CN110995487A

CN110995487A - Multi-service quality prediction method and device, computer equipment and readable storage medium

Info

Publication number: CN110995487A
Application number: CN201911218022.4A
Authority: CN
Inventors: 刘志中; 曾峰
Original assignee: Shenzhen Wuyu Zhilian Technology Co ltd
Current assignee: Shenzhen Wuyu Zhilian Technology Co ltd
Priority date: 2019-12-03
Filing date: 2019-12-03
Publication date: 2020-04-10
Anticipated expiration: 2039-12-03
Also published as: CN110995487B

Abstract

The invention is suitable for the technical field of computers, and provides a multi-service quality prediction method, a device, computer equipment and a readable storage medium, wherein the multi-service quality prediction method comprises the following steps: constructing a multi-dimensional scene perception model according to the multi-dimensional scene information and the multi-service quality value corresponding to the multi-dimensional scene information; based on the multi-dimensional context awareness model, training and optimizing a multi-task deep neural network according to data fusion and a loss function, and establishing a multi-task deep neural network model; and determining a predicted value of the multi-service quality according to the current multi-dimensional scene information and the multi-task deep neural network model. The invention realizes the accurate prediction of multi-QoS attributes of multi-dimensional context awareness and improves the application of the multi-QoS attributes in the service recommendation field; meanwhile, the method is proved to have better prediction capability and good expansion capability based on the experiment of the edge computing scene simulation platform.

Description

Multi-service quality prediction method and device, computer equipment and readable storage medium

Technical Field

The invention belongs to the technical field of computers, and particularly relates to a multi-service quality prediction method, a multi-service quality prediction device, computer equipment and a readable storage medium.

Background

Mobile Edge Computing (MEC) is a new network architecture concept that deploys Computing and storage resources at the Edge of a Mobile network, pushing applications, data and services from a centralized point (cloud) to the logical extremes of the network. Due to the fact that the MEC is closer to a user, the MEC can provide a service environment with ultra-low delay, high bandwidth and real-time access, and the MEC becomes a new technology for making up for the deficiency of a cloud computing architecture. MECs offer rich service distribution and usage opportunities for service providers and users. A large number of computing services (such as ultra-high-definition video, augmented reality, virtual reality, image processing and the like) can be deployed on a server of the MEC, so that flexible deployment of the services is realized, and the reliability of the services is improved.

With the rapid development of the MEC technology, a great amount of edge computing services with rich functions from different fields appear on the network, and great convenience is provided for the majority of users. Generally, when a user uses a service, certain constraints are often imposed on a plurality of QoS attributes of the service. For example, the reaction time is less than 1 hour, the availability is greater than 85%, the reliability is greater than 90%, etc. Therefore, the service system needs to select or aggregate edge computing services for the user that can satisfy the user multi-QoS constraints.

However, in a dynamically complex network environment, the QoS of the edge computing service is dynamically changing. The service selected based on the static QoS or the built combined service often violates the multi-QoS constraint proposed by the user with a large probability at runtime, thereby causing service application failure and seriously affecting the satisfaction of the user and the utilization rate of service resources. Therefore, in the MEC environment, how to accurately predict a plurality of QoS attribute values of the edge computing service becomes one of the key problems to be solved in the MEC field.

In recent years, scholars at home and abroad carry out a great deal of research work aiming at the QoS prediction problem and provide some effective methods, which mainly comprise: a Collaborative Filtering (CF) based method, a Matrix Factorization (MF) based method. Among them, the CF-based QoS prediction method mainly uses a history service invocation record of a user to identify similar users, and further uses such similarity to predict the QoS of a target user. However, the QoS prediction method based on CF is difficult to overcome the difficulty caused by the sparsity of QoS data, and the prediction accuracy needs to be improved. Some researches propose a model-based MF method to predict QoS, and research results show that the QoS prediction method based on MF can effectively improve QoS prediction accuracy under the condition of high sparsity. Although scholars at home and abroad obtain remarkable and effective research results in the aspect of QoS prediction, the existing research work mainly focuses on the prediction of the QoS of Web service or cloud service, does not deeply consider the characteristics of the QoS prediction problem in the MEC environment, and cannot meet the new challenges brought by the QoS prediction in the MEC environment.

Therefore, the existing QoS prediction method generally has the problems that the prediction precision is low, and new challenges brought by QoS prediction in the MEC environment cannot be met.

Disclosure of Invention

The embodiment of the invention aims to provide a multi-service quality prediction method, and aims to solve the problems that the existing QoS prediction method is low in prediction precision and cannot meet new challenges brought by QoS prediction in an MEC environment.

The embodiment of the invention is realized in such a way that a multi-service quality prediction method comprises the following steps:

constructing a multi-dimensional scene perception model according to the multi-dimensional scene information and the multi-service quality value corresponding to the multi-dimensional scene information;

based on the multi-dimensional context awareness model, training and optimizing a multi-task deep neural network according to data fusion and a loss function, and establishing a multi-task deep neural network model;

and determining a predicted value of the multi-service quality according to the current multi-dimensional scene information and the multi-task deep neural network model.

Another objective of an embodiment of the present invention is to provide a multi-qos prediction apparatus, including:

the perception model building unit is used for building a multi-dimensional scene perception model according to the multi-dimensional scene information and the multi-service quality value corresponding to the multi-dimensional scene information;

the neural network model establishing unit is used for training and optimizing the multitask deep neural network according to data fusion and a loss function based on the multidimensional context awareness model and establishing a multitask deep neural network model; and

and the predicted value determining unit is used for determining the predicted value of the multi-service quality according to the current multi-dimensional scene information and the multi-task deep neural network model.

It is a further object of an embodiment of the present invention a computer arrangement comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to carry out the steps of the multi-quality of service prediction method.

Another object of an embodiment of the present invention is a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the steps of the multi-quality of service prediction method.

The multi-service quality prediction method provided by the embodiment of the invention provides an easily-expanded multi-attribute QoS prediction universal architecture based on a deep neural network model under an MEC environment for the first time, constructs a multi-dimensional context information perception model according to multi-dimensional context information and a multi-service quality value corresponding to the multi-dimensional context information, introduces a data fusion and loss function weighting automatic updating mechanism into a multi-task deep neural network based on the multi-dimensional context information perception model, establishes an improved multi-task deep neural network model, realizes accurate prediction of multi-dimensional context-aware multi-QoS attributes according to the multi-task deep neural network model, and improves the application of the multi-dimensional context-aware multi-QoS attributes in the service recommendation field; meanwhile, the method is proved to have better prediction capability and good expansion capability based on the experiment of the edge computing scene simulation platform.

Drawings

Fig. 1 is a flowchart illustrating an implementation of a multi-qos prediction method according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating an implementation of a multi-QoS prediction method according to a second embodiment of the present invention;

fig. 3 is a flowchart illustrating an implementation of a multi-qos prediction method according to a third embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a multitasking deep neural network model according to an embodiment of the present invention;

fig. 5 is a flowchart of an implementation of a multi-qos prediction method according to a fourth embodiment of the present invention;

FIG. 6 is a comparison graph of service reliability accuracy provided by an embodiment of the present invention;

FIG. 7 is a service time accuracy comparison graph according to an embodiment of the present invention;

FIG. 8 is a service reliability accuracy rate packet comparison graph according to an embodiment of the present invention;

FIG. 9 is another service reliability accuracy packet comparison graph provided by embodiments of the present invention;

fig. 10 is a block diagram of a multi-qos prediction apparatus according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

It should be understood that although the terms first, second, etc. may be used to describe various information in embodiments of the present invention, the information should not be limited by these terms. These terms are only used to distinguish one type of information from another.

Currently, the existing QoS prediction method does not deeply consider the characteristics of the QoS prediction problem in the MEC environment, and cannot meet the new challenges brought by the QoS prediction in the MEC environment, and the method mainly includes the following two points:

(1) how to implement multidimensional context-aware QoS prediction. In the MEC environment, the QoS of an edge computing service is affected by multidimensional scenarios related to service operation, mainly including user dimensions, network dimensions, and service dimensions. Each dimension scenario includes a plurality of scenario factors, and a change in each scenario factor causes a change in QoS of the service. Therefore, how to implement multidimensional context-aware QoS prediction is an important challenge facing QoS prediction in the MEC environment.

(2) How to implement multi-QoS attribute prediction. In the MEC environment, users often put forward multiple QoS constraints, which requires simultaneous prediction of multiple QoS attribute values of edge computing services. How to realize accurate prediction of multiple QoS attributes is another key challenge facing QoS prediction in the MEC environment.

In order to realize multi-QoS prediction of multi-dimensional context awareness under an MEC environment, the embodiment of the invention firstly analyzes the influence of user dimension, network dimension and service dimension on edge calculation service QoS, determines context factors contained in each dimension of context, and constructs a multi-dimensional context awareness model; and then, incorporating a data fusion and loss function weighting automatic updating mechanism into a multitask Deep neural network, providing an Improved multitask Deep neural network model (IMTDNN), and realizing Multi-QoS attribute prediction of Multi-dimensional context awareness based on the IMTDNN model. And finally, acquiring a large amount of experimental data based on the edge computing scene simulation platform EdgeCloudSim, and verifying the effectiveness of the method provided by the invention through a large amount of experiments. The main contributions are as follows:

(1) the multi-QoS prediction idea of multi-dimensional scene perception under the MEC environment is provided, the multi-dimensional scenes which have important influence on the service QoS under the MEC environment are deeply excavated, and the multi-dimensional scene perception model is constructed.

(2) A data fusion and loss function weighting automatic updating mechanism is introduced into a multitask Deep neural Network, an Improved Deep multitask learning model (IMTDNN) is provided, and Multi-dimensional context-aware Multi-QoS prediction is achieved based on the IMTDNN.

The multi-service quality prediction method provided by the embodiment of the invention is a prediction method aiming at a plurality of Qos indexes of one service, and provides an easily-expanded multi-attribute QoS prediction general framework based on a deep neural network model under an MEC environment for the first time, wherein a multi-dimensional context information perception model is constructed according to multi-dimensional context information and multi-QoS attributes corresponding to the multi-dimensional context information, a data fusion and loss function weighting automatic updating mechanism is introduced into a multi-task deep neural network based on the multi-dimensional context information perception model, an improved multi-task deep neural network model is established, accurate prediction of multi-dimensional context-aware multi-QoS attributes is realized according to the multi-task deep neural network model, and the application of the multi-dimensional context-aware multi-QoS attributes in the service recommendation field is improved; meanwhile, the method is proved to have better prediction capability and good expansion capability based on the experiment of the edge computing scene simulation platform.

To further explain the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the embodiments, structures, features and effects according to the present invention will be given with reference to the accompanying drawings and preferred embodiments.

Fig. 1 shows an implementation flow of a multi-qos prediction method provided by an embodiment of the present invention, and for convenience of description, only the relevant parts related to the embodiment of the present invention are shown, which are detailed as follows:

in step S101, a multidimensional context awareness model is constructed according to multidimensional context information and a multi-service quality value corresponding to the multidimensional context information.

In the embodiment of the present invention, the multidimensional context information refers to a context having a strong association relationship with the operation of an edge computing service in a mobile edge computing environment, and mainly includes a user dimension context, a network dimension context, and a service dimension context, and these three-dimensional contexts respectively include a plurality of different context factors. However, a change in each scenario factor in the three-dimensional scenario causes a change in the QoS of the edge computing service. Therefore, when performing QoS prediction in the MEC environment, it is necessary to sufficiently consider the correlation between the context information and the QoS for each dimension.

In the embodiment of the present invention, the user dimension context information mainly includes: location of the User (User Location), User Device (User Device), Task type (Task type), and volume of the Task (Task volume). In the MEC environment, a user is the initiator of an edge computing service usage request, and context information associated with the user has some impact on the QoS of the service. In an MEC environment, a change in user profile information will result in a change in service QoS. Such as: for the same service, different types of service requests, the service will have different QoS expressions; for the same type of service request, the service will have different QoS performance due to the different amount of tasks that each request needs to process; in addition, the user's equipment also has certain influence on the experience of the service quality, and better service quality experience can be achieved by configuring better equipment; based on the above analysis, the definition of the user dimensional scene is given as formula (1).

UC＝＜UserLocation,UserDevice,TaksType,TaskVolume＞ (1)

In the embodiment of the present invention, the contextual factors of the network dimension mainly include: network Type (Network Type), Network rate (Network Speed), and Network Latency (Network Latency). In the MEC environment, the sending of the user service use request, the transmission of the data, and the return of the processing result all need to be completed through the network. Thus, network dimensional scenarios have a strong correlation with the QoS of edge computing services. The network dimension context information has a strong influence on the QoS of the service. First, network fluctuations often result in user failures to request and use services. The network fluctuation is mainly represented in two aspects of network speed and network delay. Secondly, the transmission rate of the network determines the interaction time of the user and the service, and different interaction times bring different service experiences to the user. The definition of the network dimensional scenario is given as shown in formula (2).

UC＝＜NetworkType,NetworkSpeed,NetworkLatency＞ (2)

In embodiments of the invention, in an MEC environment, a service is the subject of completing a task submitted by a user. When processing tasks, the context information related to service operation has strong relevance with the service QoS. The context information related to the service operation mainly includes a Load (Server Load) of the Server, a queuing number (Request _ Amount) of the service Request, a Performance (Server Performance) of the Server, and the like. The context information related to the service operation has a strong influence on the QoS of the service. Such as: when other scene information is the same and the load of the service is low, the quality of the service is better; when other scene information is the same and the configuration of the server is higher, the quality of service will be better. The definition of the service dimension scenario is shown in equation (3).

SC＝＜ServerLoad,RequestAmount,ServerPerformance＞ (3)

In the embodiment of the present invention, in the MEC environment, the user dimension scenario, the network scenario, and the service scenario have a large influence on the QoS of the service, and therefore, when the QoS prediction is performed in the MEC environment, the multidimensional scenario information needs to be considered at the same time. Under the MEC environment, the multidimensional context awareness model for service QoS prediction is shown as formula (4).

MC＝＜UC,NC,SC＞ (4)

In step S102, based on the multi-dimensional context awareness model, a multi-tasking deep neural network is trained and optimized according to data fusion and a loss function, and a multi-tasking deep neural network model is established.

In the embodiment of the present invention, based on the multidimensional context awareness model, the multitask deep neural network is trained and optimized according to data fusion and a loss function, and a multitask deep neural network model is established, which may be: according to the relation between various multi-dimensional scene information contained in the multi-dimensional scene perception model and the multi-service quality value corresponding to the corresponding multi-dimensional scene information, the multi-dimensional scene information is used as a training data set of the multi-task deep neural network, the multi-dimensional scene information is used as the input of the multi-task deep neural network, the multi-service quality predicted value of the multi-dimensional scene perception is output, based on the thought of the residual neural network, the total loss value is calculated based on the loss function according to the multi-service quality predicted value of the multi-dimensional scene perception and the multi-service quality value corresponding to the real multi-dimensional scene information, so that the multi-task deep neural network is subjected to parameter optimization processing, and the multi-task deep neural network model is.

In the embodiment of the invention, the multi-QoS prediction problem of multi-dimensional context awareness is essentially a multi-input multi-output problem, and the key for efficiently solving the problem is to design a multi-input multi-output neural network with excellent performance. In recent years, multitask learning has been successfully applied to a plurality of fields such as natural language processing, speech recognition, computer vision, and drug discovery as an important branch of the field of machine learning. The multitask neural network enables the model to better summarize the learning mode of the original task by learning and sharing the characteristics between related tasks. For multi-tasking neural networks, the performance of the model is mainly limited by feature sharing between different tasks and the loss between different tasks. In order to realize efficient multi-dimensional scene-aware multi-QoS prediction, a multi-task learning model is improved based on the idea of a residual neural network to obtain an improved multi-task deep neural network model (IMTDNN).

In step S103, a predicted value of the multi-service quality is determined according to the current multi-dimensional context information and the multitask deep neural network model.

Fig. 2 shows an implementation flow of a multi-qos prediction method provided by the second embodiment of the present invention, and for convenience of description, only the relevant parts related to the second embodiment of the present invention are shown, which are detailed as follows:

it is similar to the embodiment, except that the step S102 specifically includes:

in step S201, a training data set is acquired; the training data set comprises feature vectors of the multi-dimensional context information and multi-quality of service values exhibited by services under the multi-dimensional context information.

In the embodiment of the invention, for a multi-QoS prediction problem of multi-dimensional context awareness, a training data set of an IMTDNN model is defined as D, D { (X)₁,Y₁),...(X_i,Y_i),...,(X_n,Y_n)}. Wherein, X_iFeature vector, X, representing multi-dimensional scene information in the ith sample data_i＝{x_i1,...,x_ij,...,x_it}，x_ijRepresenting the jth scene feature. Y is_iExpressed in multi-dimensional scene information X_iMultiple QoS values, Y, exhibited by the lower service_i＝{y_i1,...,y_ih,...,y_ik}，y_ihA value representing the h-th QoS attribute of the service; n represents the number of sample data.

In step S202, a predicted value of a multi-quality-of-service attribute of the multi-dimensional context awareness is determined according to the feature vector of the multi-dimensional context information and the multitask deep neural network.

In the embodiment of the invention, the phenomenon of information loss exists in the information transmission process between layers of the multitask deep neural network, so the thought of a residual error neural network is introduced into the multitask deep neural network, and the output of each layer is fused with the input data of an input layer. For example, the multitask deep neural network model structure comprises an input layer, a sharing layer and a specific task layer, the feature vector of the multidimensional situation information is input into the sharing layer, the output of each layer of the sharing layer is fused with the input data of the input layer, and fused data are obtained; and performing regression processing on the fusion data based on a specific task layer to determine a predicted value of the multi-service quality attribute of the multi-dimensional context awareness.

In step S203, a total loss value is calculated based on a loss function according to the predicted value of the multidimensional context-aware multi-service quality attribute and the multi-service quality value represented by the service under the multidimensional context information.

In the embodiment of the present invention, the commonly used loss functions are Mean Absolute Error (MAE) and Mean Square Error (MSE). Wherein, MAE refers to the average value of the distance between the predicted value of the model and the actual value of the sample; MSE refers to the average of the squares of the distances between the predicted values of the model and the actual values of the samples. Since MSE usually scales up or down with the limit that the error between the predicted value and the true value is equal to 1, MSE gives more weight to the abnormal value in the sample, thereby affecting the prediction effect of normal sample data, while MAE does not have such a disadvantage. In QoS prediction, the QoS attribute values are scaleless, resulting in a large number of outliers in the data set, and therefore MAE is taken as the task-layer-specific loss function, and the loss function of IMTDNN is defined as a weighted sum of a plurality of task-layer-specific loss functions MAE. However, the weights occupied by the loss functions of each specific task layer are different, and the values of the weights usually need to be manually adjusted, so that a large amount of human resources are consumed, and the performance of the model is influenced.

In step S204, according to the total loss value, performing parameter optimization processing on the multitask deep neural network, and establishing a multitask deep neural network model.

In the embodiment of the present invention, the classical optimization algorithms for parameter optimization mainly include a Stochastic Gradient Descent (SGD) method, a Momentum Gradient Descent (Momentum) method, and a Nesterov Momentum Stochastic Gradient Descent method, and some improvements based on the methods, such as AdaGrad, RMSProp, and Adam optimization algorithms, are applied to the adaptive parameter optimization problem. In the deep neural network, it is very important to select a proper optimization algorithm to optimize the parameters of the model. Adam is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process, and can iteratively update the weights of the neural network based on training data. Moreover, the diagonal scaling of the Adam algorithm gradient has invariance, so that the method is very suitable for solving the problem of large-scale data or parameters, the algorithm is also suitable for solving the unsteady problem of large noise and sparse gradient, the data containing noise and having sparsity has good interpretability, and the problem of the sparsity band of QoS data can be well overcome. Therefore, Adam is adopted as an optimization algorithm of the IMTDNN model in the embodiment of the invention.

In each training of the model, the Adam algorithm randomly selects a training example x, calculates the first moment deviation and the second moment deviation of the parameter, corrects the first moment deviation and the second moment deviation, and then moves along the opposite direction of the parameter gradient, as shown in formula (5).

Where θ indicates a trainable parameter, t represents the last time step, α is the learning rate,

the corrected first moment deviation and the corrected second moment deviation are respectively obtained, and epsilon is a constant.

According to the multi-service quality prediction method provided by the embodiment of the invention, the predicted value of the multi-service quality attribute of multi-dimensional context awareness is determined according to the feature vector of the multi-dimensional context information and the multi-task deep neural network, and the total loss value is calculated based on a loss function according to the predicted value of the multi-service quality attribute of the multi-dimensional context awareness and the multi-service quality value represented by service under the multi-dimensional context information; and performing parameter optimization processing on the multitask deep neural network according to the total loss value, so that the obtained improved multitask deep neural network model realizes accurate prediction of multi-QoS attributes of multi-dimensional context awareness and improves the application of the model in the field of service recommendation.

Fig. 3 shows an implementation flow of a multi-qos prediction method provided by the third embodiment of the present invention, and for convenience of description, only the relevant parts related to the third embodiment of the present invention are shown, which are detailed as follows:

it is similar to the second embodiment, except that the multitasking deep neural network comprises a shared layer and a task-specific layer.

In the multitask deep neural network, a sharing layer is formed by multi-layer full connection with a plurality of neuron nodes, and the multitask deep neural network has the phenomenon of information loss in the information transmission process between layers. In order to overcome the defect, the embodiment of the invention introduces the thought of the residual error neural network into the multitask deep neural network and proposes to fuse the output of each layer of the sharing layer with the input data of the input layer. And the loss of information is effectively compensated through data fusion. The multitask deep neural network model structure provided by the embodiment of the invention is shown in FIG. 4.

The step S102 specifically includes:

in step S301, a data fusion operation is performed on the feature vector of the multi-dimensional scene information and the output vector of the shared layer to obtain fusion data.

In the embodiment of the present invention, the feature vector of the multi-dimensional context information and the output vector of the shared layer are subjected to data fusion to obtain fusion data, and the implementation manner may be: determining the output vector of the shared layer according to the serial number, the activation function, the weight matrix and the offset value of the shared layer; and performing data fusion operation on the feature vector of the multi-dimensional scene information and the sharing layer output vector to obtain fusion data.

Let L be the number of shared layers of the multitask deep neural network, each layer has m neuron nodes, and the output vector of the L-th layer is H_l＝{h_l1,...,h_lmL is more than or equal to 1 and less than or equal to L). Let the number of neuron nodes of the input layer of the multitask deep neural network be m, and the current input data be X_i＝{x_i1,...,x_it}. The invention defines the data fusion operation as the transverse splicing fusion on a one-dimensional vector by using symbols

Representing a data fusion operation. The data fusion operation of the ith sharing layer is shown as formula (6).

Wherein

H_l＝δ_l(w_lM_l-1+b_l),l＝1,...,L，

l denotes the number of the shared layer, δ_lRepresents the activation function of the l-th shared layer, w_lWeight matrix representing the l-th shared layer, b_lA bias value representing the l-th shared layer; in particular, M₀＝X_i。

In step S302, regression processing is performed on the fusion data based on the specific task layer, and a predicted value of the multi-dimensional context-aware multi-quality-of-service attribute is determined.

In the embodiment of the present invention, the fusion data is subjected to regression processing based on the specific task layer, and a predicted value of a multi-dimensional context-aware multi-service quality attribute is determined, which may be implemented as follows: and determining a predicted value of the multi-service quality attribute of the multi-dimensional context awareness according to the fusion data and the number of layers, the activation function, the weight matrix and the bias value of the specific task layer.

Since different QoS attributes have different sensitivities to different contextual factors, a specific task awareness layer is required to perceive different prediction tasks. In the model structure shown in fig. 4, each layer of the specific task layer is formed by a fully-connected layer of the neural network, and the data processing formula of each layer is shown as formula (7):

y_p＝σ_p(β_py_p-1+c_p),p＝1,...,P (7)

where p denotes the number of layers of a particular task layer, σ_pRepresenting the activation function of the p-th task-specific layer, β_pA weight matrix representing the p-th task-specific layer, c_pIndicating the bias value of the pth task-specific layer. Specifically, when the number p of task-specific layers is 1, y is input_p-1For the output M of the last layer of feature sharing layer_L。

The predicted value of the multi-QoS attribute of the multi-dimensional scene perception can be obtained through regression of a specific task layer. The prediction formula of the kth task is shown as formula (8).

Wherein the content of the first and second substances,

for the output of the k-th task-specific layer,

d_krespectively, the weight and the bias value when predicting the k-th task.

According to the multi-service quality prediction method provided by the embodiment of the invention, the feature vector of the multi-dimensional scene information and the output vector of the sharing layer are subjected to data fusion operation to obtain fusion data; and performing regression processing on the fusion data based on the specific task layer to determine a predicted value of the multi-dimensional context-aware multi-service quality attribute, namely fusing the output of each layer of the shared layer with the input data of the input layer, and effectively compensating the loss of information through data fusion, thereby being beneficial to subsequently improving the prediction precision of the multi-dimensional context-aware multi-QoS attribute and improving the application of the multi-dimensional context-aware multi-QoS attribute in the field of service recommendation.

Fig. 5 shows an implementation flow of a multi-qos prediction method provided by the fourth embodiment of the present invention, and for convenience of description, only the relevant parts related to the fourth embodiment of the present invention are shown, which are detailed as follows:

it is similar to the second embodiment, except that the step S203 specifically includes:

in step S501, a total loss value is calculated according to an absolute difference between a predicted value of the multidimensional context-aware multi-service quality attribute and the multi-service quality value represented by the service under the multidimensional context information, the number of specific task layers, and a weight of a specific task layer loss function.

In the embodiment of the invention, in order to enable the IMTDNN model to have better performance, selection is requiredA suitable loss function guides the learning of the parameters by the model. For a single-task deep neural network, the commonly used loss functions are Mean Absolute Error (MAE) and Mean Square Error (MSE). Wherein MAE refers to the predicted value of the model

Average of the distance from the true value y of the sample. If the number of data samples is n, the MAE formula is shown in equation (9):

MSE refers to the predicted value of the model

And the mean of the squares of the distances between the real values of the samples y. If the number of sample data is n, the MSE is shown as equation (10):

MSE is usually predicted as a value

And the error of the true value y is equal to 1 to limit the error to be amplified or reduced, so that the MSE gives more weight to the abnormal value in the sample, thereby influencing the prediction effect of the normal sample data, while the MAE does not have the defect. In QoS prediction, the QoS attribute values are non-scale, resulting in a large number of outliers in the data set, therefore, the present invention adopts MAE as the task-specific loss function, and defines the loss function of IMTDNN as a weighted sum of a plurality of task-specific loss functions MAE, as shown in equation (11).

Where k denotes the number of task-specific layers, λ_iRepresenting the weight of the i-th task-specific layer loss function MAE. Each timeThe weights occupied by the loss functions of the specific task layers are different, the values of the weights generally need to be adjusted manually, the mode consumes a large amount of human resources and influences the performance of the model, the weights are set as variables, and the appropriate values are learned along with the training of the model.

According to the multi-service quality prediction method provided by the embodiment of the invention, a total loss value is calculated according to the absolute difference value of the predicted value of the multi-service quality attribute perceived by the multi-dimensional scene and the multi-service quality value represented by the service under the multi-dimensional scene information, the number of specific task layers and the weight of a loss function of the specific task layers; namely, MAE is adopted as a loss function of a specific task layer, and the loss function of IMTDNN is defined as the weighted sum of a plurality of loss functions MAE of the specific task layer, so that the accurate prediction effect of data is ensured.

The following is a multi-dimensional context-aware multi-QoS prediction algorithm in a mobile edge computing environment according to an embodiment of the present invention, as shown below.

The algorithm is as follows: IMDNN-based multi-dimensional context-aware multi-QoS prediction

Stage 1.IMDNN model training

Inputting: d// data set for IDMNN training

1. Initializing parameters of an IDMNN model;

2.Repeat

FOR i to n DO// n is the number of sample data

3. Input sample data < X_i,Y_i＞

4.FOR(l＝1 to L)

5. Inputting fusion data of the l-1 sharing layer;

obtaining fusion data M of the first sharing layer according to the formula (6)_l；

6.END FOR

FOR j ═ 1 to k// k is the number of specific tasks;

8. calculating the output of the specific task layer according to the formula (7);

9. calculating the predicted value of task j according to formula (8)

11. Calculating a loss value of the jth task according to formula (9);

12.END FOR

13. calculating a total loss value of the IMDNN model according to the formula (11);

14. updating parameters in the model according to an Adam algorithm;

15.END FOR

until { satisfies a model training end condition, ends training and outputs a trained IMDNN model }

Stage 2. multiple QoS prediction based on IMDNN model

17. Inputting X'_cur＝{x′₁,x′₂,...,x′_t-1,x′_tCurrent multidimensional context information

Executing the IMDNN model;

18. outputting a predicted value of multiple QoS:

the multi-service quality prediction method provided by the embodiment of the invention is experimentally analyzed as follows:

experimental data preparation:

based on the edge computing simulation platform EdgeCloudSim, a scene that students use an online video processing service S in a university campus is simulated, and configuration information of an experimental platform is shown in table 1. In the experiment, each piece of data records the situation information when the student uses the service S and a plurality of QoS values expressed by the service S. The invention mainly considers 8 kinds of scene information and two kinds of QoS attributes. In this experiment, about 21 ten thousand pieces of service usage data were extracted by the present invention, and the information recorded in sequence for each piece of data is shown in table 2.

TABLE 1 Experimental platform Environment configuration

TABLE 2 Experimental data field information

Evaluation indexes are as follows:

from the experimental data, 30% of the data were randomly extracted as test data, and the rest were used as training data. The deviation between the predicted value and the true value of the QoS attribute is calculated by adopting the average absolute error MAE, the mean square error MSE and the accuracy Acc, so that the effectiveness of the prediction method provided by the embodiment of the invention is measured. The calculation of MAE and MSE is shown in (12) in the formulas (9) and (10) and the calculation formula of Acc.

Wherein, the smaller the values of MAE and MSE are, the higher the prediction accuracy of the model is, and the larger the value of Acc is, the higher the prediction accuracy of the model is.

Setting parameters:

in order to verify the effectiveness of the proposed method, the invention selects four typical multiple-input multiple-output prediction methods for performance comparison with the proposed method. The four methods are: multitasking deep neural networks (MTDNN), case inference (CBR), Cross-batch, U-MTL.

For the case inference method, a local similarity calculation method based on Manhattan distance is used for calculating the global similarity of a target case and a historical case, wherein the similarity weight w of each dimension scene information determines the global similarity; finding the first k similar cases based on the global similarity of the target case and the historical cases, setting the similarity weight of each dimension scene information to be the same, namely w is 1/8, and selecting 3 similar historical cases to calculate the similarity.

For the Cross-batch method, different prediction tasks respectively adopt the same full-connection deep network structure, and feature sharing of different tasks is performed by using a Cross unit behind an implicit layer. The number of hidden layers is set to be 5, a cross unit is used after the 5 th layer, the number of nodes of the hidden layers is 128, MAE loss is adopted for two QoS prediction tasks, an Adam optimization algorithm is adopted, and the initial learning rate is 0.0001.

For the U-MTL method, a hard sharing mode of multitask deep learning is adopted, a feature sharing layer and a specific task layer are set to be 4 layers and 1 layer, the number of corresponding layer nodes is 128 and 64 respectively, a loss function is a special method provided in the text, Adam is adopted in an optimization algorithm, and the initial learning rate is 0.001.

For the multi-task deep neural network MTDNN, the number of the feature sharing layer and the specific task layer is set to be 4 and 1, the number of corresponding layer nodes is 128 and 64 respectively, the loss function is MAE, the optimization algorithm is Adam, and the initial learning rate is 0.01.

For the IMTDNN model provided by the invention, the number of layers of a feature sharing layer and a specific task layer is set to be 4 and 1 respectively; the number of points of each layer of the sharing layer is 128, the number of nodes of each layer of the specific task layer is 64, loss weights of 1 are respectively initialized for the two QoS prediction tasks, the optimization algorithm is Adam, and the initialized learning rate is 0.01. In order to ensure the comparative fairness, the maximum iteration times of the multitask deep neural network Cross-batch, the U-MTL and the IMTDNN are set to be 100, and a batch training strategy with the size of 256 is used.

And (3) comparing the performances:

in order to fully verify the effectiveness of the method provided by the invention and avoid the fluctuation influence of prediction caused by data difference, the invention respectively adopts 10%, 20% and 30% of total data samples as test sets (respectively marked as TDS 10%, TDS 20% and TDS 30%), and calculates three evaluation indexes of each method. For each method, the final results were averaged over 10 experimental results. The results of the experiment are shown in table 3.

TABLE 3 prediction accuracy of different methods on different test sets

As can be seen from table 3, CBR performs the worst of the five algorithms in predicting two QoS attributes, reliability of service and service time. For the other four deep learning algorithms, as can be seen from table 3, the IMTDNN model provided by the present invention is superior to the other three methods in the evaluation indexes Acc, MAE, and MSE when predicting two QoS attributes. In particular, on the test data set with TDS of 30%, the IMTDNN model is superior to the optimal results of other methods by 0.8% and 1.1% respectively on the evaluation index Acc when predicting two QoS attributes; the indexes MAE respectively lead the suboptimal result by 8.2 percent and 6.5 percent.

The IMTDNN model provided by the invention can effectively capture different influences of scene information on QoS prediction based on a layered processing structure of shared characteristics and task characteristics; meanwhile, through data fusion of each sharing layer, loss in the characteristic extraction process can be effectively reduced; on the other hand, the autonomous updating of the weight of the loss function is beneficial to achieving the optimal parameter value.

In order to verify the convergence of the proposed method, 70% of experimental data is adopted as training data, 30% of the experimental data is adopted as test data, and four deep learning algorithms are trained and tested. The experimental results are shown in fig. 6 and 7, where fig. 6 shows the results of predicting the service failure rate and fig. 3 shows the results of predicting the service time. The ordinate of fig. 6 and 7 represents the prediction accuracy, and the abscissa represents the number of iterations of the algorithm.

As can be seen from fig. 6 and 7, the accuracy rates of the four deep learning algorithms all rise with the increase of the number of iterations. The IMTDNN model provided by the invention has better prediction accuracy and better convergence rate than other three algorithms. The IMTDNN model has good learning ability, can obtain higher accuracy with less iteration and has higher convergence rate.

In order to verify the stability of the prediction capability of the IMTDNN model, the test set is randomly and differently grouped. The specific operation method comprises the following steps: randomly extracting 50% of data from a test set with TDS (total dissolved solids) of 30% as a sub-test set, and extracting 10 times to form 10 different sub-test sets; then, testing is performed on different test data sets by using IMTDNN and other four deep learning algorithms. Experimental results the test results are shown in fig. 8 and 9. FIG. 8 illustrates the prediction accuracy of four different methods for service failure rate; fig. 9 shows four different methods for prediction accuracy of service time. The ordinate of fig. 8 and 9 indicates the prediction accuracy, and the abscissa indicates the serial number of the test data. As can be seen from fig. 8 and 9, in the QoS reliability and the service time prediction results, the accuracy of the IMTDNN model is stabilized at about 0.88 and 0.95, respectively, and it can be found that the IMTDNN model provided by the present invention has good prediction stability.

Parameter analysis:

in the multi-task deep learning neural network model, a sharing layer is used for carrying out higher-level abstraction on basic features, meanwhile, the sharing layer is also used for learning common features of different prediction tasks, and the sharing layer has important influence on the accuracy of the model. Therefore, the present invention sets different number of sharing layers in the IDMM model to check its influence on QoS prediction. The results of the experiment are shown in table 4.

TABLE 4 Effect of the number of shared layers on the results

Experimental results show that the addition of the number of sharing layers is beneficial to reducing errors and improving the model prediction accuracy. However, the degree of lift is limited. However, as the number of hidden layers increases, more parameters will be brought to the model, thereby greatly increasing the overhead of model training and the risk of overfitting. Therefore, a proper number of hidden layers needs to be selected, and based on the experimental result, the performance and training consumption of the model are comprehensively considered, and the number of the hidden layers is determined to be 4 by the method.

In addition, in the deep neural network model, the number of nodes of the hidden layer also determines the learning ability of the hidden layer. In order to enable the model to obtain better learning capability, the number of the nodes of the sharing layer is set to be 32, 64, 96, 128, 192 and 256 in the IMTDNN model respectively, and the influence of different node values on the model prediction performance is observed. The results of the experiment are shown in Table 5. Similarly, increasing the number of nodes in the hidden layer increases the training amount and the risk of overfitting while improving the model performance. Based on the experimental result, the invention sets the node number of the model hidden layer to be 128.

TABLE 5 Effect of number of shared tier nodes on results

In summary, in order to realize accurate prediction of multiple QoS attributes and improve application of the QoS attributes in the service recommendation field, embodiments of the present invention provide an easily extensible generic architecture for multiple attribute QoS prediction based on a deep neural network model in an MEC environment for the first time. Experiments prove that the method has better prediction capability and good expansion capability.

Fig. 10 shows a structure of a multi-qos prediction apparatus according to an embodiment of the present invention, and for convenience of description, only a part related to the embodiment of the present invention is shown.

A multi-quality-of-service prediction apparatus comprising: the system comprises a perception model building unit 100, a neural network model building unit 200 and a predicted value determining unit 300.

The perceptual model building unit 100 is configured to build a multidimensional context perceptual model according to the multidimensional context information and the multi-service quality value corresponding to the multidimensional context information.

UC＝＜UserLocation,UserDevice,TaksType,TaskVolume＞ (5)

UC＝＜NetworkType,NetworkSpeed,NetworkLatency＞ (6)

SC＝＜ServerLoad,RequestAmount,ServerPerformance＞ (7)

MC＝＜UC,NC,SC＞ (8)

And the neural network model establishing unit 200 is configured to train and optimize the multitask deep neural network according to data fusion and a loss function based on the multi-dimensional context awareness model, and establish a multitask deep neural network model.

In the embodiment of the present invention, based on the multidimensional context awareness model, the multi-task deep neural network is trained and optimized according to data fusion and a loss function, and a multi-task deep neural network model is established, specifically: according to the relation between various multi-dimensional scene information contained in the multi-dimensional scene perception model and the multi-service quality value corresponding to the corresponding multi-dimensional scene information, the multi-dimensional scene information is used as a training data set of the multi-task deep neural network, the multi-dimensional scene information is used as the input of the multi-task deep neural network, the multi-service quality predicted value of the multi-dimensional scene perception is output, based on the thought of the residual neural network, the total loss value is calculated based on the loss function according to the multi-service quality predicted value of the multi-dimensional scene perception and the multi-service quality value corresponding to the real multi-dimensional scene information, so that the multi-task deep neural network is subjected to parameter optimization processing, and the multi-task deep neural network model is.

And a predicted value determining unit 300, configured to determine a predicted value of multiple service qualities according to the current multi-dimensional context information and the multitasking deep neural network model.

The multi-service quality prediction device provided by the embodiment of the invention provides an easily-expanded multi-attribute QoS prediction general framework based on a deep neural network model under an MEC environment for the first time, the multi-dimensional context information perception model is constructed according to multi-dimensional context information and a multi-service quality value corresponding to the multi-dimensional context information, a data fusion and loss function weighting automatic updating mechanism is introduced into a multi-task deep neural network based on the multi-dimensional context information perception model, an improved multi-task deep neural network model is established, the accurate prediction of the multi-QoS attribute of multi-dimensional context perception is realized according to the multi-task deep neural network model, and the application of the multi-attribute in the service recommendation field is improved; meanwhile, the method is proved to have better prediction capability and good expansion capability based on the experiment of the edge computing scene simulation platform.

In one embodiment, a computer device is proposed, the computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:

step S101, a multi-dimensional context perception model is constructed according to multi-dimensional context information and a multi-service quality value corresponding to the multi-dimensional context information;

step S102, based on the multi-dimensional context awareness model, training and optimizing a multi-task deep neural network according to data fusion and a loss function, and establishing a multi-task deep neural network model;

and step S103, determining a predicted value of the multi-service quality according to the current multi-dimensional scene information and the multi-task deep neural network model.

In one embodiment, a computer readable storage medium is provided, having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the steps of:

It should be understood that, although the steps in the flowcharts of the embodiments of the present invention are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in various embodiments may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims

1. A method for multi-quality of service prediction, comprising:

2. The method for predicting multi-service quality according to claim 1, wherein the step of training and optimizing the multi-task deep neural network according to a data fusion and a loss function based on the multi-dimensional context awareness model to establish the multi-task deep neural network model specifically comprises:

acquiring a training data set; the training data set comprises a feature vector of the multi-dimensional scene information and a multi-service quality value represented by the service under the multi-dimensional scene information;

determining a predicted value of a multi-service quality attribute of multi-dimensional context awareness according to the feature vector of the multi-dimensional context information and a multi-task deep neural network;

calculating a total loss value based on a loss function according to the predicted value of the multi-service quality attribute of the multi-dimensional context awareness and the multi-service quality value represented by the service under the multi-dimensional context information;

and performing parameter optimization processing on the multitask deep neural network according to the total loss value, and establishing a multitask deep neural network model.

3. The multi-service quality prediction method of claim 2 wherein the multitasking deep neural network comprises a shared layer and a task-specific layer;

the step of determining the predicted value of the multi-dimensional context-aware multi-service quality attribute according to the feature vector of the multi-dimensional context information and the multi-task deep neural network specifically includes:

performing data fusion operation on the feature vector of the multi-dimensional scene information and the output vector of the sharing layer to obtain fusion data;

and performing regression processing on the fusion data based on the specific task layer to determine a predicted value of the multi-dimensional context-aware multi-service quality attribute.

4. The method according to claim 3, wherein the step of performing data fusion operation on the feature vector of the multi-dimensional context information and the output vector of the shared layer to obtain fused data specifically comprises:

determining the output vector of the shared layer according to the serial number, the activation function, the weight matrix and the offset value of the shared layer;

and performing data fusion operation on the feature vector of the multi-dimensional scene information and the sharing layer output vector to obtain fusion data.

5. The method according to claim 3, wherein the step of performing regression processing on the fused data based on the specific task layer to determine the predicted value of the multidimensional context-aware multi-service quality attribute specifically comprises:

and determining a predicted value of the multi-service quality attribute of the multi-dimensional context awareness according to the fusion data and the number of layers, the activation function, the weight matrix and the bias value of the specific task layer.

6. The method according to claim 2, wherein the step of calculating a total loss value based on a loss function according to the predicted value of the multidimensional context-aware multi-quality-of-service attribute and the multi-quality-of-service value exhibited by the service under the multidimensional context information specifically comprises:

and calculating a total loss value according to the absolute difference value of the predicted value of the multi-service quality attribute perceived by the multi-dimensional scene and the multi-service quality value represented by the service under the multi-dimensional scene information, the number of specific task layers and the weight of a specific task layer loss function.

7. The multi-qos prediction method of claim 1, wherein the multi-dimensional context information comprises user-dimensional context information, network context information, and service context information.

8. A multi-quality of service prediction apparatus, comprising:

9. A computer arrangement, comprising a memory and a processor, the memory having stored thereon a computer program that, when executed by the processor, causes the processor to carry out the steps of the multi quality of service prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, causes the processor to carry out the steps of the multi-quality of service prediction method according to any one of claims 1 to 7.