CN107734579B

CN107734579B - Mobile platform energy consumption optimization method based on Markov decision process

Info

Publication number: CN107734579B
Application number: CN201710962155.7A
Authority: CN
Inventors: 高岭; 郭延超; 王海; 杨旭东; 郑杰
Original assignee: Northwestern University
Current assignee: Northwestern University
Priority date: 2017-10-16
Filing date: 2017-10-16
Publication date: 2021-02-05
Anticipated expiration: 2037-10-16
Also published as: CN107734579A

Abstract

A mobile platform energy consumption optimization method based on a Markov decision process considers network states of different environments, selects an optimal strategy in a decision table generated in the Markov decision process according to the current electric quantity of a mobile phone and the refreshing frequency of a user, and dynamically selects the optimal network access and refreshes and downloads the optimal picture format. The method not only reduces the refreshing time, but also can reduce the energy consumption of the mobile platform. The experimental result shows that compared with a picture refreshing mode using a single network, the energy consumption optimization method provided by the invention can reduce the energy consumption by about 12.1% on the premise of ensuring that the refreshing times of users are not reduced.

Description

Mobile platform energy consumption optimization method based on Markov decision process

Technical Field

The invention belongs to the technical field of computer network communication, and particularly relates to a mobile platform energy consumption optimization method based on a Markov decision process.

Background

With the continuous development of science and technology, the market occupation of intelligent mobile equipment is continuously increased. Meanwhile, with the development and improvement of the wireless access network technology, the mobile internet becomes the fifth development cycle after the mainframe, the mini-machine, the personal computer, and the desktop internet. By the year of 2016, the number of domestic mobile phone netizens reaches 6.56 hundred million, the number of mobile phone netizens surfing the internet is improved to 92.5% from 90.1% at the end of 2015, the number of mobile phone netizens surfing the internet is 24.5% only by using mobile phones, and internet surfing equipment of people is more concentrated on mobile terminals.

As the network service form using human social as a core, the social network is gradually becoming the mainstream of the network with the continuous development of the internet. In daily life, each person in the country spends about 200 minutes in mobile phone use every day, and most frequently used social software comprises WeChat, QQ, Xinlang microblog and the like. Since the inner test of the Sina microblog in 2009, the Sina microblog becomes one of the most influential social platforms in China. According to the domestic social media sharing arrangement released in 2011, the percentage of users who surf the green microblog is 31.29%, and the rank is first. The Sina microblog is one of the most fierce social platforms at present, registered users exceed 5 hundred million at 12 months in 2012, and active users break through 2.12 million in 2015 until the active users occupy 86% in the microblog month. The microblog produces 10 million articles each day, ten thousand videos, and 2100 ten thousand pictures. In the microblog information, the pictures become a main transfer mode and are an indispensable part in the microblog.

For example, in a Sina microblog open platform, the formats of pictures contained in the microblog can be checked to be different types including original pictures, high-definition pictures and thumbnails, and experimental results show that the different formats of pictures have different energy consumption and obvious difference in the refreshing process. In actual life, the mobile phone downloading power and the downloading time are different under different environment states of a user, and the energy consumption for refreshing the picture microblog is high. Aiming at the problems, the invention provides that the mobile phone executes different refreshing actions under different network environment states, refreshes the optimal picture format under the current environment state, and achieves the maximum energy consumption optimization.

In the use process of the mobile client, the network environment where the user is located is constantly changed, the network signal strength is constantly and dynamically changed, the changes of the 4G and Wi-Fi network signals are mutually independent, the next signal strength is not correlated with the previous signal strength, and in order to cope with the dynamic change process, a Markov decision process is proposed. The markov decision process selects an action from the available set of actions and makes a decision based on the observed state at each time instant, and the state of the system's next step is randomly varied. In daily life, many processes are very similar to Markov decision processes, such as a queuing system and a population change process, the invention finds that the 4G network signal and Wi-Fi network signal change processes are also very similar to the problems, the state conversion of the 4G network signal and the Wi-Fi network signal change processes is changed randomly, different reward values are obtained through different actions, and finally an optimal strategy decision table is calculated.

Disclosure of Invention

Aiming at the self-downloaded content of the mobile social application, in the running process of the mobile social application, according to the network signal strength level value (mainly 4G, Wi-Fi) of the current user, the environment of the current user is constantly changed, the time interval value and the residual electric quantity value, different formats of pictures in a microblog are selected, including an original picture, a high definition picture, a thumbnail picture and a no picture, the environment state change process is described by using a continuous-time MDP model, the state transition probability table shows the state transition randomness, then mathematical modeling is carried out in the user decision making process, the optimal strategy decision table is obtained through simulation optimization, the energy consumption of the mobile platform is reduced, and meanwhile, the refreshing time of the user is reduced.

In order to achieve the purpose, the invention adopts the technical scheme that:

a mobile platform energy consumption optimization method based on a Markov decision process comprises the following steps:

1) the android application is autonomously developed by using an API (application programming interface) provided by a Sina microblog open platform, and the application is used for testing the APP and has the functions of logging in, refreshing information, releasing information, browsing information, logging out and the like by a Sina microblog mobile client user;

2) the method comprises the steps that a Mix test APP automatically collects an environment state value of a user, the environment state value comprises a network signal, the network signal is a 4G or Wi-Fi network, a 4G or Wi-Fi network strength grade value, a time interval value, a mobile phone residual electric quantity value and a microblog frequency value refreshed by the user, meanwhile, a Wi-Fi network disconnection threshold value is set according to the network signal strength grade value, and the mobile terminal automatically switches a 4G and Wi-Fi communication network; the Markov decision process is a set of four-tuple states comprisingS，A，P.(.,.)，R.)) of the environmental state values collected in step 2)When the user uses the mobile terminal, the mobile terminal is preferentially connected with a Wi-Fi network, then a network switching threshold value is set, the method becomes a Mix network switching mode, the network signal level value is divided into 5 levels according to an LTE signal strength level value dividing mode defined in an Android open source API, each network signal level is divided into 5 levels which are respectively 5, 4, 3, 2 and 1, when the Wi-Fi network signal strength level value is not more than 2, the user cannot finish the microblog refreshing action, and at the moment, the network switching threshold value is reached, and the application is automatically switched to a 4G communication network;

3) establishing a continuous time MDP mathematical model, taking the environment state value acquired in the step 2) as a state space of the MDP model, then establishing a dynamic transition probability of the MDP model, and finally establishing an action space of the MDP model, taking the optimal reward value as a target function, simulating the change process of the MDP model, and acquiring an optimal strategy decision table;

4) using an MATLAB simulation MDP model, taking all state sets as input values, and solving the model to obtain an optimal strategy decision table;

5) in the actual use process, the Mix test APP selects the optimal communication network and the optimal picture format according to the environment state value and the optimal strategy decision table in the step 4), and the microblog refreshing action is completed.

The method comprises the steps of 1), carrying out Mix test APP in the step 1), wherein the Mix test APP comprises user login, information refreshing, information releasing, browsing information and exit login, in the actual information refreshing process, a request is sent to a server through network connection, relevant contents are downloaded according to a returned URL address, the contents comprise character information and picture information, in the optimization method, a picture format is dynamically selected, an optimal picture format is selected according to a corresponding picture format URL address, and a microblog is downloaded and refreshed.

The step 3) arranges the environmental state values collected in the step 2) to generate a probability setPAnd a set of related actionsAReward setRAs an MDP model input value, the action set A comprises a refresh picture format corresponding to the network signal grade value, and the action set A is subdivided into two network action sets of 4G and Wi-FiUnder 4G network connection, when the signal intensity grade value is 1, the network is considered to be unable to be connected, because a large amount of energy consumption is needed to download and refresh microblogs, the microblogs cannot be refreshed under the grade value, the corresponding downloaded picture formats from high to low are the original picture, the high-definition picture, the thumbnail and the non-picture format under other grade values, and under Wi-Fi network connection, when the signal intensity grade value is not more than 2, the network is considered to be disconnected, the microblog cannot be downloaded and refreshed, and the corresponding downloaded picture formats from high to low are the original picture, the high-definition picture and the thumbnail picture respectively.

The invention has the beneficial effects that:

the core of the method is to continuously and dynamically adjust a user decision list according to the environment state value of the user. In the actual use process, a Mix test APP and a Mix network switching mode are used, the best network is used, the best picture format is selected, the network disconnection time is shortened, and the energy consumption of the application is reduced.

Drawings

FIG. 1 illustrates the social APP model of the present invention.

FIG. 2 is a usage scenario diagram of the present invention

FIG. 3 is a flow chart of the method of the present invention.

Detailed Description

The technical solution of the present invention is described in detail below with reference to the embodiments and the drawings of the specification, but is not limited thereto.

As shown in fig. 1, a user logs in a Mix test APP, the application automatically jumps to a user authentication interface of a new microblog, an account password is input to obtain a Token secret order of the user, the Token secret order is valid within a certain time, and the user logs in next time without verifying again.

After a user logs in a Mix test APP, the application automatically collects the environmental state value of the user, the environmental state value comprises a network signal (mainly 4G and Wi-Fi network) strength grade value, a time interval value, a mobile phone residual electric quantity value and a microblog frequency value refreshed by the user, the collected environmental state value is sorted and sequenced in a format set by a database, and is stored in a mobile phone local database, and finally, data is sent to a designated mailbox in a mail mode.

And (3) establishing a continuous time MDP mathematical model, taking the environment state value acquired in the step (2) as a state space of the MDP model, calculating the probability value between the environmental state transitions according to the data, then establishing a dynamic transition probability matrix of the MDP model, and finally establishing an action set of the MDP model, taking the optimal reward value as a target function, simulating the change process of the MDP model, acquiring an optimal strategy decision table, and outputting the optimal strategy decision table to the local. In general, a markov decision process is a set of quadruple states, including (S, a, P. (,), R.)). The environmental state value collected in step 2 is the state contained in the finite state set S. When the user uses the mobile terminal, the mobile terminal is preferentially connected with the Wi-Fi network, and then a network switching threshold value is set, so that the method becomes a Mix network switching mode. And dividing the network signal grade value into 5 grades according to an LTE signal strength grade value dividing mode defined in the Android open source API, wherein the grade of each network signal is 5, 4, 3, 2 and 1. When the Wi-Fi network signal strength grade value is not larger than 2, the user cannot finish the action of refreshing the microblog, at the moment, the application automatically switches to the 4G communication network to finish the microblog refreshing when the network switching threshold value is reached.

And (3) utilizing an MATLAB simulation MDP model to solve the model to obtain an optimal strategy decision table, and embedding the optimal strategy decision table into a Mix test APP mobile phone so as to search the optimal strategy when a user actually uses the mobile phone. The action set A comprises a refreshing picture format corresponding to the network signal grade value, and the action set A is subdivided into two network action sets of 4G and Wi-Fi. Under 4G network connection, when the signal intensity grade value is 1, the network is considered to be unable to be connected with the network, and microblogs are unable to be refreshed under the grade value because a large amount of energy consumption is needed to download and refresh the microblogs at the time. And at other grade values, the corresponding downloaded picture formats from high to low are original pictures, high-definition pictures, thumbnails and non-picture formats. And under Wi-Fi network connection, when the signal intensity grade value is not more than 2, the network is regarded as disconnected, the microblog cannot be downloaded and refreshed, and the corresponding downloaded picture formats of other high-level and low-level values are respectively an original picture, a high-definition picture and a thumbnail picture.

In the actual use process, the Mix test APP searches for an optimal strategy in an optimal strategy decision table stored locally through the acquired environment state value, an optimal communication network (mainly 4G and Wi-Fi) is switched, an optimal picture format is downloaded, and the microblog refreshing action is completed.

As shown in figure 2, the use scene of the method is shown in the figure, a user serves as a user, the user holds mobile terminal equipment to freely move in an LTE base station and a Wi-Fi coverage area, the terminal equipment starts 4G to be connected with a Wi-Fi network, the terminal is firstly connected with the Wi-Fi network, and the user checks and refreshes a microblog in the area according to behavior habits of the user.

When the user uses the mobile terminal device for the first time, the mobile terminal device automatically collects the environmental state value of the user, and stores the collected state value to the local in a set format in the database, as shown in fig. 2. Then, when the data volume reaches the specified magnitude, the terminal device will automatically send to the specified mailbox by way of mail. And after the server receives the data collected by the user, storing the data in a designated position of the server, and fitting the data by operating a data processing program. The specific process of data processing is as follows:

all the possible state values are derived and recorded, and a matrix is set up in which the individual state values are converted into one another.

And analyzing and calculating the occurrence times of two adjacent state values according to actually received data, and recording the occurrence times into a state transition matrix.

And counting the total times of state values in the actual data, and dividing the times of each state transition in the state transition matrix by the total times to finally obtain a state transition probability matrix.

Through a simulation tool, the processed probability matrix takes the set reward function as an input value, and an MDP mathematical model is established, wherein the specific process is as follows:

and setting a related reward function R, and setting a related reward value matrix according to the reward function.

And setting a refresh picture format corresponding to the network signal grade value in the action set A, and subdividing the action set A into two network action sets of 4G and Wi-Fi. Under 4G network connection, when the signal intensity grade value is 1, the network is considered to be unable to be connected with the network, and microblogs are unable to be refreshed under the grade value because a large amount of energy consumption is needed to download and refresh the microblogs at the time. And at other grade values, the corresponding downloaded picture formats from high to low are original pictures, high-definition pictures, thumbnails and non-picture formats. And under the Wi-Fi network connection, when the signal intensity grade value is not more than 2, the network is regarded as disconnected, and the picture format is that the microblog cannot be downloaded and refreshed. The network signal intensity grade values are arranged from high to low and are three effective grade values of 5, 4 and 3, and the downloaded picture formats corresponding to the optimal other values from high to low are respectively an original picture, a high-definition picture and a thumbnail picture.

The state transition probability matrix and the reward value matrix are brought into the simulation tool, the optimal strategy decision table obtained by the simulation tool is obtained through the related optimal strategy decision table, a data sorting program runs and outputs a format which can be identified by a Mix test APP, then the sorted decision table is sent to an application in a mail form, a user can freely move in a 4G and Wi-Fi coverage range at the moment, a microblog is refreshed according to the self use habit, the overall flow chart is shown in figure 3, and the specific use process at the moment of the application is as follows:

when the user refreshes the microblog, the mobile terminal is preferentially connected with the Wi-Fi network, and the user automatically collects the environment state value.

If the Wi-Fi network signal intensity grade value is lower than the set threshold value, namely the signal intensity grade value is not greater than 2, a Mix network switching mode is used, and the Wi-Fi network is automatically switched to the 4G network.

And if the Wi-Fi network does not reach the switching threshold value, selecting the optimal picture format according to the optimal strategy decision table, and downloading and refreshing the microblog.

And if the network is switched to the 4G network and the network does not reach the disconnection threshold value, selecting the optimal picture format according to the optimal strategy decision table, and downloading and refreshing the microblog. If the network connection threshold is reached, the user is notified that the refresh is not possible.

Claims

1. A mobile platform energy consumption optimization method based on a Markov decision process is characterized by comprising the following steps:

1) using an android application autonomously developed based on an API (application programming interface) provided by a Sina microblog open platform, carrying out Mix test on the APP, wherein the application has the functions of logging in, refreshing information, releasing information, turning over information and logging out of a user at a Sina microblog mobile client;

2) the method comprises the steps that a Mix test APP automatically collects an environment state value of a user, the environment state value comprises an intensity grade value and a time interval value of a network signal, a mobile phone residual electric quantity value, a user refreshes a microblog frequency value, and the network signal is a 4G or Wi-Fi network; meanwhile, a Wi-Fi network disconnection threshold value is set according to the network signal intensity grade value, and the mobile terminal automatically switches between the 4G communication network and the Wi-Fi communication network; the Markov decision process is a set of four-tuple states comprisingS，A，P.(.,.)，R()), wherein S is a finite state set, P is a probability set, A is a related action set, and R is a reward set;

the environmental state value acquired in the step 2) is a state contained in the finite state set S, when a user uses the mobile terminal, the mobile terminal is preferentially connected with a Wi-Fi network, then a network switching threshold value is set, the method becomes a Mix network switching mode, the network signal level value is divided into 5 levels according to an LTE signal strength level value dividing mode defined in an Android open source API, each network signal level is divided into 5 levels which are respectively 5, 4, 3, 2 and 1, when the Wi-Fi network signal strength level value is not greater than 2, the user cannot finish microblog refreshing action, and at the moment, the network switching threshold value is reached, and the application is automatically switched to a 4G communication network;

2. The markov decision process-based mobile platform energy consumption optimization method of claim 1, wherein in the step 1), Mix test APP comprises user login, refresh information, release information, browsing information and logout, in the actual information refresh process, a request needs to be sent to a server through network connection, relevant contents are downloaded according to a returned URL address, wherein the contents comprise text information and picture information, and in the optimization method, the picture format is dynamically selected, and the optimal picture format is selected according to the corresponding picture format URL address, so that a refreshed microblog is downloaded.

3. The markov decision process-based mobile platform energy consumption optimization method of claim 1, wherein the step 3) is to collate the environmental state values collected in the step 2) to generate a probability setPAnd a set of related actionsAReward setRThe method comprises the steps that as an MDP model input value, an action set A comprises a refresh picture format corresponding to a network signal grade value, the action set A is subdivided into two network action sets of 4G and Wi-Fi, under 4G network connection, when the signal strength grade value is 1, the network is considered to be unable to be connected with the network, the microblog cannot be refreshed under the grade value because a large amount of energy consumption is needed to download the refresh microblog at the moment, under other grade values, the picture formats corresponding to the high level and the low level are an original picture, a high-definition picture, a thumbnail and a non-picture format, under Wi-Fi network connection, when the signal strength grade value is not more than 2, the network is considered to be disconnected, the refresh microblog cannot be downloaded, and under the Wi-Fi network connection, the picture formats corresponding to the high level and the low level are the original picture, the high-definition picture.