CN105338186A - Context awareness-based Android mobile terminal power management method - Google Patents

Abstract

The invention discloses a context awareness-based Android mobile terminal power management method. context awareness information of a mobile terminal is used for decision making to predict whether to need a network interface currently, dynamic management is intelligently carried out, and system power consumption in a standby mode is reduced. Related context information is acquired via the system interface of the mobile terminal, pre-treatment is carried out on complicated context information, and a method of acquiring positioning data in an optimization mode is included. The decision making algorithm is an improved k-nearest neighbor algorithm, calculation of a sample distance adopts a weighted Euclidean distance formula, and information gains of an attribute are used for selecting a context information attribute and determining the weight of the attribute. A sample cutting method based on the maximum attribute range is brought forward at the same time, thereby reducing the calculation sample amount in the case of decision making. While the network interface is intelligently managed, power consumption of the Android mobile terminal system in the standby mode can be saved.

Description

Based on the Android mobile terminal method for managing power supply of context aware

Technical field

The invention belongs to mobile terminal technical field of power management, be specifically related to a kind of method for managing power supply of the intelligent management Android mobile terminal network interface based on context aware.

Background technology

Along with the high speed development of mobile terminal software and hardware and the universal of new generation network 3G and 4G, increasing people use mobile intelligent terminal and enjoy its convenient service brought and amusement and recreation.Current global smart phone user quantity rapid growth, and android system is in occupation of most of market share of mobile intelligent terminal system.The simultaneously high-performance of mobile terminal and complicated applications and severe game also brings higher power consumption, but battery technology slower development, how capacity obviously by himself volume restriction conclusion, reduces the problem demanding prompt solution that power consumption has become mobile terminal development.

Dynamic power management (DynamicPowerManagement, DPM) be a kind of system-level Low-power Technology, it is according to request service and performance requirement, by the dynamic-configuration to system unit, dynamic translation is carried out to the state of system unit, start the least possible parts or allow it be in suitable power consumption state, reaching the effective utilization to energy consumption with this.Traditional strategy comprises Timeout policy, predicting strategy and stochastic model strategy, and they have respective feature and deficiency.

Android system also provides self power management scheme, and comprising the power supply managing based on Wakelock, as far as possible object makes it enter resting state in the standby state.But research shows that the standby-state energy consumption of Android mobile terminal accounts for the over half of whole energy consumption, this is because the opening of android system causes the third-party backstage that is applied in abuse Wakelock resource, constantly carries out transfer of data by network interface.For the high energy consumption problem under holding state, current method mainly some regular times of static state setting and other conditions opens and closes network interface.

Summary of the invention

In order to power consumption that is more intelligent and that effectively reduce under holding state, the present invention proposes the Android mobile terminal method for managing power supply based on context aware, the context aware information of mobile terminal is utilized to carry out decision-making current the need of network interface to predict, and intelligently dynamic management is carried out to it, reduce the system energy consumption under holding state.The present invention has carried out preliminary treatment to the context information of complexity, and optimizes the method obtaining locator data.Decision making algorithm is the K nearest neighbor algorithm based on improving, and sample distance calculates and adopts weighted euclidean distance formula, and utilizes information gain value to choose context information attribute and the weight determining attribute.The present invention proposes a kind of sample method of cutting out based on maximum attribute codomain in addition, reduce calculating sample size during decision-making.

Concrete technical scheme of the present invention is as follows.

A kind of Android mobile terminal method for managing power supply based on context aware, it comprises: utilize the context aware information of mobile terminal to carry out decision-making current the need of network interface to predict, and intelligently dynamic management is carried out to mobile terminal, reduce the system power dissipation under holding state, and obtain relevant context information by the system interface of mobile terminal.

Further, preliminary treatment is carried out to the most complex scenarios information got, comprising to the determination of locator data accuracy and the method for optimization acquisition locator data, reduced the energy of excess loss.

Further, based on the K nearest neighbor algorithm improved, sample distance calculates and adopts weighted euclidean distance formula, and utilizes the information gain of attribute to choose context information attribute and the weight determining attribute.

Further, based on the sample method of cutting out of maximum attribute codomain, reduce calculating sample size during decision-making.

Further, the described most complex scenarios information to getting has been carried out preliminary treatment and has been comprised the steps:

(1) primary data is obtained, mainly for the geographical position of longitude and latitude at sight acquisition layer;

(2) result of (1) is utilized to adopt different accuracy to make its unique corresponding geographical position;

(3) utilize the result of (2) the longitude and latitude data of same position to be merged, obtain the Annual distribution situation of different accuracy upper/lower positions;

(4) utilize the result of (3) to carry out data processing and inversion, choose the longitude and latitude scheme that accuracy equals 4.

Further, the described K nearest neighbor algorithm based on improving comprises the steps:

(1) sample Y to be sorted and each training sample X is calculated _idistance;

(2) utilize the result of (1) to carry out ascending sort, choose the individual nearest sample set Z of front k;

(3) the number N (c) of each classification c in the nearest sample set Z of result statistics k of (2) is utilized;

(4) result of (3) is utilized to choose the classification C (Y) of the maximum classification c of N (c) in sample set as sample to be sorted.

Further, the described sample method of cutting out based on maximum attribute codomain comprises the steps:

(1) in training sample set X, the distance of the sample P sample Y identical with its time property value is calculated, and the distance of the sample P sample Q not identical with its time property value;

(2) result of (1) is utilized to represent the relative position of each sample relative to sample Y;

(3) result of (2) is utilized to obtain K the nearest sample X of sample Y _iscope;

(4) utilize the result of (3) to gather training sample in units of one week, to determine the K value in formula with the K value of KNN algorithm divided by 7.

Compared with prior art, tool of the present invention has the following advantages and technique effect:

Decision making algorithm of the present invention is the K nearest neighbor algorithm based on improving, sample distance calculates and adopts weighted euclidean distance formula, and utilizing the information gain of attribute to choose context information attribute, the information gain simultaneously based on attribute devises a kind of method determining attribute weight.The present invention is based on the sample method of cutting out of maximum attribute codomain to reduce calculating sample size during decision-making.Specifically, the present invention is based on time attribute to determine the possible range of K nearest sample, thus eliminating can not become individual other the nearest samples of K.The present invention is also implemented to mobile terminal in the form of services, carrys out the network interface of managing mobile terminal intelligently, reduces the system power dissipation under holding state, extends the available time of battery, improves the Consumer's Experience of mobile terminal.

Accompanying drawing explanation

Fig. 1 is the frame diagram of context aware strategy.

Fig. 2 is the context information attribute instance figure needing to obtain.

Fig. 3 is the position distribution schematic diagram produced under different accuracy.

Fig. 4 is the information gain value instance graph of different attribute.

Fig. 5 is the weighted value instance graph of different attribute.

Fig. 6 is the position view of first possible nearest sample relative to sample Y.

Fig. 7 is the position view of the individual possible nearest sample of K relative to sample Y.

Fig. 8 is the accuracy comparison diagram of different K values.

Fig. 9 is the system power dissipation comparison diagram of experimental result.

Embodiment

In order to make technical scheme of the present invention and advantage clearly understand, below in conjunction with accompanying drawing, be described in further detail, but enforcement of the present invention and protection are not limited thereto.If have process or the symbol of not special detailed description it is noted that following, be all that those skilled in the art can understand with reference to prior art or realize.

Fig. 1 is the frame diagram of the method for managing power supply (hereinafter referred to as context aware strategy) based on context aware, is divided into hardware device level, sight acquisition layer, sight processing layer and policy control layer from top to down.Wherein sight processing layer and policy control layer are the core places of whole policy framework.To set forth every one deck below.

Hardware device level refers to all hardware equipment of mobile terminal, comprises system component and transducer.System component has mobile network's interface, WiFi interface, GPS location, screen, battery etc., and transducer has acceleration transducer, light sensor, gravity sensor etc.What these hardware devices context information that has been the present invention obtained provides hardware foundation.

Context aware layer, then on the basis of hardware device level, utilizes the development interface that android system framework provides, and obtains various context information.The present invention obtains context information by this one deck, and is saved in as historical data in the middle of local data base, for the use habit studying user in depth carries out data encasement below.

Sight processing layer mainly processes the historical data of the collection of context aware layer, obtains the useful information that decision model needs.Wherein pretreatment module is responsible for carrying out simplify processes to some complex informations of context aware layer, comprises geographical location information; The attribute that characteristic selecting module is then selected in the middle of numerous context information attributes and result of decision correlation is stronger, improves the decision-making results of decision model.

The validity feature that policy control layer utilizes the real-time context data of context aware layer and sight processing layer to provide carries out decision-making to network interface components, finally carrys out the state of control assembly according to the result of decision.Wherein policy decision module carries out decision-making according to KNN algorithm decision model, and assembly control module then performs the last result of decision.

Mobile terminal is to the context information that the invention provides a lot of system and user, and Fig. 2 lists the attribute that the present invention needs the context information obtained.Mainly contain four kinds of scenario type, comprise time type, space type, device type and target.

Time type is the attribute be significant, and can embody the behavioural characteristic of user at different time.Space type represents place, space residing for mobile terminal and user and environment, and different places, space and environment often also have different user's service conditions.Device type mainly reflects the service condition of current mobile terminal, comprises electricity size, battery status, screen state, mobile network's state, WiFi state and CPU usage.Target information is current the need of network interface, and the present invention is by judging the access rights of the current foreground application run whether determine by log on connection.

Context information preliminary treatment.In order to improve the accuracy of decision making algorithm KNN, the present invention needs to carry out preliminary treatment to some primary datas, mainly for the geographical position of longitude and latitude.The localization method that Android provides can get 7 even more multidigits after longitude and latitude numeral decimal, accepts or rejects different accuracy and longitude and latitude data can be made to produce the geographical position of different number.In addition, the present invention can screen by arranging some conditions to carry out and process, and such as deletes the position data that the time residing for those users is less than 10 minutes, reduces longitude and latitude data volume.

According to the context information data that Fig. 2 lists, the present invention have recorded the context data of a real user in one week, comprising longitude and latitude position data.Finally obtain the Annual distribution situation of different accuracy upper/lower positions, as shown in Figure 3, the position that each Color pair should be unique, the accounting of the size of flabellum represents user residing for correspondence position time.Through process and the analysis of data, the present invention determines the scheme adopting accuracy to equal 4, and it can provide proper position distribution.

Obtain the optimization of locator data.In order to reduce the loss of locating and bringing, just the present invention can position in the case of necessary, the number of times obtaining position data can be greatly reduced like this.Specifically, the present invention can by judging whether subscriber equipment moves to another one place to select the need of reorientating, in addition, in general geographical position due to same WiFi environment is fixing, so the present invention can preserve positional information corresponding to each WiFi, after only need to inquire about corresponding position by WiFi information.

The concrete steps obtaining locator data are as follows:

(1) whether judgment device is in mobile status, if equipment does not move, goes to step (2), if equipment moving, then waits for that its stopping goes to step (3) after mobile.

(2) current geographic position is constant, adopts nearest positional information conduct.(1) is turned after waiting for the fixed cycle.

(3) need to reorientate, if current network disconnects, open network interface.Judge current network environment, if under being in mobile network environment, turn (4); If WiFi environment then turns (5).

(4) current is mobile network environment, obtains positional information by mobile base station positioning method.(1) is turned after waiting for the fixed cycle.

(5) current is WiFi network environment, judges whether to preserve this WiFi correspondence position information.If preserve the longitude and latitude data that current WiFi is corresponding, directly obtain correspondence position information, otherwise WiFi network location obtains position, and preserve WiFi information and correspondence position.(1) is turned after waiting for the fixed cycle.

K nearest neighbor algorithm in decision making algorithm.K nearest neighbor algorithm (KNN) is one of large algorithm of machine learning ten, and its basic thought is that then this sample also belongs to this classification if the great majority in the k of sample in a feature space the most adjacent individual sample belong to some classifications.Performing step is as follows: suppose there is a training set X containing m sample, each sample is made up of n attribute, and target classification has w, is expressed as X for each training sample _i=(X _i1, X _i2, X _i3..., X _in), wherein i=(1,2,3 ..., m).An existing sample Y to be sorted, is expressed as Y=(Y ₁, Y ₂, Y ₃..., Y _n).

(1) sample Y to be sorted and each training sample X is calculated _idistance, here is the Euclidean distance formula of weighting

$D(Y,X_i)=\sqrt{{\mathrm{\Σ}}_{j=1}^n{\mathrm{\λ}}_j{(Y_i-X_{ij})}^2}---\left(1\right)$

Wherein λ _jit is the weight of attribute j.

(2) ascending sort is carried out to all distance results, choose the individual nearest sample set Z of front k, be expressed as Z _i=(z _i1, z _i2, z _i3..., z _in), wherein i=(1,2,3 ..., m).

(3) the number N (c) of each classification c in k nearest sample set Z is added up

$N\left(c\right)={\mathrm{\Σ}}_{i=1}^{k}G(c,Z_i)---\left(2\right)$

$G(c,Z_i)=\left\{\begin{array}{ccc}1& if& C\left(Z_i\right)=c\\ 0& if& C\left(Z_i\right)\≠c\end{array}\right.---\left(3\right)$

Wherein C (Z _i) represent sample Z _iclassification.

(4) the classification C (Y) of sample to be sorted is maximum that the classification c of N (c) in k recently sample set

C(Y)＝argmax _0<c≤w(N(c))(4)

The basis of the KNN algorithm in decision making algorithm calculates the distance between sample, if the attribute information chosen not quite will affect the result of classification on classification results relevance, therefore the present invention needs to choose attribute.Information gain can reflect that attribute distinguishes the ability of training sample set conjunction, and the present invention can be used as weighing attribute to the correlation of sample class with it.

Suppose there is a sample set S, its target category attribute contains c different value, and so S set is defined as follows relative to the entropy of c category attribute value:

$E\left(S\right)={\mathrm{\&Sigma;}}_{i=1}^c-p_i{\log }_2{p}_i---\left(5\right)$

Wherein p _irepresent the ratio of classification shared by i in sample set S.

An attribute A in sample set S can be expressed as G (S, A) relative to the information gain of S, is defined as follows:

$G(S,A)=E\left(S\right)-{\mathrm{\&Sigma;}}_{v\&Element;V\left(A\right)}\frac{\left|S_v\right|}{\left|S\right|}E\left(S_v\right)---\left(6\right)$

Wherein V (A) is the set of all values inside attribute A, S _vall subsamples set that the value of attribute A inside sample set S equals v, | S _v| represent S _vthe size of subsample set, be S _vsubsample is integrated into the ratio shared by S sample set, E (S _v) be subsample S set _ventropy.

According to formula (6) the present invention, all properties has been carried out to the calculating of information gain value, result as shown in Figure 4.Through experiment with analyze, the attribute that the present invention is greater than 0.04 using information gain value is as the characteristic attribute of final KNN algorithm, and then giving up the attribute being less than 0.04 need not.Front nine attributes inside final form are using the characteristic attribute as sample in KNN algorithm.

The weight of attribute directly affects the distance between sample, and weight is larger, and the difference of corresponding attribute will more be amplified, and classification results will vary widely.Therefore, the ability that the weight of attribute should be sub-category with attribute area is directly proportional, and namely the information gain of attribute is larger, and its weight of distributing is larger, with the weight λ of equation expression attribute A _aas follows with the relation of information gain G (S, A):

λ _A∝G(S,A)(7)

According to above analysis, the present invention proposes a kind of method determining the attribute weight of Euclidean distance based on information gain.Suppose that the sample in training sample set S has n attribute, be designated as A _i(i=1,2,3 ..., n), according to information gain value G (S, the A of a formulae discovery n attribute of above-mentioned information gain _i) (i=1,2,3 ..., n), then each attribute weight coefficient lambda _idetermined by formula below:

${\mathrm{\&lambda;}}_i=\frac{G(S,A_i)}{{\min }_{0<j\&le;n}G(S,A_j)}---\left(8\right)$

The Attribute Weight weight values finally determined as shown in Figure 5.

Sample cutting.One of them deficiency of K nearest neighbor algorithm needs the distance calculating sample and each training sample when being exactly decision-making, amount of calculation causes very greatly the algorithm efficiency of decision-making low.How to raise the efficiency and mainly contain two kinds of modes: a kind of is reduce the time calculated ^[10-12]; Another is the quantity reducing sample.The present invention proposes a kind of sample method of cutting out based on maximum attribute codomain, reduce calculating sample size during KNN decision-making, thus improve the execution efficiency of algorithm.Specifically, the present invention is based on time attribute to determine the possible range of K nearest sample, thus eliminating can not become individual other the nearest samples of K.

Suppose that training set X contains m sample, each sample has n different attribute, is then expressed as X for each training sample _i=(x _i1, x _i2, x _i3..., x _in), wherein i=(1,2,3 ..., m), if time attribute is c attribute (0≤c≤n).An existing sample Y to be sorted, is expressed as Y=(y ₁, y ₂, y ₃..., y _n).

(1) the 1st the nearest range of the sample of sample Y

If there is a sample P, its time attribute value and the identical of sample Y in training set X, other property values are all different and difference is maximum, then the distance of sample Y and sample P is:

$D(Y,P)=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^{c-1}{\mathrm{\&lambda;}}_j{(Y_i-X_{ij})}^2+{\mathrm{\&Sigma;}}_{j=c+1}^n{\mathrm{\&lambda;}}_j{(Y_i-X_{ij})}^2}---\left(9\right)$

D (Y, P) represents except time attribute, and the maximum of two sample distances, is designated as D _max(Y _c=X _ic), its value is a constant.

If training set X exists another one sample Q, its time attribute value and the difference of sample Y, other property values are all identical, then the distance of sample Y and sample Q is:

$D(Y,Q)=\sqrt{{\mathrm{\&lambda;}}_c{(Y_c-Q_c)}^2}---\left(10\right)$

If D (Y, Q) > D (Y, P), then

$Y_c-Q_c>\sqrt{\frac{D_{max}{(Y_c=X_{ic})}^2}{{\mathrm{\&lambda;}}_c}}---\left(11\right)$

Because for sample Y, existence sample Z certainly, the time attribute value of two samples is all the same, and this is that the historical circumstances information data gathered by the present invention decides.In addition, sample Y is identical with other property values possibilities of sample Z also may be different, therefore D (Y, P) >=D (Y, Z), and D (Y, Q) > D (Y, P), can obtain D (Y, Q) > D (Y, Z).

Below with time attribute axis represent each sample relative to sample Y position as shown in Figure 6.

So time attribute difference is greater than and the sample that equals sample Q is the 1st the nearest sample that impossible become sample Y, therefore for the 1st the nearest sample of sample Y, its possible sample X _iscope is:

$Y_c-X_{ic}\&le;\sqrt{\frac{D_{\max }{\left(Y_c=X_{ic}\right)}^2}{{\mathrm{\&lambda;}}_c}}---\left(12\right)$

(2) K of sample Y nearest range of the sample

If there is a sample V in training sample set X, its time attribute value and the difference K-1 of sample Y, other property values are all different and difference is maximum, then the distance of sample Y and sample V is:

$\begin{array}{c}D\left(Y,V\right)=\sqrt{\underset{j=1}{\overset{c-1}{\&Sigma;}}{\mathrm{\&lambda;}}_j{\left(Y_i-X_{ij}\right)}^2+{\mathrm{\&lambda;}}_c{\left(K-1\right)}^2+\underset{j=c+1}{\overset{n}{\&Sigma;}}{\mathrm{\&lambda;}}_j{\left(Y_i-X_{ij}\right)}^2}\\ =\sqrt{{\mathrm{\&lambda;}}_c{\left(K-1\right)}^2+D_{\max }{\left(Y_c=X_{ic}\right)}^2}\end{array}---\left(13\right)$

If there is another one sample W in training sample set X, its time attribute value and the difference of sample Y, other property values are all identical, then the distance of sample Y and sample W is:

$D(Y,W)=\sqrt{{\mathrm{\&lambda;}}_c{(Y_c-W_c)}^2}---\left(14\right)$

If D (Y, W) > D (Y, V), then

$Y_c-W_c\&le;\sqrt{\frac{{\mathrm{\&lambda;}}_c{\left(K-1\right)}^2+D_{\max }{\left(Y_c=X_{ic}\right)}^2}{{\mathrm{\&lambda;}}_c}}=\sqrt{{\left(K-1\right)}^2+\frac{D_{\max }{\left(Y_c=X_{ic}\right)}^2}{{\mathrm{\&lambda;}}_c}}---\left(15\right)$

Because for sample Y, existence sample U certainly, the time attribute value difference K-1 of two samples.In addition, sample Y is identical with other property values possibilities of sample U also may be different, therefore D (Y, V) >=D (Y, U), and D (Y, W) > D (Y, V), can obtain D (Y, W) > D (Y, U).Add D (Y, V) > D (Y, and D (Y P), P) >=D (Y, Z), so D (Y, W) > D (Y, Z), that is sample W can not become the individual nearest sample of K (1+K-1) of sample Y.

Below with time attribute axis represent each sample relative to sample Y position as shown in Figure 7.

In like manner, the individual nearest sample of the K for sample Y, its possible sample X _iscope is:

$Y_c-X_{ic}\&le;\sqrt{\frac{{\mathrm{\&lambda;}}_c{\left(K-1\right)}^2+D_{\max }{\left(Y_c=X_{ic}\right)}^2}{{\mathrm{\&lambda;}}_c}}=\sqrt{{\left(K-1\right)}^2+\frac{D_{\max }{\left(Y_c=X_{ic}\right)}^2}{{\mathrm{\&lambda;}}_c}}---\left(16\right)$

Because the present invention gathered training sample in units of one week, at least can find the training sample that 7 time attribute are identical for the present invention current certain sample Y, the K value that therefore the K value of above-mentioned formula should be KNN algorithm is determined divided by 7.

4. description of test

For KNN algorithm, first the present invention needs the size of true defining K value.K gets different values respectively, then tests the forecasting accuracy in training sample respectively, final selection the highest preparatory K value.In the test of each K value, the present invention adopts the method for 10 folding cross validations to calculate corresponding accuracy, the more representative and fairness of the accuracy obtained like this.

The present invention acquires user's context information data of a week, and the time that every day gathers is 8 o'clock to 24 o'clock, because be the length of one's sleep of user during this period of time substantially in morning to morning 8, and can the network interface of direct turning-off mobile terminal.And then the data of every day are divided into 10 parts at random, more respectively from 10 numbers of every day according to Stochastic choice a, 7 parts that elect like this, as a sub-sample set, finally obtain 10 random sub-sample sets.The present invention utilizes these 10 sub-sample sets to carry out 10 folding cross validations for each K value, obtains 10 predictablity rates of each K value and calculates its Average Accuracy.Preferably select the K value of K value as KNN algorithm of the highest correspondence of accuracy rate.Because the data of training sample set by 7 days form, the range of choices of K value is set to 7,14,21 ..., 91,98.

Fig. 8 is the experimental result of different K values being carried out to forecasting accuracy, and wherein during K=63, accuracy is the highest, so the present invention selects it as the K value of KNN algorithm.

Following the present invention will carry out the experiment of context aware strategy, specific experiment method: the electricity service condition of Android mobile terminal in one week under holding state being carried out recording user by the background service of a lightweight, start context aware strategy of the present invention afterwards to rerun 4 weeks above, record the electricity service condition in this surrounding under holding state equally, finally compare respectively first week and surrounding respective power consumption in the standby state below.

Fig. 9 is first week average power consumption to the 5th week system, wherein within first week, does not use context aware strategy, within second week, then starts context aware strategy by the 5th week.Can find out, less than first week of the system power dissipation of second week to the 5th week, saves the system power dissipation of 19.5%, 21.2%, 18.7% and 22.6% respectively.The system power dissipation of rear surrounding fluctuates in certain scope, and this and user weekly service condition have certain relation, belong to normal condition.

Claims (7)

1. the Android mobile terminal method for managing power supply based on context aware, it is characterized in that comprising: utilize the context aware information of mobile terminal to carry out decision-making current the need of network interface to predict, and intelligently dynamic management is carried out to mobile terminal, reduce the system power dissipation under holding state, and obtain relevant context information by the system interface of mobile terminal.

2. the Android mobile terminal method for managing power supply based on context aware according to claim 1, it is characterized in that the most complex scenarios information to getting has carried out preliminary treatment, comprising to the determination of locator data accuracy and the method for optimization acquisition locator data, reduce the energy of excess loss.

3. the Android mobile terminal method for managing power supply based on context aware according to claim 2, it is characterized in that the K nearest neighbor algorithm based on improving, sample distance calculates and adopts weighted euclidean distance formula, and utilizes the information gain of attribute to choose context information attribute and the weight determining attribute.

4. the Android mobile terminal method for managing power supply based on context aware according to claim 3, is characterized in that the sample method of cutting out based on maximum attribute codomain, reduces calculating sample size during decision-making.

5. the Android mobile terminal method for managing power supply based on context aware according to claim 2, is characterized in that the described most complex scenarios information to getting has been carried out preliminary treatment and comprised the steps:

(1) primary data is obtained, mainly for the geographical position of longitude and latitude at sight acquisition layer;

(2) result of (1) is utilized to adopt different accuracy to make its unique corresponding geographical position;

(3) utilize the result of (2) the longitude and latitude data of same position to be merged, obtain the Annual distribution situation of different accuracy upper/lower positions;

(4) utilize the result of (3) to carry out data processing and inversion, choose the longitude and latitude scheme that accuracy equals 4.

6. the Android mobile terminal method for managing power supply based on context aware according to claim 3, is characterized in that the described K nearest neighbor algorithm based on improving comprises the steps:

(1) sample Y to be sorted and each training sample X is calculated _idistance;

(2) utilize the result of (1) to carry out ascending sort, choose the individual nearest sample set Z of front k;

(3) number of each classification c in the nearest sample set Z of result statistics k of (2) is utilized ;

(4) result of (3) is utilized to choose in sample set maximum classification c is as the classification C (Y) of sample to be sorted.

7. the Android mobile terminal method for managing power supply based on context aware according to claim 4, is characterized in that the described sample method of cutting out based on maximum attribute codomain comprises the steps:

(1) in training sample set X, the distance of the sample P sample Y identical with its time property value is calculated, and the distance of the sample P sample Q not identical with its time property value;

(2) result of (1) is utilized to represent the relative position of each sample relative to sample Y;

(3) result of (2) is utilized to obtain K the nearest sample of sample Y scope;

(4) utilize the result of (3) to gather training sample in units of one week, determine the K value of the individual nearest sample of K with the K value of KNN algorithm divided by 7.