CN116822340A - Terminal electric quantity optimization method based on software energy consumption - Google Patents

Terminal electric quantity optimization method based on software energy consumption Download PDF

Info

Publication number
CN116822340A
CN116822340A CN202310694825.7A CN202310694825A CN116822340A CN 116822340 A CN116822340 A CN 116822340A CN 202310694825 A CN202310694825 A CN 202310694825A CN 116822340 A CN116822340 A CN 116822340A
Authority
CN
China
Prior art keywords
target
energy consumption
software
individuals
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202310694825.7A
Other languages
Chinese (zh)
Other versions
CN116822340B (en
Inventor
王含
李光春
郭展威
张辰
王秀敏
钟孝琴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhongke Software Evaluation Guangzhou Co ltd
Original Assignee
Zhongke Software Evaluation Guangzhou Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhongke Software Evaluation Guangzhou Co ltd filed Critical Zhongke Software Evaluation Guangzhou Co ltd
Priority to CN202310694825.7A priority Critical patent/CN116822340B/en
Publication of CN116822340A publication Critical patent/CN116822340A/en
Application granted granted Critical
Publication of CN116822340B publication Critical patent/CN116822340B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Primary Health Care (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Human Resources & Organizations (AREA)
  • Algebra (AREA)
  • Water Supply & Treatment (AREA)
  • Databases & Information Systems (AREA)
  • Public Health (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Stored Programmes (AREA)

Abstract

The invention discloses a terminal electric quantity optimization method based on software energy consumption, which can effectively prolong the service time of a terminal by acquiring the software energy consumption and optimizing the electric quantity based on the software energy consumption, and calculate the residual electric quantity, so that the user experience can be ensured when the electric quantity is sufficient, the running of common software of the user can be preferentially ensured when the electric quantity is insufficient, and the user experience can be ensured when the electric quantity is optimized.

Description

Terminal electric quantity optimization method based on software energy consumption
Technical Field
The invention belongs to the technical field of electric quantity optimization, and particularly relates to a terminal electric quantity optimization method based on software energy consumption.
Background
With the rapid development of internet technology, the processing capability of terminals such as smartphones is stronger and stronger, and various application programs can be supported, however, the power consumption is larger and larger, and the standby time of the terminals is greatly shortened, for example, the standby time of smartphones is not longer than one day, so that users need to charge the terminals frequently to ensure normal use. However, a large part of the electric quantity is consumed in some unnecessary application programs hidden in the background, and the application programs misuse the wake-up locking mechanism for the purpose of enabling the application programs to push information to users in real time, so that the CPU of the terminal is always in an operation state and cannot enter a power-saving sleep mode, and the electric quantity of the battery is consumed quickly. Therefore, the application programs need to be optimized to optimize the terminal electric quantity, and the service time of a user is prolonged.
Disclosure of Invention
The invention provides a terminal electric quantity optimization method based on software energy consumption, which is used for solving the problems in the prior art.
A terminal electric quantity optimization method based on software energy consumption comprises the following steps:
acquiring source programs of a plurality of pieces of software on a target terminal, compiling all the source programs into target languages corresponding to a target platform, and obtaining a plurality of target programs; the target platform is used for running the target program and determining the energy consumption of the target program;
uploading a plurality of target programs to a target platform, aiming at the plurality of target programs, and acquiring code energy consumption tables corresponding to all the target programs through the target platform;
when the target terminal runs the target software, determining the execution conditions of various interface methods in the target software, and determining the energy consumption of the target software based on the code energy consumption table and the execution conditions of various interface methods; the target software represents the running software on the target terminal;
and scheduling the software operation in the target terminal by using the energy consumption of the target software to finish the optimization of the electric quantity of the target terminal.
Further, uploading the plurality of target programs to a target platform, aiming at the plurality of target programs, and acquiring code energy consumption tables corresponding to all the target programs through the target platform, wherein the method comprises the following steps of:
uploading a plurality of target programs to a target platform, extracting various sentences in the target programs through the target platform, and removing repeated sentences to obtain a plurality of target sentences;
and analyzing the multi-item slogans by adopting a regression analysis method, determining the energy consumption of different target sentences, and acquiring code energy consumption tables corresponding to all target programs.
Further, a regression analysis method is adopted to analyze the multi-entry tagline sentence, the energy consumption of different target sentences is determined, and the code energy consumption tables corresponding to all target programs are obtained, including:
determining various types of basic codes in all target sentences to obtain n types of basic codes; each target sentence is composed of a plurality of basic codes;
for n basic codes, j test tasks are operated, and the execution frequency matrix is determined as follows:
wherein VCn represents an nth base code, N (VC) represents an execution count matrix, N k (VCm) represents the number of runs of the mth base code during the kth test, k=1, 2, …, j, m=1, 2, …, n;
in j test tasks, the total energy consumption matrix is obtained as follows:
E(M)＝(E(M1),E(M2),…,E(Mj)) T
wherein E (M) represents an energy consumption matrix, and E (MK) represents the total energy consumption value in the kth test process;
according to the total energy consumption matrix and the execution frequency matrix, determining a reference energy consumption sequence E (VC) corresponding to n basic codes by adopting a regression analysis method as follows:
E(VC)＝(N T (VC)N(VC)) -1 N T (VC)E(M)
wherein T represents a transpose;
based on a reference energy consumption sequence E (VC) corresponding to the n basic codes, determining the energy consumption of each target sentence, traversing all target sentences, and determining the energy consumption of each target sentence;
and determining a code energy consumption table of the mutual mapping of the target sentences and the energy consumption according to the energy consumption of all the target sentences.
Further, when the target terminal runs the target software, determining the execution conditions of various interface methods in the target software, and determining the energy consumption of the target software based on the code energy consumption table and the execution conditions of various interface methods, including:
when a target terminal runs target software, tracking the execution conditions of a basic interface method, a dependent interface method and an independent interface method which are run in the target software to obtain an energy consumption matrix of the basic interface method, an energy consumption matrix of the dependent interface method and an energy consumption matrix of the independent interface method, and simultaneously obtaining a calling matrix of the basic interface method, a calling matrix of the dependent interface method and a calling matrix of the independent interface method;
and determining the energy consumption of the target software based on the energy consumption matrix of the basic interface method, the energy consumption matrix of the dependent interface method, the energy consumption matrix of the independent interface method, the calling matrix of the basic interface method, the calling matrix of the dependent interface method and the calling matrix of the independent interface method.
Further, based on the energy consumption matrix of the basic interface method, the energy consumption matrix of the dependent interface method, the energy consumption matrix of the independent interface method, the call matrix of the basic interface method, the call matrix of the dependent interface method and the call matrix of the independent interface method, the energy consumption of the target software is determined as follows:
E all ＝EM(IF fix )NM T (IF fix )+EM(IF dep )NM T (IF dep )+EM(IF indep )NM T (IF indep )
wherein EM (IF fix ) Energy consumption matrix, NM, representing basic interface method T (IF fix ) Transpose of call matrix representing basic interface method, EM (IF dep ) Energy consumption matrix representing interface-dependent method, NM T (IF dep ) Transpose matrix representing call matrix of dependent interface method, EM (IF indep ) Energy consumption matrix, NM, representing independent interface method T (IF indep ) A transpose of the call matrix representing the independent interface method.
Further, scheduling the software operation in the target terminal with the energy consumption of the target software to complete the optimization of the electric quantity of the target terminal, including:
collecting operation data of a target terminal where target software is located, and calling a pre-trained residual electric quantity estimation model to process the operation data so as to obtain residual electric quantity;
obtaining the residual electric quantity once in each preset period, and determining the residual electric quantity difference value of two adjacent periods;
judging whether the residual electric quantity is smaller than a set electric quantity threshold value or whether at least one residual electric quantity difference value is larger than a set electric quantity reduction threshold value, if so, optimizing the electric quantity of the target terminal, otherwise, keeping the normal operation of target software on the target terminal;
determining the foreground times and the current foreground state of target software, wherein the foreground times represent the times of switching from a background to a foreground, and the foreground times represent whether the target software is positioned in the foreground or not, and the foreground times comprise the position in the foreground or the position in the background;
keeping the current foreground state as the normal operation of the target software positioned at the foreground, and determining the optimizable software of which the foreground times are smaller than a set threshold value from the rest target software;
arranging the optimizable software according to the energy consumption from large to small, and sequentially closing the optimizable software until the difference value of the residual electric quantity difference value minus the energy consumption of the closed optimizable software is smaller than the electric quantity reduction threshold value, so as to finish the optimization of the electric quantity of the target terminal;
and when the difference value obtained by subtracting the energy consumption comprehensive of all the optimizable software from the difference value of the residual electric quantity is larger than the electric quantity reduction threshold value, directly closing all the optimizable software.
Further, the training method of the residual electric quantity estimation model comprises the following steps:
a1, constructing a residual electric quantity estimation model by adopting a neural network, initializing network parameters of the residual electric quantity estimation model, forming vectors by all the network parameters to obtain a training individual, and repeatedly obtaining multiple training individuals to obtain a training population;
a2, acquiring fitness values of all individuals in the training population, determining the training individuals with the largest fitness values as optimal individuals, taking N training individuals with the largest fitness values in the rest individuals as searching individuals, and taking M training individuals with the smallest fitness values as following individuals; wherein M+N+1 is the total number of training individual values in the training population;
a3, executing the walk behavior on all the search individuals, acquiring updated search individuals, and updating the optimal individuals based on the updated search individuals to obtain updated optimal individuals;
a4, executing calling behaviors on the following individuals on the basis of the updated optimal individuals to obtain updated following individuals;
a5, executing a tapping behavior on the updated following individuals, and determining the following individuals after secondary updating;
a6, determining whether the maximum fitness value in the current training population is larger than a preset threshold value or the training times reach an upper limit, if so, taking the training individuals corresponding to the maximum fitness value as final network parameters of a residual electric quantity estimation model, finishing the training of the residual electric quantity estimation model, otherwise, eliminating L training individuals with the minimum fitness value in the current training population, generating L new individuals, adding the L new individuals into the training population, and returning to the step A2.
Further, executing the walk behavior on all the search individuals to obtain updated search individuals, including:
executing the walk behavior, and determining an updated value of each latitude in the searching individual as follows:
wherein ,xnd Represents the D-th dimension of the network parameter in the nth search individual, d=1, 2, …, D represents the total dimension of the network parameter,representing the updated value of the d-th dimensional network parameter in the nth search individual, pi representing the circumference ratio, p representing the search direction, p=1, 2, …, h, h representing the total number of search directions, T representing the current training number, ρ representing the adjustment coefficient, T max Represents the upper limit of the training times,/->Representing the wander step length of the d-th dimension network parameter in wander behavior;
judging d-th dimension network parameter x nd Whether the fitness value corresponding to the searching individual is increased or not after updating by adopting the updated value, if so, the d-th dimension network parameter x is calculated nd Updating, otherwise, not updating;
and traversing all network parameters in the nth searching individual to finish updating the nth searching individual.
Further, based on the updated optimal individual, executing a calling action on the following individual to obtain the updated following individual, including:
b1, executing the attack update on the following individuals as follows:
wherein ,represents the d-th dimension network parameter in the mth following individual,>representing updated->Representing the step size of the d-th dimension network parameter in the summoning behavior, +.>Representing a d-th dimensional network parameter in the optimal individual;
b2, judging whether the fitness value corresponding to the updated following individual is larger than the fitness value corresponding to the optimal individual, if so, exchanging the positions of the following individual and the optimal individual, and entering a step B3, otherwise, directly entering the step B3;
b3, judging whether the distances between all following individuals and the optimal individual are smaller than a distance threshold valueIf yes, finishing updating to obtain updated following individuals, otherwise, continuing updating the following individuals larger than the distance threshold value, and returning to the step B1;
wherein ω represents a distance determination factor, max d Represents the maximum value, min, of the d-th-dimension network parameter d Representing the minimum value of the d-th dimensional network parameter.
Further, performing a tapping action on the updated following individual to determine a secondary updated following individual, comprising:
wherein ,represents the d-th dimensional network parameter in the following individuals after the mth update in the t-th training process,/for>Representing updated->Lambda represents the random constant between (-1, 1), ζ t Representing the adjustment factor during the t-th training; t=0, ζ t =1, and ζ t+1 ＝α*ζ t Alpha represents a random constant between (0.9,1), -a->Representing the step size of the attack of the d-th dimension network parameter in the attack behavior.
According to the terminal electric quantity optimization method based on software energy consumption, the service time of the terminal can be effectively prolonged by acquiring the software energy consumption and optimizing the electric quantity based on the software energy consumption, the residual electric quantity is calculated, user experience can be guaranteed when the electric quantity is sufficient, running of common software of a user is preferentially guaranteed when the electric quantity is insufficient, and user experience is guaranteed when the electric quantity is optimized.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
Fig. 1 is a flowchart of a terminal power optimization method based on software power consumption according to an embodiment of the present invention.
Specific embodiments of the present invention have been shown by way of the above drawings and will be described in more detail below. The drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but rather to illustrate the inventive concepts to those skilled in the art by reference to the specific embodiments.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with aspects of the invention as detailed in the accompanying claims.
Embodiments of the present invention are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, a terminal electric quantity optimizing method based on software energy consumption includes:
s101, acquiring source programs of a plurality of pieces of software of a target terminal, and compiling all the source programs into target languages corresponding to a target platform to obtain a plurality of target programs. The target platform is used for running the target program and determining the energy consumption of the target program.
Each object pair may use multiple interface methods, such as: the calling times of the basic interface method are fixed constants, can be a constructor and a destructor, and are the operations of establishing connection and closing connection when the database is operated, and the like. The dependent interface method may be an operation dependent on other methods, for example, the number of times a read-write stream is turned off is dependent on an open method, and the number of times the dependent interface method is executed may be inferred by counting the execution characteristics of the dependent method. The independent interface method refers to a method in which an operation is executed independently of other interfaces and the number of times of being called is not fixed. The execution condition is only determined by the input data of the class, namely the data holding condition and the input data condition of the corresponding object. It should be noted that, here, only a part of examples are provided, and the present embodiment is not illustrated one by one due to the numerous interface methods.
S102, uploading a plurality of target programs to a target platform, aiming at the plurality of target programs, and acquiring code energy consumption tables corresponding to all the target programs through the target platform.
Each interface method can be composed of at least one statement, such as an arithmetic and logic expression, a jump statement, a loop statement, a call statement, a dynamic storage management statement and the like, each statement can be composed of a plurality of codes, and based on the statement, the execution energy consumption of each statement can be determined first, so that a code energy consumption table corresponding to a target program is determined. It should be noted that, here, only a part of examples are provided, and this embodiment is not illustrated one by one due to the numerous sentences.
S103, when the target terminal runs the target software, determining the execution conditions of various interface methods in the target software, and determining the energy consumption of the target software based on the code energy consumption table and the execution conditions of various interface methods. Wherein the target software represents software running on the target terminal.
And S104, scheduling the software operation in the target terminal by using the energy consumption of the target software to finish the optimization of the electric quantity of the target terminal.
In this embodiment, uploading a plurality of target programs to a target platform, and obtaining code energy consumption tables corresponding to all the target programs through the target platform, where the code energy consumption tables include:
uploading a plurality of target programs to a target platform, extracting various sentences in the target programs through the target platform, and removing repeated sentences to obtain a plurality of target sentences.
And analyzing the multi-item slogans by adopting a regression analysis method, determining the energy consumption of different target sentences, and acquiring code energy consumption tables corresponding to all target programs.
In this embodiment, a regression analysis method is used to analyze multiple target sentences, determine energy consumption of different target sentences, and obtain code energy consumption tables corresponding to all target programs, including:
for all target sentences, determining various types of basic codes in all target sentences to obtain n kinds of basic codes. Each target statement is made up of a plurality of base codes.
For n basic codes, j test tasks are operated, and the execution frequency matrix is determined as follows:
wherein VCn represents an nth base code, N (VC) represents an execution count matrix, N k (VCm) represents the number of runs of the mth base code during the kth test, k=1, 2, …, j, m=1, 2, …, n.
In j test tasks, the total energy consumption matrix is obtained as follows:
E(M)＝(E(M1),E(M2),…,E(Mj)) T
wherein E (M) represents the energy consumption matrix, and E (MK) represents the total energy consumption value in the kth test process.
According to the total energy consumption matrix and the execution frequency matrix, determining a reference energy consumption sequence E (VC) corresponding to n basic codes by adopting a regression analysis method as follows:
E(VC)＝(N T (VC)N(VC)) -1 N T (VC)E(M)
wherein T represents the transpose.
And determining the energy consumption of each target sentence based on the reference energy consumption sequences E (VC) corresponding to the n basic codes, traversing all target sentences, and determining the energy consumption of each target sentence.
And determining a code energy consumption table of the mutual mapping of the target sentences and the energy consumption according to the energy consumption of all the target sentences.
Since each statement is composed of a plurality of basic codes, the reference energy consumption sequence E (VC) of the basic code of each statement can be determined first, and then the addition in the reference energy consumption sequence is determined to obtain the energy consumption of the target statement. In order to facilitate the subsequent power optimization, a period can be predetermined, and then a code power consumption table in which the target sentence and the power consumption are mapped to each other is determined by taking the period as a unit.
In summary, the scheme of this embodiment can be summarized as follows: the energy consumption of each statement in the period is determined, and when the software runs, the energy consumption of the software in the period can be determined according to the execution condition of the statement in the period, so that the electric quantity optimization can be performed based on the energy consumption.
In this embodiment, when the target terminal runs the target software, determining the execution conditions of various interface methods in the target software, and determining the energy consumption of the target software based on the code energy consumption table and the execution conditions of various interface methods, including:
when the target terminal runs the target software, the execution conditions of a basic interface method, a dependent interface method and an independent interface method which run in the target software are tracked, so that an energy consumption matrix of the basic interface method, an energy consumption matrix of the dependent interface method and an energy consumption matrix of the independent interface method are obtained, and a calling matrix of the basic interface method, a calling matrix of the dependent interface method and a calling matrix of the independent interface method are obtained.
And determining the energy consumption of the target software based on the energy consumption matrix of the basic interface method, the energy consumption matrix of the dependent interface method, the energy consumption matrix of the independent interface method, the calling matrix of the basic interface method, the calling matrix of the dependent interface method and the calling matrix of the independent interface method.
In this embodiment, the energy consumption of the target software is determined based on the energy consumption matrix of the basic interface method, the energy consumption matrix of the dependent interface method, the energy consumption matrix of the independent interface method, the call matrix of the basic interface method, the call matrix of the dependent interface method, and the call matrix of the independent interface method:
E all ＝EM(IF fix )NM T (IF fix )+EM(IF dep )NM T (IF dep )+EM(IF indep )NM T (IF indep )
wherein EM (IF fix ) Energy consumption matrix, NM, representing basic interface method T (IF fix ) Transpose of call matrix representing basic interface method, EM (IF dep ) Energy consumption matrix representing interface-dependent method, NM T (IF dep ) Transpose matrix representing call matrix of dependent interface method, EM (IF indep ) Energy consumption matrix, NM, representing independent interface method T (IF indep ) A transpose of the call matrix representing the independent interface method.
Alternatively, the energy consumption matrix may be:
wherein ,E1 (IFl) represents the energy consumption of the interface method, E, in the 1 st period of the cycle 2 (IFl) energy consumption of the interface method during the 2 nd period of the cycle, and so on.
The call matrix may be:
wherein ,N1 (IFl) represents the number of calls of the l interface method in the 1 st time period in the cycle, N 2 (IFl) represents the number of calls of the l interface method in the 2 nd period, and so on.
In this embodiment, the scheduling of the software operation in the target terminal to complete the optimization of the electric quantity of the target terminal with the energy consumption of the target software includes:
and acquiring operation data of the target terminal where the target software is located, and calling a pre-trained residual electric quantity estimation model to process the operation data so as to obtain the residual electric quantity.
Alternatively, a BP (Back Propagation) neural network or a convolutional neural network may be used as the residual capacity estimation model, and an optimization algorithm or a gradient descent method is used to train the residual capacity estimation model, so as to obtain a trained residual capacity estimation model.
The operation data may include a voltage and a current of the terminal battery, and the corresponding output data is an SOC (State of Charge) of the battery.
And obtaining the residual electric quantity once in each preset period, and determining the difference value of the residual electric quantities of two adjacent periods.
And judging whether the residual electric quantity is smaller than a set electric quantity threshold value or whether at least one residual electric quantity difference value is larger than a set electric quantity reduction threshold value, if so, optimizing the electric quantity of the target terminal, otherwise, keeping the normal operation of target software on the target terminal.
The method comprises the steps of determining the foreground times and the current foreground state of target software, wherein the foreground times represent the times of switching from a background to the foreground, and the foreground times represent whether the target software is positioned in the foreground or not, and the foreground times comprise the position in the foreground or the position in the background.
And keeping the current foreground state as the target software positioned at the foreground to normally run, and determining the optimizable software of which the foreground times are smaller than the set threshold value from the rest target software.
And arranging the optimizable software according to the energy consumption from large to small, and sequentially closing the optimizable software until the difference value of the energy consumption of the closed optimizable software subtracted by the difference value of the residual electric quantity is smaller than the electric quantity reduction threshold value, thereby completing the optimization of the electric quantity of the target terminal.
And when the difference value obtained by subtracting the energy consumption comprehensive of all the optimizable software from the difference value of the residual electric quantity is larger than the electric quantity reduction threshold value, directly closing all the optimizable software.
In this embodiment, the training method of the remaining power estimation model is as follows:
a1, constructing a residual electric quantity estimation model by adopting a neural network, initializing network parameters of the residual electric quantity estimation model, forming vectors by all the network parameters to obtain a training individual, and repeatedly obtaining the training individual for multiple times to obtain a training population.
Alternatively, the network parameters of the remaining power estimation model may be initialized by: setting an upper limit of the network parameter and a lower limit of the network parameter, and randomly generating the network parameter of the residual electric quantity estimation model between the upper limit of the network parameter and the lower limit of the network parameter, thereby obtaining a training individual.
A2, acquiring fitness values of all individuals in the training population, determining the training individuals with the largest fitness values as optimal individuals, taking N training individuals with the largest fitness values in the rest individuals as searching individuals, and taking M training individuals with the smallest fitness values as following individuals. Where M+N+1 is the total number of trained individual values in the training population.
Alternatively, the fitness function may be a negative value of a root mean square error, where the root mean square error is an error function commonly used in the BP neural network, and the smaller the error function is, the better the training individual is, so that the negative value of the error function may be used as the fitness function, and the greater the fitness value is, the better the training individual is.
A3, executing the walk behavior on all the search individuals, acquiring updated search individuals, and updating the optimal individuals based on the updated search individuals to obtain updated optimal individuals.
And A4, executing calling behaviors on the following individuals based on the updated optimal individuals to obtain the updated following individuals.
A5, executing the tapping behavior on the updated following individuals, and determining the following individuals after the secondary updating.
A6, determining whether the maximum fitness value in the current training population is larger than a preset threshold value or the training times reach an upper limit, if so, taking the training individuals corresponding to the maximum fitness value as final network parameters of a residual electric quantity estimation model, finishing the training of the residual electric quantity estimation model, otherwise, eliminating L training individuals with the minimum fitness value in the current training population, generating L new individuals, adding the L new individuals into the training population, and returning to the step A2.
During the training process, the network parameter may exceed the upper limit or the lower limit, so that after each training is finished, each individual may be subjected to out-of-range processing.
In this embodiment, the walk behavior is performed on all the search individuals, and the updated search individuals are obtained, including:
executing the walk behavior, and determining an updated value of each latitude in the searching individual as follows:
wherein ,xnd Represents the D-th dimension of the network parameter in the nth search individual, d=1, 2, …, D represents the total dimension of the network parameter,representing the updated value of the d-th dimensional network parameter in the nth search individual, pi representing the circumference ratio, p representing the search partyP=1, 2, …, h, h denotes the total number of search directions, T denotes the current training times, ρ denotes the adjustment coefficient, T max Represents the upper limit of the training times,/->Representing the walk step size of the d-th dimension network parameter in the walk behavior.
Judging d-th dimension network parameter x nd Whether the fitness value corresponding to the searching individual is increased or not after updating by adopting the updated value, if so, the d-th dimension network parameter x is calculated nd And carrying out updating, otherwise, not carrying out updating.
And traversing all network parameters in the nth searching individual to finish updating the nth searching individual.
The embodiment is improved based on the wolf's swarm algorithm, and the prior art is only related to the current position and the walking length, so that the source of information acquired by the wolf's swarm is less, real-time adjustment cannot be performed in time according to the global situation, the algorithm cannot jump out of a local optimal solution, and the optimizing precision of the algorithm is reduced. The walk behavior of the embodiment can increase the randomness of the search and can effectively avoid the algorithm from falling into the local optimal solution.
In this embodiment, based on the updated optimal individual, a calling action is performed on the following individual, so as to obtain the updated following individual, including:
b1, executing the attack update on the following individuals as follows:
wherein ,represents the d-th dimension network parameter in the mth following individual,>representing updated->Representing the step size of the d-th dimension network parameter in the summoning behavior, +.>Representing a d-th dimensional network parameter in the optimal individual;
optionally, in this embodiment, the step length of the attack of the d-th dimension network parameter in the calling behavior is set as an adaptive step length, which specifically is:
wherein η represents a random value between (0, 1), F g Representation ofAdaptation value of F m Representation->Is used for the adaptation value of the (c).
By setting the self-adaptive step length, the early iteration efficiency of the algorithm can be effectively improved, the resolution precision of the later stage is enhanced, and the integral training effect is improved.
And B2, judging whether the fitness value corresponding to the updated follow-up individual is larger than the fitness value corresponding to the optimal individual, if so, exchanging the positions of the follow-up individual and the optimal individual, and entering a step B3, otherwise, directly entering the step B3.
B3, judging whether the distances between all following individuals and the optimal individual are smaller than a distance threshold valueIf yes, finishing updating to obtain updated following individuals, otherwise, continuing updating the following individuals larger than the distance threshold value, and returning to the step B1.
Wherein ω represents a distance determination factor, max d Represents the maximum value, min, of the d-th-dimension network parameter d Representing the minimum value of the d-th dimensional network parameter.
In this embodiment, performing a tapping action on the updated following individual to determine a secondary updated following individual includes:
wherein ,represents the d-th dimensional network parameter in the following individuals after the mth update in the t-th training process,/for>Representing updated->Lambda represents the random constant between (-1, 1), ζ t Representing the adjustment factor during the t-th training. t=0, ζ t =1, and ζ t+1 ＝α*ζ t Alpha represents a random constant between (0.9,1), -a->Representing the step size of the attack of the d-th dimension network parameter in the attack behavior.
The present example introduces a regulatory factor ζ t Along with the continuous increase of the iteration times, the step length of the attack is continuously shortened, and the local optimizing capability of the algorithm is improved, so that more accurate training is realized.
According to the terminal electric quantity optimization method based on software energy consumption, the service time of the terminal can be effectively prolonged by acquiring the software energy consumption and optimizing the electric quantity based on the software energy consumption, the residual electric quantity is calculated, user experience can be guaranteed when the electric quantity is sufficient, running of common software of a user is preferentially guaranteed when the electric quantity is insufficient, and user experience is guaranteed when the electric quantity is optimized.
Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
It is to be understood that the invention is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims (10)

1. The terminal electric quantity optimizing method based on software energy consumption is characterized by comprising the following steps:
acquiring source programs of a plurality of pieces of software on a target terminal, compiling all the source programs into target languages corresponding to a target platform, and obtaining a plurality of target programs; the target platform is used for running the target program and determining the energy consumption of the target program;
uploading a plurality of target programs to a target platform, aiming at the plurality of target programs, and acquiring code energy consumption tables corresponding to all the target programs through the target platform;
when the target terminal runs the target software, determining the execution conditions of various interface methods in the target software, and determining the energy consumption of the target software based on the code energy consumption table and the execution conditions of various interface methods; the target software represents the running software on the target terminal;
and scheduling the software operation in the target terminal by using the energy consumption of the target software to finish the optimization of the electric quantity of the target terminal.
2. The method for optimizing terminal electric quantity based on software energy consumption according to claim 1, wherein uploading a plurality of target programs to a target platform, aiming at the plurality of target programs, and acquiring code energy consumption tables corresponding to all the target programs through the target platform, comprises:
uploading a plurality of target programs to a target platform, extracting various sentences in the target programs through the target platform, and removing repeated sentences to obtain a plurality of target sentences;
and analyzing the multi-item slogans by adopting a regression analysis method, determining the energy consumption of different target sentences, and acquiring code energy consumption tables corresponding to all target programs.
3. The terminal electric quantity optimizing method based on software energy consumption according to claim 2, wherein the method for analyzing the multi-item tagline sentence by adopting a regression analysis method, determining the energy consumption of different target taglines, and obtaining code energy consumption tables corresponding to all target programs comprises the following steps:
determining various types of basic codes in all target sentences to obtain n types of basic codes; each target sentence is composed of a plurality of basic codes;
for n basic codes, j test tasks are operated, and the execution frequency matrix is determined as follows:
wherein VCn represents an nth base code, N (VC) represents an execution count matrix, N k (VCm) represents the number of runs of the mth base code during the kth test, k=1, 2, …, j, m=1, 2, …, n;
in j test tasks, the total energy consumption matrix is obtained as follows:
E(M)＝(E(M1),E(M2),…,E(Mj)) T
wherein E (M) represents an energy consumption matrix, and E (MK) represents the total energy consumption value in the kth test process;
according to the total energy consumption matrix and the execution frequency matrix, determining a reference energy consumption sequence E (VC) corresponding to n basic codes by adopting a regression analysis method as follows:
E(VC)＝(N T (VC)N(VC)) -1 N T (VC)E(M)
wherein T represents a transpose;
based on a reference energy consumption sequence E (VC) corresponding to the n basic codes, determining the energy consumption of each target sentence, traversing all target sentences, and determining the energy consumption of each target sentence;
and determining a code energy consumption table of the mutual mapping of the target sentences and the energy consumption according to the energy consumption of all the target sentences.
4. The method for optimizing terminal power consumption based on software according to claim 3, wherein when the target terminal runs the target software, determining execution conditions of various interface methods in the target software, and determining energy consumption of the target software based on the code energy consumption table and the execution conditions of the various interface methods, comprises:
when a target terminal runs target software, tracking the execution conditions of a basic interface method, a dependent interface method and an independent interface method which are run in the target software to obtain an energy consumption matrix of the basic interface method, an energy consumption matrix of the dependent interface method and an energy consumption matrix of the independent interface method, and simultaneously obtaining a calling matrix of the basic interface method, a calling matrix of the dependent interface method and a calling matrix of the independent interface method;
and determining the energy consumption of the target software based on the energy consumption matrix of the basic interface method, the energy consumption matrix of the dependent interface method, the energy consumption matrix of the independent interface method, the calling matrix of the basic interface method, the calling matrix of the dependent interface method and the calling matrix of the independent interface method.
5. The method for optimizing terminal power consumption based on software energy consumption according to claim 4, wherein the energy consumption of the target software is determined based on an energy consumption matrix of a basic interface method, an energy consumption matrix of a dependent interface method, an energy consumption matrix of an independent interface method, a call matrix of the basic interface method, a call matrix of the dependent interface method, and a call matrix of the independent interface method:
E all ＝EM(IF fix )NM T (IF fix )+EM(IF dep )NM T (IF dep )+EM(IF indep )NM T (IF indep )
wherein EM (IF fix ) Energy consumption matrix, NM, representing basic interface method T (IF fix ) Transpose of call matrix representing basic interface method, EM (IF dep ) Energy consumption matrix representing interface-dependent method, NM T (IF dep ) Transpose matrix representing call matrix of dependent interface method, EM (IF indep ) Energy consumption matrix, NM, representing independent interface method T (IF indep ) A transpose of the call matrix representing the independent interface method.
6. The method for optimizing terminal power based on software energy consumption according to claim 5, wherein the scheduling the software operation in the target terminal to complete the optimization of the terminal power based on the energy consumption of the target software comprises:
collecting operation data of a target terminal where target software is located, and calling a pre-trained residual electric quantity estimation model to process the operation data so as to obtain residual electric quantity;
obtaining the residual electric quantity once in each preset period, and determining the residual electric quantity difference value of two adjacent periods;
judging whether the residual electric quantity is smaller than a set electric quantity threshold value or whether at least one residual electric quantity difference value is larger than a set electric quantity reduction threshold value, if so, optimizing the electric quantity of the target terminal, otherwise, keeping the normal operation of target software on the target terminal;
determining the foreground times and the current foreground state of target software, wherein the foreground times represent the times of switching from a background to a foreground, and the foreground times represent whether the target software is positioned in the foreground or not, and the foreground times comprise the position in the foreground or the position in the background;
keeping the current foreground state as the normal operation of the target software positioned at the foreground, and determining the optimizable software of which the foreground times are smaller than a set threshold value from the rest target software;
arranging the optimizable software according to the energy consumption from large to small, and sequentially closing the optimizable software until the difference value of the residual electric quantity difference value minus the energy consumption of the closed optimizable software is smaller than the electric quantity reduction threshold value, so as to finish the optimization of the electric quantity of the target terminal;
and when the difference value obtained by subtracting the energy consumption comprehensive of all the optimizable software from the difference value of the residual electric quantity is larger than the electric quantity reduction threshold value, directly closing all the optimizable software.
7. The method for optimizing terminal power consumption based on software according to claim 6, wherein the training method of the remaining power estimation model is as follows:
a1, constructing a residual electric quantity estimation model by adopting a neural network, initializing network parameters of the residual electric quantity estimation model, forming vectors by all the network parameters to obtain a training individual, and repeatedly obtaining multiple training individuals to obtain a training population;
a2, acquiring fitness values of all individuals in the training population, determining the training individuals with the largest fitness values as optimal individuals, taking N training individuals with the largest fitness values in the rest individuals as searching individuals, and taking M training individuals with the smallest fitness values as following individuals; wherein M+N+1 is the total number of training individual values in the training population;
a3, executing the walk behavior on all the search individuals, acquiring updated search individuals, and updating the optimal individuals based on the updated search individuals to obtain updated optimal individuals;
a4, executing calling behaviors on the following individuals on the basis of the updated optimal individuals to obtain updated following individuals;
a5, executing a tapping behavior on the updated following individuals, and determining the following individuals after secondary updating;
a6, determining whether the maximum fitness value in the current training population is larger than a preset threshold value or the training times reach an upper limit, if so, taking the training individuals corresponding to the maximum fitness value as final network parameters of a residual electric quantity estimation model, finishing the training of the residual electric quantity estimation model, otherwise, eliminating L training individuals with the minimum fitness value in the current training population, generating L new individuals, adding the L new individuals into the training population, and returning to the step A2.
8. The method for optimizing terminal power consumption based on software according to claim 7, wherein the step of performing walk behavior on all search individuals to obtain updated search individuals comprises:
executing the walk behavior, and determining an updated value of each latitude in the searching individual as follows:
wherein ,xnd Represents the D-th dimension of the network parameter in the nth search individual, d=1, 2, …, D represents the total dimension of the network parameter,representing the updated value of the d-th dimensional network parameter in the nth search individual, pi representing the circumference ratio, p representing the search direction, p=1, 2, …, h, h representing the total number of search directions, T representing the current training number, ρ representing the adjustment coefficient, T max Represents the upper limit of the training times,/->Representing the wander step length of the d-th dimension network parameter in wander behavior;
judging d-th dimension network parameter x nd Whether the fitness value corresponding to the searching individual is increased or not after updating by adopting the updated value, if so, the d-th dimension network parameter x is calculated nd Updating, otherwise, not updating;
and traversing all network parameters in the nth searching individual to finish updating the nth searching individual.
9. The method for optimizing terminal power consumption based on software according to claim 8, wherein the step of performing a calling action on the following individual based on the updated optimal individual to obtain the updated following individual includes:
b1, executing the attack update on the following individuals as follows:
wherein ,represents the d-th dimension network parameter in the mth following individual,>representing updated-> Representing the step size of the d-th dimension network parameter in the summoning behavior, +.>Representing a d-th dimensional network parameter in the optimal individual;
b2, judging whether the fitness value corresponding to the updated following individual is larger than the fitness value corresponding to the optimal individual, if so, exchanging the positions of the following individual and the optimal individual, and entering a step B3, otherwise, directly entering the step B3;
b3, judging whether the distances between all following individuals and the optimal individual are smaller than a distance threshold valueIf yes, finishing updating to obtain updated following individuals, otherwise, continuing updating the following individuals larger than the distance threshold value, and returning to the step B1;
wherein ω represents a distance determination factor, max d Representing d-th dimensional network parametersMaximum value of (min) d Representing the minimum value of the d-th dimensional network parameter.
10. The method for optimizing terminal power consumption based on software according to claim 9, wherein the step of performing a tapping operation on the updated following individual to determine the following individual after the second update includes:
wherein ,represents the d-th dimensional network parameter in the following individuals after the mth update in the t-th training process,/for>Representing updated->Lambda represents the random constant between (-1, 1), ζ t Representing the adjustment factor during the t-th training; t=0, ζ t =1, and ζ t+1 ＝α*ζ t Alpha represents a random constant between (0.9,1), -a->Representing the step size of the attack of the d-th dimension network parameter in the attack behavior.
CN202310694825.7A 2023-06-12 2023-06-12 Terminal electric quantity optimization method based on software energy consumption Active CN116822340B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310694825.7A CN116822340B (en) 2023-06-12 2023-06-12 Terminal electric quantity optimization method based on software energy consumption

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310694825.7A CN116822340B (en) 2023-06-12 2023-06-12 Terminal electric quantity optimization method based on software energy consumption

Publications (2)

Publication Number Publication Date
CN116822340A true CN116822340A (en) 2023-09-29
CN116822340B CN116822340B (en) 2024-05-28

Family

ID=88123386

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310694825.7A Active CN116822340B (en) 2023-06-12 2023-06-12 Terminal electric quantity optimization method based on software energy consumption

Country Status (1)

Country Link
CN (1) CN116822340B (en)

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104199696A (en) * 2014-08-28 2014-12-10 刘勇 Electric consumption control method and system and intelligent terminal
CN104484031A (en) * 2014-12-30 2015-04-01 北京奇虎科技有限公司 Method and device for optimizing power consumption state of terminal equipment
CN105446457A (en) * 2015-12-09 2016-03-30 北京金山安全软件有限公司 Method and device for optimizing power consumption of application program and electronic equipment
CN106610873A (en) * 2016-11-17 2017-05-03 北京大学（天津滨海）新代信息技术研究院 Method for predicting application energy consumption on Android equipment
CN106714289A (en) * 2017-03-14 2017-05-24 西安电子科技大学 Smart phone dump energy management method based on application combination power consumption prediction
CN106933323A (en) * 2015-12-30 2017-07-07 北京金山安全软件有限公司 Method and device for optimizing power consumption of application program and electronic equipment
CN107357703A (en) * 2017-06-16 2017-11-17 北京奇安信科技有限公司 A kind of terminal applies power consumption detection method and server
CN107809542A (en) * 2017-11-14 2018-03-16 广东欧珀移动通信有限公司 application control method, device, storage medium and electronic equipment
CN108089970A (en) * 2018-01-11 2018-05-29 深圳市金立通信设备有限公司 Predict method, terminal and the storage medium of remaining capacity up time
CN108279764A (en) * 2017-01-04 2018-07-13 北京搜狗科技发展有限公司 A kind of application control method, device and mobile terminal
CN110320991A (en) * 2019-06-24 2019-10-11 维沃移动通信有限公司 A kind of method for managing power supply and device of mobile terminal
US20200226433A1 (en) * 2017-09-30 2020-07-16 Guangdong Oppo Mobile Telecommunications Corp., Ltd. Method and Device for Controlling Application, Storage Medium, and Electronic Device
CN112882762A (en) * 2021-03-25 2021-06-01 广州文石信息科技有限公司 Power consumption optimization method, device, equipment and storage medium for ink screen equipment

Patent Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104199696A (en) * 2014-08-28 2014-12-10 刘勇 Electric consumption control method and system and intelligent terminal
CN104484031A (en) * 2014-12-30 2015-04-01 北京奇虎科技有限公司 Method and device for optimizing power consumption state of terminal equipment
CN105446457A (en) * 2015-12-09 2016-03-30 北京金山安全软件有限公司 Method and device for optimizing power consumption of application program and electronic equipment
CN106933323A (en) * 2015-12-30 2017-07-07 北京金山安全软件有限公司 Method and device for optimizing power consumption of application program and electronic equipment
CN106610873A (en) * 2016-11-17 2017-05-03 北京大学（天津滨海）新代信息技术研究院 Method for predicting application energy consumption on Android equipment
CN108279764A (en) * 2017-01-04 2018-07-13 北京搜狗科技发展有限公司 A kind of application control method, device and mobile terminal
CN106714289A (en) * 2017-03-14 2017-05-24 西安电子科技大学 Smart phone dump energy management method based on application combination power consumption prediction
CN107357703A (en) * 2017-06-16 2017-11-17 北京奇安信科技有限公司 A kind of terminal applies power consumption detection method and server
US20200226433A1 (en) * 2017-09-30 2020-07-16 Guangdong Oppo Mobile Telecommunications Corp., Ltd. Method and Device for Controlling Application, Storage Medium, and Electronic Device
CN107809542A (en) * 2017-11-14 2018-03-16 广东欧珀移动通信有限公司 application control method, device, storage medium and electronic equipment
CN108089970A (en) * 2018-01-11 2018-05-29 深圳市金立通信设备有限公司 Predict method, terminal and the storage medium of remaining capacity up time
CN110320991A (en) * 2019-06-24 2019-10-11 维沃移动通信有限公司 A kind of method for managing power supply and device of mobile terminal
CN112882762A (en) * 2021-03-25 2021-06-01 广州文石信息科技有限公司 Power consumption optimization method, device, equipment and storage medium for ink screen equipment

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
"史上最全！Android程序性能优化之耗电优化（一", Retrieved from the Internet <URL:https://zhuanlan.zhihu.com/p/134483128> *
"移动性能测试 Android 后台耗电分析及优化", Retrieved from the Internet <URL:https://testerhome.com/topics/27118> *
刘雪冰: "面向代码结构的软件模块能耗评估方法", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 6, 15 March 2016 (2016-03-15), pages 15 - 44 *
周文迪: "面向C程序的软件功耗评估方法研究与应用", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 2, 15 February 2022 (2022-02-15), pages 138 - 157 *
宋杰等: "面向代码的软件能耗优化研究进展", 《计算机学报》, no. 11, 31 December 2016 (2016-12-31), pages 112 - 132 *
罗刚等: "源程序级和算法级嵌入式软件功耗特性的分析与优化方法研究", 《计算机学报》, no. 09, 15 September 2009 (2009-09-15), pages 177 - 183 *

Also Published As

Publication number Publication date
CN116822340B (en) 2024-05-28

Similar Documents

Publication Publication Date Title
Su et al. Sample-efficient actor-critic reinforcement learning with supervised data for dialogue management
CN109902222B (en) Recommendation method and device
CN112668128B (en) Method and device for selecting terminal equipment nodes in federal learning system
CN112052948B (en) Network model compression method and device, storage medium and electronic equipment
CN108376028B (en) Method based on user behavior and system mode prediction mobile device battery durable
CN111275185A (en) Energy use state early warning method, device, equipment and storage medium
CN116703009B (en) Operation reference information generation method of photovoltaic power generation energy storage system
CN111008790A (en) Hydropower station group power generation electric scheduling rule extraction method
CN109925718A (en) A kind of system and method for distributing the micro- end map of game
CN111339066A (en) Database optimization method and device, electronic equipment and computer-readable storage medium
CN112990958A (en) Data processing method, data processing device, storage medium and computer equipment
CN116822340B (en) Terminal electric quantity optimization method based on software energy consumption
CN116595356B (en) Time sequence signal prediction method and device, electronic equipment and storage medium
Arunadevi et al. Comparison of feature selection strategies for classification using rapid miner
CN112633516B (en) Performance prediction and machine learning compiling optimization method and device
CN114489966A (en) Job scheduling method and device
CN114841451A (en) Driver travel subsidy method and device and storage medium
CN107480768A (en) Bayesian network structure adaptive learning method and device, storage device and terminal device
CN112070200B (en) Harmonic group optimization method and application thereof
CN111931994A (en) Short-term load and photovoltaic power prediction method, system, equipment and medium thereof
CN112489663A (en) Voice wake-up method, device, medium and equipment
CN117993544A (en) User load data prediction method and device and electronic equipment
CN117009092B (en) Method and system for dynamically distributing compiling time resources based on multiple multi-arm slot machines
CN117892667B (en) Method for setting arithmetic unit chip, computing subsystem and intelligent computing platform
Branchi et al. Recommending the optimal policy by learning to act from temporal data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant