CN108738046B - Mobile edge calculation rate maximization method based on semi-supervised learning - Google Patents
Mobile edge calculation rate maximization method based on semi-supervised learning Download PDFInfo
- Publication number
- CN108738046B CN108738046B CN201810343311.6A CN201810343311A CN108738046B CN 108738046 B CN108738046 B CN 108738046B CN 201810343311 A CN201810343311 A CN 201810343311A CN 108738046 B CN108738046 B CN 108738046B
- Authority
- CN
- China
- Prior art keywords
- wireless device
- wireless devices
- base station
- energy
- wireless
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/06—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/10—Flow control between communication endpoints
Abstract
A mobile edge calculation rate maximization method based on semi-supervised learning comprises the following steps: 1) each wireless device needs to establish contact with the base station; 2) using two non-overlapping sets M0And M1Representing all wireless devices in local compute mode and offload mode, respectively; 3) in the set M0Can collect energy and process local tasks simultaneously while in the set M1The wireless equipment in the system can only shunt the task to the base station for processing after collecting energy; 4) the mode selection of all wireless devices will be through their channel gain hiThe decision, semi-supervised learning, is to take as input their channel gains and then generate an optimal mode selection that maximizes the aggregate computation rate for all wireless devices, i.e., to decide which wireless devices have their tasks processed locally and which to offload to the base station. The invention maximizes the total calculation rate of all wireless devices on the premise of ensuring user experience.
Description
Technical Field
The invention belongs to the field of communication, and particularly relates to a communication system for mobile edge calculation and a mobile edge calculation rate maximization method based on semi-supervised learning.
Background
The recent development of internet of things technology is a key step towards real intelligence and autonomous control, and is particularly prominent in many important industrial and commercial systems. In an internet of things network, a large number of Wireless Devices (WDs) capable of communication and computing are deployed, and due to device size limitations and manufacturing cost considerations, internet of things devices (e.g., sensors) often carry batteries with limited capacity and energy-efficient low-performance processors, and therefore, the limited device lifetime and low computing power cannot support more and more sustainable new applications that require high-performance computing, such as autopilot and augmented reality. Deployment of wireless energy Transfer Systems (WPTs) can solve the two aforementioned performance problems, but frequent device battery failures not only disrupt normal personal wireless device operation but can also significantly degrade overall network performance, e.g., sensing accuracy in wireless sensor networks. Conventional wireless systems require frequent manual battery replacement, which is expensive and inconvenient, and due to severe battery capacity limitations, minimizing power consumption and extending the operational life of the wireless device is a critical design in battery-powered wireless systems. Each energy harvesting wireless device follows a binary computation offload policy, i.e., the data set for one task may be performed locally or by remote server offload. In order to maximize the total computation rate of all wireless devices, it is necessary to find the optimal individual computation mode selection.
Disclosure of Invention
In order to overcome the defect that the sum calculation rate of the existing wireless energy transmission system is low, in order to maximize the sum calculation rate of all wireless devices and find the optimal individual calculation mode selection and system transmission time allocation, the invention provides a mobile edge calculation rate maximization method based on semi-supervised learning, and the sum calculation rate of all wireless devices is maximized on the premise of ensuring user experience.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method for maximizing a moving edge computation rate based on semi-supervised learning, the method comprising the following steps:
1) in an edge computing system powered wirelessly by a base station and a plurality of wireless devices, the base station and each wireless device having a separate antenna; the radio frequency energy emitter and the edge calculation server are integrated in the base station, and the base station is assumed to have a stable energy supply and can broadcast radio frequency energy to all wireless devices; each wireless device has an energy harvesting circuit and a rechargeable battery to perform some task by storing harvested energy; in this wireless communication system, each wireless device needs to communicate withBase station establishing contact, channel gain h between wireless device i and base stationiThe calculation is as follows:
wherein, each parameter is defined as follows:
Ad: antenna gain;
pi: a circumferential ratio;
fc: a carrier frequency;
di: distance between wireless device i and base station;
de: a path loss exponent;
2) assuming that the computing tasks of each wireless device can be executed on a local low-performance microprocessor or offloaded to an edge computing server with more powerful processing power, it will process the computing tasks and then send the results back to the wireless device; suppose a wireless device employs a binary computation offload rule, i.e., a wireless device must choose either a local computation mode or an offload mode; using two non-overlapping sets
And
all wireless devices, all sets of wireless devices, representing local compute mode and offload mode, respectively
Expressed as:
3) in a collection
In a wireless deviceCapable of harvesting energy and simultaneously processing local tasks while in aggregate
The wireless device in the system can only shunt the task to the base station for processing after collecting energy, and under the condition that the computing capacity and transmission capacity of the base station are much stronger than those of the energy collection wireless device, the wireless device exhausts the energy collected by the wireless device in the task shunting process; the compute rate sum maximization problem for all wireless devices is described as:
the constraint conditions are as follows:
in the formula:
wherein, each parameter is defined as follows:
ωi: a transition weight for the ith wireless device;
μ: an energy collection efficiency;
p: radio frequency energy transmission power;
phi: the number of calculation cycles required to process each bit of data;
hi: channel gain of the ith wireless device;
ki: an energy efficiency coefficient for the ith wireless device;
a: a time coefficient;
vμ: conversion efficiency;
b: a bandwidth;
τj: a time coefficient for the jth wireless device;
N0: the number of wireless devices in the local processing mode;
4) the mode selection of all wireless devices will be through their channel gain hiDetermining, the semi-supervised learning is used for taking the channel gain of the wireless devices as input, and then generating an optimal mode selection capable of maximizing the sum computation rate of all the wireless devices, namely determining which wireless devices have tasks processed locally and which are shunted to a base station for processing; the neural network in semi-supervised learning is used for pattern prediction and is also the core of the learning process, which receives the channel gains h of all wireless devicesiThen a pattern selection is predicted, but the pattern selection is not necessarily optimal, so that a pattern is generated on the basis of the predicted pattern, then the best one of the two patterns is selected by calculation, and finally the good pattern is used as a prediction target of the neural network to optimize the neural network, so that the neural network can be predicted to be more accurate in the next prediction. In the process of repeating the process, the neural network can predict more and more accurate until convergence.
Further, in the step 4), an iterative process of semi-supervised learning is as follows:
step 4.1: initializing an evaluation neural network in semi-supervised learning, and initializing an iteration number k to 1;
step 4.2: when K is less than or equal to a given number of iterations K, the channel gains h of all wireless devices are adjustediAs input to the neural network, a preliminary mode selection is predicted;
step 4.3: generating a different mode selection based on the preliminary prediction mode selection;
step 4.4: selecting one of the two modes which enables the total rate of all wireless devices to be larger through calculation;
step 4.5: channel gain h for better mode selection and inputiPairing to form a group of labeled data for neural network learning;
step 4.6: continuously reducing the error of the neural network by using a gradient descent algorithm and data of the previous process, continuously optimizing the neural network, predicting the error more accurately, and returning to the step 4.2 when k is equal to k + 1;
step 4.7: and when K is larger than the given iteration number K, finishing the learning process to obtain the optimal mode selection.
The technical conception of the invention is as follows: first, in an internet of things network, a large number of Wireless Devices (WDs) capable of communication and computation are deployed, and due to device size constraints and manufacturing cost considerations, internet of things devices (e.g., sensors) often carry batteries with limited capacity and energy-saving low-performance processors, so that the limited device lifetime and low computing power cannot support more and more sustainable new applications requiring high-performance computation, and due to strict battery capacity constraints, in a battery-powered wireless system, minimizing energy consumption and extending the wireless device operational life cycle is a critical design. Each energy harvesting wireless device follows a binary computation offload policy, i.e., the data set for one task may be performed locally or by remote server offload. To maximize the total computation rate of all wireless devices, an optimal individual computation mode selection method is proposed.
The invention has the following beneficial effects: an optimal mode selection method is found through semi-supervised learning, the sum calculation rate of all wireless devices is maximized, the energy consumption is minimized, and the operation life cycle of the wireless devices is prolonged.
Drawings
FIG. 1 is a system model diagram.
Fig. 2 is a flow chart of a method of finding an optimal mode selection.
Detailed Description
The present invention is described in further detail below with reference to the attached drawing figures.
Referring to fig. 1 and 2, a mobile edge computation rate maximization method based on semi-supervised learning maximizes the sum computation rate of all wireless devices, minimizes energy consumption, and prolongs the wireless device operation life cycle. The present invention proposes an optimal individual computation mode selection method to decide which wireless devices will be tasked with offloading to the base station based on a system model of multiple wireless devices (as shown in fig. 1). The optimal individual calculation mode selection method comprises the following steps (as shown in fig. 2):
1) in an edge computing system powered wirelessly by a base station and a plurality of wireless devices, the base station and each wireless device having a separate antenna; the radio frequency energy emitter and the edge calculation server are integrated in the base station, and the base station is assumed to have a stable energy supply and can broadcast radio frequency energy to all wireless devices; each wireless device has an energy harvesting circuit and a rechargeable battery, and can perform some tasks by storing harvested energy; in this wireless communication system, each wireless device needs to establish contact with a base station, and the channel gain h between the wireless device i and the base stationiThe calculation is as follows:
wherein, each parameter is defined as follows:
Ad: antenna gain;
pi: a circumferential ratio;
fc: a carrier frequency;
di: is free ofThe distance between the line device i and the base station;
de: a path loss exponent;
2) assuming that the computing tasks of each wireless device are executed on a local low-performance microprocessor or offloaded to an edge computing server with greater processing power, it will process the computing tasks and then send the results back to the wireless device; suppose a wireless device employs a binary computation offload rule, i.e., a wireless device must choose either a local computation mode or an offload mode; using two non-overlapping sets
And
all wireless devices, all sets of wireless devices, representing local compute mode and offload mode, respectively
Expressed as:
3) in a collection
The wireless device in (1) is able to collect energy and process local tasks simultaneously while in the aggregate
The wireless device in (1) can only shunt the task to the base station for processing after collecting energy, and assuming that the computing power and transmission capability of the base station are much stronger than those of the energy collecting wireless device, in this case, in the task shunting process, the wireless device exhausts the energy collected by the wireless device, and the problem of maximizing the sum of the computing rates of all the wireless devices is described as follows:
the constraint conditions are as follows:
in the formula:
wherein, each parameter is defined as follows:
ωi: a transition weight for the ith wireless device;
μ: an energy collection efficiency;
p: radio frequency energy transmission power;
phi: the number of calculation cycles required to process each bit of data;
hi: channel gain of the ith wireless device;
ki: an energy efficiency coefficient for the ith wireless device;
a: a time coefficient;
vμ: conversionEfficiency;
b: a bandwidth;
τj: a time coefficient for the jth wireless device;
N0: the number of wireless devices in the local processing mode;
4) the mode selection of all wireless devices will be through their channel gain hiThe decision, semi-supervised learning, is to take as input their channel gains and then generate an optimal mode selection that maximizes the aggregate computation rate for all wireless devices, i.e., to decide which wireless devices have their tasks processed locally and which to offload to the base station. The neural network in semi-supervised learning is mainly used for mode prediction and is also the core of the learning process, and receives the channel gains h of all wireless devicesiThen a pattern selection is predicted, but the pattern selection is not necessarily optimal, so that a pattern is generated on the basis of the predicted pattern, then the best one of the two patterns is selected by calculation, and finally the good pattern is used as a prediction target of the neural network to optimize the neural network, so that the neural network can be predicted to be more accurate in the next prediction. In the process of repeating the process, the neural network can predict more and more accurate until convergence.
Further, in the step 4), an iterative process of semi-supervised learning is as follows:
step 4.1: initializing an evaluation neural network in semi-supervised learning, and initializing an iteration number k to 1;
step 4.2: when K is less than or equal to a given number of iterations K, the channel gains h of all wireless devices are adjustediAs input to the neural network, a preliminary mode selection is predicted;
step 4.3: generating a different mode selection based on the preliminary prediction mode selection;
step 4.4: selecting one of the two modes which enables the total rate of all wireless devices to be larger through calculation;
step 4.5: channel gain h for better mode selection and inputiAre paired to form a groupThe data with labels is used for neural network learning;
step 4.6: continuously reducing the error of the neural network by using a gradient descent algorithm and data of the previous process, continuously optimizing the neural network, predicting the error more accurately, and returning to the step 4.2 when k is equal to k + 1;
step 4.7: and when K is larger than the given iteration number K, finishing the learning process to obtain the optimal mode selection.
Claims (2)
1. A method for maximizing a moving edge calculation rate based on semi-supervised learning, which is characterized by comprising the following steps:
1) in an edge computing system powered wirelessly by a base station and a plurality of wireless devices, the base station and each wireless device having a separate antenna; the radio frequency energy emitter and the edge computing server are integrated in the base station, and the base station is supposed to have a stable energy supply and can broadcast radio frequency energy to all wireless devices, each wireless device is provided with an energy collecting circuit and a rechargeable battery, and the tasks are completed by storing the collected energy; in this wireless communication system, each wireless device needs to establish contact with a base station, and the channel gain h between the wireless device i and the base stationiThe calculation is as follows:
wherein, each parameter is defined as follows:
Ad: antenna gain;
pi: a circumferential ratio;
fc: a carrier frequency;
di: distance between wireless device i and base station;
de: a path loss exponent;
2) it is assumed that the computational tasks of each wireless device can be performed on a local low-performance microprocessor or split to edge computations with more powerful processing powerA server that will process the computing task and then send the results back to the wireless device; suppose a wireless device employs a binary computation offload rule, i.e., a wireless device must choose either a local computation mode or an offload mode; using two non-overlapping sets
And
all wireless devices, all sets of wireless devices, representing local compute mode and offload mode, respectively
Expressed as:
3) in a collection
The wireless device in (1) is able to collect energy and process local tasks simultaneously while in the aggregate
The wireless device in (1) can only shunt the task to the base station for processing after collecting energy, and assuming that the computing power and transmission capability of the base station are much stronger than those of the energy collecting wireless device, in this case, in the task shunting process, the wireless device exhausts the energy collected by the wireless device, and the problem of maximizing the sum of the computing rates of all the wireless devices is described as follows:
the constraint conditions are as follows:
in the formula:
wherein, each parameter is defined as follows:
ωi: a transition weight for the ith wireless device;
μ: an energy collection efficiency;
p: radio frequency energy transmission power;
phi: the number of calculation cycles required to process each bit of data;
hi: channel gain of the ith wireless device;
ki: an energy efficiency coefficient for the ith wireless device;
α: a time coefficient;
vμ: conversion efficiency;
b: a bandwidth;
τj: a time coefficient for the jth wireless device;
N0: the number of wireless devices in the local processing mode;
4) the mode selection of all wireless devices will be through their channel gain hiDetermining, the semi-supervised learning is used for taking the channel gain of the wireless devices as input, and then generating an optimal mode selection capable of maximizing the sum computation rate of all the wireless devices, namely determining which wireless devices have tasks processed locally and which are shunted to a base station for processing; the neural network in semi-supervised learning is mainly used for mode prediction and is also the core of the learning process, and receives the channel gains h of all wireless devicesiThen predicting a mode selection, but the mode selection is not necessarily optimal, so that a mode is generated on the basis of the predicted mode, then the best one of the two modes is selected through calculation, and finally the good mode is used as a prediction target of the neural network to optimize the neural network so that the neural network can be predicted to be more accurate in the next prediction; in the process of repeating the process, the neural network can predict more and more accurate until convergence.
2. The method of claim 1, wherein the method for maximizing the computation rate of the moving edge based on semi-supervised learning comprises: in the step 4), the iteration process of semi-supervised learning is as follows:
step 4.1: initializing an evaluation neural network in semi-supervised learning, and initializing an iteration number k to 1;
step 4.2: when K is less than or equal to a given number of iterations K, the channel gains h of all wireless devices are adjustediAs input to the neural network, a preliminary mode selection is predicted;
step 4.3: generating a different mode selection based on the preliminary prediction mode selection;
step 4.4: selecting one of the two modes which enables the total rate of all wireless devices to be larger through calculation;
step 4.5: channel for selecting and inputting more effective modeGain hiPairing to form a group of labeled data for neural network learning;
step 4.6: continuously reducing the error of the neural network by using a gradient descent algorithm and data of the previous process, continuously optimizing the neural network, predicting the error more accurately, and returning to the step 4.2 when k is equal to k + 1;
step 4.7: and when K is larger than the given iteration number K, finishing the learning process to obtain the optimal mode selection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810343311.6A CN108738046B (en) | 2018-04-17 | 2018-04-17 | Mobile edge calculation rate maximization method based on semi-supervised learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810343311.6A CN108738046B (en) | 2018-04-17 | 2018-04-17 | Mobile edge calculation rate maximization method based on semi-supervised learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108738046A CN108738046A (en) | 2018-11-02 |
| CN108738046B true CN108738046B (en) | 2021-04-06 |
Family
ID=63938979
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810343311.6A Active CN108738046B (en) | 2018-04-17 | 2018-04-17 | Mobile edge calculation rate maximization method based on semi-supervised learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108738046B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111988791B (en) * | 2019-05-22 | 2023-06-20 | 国网能源研究院有限公司 | Fog calculation-based wireless charging network node computing capacity improving method and system |
| CN111464208A (en) * | 2020-03-09 | 2020-07-28 | 深圳大学 | Passive edge computing system based on spread spectrum communication, task unloading method and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107708135A (en) * | 2017-07-21 | 2018-02-16 | 上海交通大学 | A kind of resource allocation methods for being applied to mobile edge calculations scene |
| CN107734558A (en) * | 2017-10-26 | 2018-02-23 | 北京邮电大学 | A kind of control of mobile edge calculations and resource regulating method based on multiserver |
| CN107846704A (en) * | 2017-10-26 | 2018-03-27 | 北京邮电大学 | A kind of resource allocation and base station service arrangement method based on mobile edge calculations |
| CN107872823A (en) * | 2016-09-28 | 2018-04-03 | 维布络有限公司 | The method and system of communication operational mode in the mobile edge calculations environment of identification |
| US9942825B1 (en) * | 2017-03-27 | 2018-04-10 | Verizon Patent And Licensing Inc. | System and method for lawful interception (LI) of Network traffic in a mobile edge computing environment |
| CN107911242A (en) * | 2017-11-15 | 2018-04-13 | 北京工业大学 | A kind of cognitive radio based on industry wireless network and edge calculations method |
-
2018
- 2018-04-17 CN CN201810343311.6A patent/CN108738046B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107872823A (en) * | 2016-09-28 | 2018-04-03 | 维布络有限公司 | The method and system of communication operational mode in the mobile edge calculations environment of identification |
| US9942825B1 (en) * | 2017-03-27 | 2018-04-10 | Verizon Patent And Licensing Inc. | System and method for lawful interception (LI) of Network traffic in a mobile edge computing environment |
| CN107708135A (en) * | 2017-07-21 | 2018-02-16 | 上海交通大学 | A kind of resource allocation methods for being applied to mobile edge calculations scene |
| CN107734558A (en) * | 2017-10-26 | 2018-02-23 | 北京邮电大学 | A kind of control of mobile edge calculations and resource regulating method based on multiserver |
| CN107846704A (en) * | 2017-10-26 | 2018-03-27 | 北京邮电大学 | A kind of resource allocation and base station service arrangement method based on mobile edge calculations |
| CN107911242A (en) * | 2017-11-15 | 2018-04-13 | 北京工业大学 | A kind of cognitive radio based on industry wireless network and edge calculations method |
Non-Patent Citations (1)
| Title |
|---|
| Computation Rate Maximization for Wireless Powered Mobile-Edge Computing With Binary Computation Offloading;Suzhi BI等;《IEEE Transactions on Wireless Communications》;20180409;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108738046A (en) | 2018-11-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108632860B (en) | Mobile edge calculation rate maximization method based on deep reinforcement learning | |
| Huang et al. | Deep reinforcement learning for online computation offloading in wireless powered mobile-edge computing networks | |
| Zhang et al. | Clustering algorithms for maximizing the lifetime of wireless sensor networks with energy-harvesting sensors | |
| Mahmood et al. | Partial offloading in energy harvested mobile edge computing: A direct search approach | |
| Leinonen et al. | Distributed joint resource and routing optimization in wireless sensor networks via alternating direction method of multipliers | |
| CN108738046B (en) | Mobile edge calculation rate maximization method based on semi-supervised learning | |
| Xie et al. | Backscatter-assisted computation offloading for energy harvesting IoT devices via policy-based deep reinforcement learning | |
| CN112835715B (en) | Method and device for determining task unloading strategy of unmanned aerial vehicle based on reinforcement learning | |
| CN112416603B (en) | Combined optimization system and method based on fog calculation | |
| CN113727362B (en) | Unloading strategy method of wireless power supply system based on deep reinforcement learning | |
| Tharini et al. | An Energy Efficient Spatial Correlation Base Data Gathering Algorithm for Wireless Sensor Networks | |
| CN113781002B (en) | Low-cost workflow application migration method based on agent model and multiple group optimization in cloud edge cooperative network | |
| CN104767833A (en) | Cloud transferring method for computing tasks of mobile terminal | |
| CN115696451A (en) | Optimization method for energy and task scheduling of edge computing system | |
| WO2022242468A1 (en) | Task offloading method and apparatus, scheduling optimization method and apparatus, electronic device, and storage medium | |
| CN108738045B (en) | Moving edge calculation rate maximization method based on depth certainty strategy gradient | |
| Al-Qurabat et al. | A Bi-level data lowering method to minimize transferring big data in the sensors of IoT applications | |
| JP2007013983A (en) | Method and apparatus for dynamic energy management in wireless sensor network | |
| CN114521023A (en) | SWIPT-assisted NOMA-MEC system resource allocation modeling method | |
| CN108650710B (en) | Resource allocation method of wireless energy supply communication network based on hybrid multiple access | |
| Acheampong et al. | A parallel computing based model for online binary computation offloading in mobile edge computing | |
| Xu et al. | Cybertwin assisted wireless asynchronous federated learning mechanism for edge computing | |
| Lin et al. | Maximum data collection rate routing for data gather trees with data aggregation in rechargeable wireless sensor networks | |
| Arunraja et al. | Distributed similarity based clustering and compressed forwarding for wireless sensor networks | |
| CN111162852B (en) | Ubiquitous power Internet of things access method based on matching learning |
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 | ||
| GR01 | Patent grant |