connection with the computing power network orchestrator; based on the long connection, receive computing power status data reported by the computing power network orchestrator; and obtain computing power status data within a first time period from the reported computing power status data.

[0019] According to a fourth aspect of the present disclosure, an electronic device is proposed, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the method for determining computing power status data as described above based on instructions stored in the memory.

[0020] According to a fifth aspect of the present disclosure, a computer-readable storage medium is proposed, on which computer program instructions are stored, and when the instructions are executed by a processor, the method for determining computing power status data as described above is implemented.

[0021] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0023] The present disclosure can be more clearly understood from the following detailed description with reference to the accompanying drawings.

[0024] Figure 1 The present invention is a flowchart of a method for determining computing power status data according to some embodiments of the present disclosure.

[0025] Figure 2 The present invention is a flowchart of encoding computing power status data to obtain an overall feature vector according to some embodiments of the present disclosure.

[0026] Figure 3The figure is a schematic diagram of the principle of weighted fusion of local feature vectors according to some embodiments of the present disclosure.

[0027] Figure 4 The present invention is a flowchart of a method for determining computing power status data according to other embodiments of the present disclosure.

[0028] Figure 5 A schematic diagram of the structure of an apparatus for determining computing power status data according to some embodiments of the present disclosure.

[0029] Figure 6 A structural diagram of a system for determining computing power status data according to some embodiments of the present disclosure.

[0030] Figure 7 It is a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure.

[0031] Figure 8 A schematic diagram of the structure of a computer system according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0032] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0033] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0034] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0035] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0036] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0037] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0038] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0039] In the related art, due to the introduction of artificial experience in the feature selection stage of computing power status data prediction, information loss may occur, which in turn affects the accuracy of computing power status data prediction results. Moreover, in order to improve the prediction accuracy of computing power status data, it is often necessary to add as much historical data as possible to feature engineering, which results in a longer running time of the computing power status data prediction algorithm.

[0040] In view of this, the present disclosure proposes a method, device and system for determining computing power status data, which can improve the prediction accuracy of computing power status data while improving the processing efficiency of computing power status data prediction.

[0041] Figure 1 FIG. 1 is a flow chart of a method for determining computing power status data according to some embodiments of the present disclosure. Figure 1 As shown, the method for determining computing power status data includes steps S110 to S130.

[0042] In step S110, computing power status data within a first time period is obtained.

[0043] In some embodiments, the method for determining computing power status data is performed by a device for determining computing power status data. In some examples, the device for determining computing power status data is part of a communication network orchestrator. In other examples, the device for determining computing power status data is a device located outside of the communication network orchestrator.

[0044] In some implementations, the computing power status data includes the available computing power resources of one or more computing power resources. For example, the computing power status data is at least one of the available amount of GPU, the available amount of CPU, the available amount of memory, the available amount of hard disk, etc. For example, the computing power status data is at least one of the usage of GPU, the usage of CPU, the usage of memory, the usage of hard disk, etc.

[0045] In some embodiments, the computing power status data of the first time period is computing power status data in a past time period, for example, the GPU availability per day in the past 30 days.

[0046] In step S120, the computing power status data within the first time period is encoded according to the autoencoder to obtain an overall feature vector of the computing power status data within the first time period.

[0047] Among them, the dimension of the overall feature vector is smaller than the dimension of the computing power status data in the first time period.

[0048] An autoencoder (AE) is a type of artificial neural network (ANN) used in semi-supervised learning and unsupervised learning. Its function is to perform representation learning on input information by taking the input information as the learning target. An autoencoder consists of two parts: encoding and decoding. The process of learning a feature vector from input data is called encoding, and the process of reconstructing the original data using the learned feature vector is called decoding. In the training phase of an autoencoder, both encoding and decoding are involved. In the use phase of an autoencoder, the encoding part is involved.

[0049] In some embodiments, the computing power status data within the first time period is input as a whole into the autoencoder for encoding to obtain an overall feature vector of the computing power status data within the first time period.

[0050] For example, let Represents a collection of computing power status data within multiple first time periods, for computing power status data s within a first time period i ,definition:

[0051] h i =f(s i ; Θ)

[0052] Where f represents the encoding function, h i represents the overall feature vector obtained by encoding, and Θ represents the set of encoding parameters.

[0053] In other embodiments, based on Figure 2 The process shown is to obtain the overall feature vector of the computing power status data in the first time period.

[0054] In the disclosed embodiment, by using an autoencoder to encode the computing power state data in the first time period, not only can the obtained overall feature vector restore the input data as much as possible, reducing the information loss in the feature engineering of computing power prediction, but also the dimension reduction of computing power state data can be achieved. In this way, the prediction efficiency of computing power state data can be improved while ensuring the prediction effect of computing power state data.

[0055] In step S130, the overall feature vector is processed according to the machine learning model to obtain computing power status data within the second time period.

[0056] The second time period is a time period after the first time period. For example, the first time period is every day in the past 30 days, and the second time period is every day in the next 30 days.

[0057] In some implementations, the machine learning model includes a deep neural network, a support vector machine, or a gradient boosted decision tree.

[0058] In some embodiments, the method for determining computing power status data further includes: before step S110, training the autoencoder and the machine learning model based on historical computing power status sample data.

[0059] In some examples, the autoencoder is trained using an unsupervised learning method based on historical computing power status sample data; the machine learning model used for computing power status data prediction is trained using a supervised learning method based on historical computing power status sample data. Furthermore, during the training of the autoencoder and the machine learning model, the backpropagation algorithm can be used to update the network weights of the machine learning model, and the Adam optimizer can be used to improve the training effect.

[0060] In the disclosed embodiment, by predicting the computing power status data in the second time period based on the computing power status data in the first time period, there is no need to manually select features during the computing power status prediction process, thus avoiding information loss that may be caused by human interference. Moreover, by encoding the computing power status data in the first time period through the autoencoder to obtain the overall feature vector, not only can the obtained overall feature vector restore the input data as much as possible, but also can achieve dimensionality reduction of the computing power status data. In this way, the prediction efficiency of the computing power status data can be improved while ensuring the prediction effect of the computing power status data.

[0061] Figure 2 The present invention is a flowchart of encoding computing power status data to obtain an overall feature vector according to some embodiments of the present disclosure. Figure 2 The process shown is an exemplary description of step S120. Figure 2 As shown, the process of encoding the computing power status data to obtain the overall feature vector includes step S121 and step S122.

[0062] In step S121, each data subsequence in the plurality of data subsequences is encoded according to the autoencoder to obtain a local feature vector of each data subsequence.

[0063] In some embodiments, the computing power status data within the first time period includes multiple data subsequences.

[0064] In some examples, each data subsequence in the multiple data subsequences contains an equal number of data. For example, the computing power status data in the first time period includes 1000 data, and the 1000 data are divided into 5 data subsequences, each data subsequence includes 200 data.

[0065] In some examples, the number of data contained in each data subsequence in the multiple data subsequences is not equal, or the number of data contained in some data subsequences is equal and the number of data contained in some data subsequences is unequal. For example, the computing power status data in the first time period includes 1000 data, and the 1000 data are divided into 5 data subsequences, and the number of data contained in the 5 data subsequences is 200, 199, 198, 201, and 202 respectively.

[0066] In some embodiments, in step S121, multiple data subsequences are successively input into the autoencoder to obtain a local feature vector of each data subsequence in the multiple data subsequences. For example, assuming that the computing power status data in the first time period is divided into s 1 to 5 These five data subsequences can be input into the autoencoder one by one to obtain the data subsequence s 1 The local feature vector, data subsequence s 2 The local feature vector, data subsequence s 3 The local feature vector, data subsequence s 4 The local feature vector of 5 The local feature vector of .

[0067] In an embodiment of the present disclosure, by dividing the computing power status data within the first time period into multiple data subsequences, and using an autoencoder to encode each data subsequence in the multiple data subsequences, compared with directly encoding the computing power status data within the first time period as a whole in other embodiments of the present disclosure, the processing efficiency of encoding the computing power status data can be further improved, thereby helping to improve the efficiency of computing power status data prediction.

[0068] In step S122, local feature vectors of all data subsequences in the multiple data subsequences are fused to obtain an overall feature vector.

[0069] In some embodiments, step S122 includes: determining the weight of each data subsequence in the multiple data subsequences based on the distribution characteristics of the multiple data subsequences; and performing weighted fusion on the local feature vectors of all the data subsequences in the multiple data subsequences based on the weights of all the data subsequences in the multiple data subsequences to obtain an overall feature vector.

[0070] In the disclosed embodiment, the weight of the data subsequence is determined by the distribution characteristics of the data subsequence, and the local feature vectors of the data subsequence are weighted and fused according to the weight of the data subsequence to obtain the overall feature vector. In this way, since the different amounts of information carried by different subsequences have different impacts on the overall feature vector, compared with the method of obtaining the overall feature vector by average weighting, the representation ability of the overall feature vector can be increased, and the accuracy of the prediction results of the computing power status data can be further improved.

[0071] In some embodiments, the distribution characteristics of the plurality of data subsequences are variance distribution characteristics.

[0072] In some embodiments, variance distribution characteristics of multiple data subsequences are determined according to steps a to d.

[0073] In step a, the variance of each data subsequence in the plurality of data subsequences is determined.

[0074] In some examples, the variance of each data subsequence is calculated according to the following formula.

[0075]

[0076]

[0077] in, represents the variance of the data subsequence, Represents a data subsequence The jth data in represents the expectation of the data subsequence, M loc Indicates the number of hashrate status data contained in the data subsequence.

[0078] In step b, multiple variance value intervals are determined according to the variances of all data subsequences in the multiple data subsequences.

[0079] In some examples, a value range between a minimum value and a maximum value in the variances of all data subsequences in a plurality of data subsequences is divided into a plurality of variance value intervals.

[0080] For example, assuming that the computing power status data in the first time period is divided into 5 data subsequences, the value range between the minimum and maximum values ​​in the variance of these 5 data subsequences is divided into L variance value intervals. Where L is an integer greater than 1. For example, L is 5, 6, 7 or other values. These L variance value intervals can be expressed as:

[0081]

[0082] in, represents the first variance value interval, represents the second variance value interval, Represents the Lth variance value interval.

[0083] In step c, the number of data subsequences in each of the multiple variance value intervals is determined.

[0084] In step d, the number of data subsequences in the variance value interval of each data subsequence in the multiple data subsequences is used as the variance distribution feature of each data subsequence.

[0085] For example, assuming that there are 100 data subsequences and 5 variance value intervals, the distribution of the variances of these 100 data subsequences in these 5 variance value intervals satisfies: there are 20 data subsequences in the first variance value interval, 30 data subsequences in the second variance value interval, 20 data subsequences in the third variance value interval, 15 data subsequences in the fourth variance value interval, and 15 data subsequences in the fifth variance value interval. Then, the variance distribution characteristics of the data subsequences falling in the first variance value interval are all 20, the variance distribution characteristics of the data subsequences falling in the second variance value interval are all 30, the variance distribution characteristics of the data subsequences falling in the third variance value interval are all 20, the variance distribution characteristics of the data subsequences falling in the fourth variance value interval are all 15, and the variance distribution characteristics of the data subsequences falling in the fifth variance value interval are all 15.

[0086] In the embodiment of the present disclosure, the variance distribution characteristics of the data subsequences are determined through steps a to d, and the weights of the respective data subsequences are determined based on the variance distribution characteristics of the data subsequences. This can better characterize the different effects of different data subsequences on the overall feature vector, further enhance the representation capability of the overall feature vector, and thus further improve the accuracy of the prediction results of the computing power status data.

[0087] After the variance distribution characteristics of each data subsequence are obtained, the weight of each data subsequence is determined according to the variance distribution characteristics of each data subsequence. In some embodiments, the weight of each data subsequence is negatively correlated with the variance distribution characteristics of each data subsequence.

[0088] In some examples, the weight of each data subsequence is in a negative exponential relationship with the variance distribution characteristic of each data subsequence. For example, the weight of each data subsequence is calculated according to the following formula:

[0089]

[0090] in, Represents a data subsequence The weight of α represents the tuning coefficient, ρl Represents a data subsequence The variance distribution characteristics, such as ρ l The specific value of The number of data subsequences in the lth variance value interval where the variance of is located.

[0091] In some embodiments of the present disclosure, the variance distribution characteristics of a data subsequence (e.g., the number of data subsequences in the variance value interval in which the variance of the data subsequence is located) can reflect the amount of information carried by the data subsequence. The smaller the variance distribution characteristics of a data subsequence, the more information the data subsequence carries. By making the weight of a data subsequence negatively correlated with the variance distribution characteristics of the data subsequence, the weight of the data subsequence with the amount of information can be increased, thereby increasing the representation ability of the overall feature vector obtained by weighted fusion, and improving the accuracy of the prediction results of the computing power status data.

[0092] In some embodiments, after obtaining the local feature vectors of each data subsequence, the overall feature vector is obtained according to the following formula:

[0093]

[0094] Among them, h n The overall feature vector representing the computing power status data in the first time period,

[0095] Represents a data subsequence The weight of represents the local feature vector of the ith data subsequence, Q n Indicates the number of data subsequences.

[0096] In the embodiments disclosed herein, on the one hand, by dividing the computing power status data within the first time period into multiple data subsequences and using an autoencoder to encode each data subsequence separately, the processing efficiency in the prediction of the computing power status data can be improved compared to encoding the computing power status data within the first time period as a whole; on the other hand, by determining the weight of each data subsequence according to the distribution characteristics of each data subsequence and weightedly fusing the local feature vectors of each data subsequence according to the weight of each data subsequence, the representation ability of the overall feature vector obtained by fusion can be enhanced, thereby helping to improve the accuracy of the prediction of the computing power status data.

[0097] Figure 3 The figure is a schematic diagram of the principle of weighted fusion of local feature vectors according to some embodiments of the present disclosure.

[0098] like Figure 3As shown, in some implementations of the present disclosure, the process of weighted fusion of local feature vectors includes steps 310 to 330.

[0099] In step 310, a variance distribution histogram is determined according to the variance of each data subsequence.

[0100] For example, multiple variance value intervals are determined according to the variance of each data subsequence, thereby determining the horizontal coordinates of each rectangle in the variance distribution histogram, and the number of data subsequences falling within each variance value interval is counted, thereby determining the vertical coordinates of each rectangle in the variance distribution histogram.

[0101] For example, the number of data subsequences falling within each variance value interval is used as the ordinate of the rectangle corresponding to the variance value interval.

[0102] For example, the ratio of the number of data subsequences falling within each variance value interval to the number of all data subsequences is used as the ordinate of the rectangle corresponding to the variance value interval.

[0103] In step 320, the weight of each data subsequence is determined using a weight function based on the variance histogram. In some embodiments, the ordinate of the rectangle where each data subsequence is located in the variance histogram is substituted into the weight function to calculate the weight of each data subsequence. In step 330, the local feature vectors of each data subsequence are fused based on the weight of each data subsequence to obtain an overall feature vector.

[0104] For example, the local feature vector of each data subsequence is expressed as These subsequences are superimposed according to the weights to obtain the overall feature vector h n .

[0105] In the embodiments of the present disclosure, by fusing the local feature vectors of each data subsequence according to the variance distribution characteristics of the data subsequences, the representation capability of the overall feature vector after fusion can be improved, which in turn helps to improve the accuracy of the computing power status data prediction results. As a result, when the user's computing power demand changes or new computing power demand arises, closed-loop control of computing power resources can be performed based on the computing power status data prediction results to better meet user needs.

[0106] Figure 4 FIG. 1 is a flow chart of a method for determining computing power status data according to other embodiments of the present disclosure. Figure 4 As shown, the method for determining computing power status data includes steps 410 to 460.

[0107] In step 410, the communication network orchestrator sends a subscription request to the computing network orchestrator.

[0108] In some embodiments, the communication network orchestrator initiates a subscription request to the computing power network orchestrator. After verifying that the subscription request is legitimate, the computing power network orchestrator sends a response to the communication network orchestrator.

[0109] Among them, the subscription request instructs the communication network orchestrator to subscribe to the computing power status data.

[0110] In step 420, the communication network orchestrator establishes a long connection with the computing network orchestrator.

[0111] In some embodiments, after receiving a response to a subscription request from the computing power network orchestrator, the communication network orchestrator establishes a long connection with the computing power network orchestrator.

[0112] In step 430, the computing power network orchestrator sends the computing power status data to the communication network orchestrator.

[0113] In some embodiments, the computing power network orchestrator periodically reports computing power status data to the communication network orchestrator. For example, every 1 second, the computing power network orchestrator sends the available amount of the GPU at that moment to the communication network orchestrator.

[0114] In step 440, the communication network orchestrator sends the computing power status data to a device for determining the computing power status data.

[0115] In some embodiments, the communication network orchestrator transmits the received computing power status data to a device for determining the computing power status data (also referred to as a data collection and analysis device or a data collection and analysis module).

[0116] In step 450, the device for determining computing power status data performs computing power resource prediction.

[0117] In some embodiments, the device for determining computing power status data intercepts computing power status data within a first time period from the received computing power status data, and predicts computing power status data within a second time period based on the method for determining computing power status data in the aforementioned embodiments of the present disclosure.

[0118] For example, the computing power status data in the first time period is the GPU availability per second in the past 60 seconds, and the computing power status data in the second time period is the GPU availability per second in the next 60 seconds.

[0119] In step 460, the device that determines the computing power status sends the computing power resource prediction result to the communication network orchestrator.

[0120] In some embodiments, after obtaining the computing power resource prediction result (e.g., computing power status data within the second time period), the device for determining the computing power status sends the computing power resource prediction result to the communication network orchestrator actively or in response to a request from the communication network orchestrator.

[0121] In the disclosed embodiment, the above method is used to realize the prediction of computing power status data. In this method, the computing power status data of the past or current moment provided by the computing power network can be obtained in a timely manner through the coordination between the communication network orchestrator and the computing power network orchestrator. Furthermore, by using the method for determining computing power status data as described above, the computing power status data of the past or current moment is processed, and the prediction results of the computing power status data of the future moment can be obtained accurately and efficiently, so as to ensure computing power when the user computing power demand changes or new computing power demand is generated.

[0122] Figure 5 FIG. 1 is a schematic diagram of a structure of an apparatus for determining computing power status data according to some embodiments of the present disclosure. Figure 5 As shown, the device 500 for determining computing power status data includes an acquisition module 510, an encoding module 520 and a determination module 530.

[0123] The acquisition module 510 is configured to acquire computing power status data within a first time period.

[0124] The encoding module 520 is configured to encode the computing power status data in the first time period according to the autoencoder to obtain an overall feature vector of the computing power status data in the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data in the first time period.

[0125] In some embodiments, the encoding module 520 is based on Figure 2 The process shown obtains the overall feature vector.

[0126] The determination module 530 is configured to process the overall feature vector according to the machine learning model to obtain the computing power status data within a second time period, and the second time period is a time period after the first time period.

[0127] In the embodiments of the present disclosure, the above device can improve the prediction efficiency of computing power status data while ensuring the prediction effect of computing power status data.

[0128] Figure 6 FIG. 1 is a schematic diagram of a system for determining computing power status data according to some embodiments of the present disclosure. Figure 6 As shown, the system 600 for determining computing power status data includes a communication network orchestrator 610 and a device 620 for determining computing power status data.

[0129] The communication network orchestrator 610 is configured to send the computing power status data within the first time period to the device 620 for determining the computing power status data.

[0130] The device 620 for determining the computing power status data is configured to encode the computing power status data in the first time period according to the autoencoder to obtain an overall feature vector of the computing power status data in the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data in the first time period.

[0131] In some embodiments, the device 620 for determining the computing power status data is based on Figure 2 The process shown encodes the computing power status data within the first time period to obtain an overall feature vector.

[0132] The device 620 for determining computing power status data is further configured to process the overall feature vector according to the machine learning model to obtain computing power status data in a second time period, wherein the second time period is a time period after the first time period.

[0133] In some embodiments, the communication network orchestrator 610 is further configured to, before sending the computing power status data within the first time period to the device 620 for determining the computing power status data, send a subscription request to the computing power network orchestrator, the subscription request being used to indicate a subscription to the computing power status data; establish a long connection with the computing power network orchestrator; based on the long connection, receive the computing power status data reported by the computing power network orchestrator; and obtain the computing power status data within the first time period from the reported computing power status data.

[0134] In the disclosed embodiment, the prediction of computing power status data is realized through the above system. In this system, the computing power status data of the past or current moment provided by the computing power network can be obtained in a timely manner through the coordination of the communication network orchestrator and the computing power network orchestrator. Furthermore, according to the computing power status data of the past or current moment, the autoencoder is used for dimensionality reduction encoding to obtain the overall feature vector, which can efficiently and accurately predict the computing power status data at the future moment, and then can provide computing power guarantee when the user computing power demand changes or new computing power demand is generated.

[0135] Figure 7 It is a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure.

[0136] like Figure 7 As shown, the electronic device 700 includes a memory 710; and a processor 720 coupled to the memory 710. The memory 710 is used to store instructions for executing the method for determining computing power status data corresponding to the embodiment. The processor 720 is configured to execute the method for determining computing power status data in any of the embodiments of the present disclosure based on the instructions stored in the memory 710.

[0137] Figure 8 A schematic diagram of the structure of a computer system according to some embodiments of the present disclosure.

[0138] like Figure 8 As shown, the computer system 800 may be embodied in the form of a general-purpose computing device. The computer system 800 includes a memory 810, a processor 820, and a bus 830 that connects various system components.

[0139] The memory 810 may include, for example, a system memory, a non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application, a boot loader, and other programs. The system memory may include a volatile storage medium, such as a random access memory (RAM) and / or a cache memory. The non-volatile storage medium may store, for example, instructions of a corresponding embodiment of at least one method for determining computing power status data being executed. The non-volatile storage medium includes, but is not limited to, a disk storage, an optical storage, a flash memory, and the like.

[0140] The processor 820 can be implemented by a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistors, etc. Discrete hardware components. Accordingly, each module such as an acquisition module, an encoding module, and a determination module can be implemented by a central processing unit (CPU) running instructions in a memory that execute corresponding steps, or can be implemented by a dedicated circuit that executes corresponding steps.

[0141] The bus 830 may use any of a variety of bus architectures, including, but not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, and a Peripheral Component Interconnect (PCI) bus.

[0142] The computer system 800 interfaces 840, 850, 860, the memory 810 and the processor 820 can be connected via a bus 830. The input / output interface 840 can provide a connection interface for input / output devices such as a display, a mouse, and a keyboard. The network interface 850 provides a connection interface for various networked devices. The storage interface 860 provides a connection interface for external storage devices such as a floppy disk, a USB flash drive, and an SD card.

[0143] Here, various aspects of the present disclosure are described with reference to flowcharts and / or block diagrams of methods, devices, and computer program products according to embodiments of the present disclosure. It should be understood that each frame of the flowchart and / or block diagram and the combination of frames can be implemented by computer-readable program instructions.

[0144] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable device to produce a machine, so that the processor executes the instructions to produce means for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0145] These computer-readable program instructions may also be stored in a computer-readable memory, which cause the computer to work in a specific manner to produce an article of manufacture, including instructions for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0146] The present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.

[0147] By using the method, device and system for determining computing power status data in the above embodiments, it is possible to accurately and efficiently predict computing power status data in a future time period.

[0148] So far, the method, device and system for determining computing power status data according to the present disclosure have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed herein.

Claims

1. A method for determining computing power status data, comprising: Obtain computing power status data within a first time period; Encoding the computing power status data within the first time period according to the autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; The overall feature vector is processed according to the machine learning model to obtain computing power status data in a second time period, where the second time period is a time period after the first time period.

2. The method for determining computing power status data according to claim 1, wherein: The computing power status data in the first time period includes a plurality of data subsequences, and encoding the computing power status data in the first time period according to the autoencoder to obtain an overall feature vector includes: Encoding each data subsequence in the multiple data subsequences according to the autoencoder to obtain a local feature vector of each data subsequence; The local feature vectors of all the data subsequences in the multiple data subsequences are fused to obtain the overall feature vector.

3. The method for determining computing power status data according to claim 2, wherein: The fusing the local feature vectors of all the data subsequences in the multiple data subsequences to obtain the overall feature vector comprises: Determining a weight of each data subsequence in the multiple data subsequences according to distribution characteristics of the multiple data subsequences; According to the weights of all the data subsequences in the multiple data subsequences, weighted fusion is performed on the local feature vectors of all the data subsequences in the multiple data subsequences to obtain the overall feature vector.

4. The method for determining computing power status data according to claim 3, wherein: The distribution characteristics of the multiple data subsequences are variance distribution characteristics of the multiple data subsequences.

5. The method for determining computing power status data according to claim 4, wherein: The variance distribution characteristics of the multiple data subsequences are determined in the following manner: determining a variance of each data subsequence in the plurality of data subsequences; Determining multiple variance value intervals according to the variances of all data subsequences in the multiple data subsequences; Determine the number of data subsequences in each variance value interval in the plurality of variance value intervals; The number of data subsequences in the variance value interval of each data subsequence in the multiple data subsequences is used as the variance distribution feature of each data subsequence.

6. The method for determining computing power status data according to claim 5, wherein: The weight of each data subsequence is negatively correlated with the variance distribution characteristics of each data subsequence.

7. The method for determining computing power status data according to claim 6, wherein: The weight of each data subsequence is in an exponential relationship with the variance distribution characteristic of each data subsequence.

8. The method for determining computing power status data according to claim 2, wherein: The number of data contained in each data subsequence of the multiple data subsequences is equal.

9. The method for determining computing power status data according to any one of claims 1 to 8, wherein: The computing power status data includes the available computing power resources of one or more computing power resources or the used computing power resources of one or more computing power resources.

10. The method for determining computing power status data according to any one of claims 1 to 8, wherein: The machine learning model includes a deep neural network, a support vector machine or a gradient boosting decision tree.

11. The method for determining computing power status data according to any one of claims 1 to 8, further comprising: Before encoding the computing power status data within the first time period according to the autoencoder, the autoencoder and the machine learning model are trained based on historical computing power status sample data.

12. A device for determining computing power status data, comprising: An acquisition module is configured to acquire computing power status data within a first time period; an encoding module, configured to encode the computing power status data within the first time period according to the autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; A determination module is configured to process the overall feature vector according to a machine learning model to obtain computing power status data within a second time period, where the second time period is a time period after the first time period.

13. A system for determining computing power status data, comprising: A communication network orchestrator is configured to send computing power status data within a first time period to a device for determining computing power status data; The device for determining computing power status data is configured to: Encoding the computing power status data within the first time period according to the autoencoder to obtain an overall feature vector of the computing power status data within the first time period, wherein the dimension of the overall feature vector is smaller than the dimension of the computing power status data within the first time period; The overall feature vector is processed according to the machine learning model to obtain computing power status data in a second time period, where the second time period is a time period after the first time period.

14. The system for determining computing power status data according to claim 13, wherein: The communication network orchestrator is further configured to: Send a subscription request to the computing power network orchestrator, where the subscription request is used to indicate a subscription to computing power status data; Establishing a persistent connection with the computing power network orchestrator; Based on the long connection, receiving computing power status data reported by the computing power network orchestrator; Obtain computing power status data within a first time period from the reported computing power status data.

15. An electronic device, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the method for determining computing power status data as described in any one of claims 1 to 11 based on instructions stored in the memory.

16. A computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the method for determining computing power status data as described in any one of claims 1 to 11.

Images