CN111930524A

CN111930524A - Method and system for distributing computing resources

Info

Publication number: CN111930524A
Application number: CN202011073741.4A
Authority: CN
Inventors: 卢国鸣
Original assignee: Shanghai Xingrong Information Technology Co ltd
Current assignee: Xingrong Shanghai Information Technology Co ltd
Priority date: 2020-10-10
Filing date: 2020-10-10
Publication date: 2020-11-13
Anticipated expiration: 2040-10-10
Also published as: CN111930524B

Abstract

The embodiment of the application discloses a method and a system for allocating computing resources, wherein the method comprises the following steps: acquiring analysis data at N time points in an application scene of computing resources to be determined; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located. Inputting the analysis data into one or more trained resource allocation models, and predicting one or more resource allocation demand types at the N +1 time point. Determining resource allocation instructions based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

Description

Method and system for distributing computing resources

Technical Field

The present application relates to the field of communications, and in particular, to a method and system for allocating computing resources.

Background

With the improvement of public service capability, intelligent equipment is equipped in more and more public places (such as airports, stations, shopping malls and the like) to provide services for workers. As the traffic in public places changes dynamically, the demand for public services also changes. For example, the greater the traffic, the greater the demand for public services, and the more computing resources are required to support the provision of public services. In many application scenarios, there is a problem of how to allocate reasonable resources for dynamic people flow.

Accordingly, a method and system for computing resource allocation is needed.

Disclosure of Invention

One aspect of the present application provides a method of computing resource allocation, the method comprising: acquiring analysis data at N time points in an application scene of computing resources to be determined, wherein N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located; acquiring the user position characteristics based on Bluetooth beacon equipment; acquiring the user attribute features and the number features based on a reflected signal of a laser beam emitted by a laser radar and/or based on frame data extracted from a video frame; acquiring the network connection quantity characteristic based on wifi connection data; acquiring the user behavior characteristics based on a preset buried point; inputting the analysis data into one or more trained resource allocation models, and predicting one or more resource allocation demand types at the N +1 time point; the resource allocation model predicts one or more resource allocation demand types at the time point of N +1, and comprises the following steps: the resource allocation model comprises a feature vector extraction submodel and a resource allocation demand forecasting submodel; acquiring feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data; and the resource allocation demand forecasting sub-model outputs the resource allocation demand type based on the characteristic vector. Determining resource allocation instructions based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

Another aspect of the application provides a system for computing resource allocation, the system comprising: the acquisition module may be configured to acquire analysis data at N time points in an application scenario of a computing resource to be determined, where N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located; acquiring the user position characteristics based on Bluetooth beacon equipment; acquiring the user attribute features and the number features based on a reflected signal of a laser beam emitted by a laser radar and/or based on frame data extracted from a video frame; acquiring the network connection quantity characteristic based on wifi connection data; and acquiring the user behavior characteristics based on a preset buried point. The prediction module can be used for inputting the analysis data into one or more trained resource allocation models and predicting one or more resource allocation demand types at the time point of N + 1; the resource allocation model predicts one or more resource allocation demand types at the time point of N +1, and comprises the following steps: the resource allocation model comprises a feature vector extraction submodel and a resource allocation demand forecasting submodel; acquiring feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data; and the resource allocation demand forecasting sub-model outputs the resource allocation demand type based on the characteristic vector. A determining module operable to determine resource allocation instructions based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

Another aspect of the present application provides an apparatus for computing resource allocation, comprising at least one storage medium and at least one processor; the at least one storage medium is configured to store computer instructions; the at least one processor is configured to execute the computer instructions to implement a method of computing resource allocation.

Another aspect of the application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement a method of computing resource allocation.

Drawings

The present application will be further explained by way of exemplary embodiments, which will be described in detail by way of the accompanying drawings. These embodiments are not intended to be limiting, and in these embodiments like numerals are used to indicate like structures, wherein:

FIG. 1 is a schematic diagram of an application scenario of a computing resource allocation system according to some embodiments of the present application;

FIG. 2 is an exemplary flow diagram of a computing resource allocation method according to some embodiments of the present application;

FIG. 3 is an exemplary block diagram of a resource allocation model according to some embodiments of the present application.

Detailed Description

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings used in the description of the embodiments will be briefly introduced below. It is obvious that the drawings in the following description are only examples or embodiments of the application, from which the application can also be applied to other similar scenarios without inventive effort for a person skilled in the art. Unless otherwise apparent from the context, or otherwise indicated, like reference numbers in the figures refer to the same structure or operation.

It should be understood that "system", "device", "unit" and/or "module" as used in this application is a method for distinguishing different components, elements, parts or assemblies at different levels. However, other words may be substituted by other expressions if they accomplish the same purpose.

As used in this application and the appended claims, the terms "a," "an," "the," and/or "the" are not intended to be inclusive in the singular, but rather are intended to be inclusive in the plural unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that steps and elements are included which are explicitly identified, that the steps and elements do not form an exclusive list, and that a method or apparatus may include other steps or elements.

Flow charts are used herein to illustrate operations performed by systems according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in the exact order in which they are performed. Rather, the various steps may be processed in reverse order or simultaneously. Meanwhile, other operations may be added to the processes, or a certain step or several steps of operations may be removed from the processes.

FIG. 1 is a schematic diagram of an application scenario of a computing resource allocation system according to some embodiments of the present application.

As shown in fig. 1, the application scenario 100 to which the present application relates may include a first computing system 130 and/or a second computing system 160.

The first computing system 130 may be configured to automatically determine the resource type of the resource allocation requirement and issue a resource allocation instruction, such as resource allocation instruction 140. The first computing system 130 can be applied to any given scenario, such as public occasions like airports, stations, shopping malls, etc., or to cells, schools, etc., or to specific occasions with any given range size. When the flow of people in an application situation changes (such as the number of people increases), the demands for computing resources (for example, container resources for supporting service operation) and service resources (for example, the number of staff, the situation attention degree) required by the service equipment in the application situation also change (for example, increase), and the first computer system 130 can predict the demands for various resources in the application situation at the next time through the resource allocation model based on the analysis data 120 and issue the resource allocation instruction 140, so as to avoid the situations of resource waste and resource shortage in the application situation.

The first computing system 130 may obtain the analysis data 120, where the analysis data 120 includes user location characteristics, user attribute characteristics, people number characteristics, network connection number characteristics, and user behavior characteristics of a target user in a preset size area where a terminal using the computing resource is located. The analytical data 120 may be acquired by the acquisition means 110. The obtaining means 110 may include various means for obtaining data, for example, the user attribute feature and the people number feature may be obtained through the laser radar 110-1 and the video monitor 110-2; the user position characteristics can be acquired through the Bluetooth beacon device 110-3; the analysis data 120 may also be obtained in other manners, for example, the network connection quantity characteristic may be obtained through a wifi device (not shown), and the user behavior characteristic may be obtained through a preset buried point (not shown). The analytical data 120 may enter the first computing system 130 in a variety of common ways. Through the model 132 in the first computing system 130, the resource allocation requirement type may be output, and the first computing system 130 may in turn determine the resource allocation instructions 140 according to the resource allocation requirement type. The first computing system 130 may adjust the resources for the instance using the computing resources according to the determined resource allocation instructions, e.g., allocate more container resources for supporting various services (e.g., computing services, push services, etc.) for the instance; more staff are called for the occasion, etc.

The parameters of the model 132 may be obtained by training. The second computing system 160 may obtain multiple sets of sample data 150, where each set of training samples includes analysis data at N time points, for example, user location characteristics, user data characteristics, people number characteristics, network connection number characteristics, user behavior characteristics, and the like at the N time points. The second computing system 160 updates the parameters of the model 162 with the sets of sample data 150 to obtain a trained model. The parameters of the model 132 are derived from the trained model 162. Wherein the parameters may be communicated in any common manner.

A model (e.g., model 132 or/and model 162) may refer to a collection of several methods performed based on a processing device. These methods may include a number of parameters. When executing the model, the parameters used may be preset or may be dynamically adjusted. Some parameters may be obtained by a trained method, and some parameters may be obtained during execution. For a detailed description of the model referred to in this application, reference is made to the relevant part of the application.

The first computing system 130 and the second computing system 160 may be the same or different. The first computing system 130 and the second computing system 160 refer to systems with computing capability, and may include various computers, such as a server and a personal computer, or may be computing platforms formed by connecting a plurality of computers in various structures.

Processing devices may be included in first computing system 130 and second computing system 160, and may execute program instructions. The Processing device may include various common general purpose Central Processing Units (CPUs), Graphics Processing Units (GPUs), microprocessors, application-specific integrated circuits (ASICs), or other types of integrated circuits.

First computing system 130 and second computing system 160 may include storage media that may store instructions and may also store data. The storage medium may include mass storage, removable storage, volatile read-write memory, read-only memory (ROM), and the like, or any combination thereof.

The first computing system 130 and the second computing system 160 may also include a network for internal connections and connections with the outside. Terminals for input or output may also be included. The network may be any one or more of a wired network or a wireless network.

For more details on the analysis data and the resource allocation model, reference may be made to the related descriptions of fig. 2 to fig. 3, which are not described herein again.

In some embodiments, an acquisition module, a prediction module, and a determination module may be included in the system (e.g., first computing system 130 or second computing system 160).

The acquisition module may be configured to acquire analysis data at N time points in an application scenario of a computing resource to be determined, where N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located; acquiring the user position characteristics based on Bluetooth beacon equipment; acquiring the user attribute features and the number features based on a reflected signal of a laser beam emitted by a laser radar and/or based on frame data extracted from a video frame; acquiring the network connection quantity characteristic based on wifi connection data; and acquiring the user behavior characteristics based on a preset buried point. For more details, reference may be made to the related description of step 202, which is not repeated here.

The prediction module can be used for inputting the analysis data into one or more trained resource allocation models and predicting one or more resource allocation demand types at the time point of N + 1; the resource allocation model predicts one or more resource allocation demand types at the time point of N +1, and comprises the following steps: the resource allocation model comprises a feature vector extraction submodel and a resource allocation demand forecasting submodel; acquiring feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data; and the resource allocation demand forecasting sub-model outputs the resource allocation demand type based on the characteristic vector. For more details, reference may be made to the related description of step 204, which is not repeated herein.

A determining module operable to determine resource allocation instructions based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction. For more details, reference may be made to the description related to step 206, which is not described herein again.

FIG. 2 is an exemplary flow diagram of a computing resource allocation method according to some embodiments of the present application. As shown in fig. 2, the process 200 may include:

in some embodiments, the first computing system 130 may define a plurality of processing layers by the neural network model, respectively process a plurality of features of the analysis data to obtain a plurality of processing results, then fuse the plurality of processing results, output a predicted resource allocation demand type, and further determine the resource allocation instruction according to the resource demand type.

In some embodiments, the first computing system 130 may determine the number of times the plurality of features of the analysis data are acquired based on the time intervals and the frequency at which the plurality of features are acquired.

Step 202, acquiring analysis data at N time points in an application scene of computing resources to be determined; wherein N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located.

In particular, step 202 may be performed by a prediction module.

In some embodiments, a computing resource may refer to a container resource used by a container (docker) deployed in a terminal/device at runtime. The container resource refers to a hardware resource, such as a memory, a central processing unit, and the like, required by the docker container during operation after the application and the runtime library are packaged by the docker container.

The time point is a certain time, and the N time points are N times after the resource allocation demand type is determined last time. N is an integer greater than 0, e.g., 5, 8, 10, etc. The time point may be an instant, or may be a time period having a certain time length, for example, 0.5 second, 1 second, or the like.

In some embodiments, the interval time between the N time points may be the same. For example, every 300 seconds is a time point. In some embodiments, the N time points may be separated by different time intervals, for example, the intervals between the time points are continuously increased or decreased, or any random time intervals, etc.

In some embodiments, the interval between the N time points may be determined based on the distribution characteristics of the human traffic over time, for example, the time point interval of the human traffic peak period may be 120 seconds, and the time point interval of the human traffic valley period may be 600 seconds.

In some embodiments, the N time points may be obtained by manually presetting the interval time. And the interval time can be automatically adjusted by an algorithm to obtain N time points based on the characteristics of the number of people obtained at the last time point. The N time points may also be obtained in other manners, and this embodiment is not limited.

The analysis data may refer to one or more types of data used to determine resource allocation requirements. The analysis data may include user location characteristics, user attribute characteristics, population characteristics, network connection number characteristics, and user behavior characteristics of the target user in a preset size area where the terminal using the computing resource is located. The terminal may be a smart mobile device (e.g., a cell phone, tablet, etc.), a laptop, a computer or a server, etc. The preset-size area may be a certain area of any size previously specified by the user, for example, a certain area of 100 square meters may be specified as the preset-size area; a certain area with a limited range size having specific attributes may also be designated as a preset size, for example, a station, a mall, a school, a cell or a shop where the terminal is located, a 100-meter square circle with the terminal as a center, and the like, which is not limited in this embodiment. The target user may refer to an active person in the preset size area, for example, all passengers in a station, a shopper in a mall, and a worker in the corresponding area.

The user location feature may refer to a location of a target user located in the preset size area, for example, in an XX train station, the user is located at a north station entrance of the train station; for another example, in the XX mall, the target users are located in the fresh area, the vegetable area, and the like.

In some embodiments, the user location characteristics may be obtained based on a bluetooth beacon device. The Bluetooth beacon device not only can acquire the position of a user by transmitting a signal through the base station, but also can send a push message to a target user. For example, when a target user carries a mobile device and enters a signal coverage range of a bluetooth beacon device, a corresponding program of the bluetooth beacon device can actively prompt the user whether to access a network, send a notification to the user, introduce product information, preferential information and the like.

In some embodiments, the user location characteristics may also be obtained based on wifi devices. The Wifi signal can be along with keeping away from the distance of Wifi equipment and constantly attenuate, consequently, can be based on the signal strength that contains in the data packet that the signal source (Wifi equipment) sent, along with mobile terminal's removal, the principle that signal strength numerical value is changing always realizes obtaining user position characteristic based on Wifi equipment. For example, a preset size area may be divided into a map with a number of grids, and a plurality of wifi signal sources with different positions are set, each signal source periodically emitting a wireless signal. When the target user is located in a certain cell in the preset size area, the mobile terminal carried by the target user can receive wifi signals sent by each signal source, the strength of the wireless signals sent by each signal source reaching the target user is different, namely, the wifi signal strength value of each signal source, which can be acquired by each map cell, is unique, and therefore the user position characteristics can be accurately determined based on the signal strength of each signal source received by the mobile terminal.

In some embodiments, the user location characteristics may be obtained based on the bluetooth beacon device and the wifi device simultaneously.

The user attribute feature may refer to an own attribute that the target user has. Such as the target user's age, gender, height, weight, whether it is a potential customer of a particular service, etc.

The population characteristic may refer to the number of target users and/or workers located within the preset size area. For example, the target user 20, the staff member 30, the total number of people 50, etc.

In some embodiments, the user attribute features and the people number features may be obtained based on a reflected signal of a laser beam emitted by a laser radar, and/or based on frame data extracted from a video frame.

The laser radar is a radar system for detecting a target object by emitting a laser beam, and includes a laser emitting system, a laser receiving system, and an information processing system. The target objects are persons (such as target users) and objects (such as commodities in shopping malls) in the detection range of the laser radar, and can reflect laser beams emitted by the laser radar. The laser beam is an optical signal emitted by the laser emitting system in real time.

In some embodiments, the lidar may include, but is not limited to: mechanical lidar, solid state lidar, Micro-Electro-Mechanical systems (MEMS) lidar, Flash area Array lidar, Optical Phased Array (OPA) solid state lidar, and hybrid solid state lidar.

In a laser emission system, an excitation source periodically drives a laser to emit laser pulses, and a laser modulator emits one or more laser beams to a target object by controlling the direction and the number of lines of the emitted laser beams through a beam controller. In some embodiments, the laser scanning System may control the direction of the laser beam by moving one or a combination of a laser emission System, a Micro-Electro-Mechanical System (MEMS) Micro-mirror, and an Optical Phased Array (OPA) to sequentially emit the laser beam to the target object in multiple regions. Such as mechanical lidar, solid state lidar, MEMS lidar and OPA solid state lidar. In some embodiments, the laser emitting system can also emit laser beams to the respective areas simultaneously to achieve light coverage. Such as Flash area array lidar.

The signal reflected by the laser beam includes a laser beam round trip time, a laser beam frequency change, a laser beam signal attenuation degree, a laser beam angle and the like. In some embodiments, the lidar may receive the laser beam via a laser receiving system and generate a reflected signal. For example, the acquisition module may acquire signals reflected by a laser beam of the lidar at N time points: signal of laser beam emission at 1 st time point, signal of laser beam emission at 2 nd time point, signal of laser beam emission at … … nth time point.

In some embodiments, a laser beam may be emitted using a laser radar, and characteristic data of a plurality of targets may be acquired based on a reflected signal of the laser beam.

The characteristic data of the object may include shape, absorption rate and direction. Taking the feature data "shape" of the target object as an example, in the known feature data of the human body, the range of the shape may be: the height is 0.5-2.5 m, the width is 0.2-1 m, the shape of the target object is 1.8 m, the width is 0.2-0.5 m, and the corresponding probability of the shape is 1; similarly to the "shape", the probabilities corresponding to the "absorptance" and the "direction" are 0.7 and 0.8, respectively; multiplying 1, 0.7 and 0.8 to obtain the probability that the object is a human being, which is 0.56, and is greater than the threshold value of 0.5, the object can be confirmed to be a human being.

Further, the user attribute features and the number of people features can be determined based on the feature data of the plurality of target objects by using the trained first user recognition model.

In some embodiments, the first user identification model may be a neural network model. The feature data of the target object may be input into a first user recognition model, which outputs the user attribute features and the demographic features. Specifically, the first recognition model may map the input feature data of the target object to a probability that the target object is confirmed as a person, and may determine whether the target object is a person based on the probability.

In some embodiments, the first user identification model may also be another model, such as a logistic regression model, and the like, and the embodiment is not limited thereto.

Further, the first user recognition model may count the number of the target objects confirmed as the person, to obtain the population characteristics. For example, the first user identification model identifies 250 persons in the multiple objects based on the feature data of the multiple objects at the 1 st time point, and the first person feature at the 1 st time point is 250. The first user identification model may obtain user attribute characteristics of the target user based on the characteristic data of the target object. For example, the first user identification model may determine user attribute characteristics of the target user based on a plurality of characteristics of the target object, such as height, weight, hair length, clothing color, shape, and the like. For example, if the target user is 160cm in height, 50cm in hair length, orange-red in clothes, and in conformity with female characteristics in body shape, it may be determined that the target user is female in gender.

In some embodiments, the first user recognition model may be trained based on a number of training samples with identifications. Specifically, a training sample with an identifier may be input into the first user recognition model, and parameters of the first user recognition model may be updated through training.

In some embodiments, the training sample may be feature data of the target object. The identification may be a user attribute feature.

In some embodiments, the identification of the training samples may be obtained by manual labeling.

In some embodiments, frame data may be extracted based on the video data; and extracting at least one target object in the frame data by using the trained second user identification model, and determining the user attribute characteristics and/or the people number characteristics based on the at least one target object.

Video data is a moving image recorded as an electrical signal and composed of a plurality of temporally successive still images. Wherein each still image is a frame of video data. The frame data is one or more frames of still images extracted from the video data. In some embodiments, video data for a point in time may contain multiple still images.

In some embodiments, the format of the video data may include, but is not limited to: one or more combinations of Digital Video Disks (DVDs), streaming Media formats (Flash videos, FLVs), Motion Picture Experts Group (MPEG), Audio Video Interleaved (AVI), Video Home Systems (VHS), and Video container file formats (RM). In some embodiments, the obtaining module may obtain the video data by reading shooting data of a monitor or a camera, calling an associated interface, or other means.

For example, the obtaining module may obtain N video data at N time points, for example, the video data at the 1 st time point, the video data at the 2 nd time point, … …, and the video data at the N time point, respectively.

After frame data is extracted from the video data, at least one target object in the frame data can be extracted by using a trained second user identification model, and the user attribute feature and the number of people feature can be determined based on the at least one target object. The target object is an image block corresponding to a person or an object in the frame data. The frame data of N time points, for example, the frame data of the 1 st time point, the frame data of the 2 nd time point, … …, and the frame data of the nth time point, may be extracted based on the video data of N time points.

In some embodiments, the trained second user recognition model may determine whether the target object in the frame data is an image of a person, and determine the user attribute feature and the person number feature based on the determination result. Specifically, the second user identification model may count the number of the target object being a person based on the determination result to obtain a person number characteristic; and extracting image features of the target object from the image of the target object being a person, and obtaining user attribute features based on the image features.

In some embodiments, the second user identification model may include an extraction layer, a feature acquisition layer, a recognition layer, and a fusion layer.

In some embodiments, the extraction layer may extract the plurality of target objects from the frame data by a Sliding-window (multi-scale), Selective Search (Selective Search), neural network, or other method.

For example, the frame data may be a static image of 200 × 200 pixels, and the frame data is first passed through a sliding window of 10 × 10 pixels and slid by step 1 to obtain 190 × 190 image blocks from the frame data; then, through a sliding window of 20 × 20 pixels, sliding is performed by step size 1, and 180 × 180 image blocks are obtained from frame data; … …, respectively; finally 190 x 190+180 x 180+ … … image blocks are obtained. Wherein one or more of the obtained image blocks may contain the target object. It is understood that if a person or an object is included in one frame data, a plurality of target objects may be determined from a plurality of image blocks obtained based on the frame data. For example, a plurality of target objects may be determined from a plurality of image blocks by an image recognition technique (e.g., an image recognition model).

For example, the frame data may be directly cut, and an image portion corresponding to a person or an object included in the frame data may be cut as a target object.

In some embodiments, the feature acquisition layer may acquire a feature vector for each target object. Specifically, the feature acquisition layer acquires a plurality of image features of the target object, and then fuses the plurality of image features to obtain a feature vector of the target object.

In some embodiments, the image features of the target object include, but are not limited to: haar (Harr) features, Histogram of Oriented Gradients (HOG) features, Local Binary Patterns (LBP) features, edge-small (Edgelet) features, Color-Similarity (CSS) features, Integral channel features, and center Transform histograms (CENTRIST) features, among others.

In some embodiments, the feature acquisition layer may be a neural network model. Preferably, the feature acquisition layer is a Convolutional Neural Network (CNN).

In some embodiments, the recognition layer may determine whether the target object is an image containing a complete person based on the feature vector of each target object. In some embodiments, the recognition layer may contain a classification model. Specifically, the input feature vector of each target object is mapped to a numerical value or probability, and whether the target object is an image of a person is determined based on the numerical value or probability.

In some embodiments, the classification model may be, but is not limited to, a support vector machine model, a Logistic regression model, a naive bayes classification model, a gaussian distributed bayes classification model, a decision tree model, a random forest model, a KNN classification model, a neural network model, or the like.

In some embodiments, the fusion layer determines the user attribute features and the demographic features based on the determination of the at least one target object. Illustratively, if 300 target objects in 400 target images recognized by the second user recognition model in the 1 st time point frame data are people, the number of people at the 1 st time point is 300, and the user attribute feature is determined based on the images of the target images. Similarly, the second user recognition model outputs a 2 nd point in time people count feature, … …, the N nd point in time people count feature.

In some embodiments, the second user recognition model may be trained based on a number of training samples with identifications. Specifically, a training sample with the identifier is input into a second user recognition model, and parameters of the second user recognition model are updated through training.

In some embodiments, the training samples may be frame data and the identifications may be user attribute features and demographic features.

In some embodiments, the identifier of the frame data may be obtained by manual tagging.

In some embodiments, the user attribute feature and the people number feature may be obtained by using a laser radar method or a video data-based method, or may be obtained by using both methods. It will be appreciated that the use of both approaches simultaneously may increase processing speed.

The network connection number characteristic may refer to the number of connected wireless networks covered in a preset size area. For example, a mall is covered with a wireless wifi network, and the number of people accessing the wireless wifi network is 100, 200, 300, and the like.

In some embodiments, the network connection quantity characteristic may be obtained based on wifi connection data. The Wifi connectivity data may include the number of current network connected users, Wifi traffic data, Wifi signal patterns (e.g., 2.4G network or 5G network). The obtaining module can directly obtain the network connection quantity characteristics from wifi connection data.

User behavior characteristics may refer to the manner in which a target user behaves when using an application or accessing a web page. User behavior characteristics may include applications (e.g., hundredths, kyoto, WeChat, etc.) accessed by the target user, web pages, operations while accessing (e.g., clicking, closing, favorites, etc.), access times, and so forth.

In some embodiments, the user behavior feature may be obtained based on a preset buried point. A buried point is a data collection method that can be analyzed by a user's website or application. Usage of applications, such as which items a target user browses on a shopping app in a mall, items joined in a shopping cart, items of interest, coupons picked up, etc., can be tracked by the landed points. The user data collected by the embedded point technology may be used to further optimize products or provide data support for operation (for example, in the present application, the data support is used to obtain resource allocation demand types), including access numbers (Visits), Visitor numbers (Visitor), dwell Time (Time On Site), Page view numbers (Page Views), and jump rates (Bounce Rate).

The predetermined buried point may be implemented in various ways. For example, relevant codes can be injected into codes of products during development of the products, and the relevant codes can also be realized through statistical tools, such as alliances, policies, Talkingdata, growth io and the like.

Step 204, inputting the analysis data into the trained one or more resource allocation models, and predicting one or more resource allocation demand types at the N +1 time point.

In particular, step 204 may be performed by a prediction module.

The resource allocation requirement refers to the requirement for various resources when the terminal needing to use the computing resource is located in a preset size area to provide services for the target user. The various resources include container resources (computing resources), human resources, service resources, and the like. As described in step 202, the container resource refers to a hardware resource, such as a memory, a cpu, and the like, required by the docker container during runtime after the application and the runtime library are packaged by the docker container. The human resources are the number of workers for distributing the service provided for the target user only in the preset size area. The service resource refers to the attention degree of a person scheduling and allocating the resource to the preset size area.

The resource allocation requirement types comprise increasing, decreasing, maintaining, withdrawing resources and the like. Such as increasing container (docker) resources, reducing staff, maintaining focus, etc.

The resource allocation model may be a machine learning model. The input of the resource allocation model is the analysis data at N time points, i.e. the analysis data from time point 1 to time point N, and the output is the resource allocation demand type predicted by the resource allocation model at time N + 1. Therefore, the time interval from the time point N to the time point N +1 can be used for allocating the resources required by the time point N +1, for example, when the number of workers needs to be increased or decreased, the workers can be scheduled in advance, so that the resources in the preset size area can meet the requirement at the time point N + 1.

The N +1 time point refers to the next time point after the nth time point.

In some embodiments, the resource allocation model includes a feature vector extraction submodel and a resource allocation demand predictor submodel. For the structure and the training mode of the resource allocation model, reference may be made to the related description of fig. 3, which is not repeated herein.

The resource allocation model predicts one or more resource allocation demand types at the N +1 time point, and comprises the following steps:

and acquiring the feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data.

The feature vector is a vector obtained by converting the analysis data. Specifically, the analysis data may be input to the feature vector extraction submodel, which outputs the feature vector.

In some embodiments, the analysis data at the N time points may be transformed to obtain numerical features before inputting the analysis data to the feature vector extraction submodel. The transformation of the analysis data into numerical features can be performed in a variety of ways, including but not limited to normalization, statistics, discretization, log changes, and higher-order and fourth-order operational features.

Further, a binning operation may be performed on the numerical signature at each time point, resulting in one or more input signature data. The sub-bucket can be used for discretizing numerical variables of the numerical features, and then the numerical features can be converted into input feature data which is easy to utilize by a machine learning model in a one-hot encoding mode by using binarization. The input feature data is input data to be input to the feature vector extraction submodel.

In some embodiments, the manner of binning includes, but is not limited to, equidistant binning, equal frequency binning, model binning, and the like. Taking equidistant sub-bucket as an example, the people number characteristic and the network connection number characteristic in the analysis data can be sub-bucket, for example, the group 1 is 50 or less, the group 2 is 50 to 100, the group 3 is 101-. By bucket separation of numerical type features obtained by converting the analysis data, sparse vectors can be obtained, the operation speed of the feature vector extraction submodel during utilization can be higher, and the robustness on abnormal data can be stronger.

After the sub-bucket operation is completed, the obtained input feature data may be input to the feature vector extraction submodel, and the feature vector extraction submodel outputs the feature vector. For more details of the feature vector extraction submodel, reference may be made to the related description of fig. 3, which is not described herein again.

And the resource allocation demand forecasting sub-model outputs the resource allocation demand type based on the characteristic vector. Specifically, the feature vector is input to the resource allocation demand predictor model, and the resource allocation demand predictor model outputs the resource allocation demand type. For more details on the resource allocation demand forecasting submodel, reference may be made to the related description of fig. 3, which is not described herein again.

In some embodiments, in the structure of one resource allocation model, the resource allocation demand predictor model may be one. Multiple types of resource allocation demand types may be predicted using multiple of the resource allocation models, e.g., resource allocation model 1, resource allocation model 2, … …, resource allocation model n. For example, the resource allocation model 1 is used for predicting the container resource demand type, the resource allocation model 2 is used for predicting the human resource demand type, and the resource allocation model 3 is used for predicting the service resource demand type. Illustratively, the structure of the demand forecast submodel for a resource in the resource allocation model is shown in fig. 3, and more details can be seen in fig. 3.

In some embodiments, in the structure of one resource allocation model, the resource allocation demand prediction submodel may be multiple (not shown), the feature vectors are respectively input to the multiple resource allocation demand prediction submodels, and different resource allocation demand prediction submodels output different resource allocation demand types. For example, the resource allocation demand forecast submodel may include a resource allocation demand forecast submodel 1, a resource allocation demand forecast submodel 2, … …, and a resource allocation demand forecast submodel n. The resource allocation demand forecasting submodel 1 can be used for outputting container resource demand types, the resource allocation demand forecasting submodel 1 can be used for outputting human resource demand types, and the resource allocation demand forecasting submodel 3 can be used for outputting service resource demand types and the like.

Step 206, determining a resource allocation instruction based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

In particular, step 206 may be performed by the determination module.

A container resource allocation instruction may refer to increasing, decreasing, or maintaining a docker resource. For example, when the flow of people in the preset size area where the terminal using the computing resource is located increases, the load of the docker container during operation also increases correspondingly, and the demand for the resource allocated to the docker container by the system increases, then the container resource allocation instruction may be to increase the docker resource.

The staff allocation instruction may be to increase or decrease a certain number of staff in a certain area (e.g., a certain area within a preset size area, or the whole preset size area), or to keep unchanged. For example, when the flow of people in the preset size area increases, the number of workers needs to be increased, and the worker allocation instruction may be to increase 10 workers.

The attention allocation instruction may be to increase, decrease, or maintain the attention to a certain area (e.g., a certain area within a preset-size area, or the entire preset-size area). For example, in step 204, the resource allocation model predicts that the human traffic in the preset size area is relatively stable, and the attention to the human traffic can be reduced for less variation of the resource demand.

The resource allocation model precision adjustment instruction may be an adjustment instruction to the precision of the resource allocation model. For example, the resource allocation model switches the precision of the data type between integer and floating point. For example, in step 204, if the prediction result of the resource allocation model has a relatively large variation, the precision of the model may be converted from floating point (float) type to integer (int) type to increase the processing speed of the model.

In predicting the type of resource allocation requirement, various factors may influence the final result. For example, the number of people is an obvious influence factor, but the real-time number of people cannot actually and accurately reflect the demand of the resource, and the behavior and the number of people may not be real-time in response to the demand of the resource. For example, in a shopping mall area, the flow speed of people may be high, and some people may just pass by the mall and may not select goods, so that the adjustment of various resources is difficult to grasp. And because these pieces of information are mixed, it is difficult to establish a clear rule to obtain a predicted result from each type of information. And by means of a machine learning mode, a predictable model can be formed through automatic data learning, and high accuracy is obtained.

Therefore, the accuracy can be effectively improved by predicting the resource demand according to the pedestrian flow conditions of the N time points. However, the accuracy of counting the people flow situation in a single mode is not sufficient, for example, other factors are difficult to consider, and the situation that multiple factors exist is complicated.

In the embodiment of the application, various forms of analysis data are acquired in various ways, and further, the resource demand type at the N +1 time point is predicted by using the data processing capacity of the machine learning model, so that the resource allocation instruction is determined according to the prediction result, the functions of various information (such as user position characteristics, user attribute characteristics, people number characteristics and the like) are integrated, and the accuracy in resource allocation prediction is effectively improved.

On the other hand, due to the fact that the related information features are more, the adoption of various standard machine learning models can cause the problems that the model parameter quantity is too much, the requirement on the training data quantity is high, overfitting is easy to happen and the like. In some embodiments of the application, a mode of self-defining a model structure is adopted, features of a plurality of time points of a sub-model are extracted through a feature vector to comprehensively extract feature vectors, then the sub-model is predicted through a resource allocation demand prediction sub-model, and finally a resource allocation instruction is determined according to a predicted resource allocation demand type. Compared with the mode of applying machine learning models of various standards, the scheme provided by the application can better adapt to the characteristics of the used information and the problem to be solved, and the problems of low operation efficiency, overlarge training data requirement or overfitting caused by excessive model parameters are avoided.

To sum up, aiming at the problem of computing resource allocation in a complex scene, the scheme provided by the application can more fully acquire data and obtain information helpful for prediction, and a self-defined machine learning model structure is adopted according to the characteristics of the information so as to obtain better operation efficiency and prediction effect.

The resource allocation model may be any existing model or combination of existing models that can implement processing of multiple features, such as CNN (Convolutional Neural Networks), DNN (Deep Neural Networks), LSTM (Bi-Long Short-Term Memory), BI-LSTM (Bi-Long Short-Term Memory), etc. The resource allocation model may also be a model customized according to needs, which is not limited in the embodiment of the present application.

In some embodiments, the feature vector extraction sub-model included in the resource allocation model is a deep neural network model, and the resource allocation demand prediction sub-model is an LSTM model.

As shown in FIG. 3, a feature vector extraction submodel 302 and a resource allocation demand forecasting submodel 304 may be included in the resource allocation demand model 300.

The input of the resource allocation demand model is analysis data at N time points, wherein in the resource allocation demand model, the analysis data at each time point respectively enters different feature vector extraction submodels, for example, the analysis data at a first time point enters DNN-1, the analysis data at a second time point enters DNN-2, … …, the analysis data at an nth time point enters DNN-N, and then the feature vectors at each time point are output by the feature vector extraction submodel, for example, the feature vector of the 1 st time point analysis data is output by DNN-1, the feature vector of the 2 nd time point analysis data is output by DNN-2, and the feature vector of the nth time point analysis data is output by DNN-N; and outputting the characteristic vectors of the analysis data of each time point as the input of a resource allocation demand prediction sub-model, and predicting and outputting the resource allocation demand type at the N +1 time point by the resource allocation demand prediction sub-model according to the input characteristic vectors.

In some embodiments, the types of the plurality of feature vector extraction submodels may be the same. For example, the types of the plurality of feature vector extraction submodels may all be DNN, with only parameters different between the respective models. In some embodiments, the types of the plurality of feature vector extraction submodels may be different. For example, the type of the partial feature vector extraction submodel is DNN, and the type of the partial feature vector extraction submodel is other types of models.

It should be noted that the plurality of feature vector extraction submodels included in the resource allocation model may be structured in parallel. In general, the number of feature vector extraction submodels is the same as the number of time points of the input analysis data, for example, if the number of feature vector extraction submodels is 5, the input analysis data should be 5 time points of analysis data.

In some embodiments, the number of feature vector extraction submodels may also be different from the number of time points.

In some embodiments, the resource allocation model may be trained based on a large number of training samples with identities. Specifically, a training sample with an identifier is input into the resource allocation model, and parameters of the resource allocation model are updated through training.

In some embodiments, historical analysis data may be used as a training sample, and a historical resource allocation requirement type is used as a sample identification.

In some embodiments, the sample identifier may be obtained by labeling in a manual labeling manner according to historical analysis data of historical time points and historical real resource allocation requirements.

In some embodiments, the feature vector extraction submodel and the resource allocation demand prediction submodel may be jointly trained in an end-to-end learning manner to obtain the resource allocation model.

In this embodiment, joint training is performed in an end-to-end learning manner, training of the resource allocation model can be completed only by labeling the historical analysis data once, and when the resource allocation model includes a plurality of sub-model structures, the training data does not need to be labeled for many times, so that the training efficiency of the model can be improved, and the workload for labeling samples can be reduced.

In some embodiments, training may be performed in a conventional manner, such as a gradient descent method.

It should be understood that the internal DNN and LSTM structures of the resource allocation requirement model shown in fig. 3 are merely examples, and the specific composition of the resource allocation model is not limited in this application.

An apparatus for controlling allocation of computing resources of a bandwidth is also provided in an embodiment of the present application, including at least one storage medium and at least one processor; the at least one storage medium is configured to store computer instructions; the at least one processor is configured to execute the computer instructions to implement the aforementioned method of computing resource allocation.

The embodiment of the application also provides a computer readable storage medium. The storage medium stores computer instructions, and after the computer reads the computer instructions in the storage medium, the computer realizes the method for allocating the computing resources.

Having thus described the basic concept, it will be apparent to those skilled in the art that the foregoing detailed disclosure is to be considered merely illustrative and not restrictive of the broad application. Various modifications, improvements and adaptations to the present application may occur to those skilled in the art, although not explicitly described herein. Such modifications, improvements and adaptations are proposed in the present application and thus fall within the spirit and scope of the exemplary embodiments of the present application.

Also, this application uses specific language to describe embodiments of the application. Reference throughout this specification to "one embodiment," "an embodiment," and/or "some embodiments" means that a particular feature, structure, or characteristic described in connection with at least one embodiment of the present application is included in at least one embodiment of the present application. Therefore, it is emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various places throughout this application are not necessarily all referring to the same embodiment. Furthermore, some features, structures, or characteristics of one or more embodiments of the present application may be combined as appropriate.

Moreover, those skilled in the art will appreciate that aspects of the present application may be illustrated and described in terms of several patentable species or situations, including any new and useful combination of processes, machines, manufacture, or materials, or any new and useful improvement thereon. Accordingly, various aspects of the present application may be embodied entirely in hardware, entirely in software (including firmware, resident software, micro-code, etc.) or in a combination of hardware and software. The above hardware or software may be referred to as "data block," module, "" engine, "" unit, "" component, "or" system. Furthermore, aspects of the present application may be represented as a computer product, including computer readable program code, embodied in one or more computer readable media.

The computer storage medium may comprise a propagated data signal with the computer program code embodied therewith, for example, on baseband or as part of a carrier wave. The propagated signal may take any of a variety of forms, including electromagnetic, optical, etc., or any suitable combination. A computer storage medium may be any computer-readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code located on a computer storage medium may be propagated over any suitable medium, including radio, cable, fiber optic cable, RF, or the like, or any combination of the preceding.

Computer program code required for the operation of various portions of the present application may be written in any one or more programming languages, including an object oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C + +, C #, VB.NET, Python, and the like, a conventional programming language such as C, Visual Basic, Fortran2003, Perl, COBOL2002, PHP, ABAP, a dynamic programming language such as Python, Ruby, and Groovy, or other programming languages, and the like. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or processing device. In the latter scenario, the remote computer may be connected to the user's computer through any network format, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet), or in a cloud computing environment, or as a service, such as a software as a service (SaaS).

Additionally, the order in which elements and sequences of the processes described herein are processed, the use of alphanumeric characters, or the use of other designations, is not intended to limit the order of the processes and methods described herein, unless explicitly claimed. While various presently contemplated embodiments of the invention have been discussed in the foregoing disclosure by way of example, it is to be understood that such detail is solely for that purpose and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements that are within the spirit and scope of the embodiments herein. For example, although the system components described above may be implemented by hardware devices, they may also be implemented by software-only solutions, such as installing the described system on an existing processing device or mobile device.

Similarly, it should be noted that in the preceding description of embodiments of the application, various features are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure aiding in the understanding of one or more of the embodiments. This method of disclosure, however, is not intended to require more features than are expressly recited in the claims. Indeed, the embodiments may be characterized as having less than all of the features of a single embodiment disclosed above.

Numerals describing the number of components, attributes, etc. are used in some embodiments, it being understood that such numerals used in the description of the embodiments are modified in some instances by the use of the modifier "about", "approximately" or "substantially". Unless otherwise indicated, "about", "approximately" or "substantially" indicates that the number allows a variation of ± 20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximations that may vary depending upon the desired properties of the individual embodiments. In some embodiments, the numerical parameter should take into account the specified significant digits and employ a general digit preserving approach. Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the range are approximations, in the specific examples, such numerical values are set forth as precisely as possible within the scope of the application.

The entire contents of each patent, patent application publication, and other material cited in this application, such as articles, books, specifications, publications, documents, and the like, are hereby incorporated by reference into this application. Except where the application is filed in a manner inconsistent or contrary to the present disclosure, and except where the claim is filed in its broadest scope (whether present or later appended to the application) as well. It is noted that the descriptions, definitions and/or use of terms in this application shall control if they are inconsistent or contrary to the statements and/or uses of the present application in the material attached to this application.

Finally, it should be understood that the embodiments described herein are merely illustrative of the principles of the embodiments of the present application. Other variations are also possible within the scope of the present application. Thus, by way of example, and not limitation, alternative configurations of the embodiments of the present application can be viewed as being consistent with the teachings of the present application. Accordingly, the embodiments of the present application are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. A method of computing resource allocation, comprising:

acquiring analysis data at N time points in an application scene of computing resources to be determined, wherein N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located; wherein the content of the first and second substances,

acquiring the user position characteristics based on Bluetooth beacon equipment;

acquiring the user attribute features and the number features based on a reflected signal of a laser beam emitted by a laser radar and/or based on frame data extracted from a video frame;

acquiring the network connection quantity characteristic based on wifi connection data;

acquiring the user behavior characteristics based on a preset buried point;

inputting the analysis data into one or more trained resource allocation models, and predicting one or more resource allocation demand types at the N +1 time point; the resource allocation model predicts one or more resource allocation demand types at the time point of N +1, and comprises the following steps:

the resource allocation model comprises a feature vector extraction submodel and a resource allocation demand forecasting submodel;

acquiring feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data;

the resource allocation demand forecasting sub-model outputs the resource allocation demand type based on the feature vector;

determining resource allocation instructions based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

2. The method of claim 1, wherein obtaining feature vectors corresponding to the analysis data at the N time points through the feature vector extraction submodel based on the analysis data comprises:

converting the analysis data under the N time points to obtain numerical characteristics;

performing barrel dividing operation on the numerical type characteristics at each time point to obtain one or more input characteristic data;

and inputting the one or more input feature data into the feature vector extraction submodel to obtain the feature vector.

3. The method of claim 1, wherein the obtaining the user attribute feature and the people number feature based on a reflected signal of a laser beam emitted by a laser radar and/or based on frame data extracted from a video frame comprises:

the method comprises the steps of emitting laser beams by using a laser radar, and acquiring characteristic data of a plurality of target objects based on reflected signals of the laser beams;

determining the user attribute features and the number features based on the feature data of the plurality of target objects by using the trained first user identification model; and/or

Extracting frame data based on the video data;

and extracting at least one target object in the frame data by using the trained second user identification model, and determining the user attribute characteristics and/or the people number characteristics based on the at least one target object.

4. The method of claim 1, wherein the feature vector extraction submodel is a deep neural network model and the resource allocation demand predictor submodel is an LSTM model;

and taking historical analysis data as a training sample, taking a historical resource allocation demand type as a sample identifier, and performing combined training on the feature vector extraction submodel and the resource allocation demand prediction submodel in an end-to-end learning mode to obtain the resource allocation model.

5. A system for computing resource allocation, comprising:

the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring analysis data at N time points in an application scene of computing resources to be determined, wherein N is an integer greater than 0; the analysis data comprises user position characteristics, user attribute characteristics, number of people characteristics, network connection quantity characteristics and user behavior characteristics of target users in a preset size area where the terminal using the computing resources is located; wherein the content of the first and second substances,

acquiring the user behavior characteristics based on a preset buried point;

the prediction module is used for inputting the analysis data into one or more trained resource allocation models and predicting one or more resource allocation demand types at the time point of N + 1; the resource allocation model predicts one or more resource allocation demand types at the time point of N +1, and comprises the following steps:

a determining module, configured to determine a resource allocation instruction based on the one or more resource allocation demand types; the resource allocation instruction comprises a container resource allocation instruction, a worker allocation instruction, an attention degree allocation instruction and a resource allocation model precision adjustment instruction.

6. The system of claim 5, wherein the prediction module is further configured to:

7. The system of claim 5, wherein the acquisition module is further configured to:

Extracting frame data based on the video data;

8. The system of claim 5, wherein the feature vector extraction submodel is a deep neural network model and the resource allocation demand predictor submodel is an LSTM model;

9. An apparatus for computing resource allocation, comprising at least one storage medium and at least one processor, the at least one storage medium configured to store computer instructions; the at least one processor is configured to execute the computer instructions to implement the method of any of claims 1-4.

10. A computer-readable storage medium, wherein the storage medium stores computer instructions, which when executed by a processor, implement the method of any one of claims 1 to 4.