CN112418342A - Mobile equipment calculation force prediction method and device under federated learning scene - Google Patents
Mobile equipment calculation force prediction method and device under federated learning scene Download PDFInfo
- Publication number
- CN112418342A CN112418342A CN202011418124.3A CN202011418124A CN112418342A CN 112418342 A CN112418342 A CN 112418342A CN 202011418124 A CN202011418124 A CN 202011418124A CN 112418342 A CN112418342 A CN 112418342A
- Authority
- CN
- China
- Prior art keywords
- information
- mobile equipment
- network
- resource state
- equipment
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The embodiment of the application discloses a method and a device for predicting computing power of mobile equipment in a federated learning scene, wherein the method comprises the steps of collecting data information on the mobile equipment; clustering the position information to obtain the hot spot position of the mobile equipment; recording the stay time of the mobile equipment at the hot spot position through the time information; calculating the average network state and the average resource state of the mobile equipment at the hotspot positions on different dates; and inputting the preprocessed data serving as training data into a recurrent neural network model for prediction to obtain and output network information and resource state information of the mobile equipment after a preset time period. Based on the mode provided by the application, the federal learning parameter server can predict the equipment resource state in the future stage, effectively improve the proportion of the effective participating equipment in the federal learning scene and reduce the time cost of the federal learning task.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to a method and a device for predicting computing power of mobile equipment in a federated learning scene.
Background
Federal learning is a popular distributed machine learning technology with a wide development prospect at present, and the privacy effect of effectively protecting the local data set of each participated device can be achieved by downloading the global model to the local by each participated device and uploading the result after training by using the local data set.
However, federal learning faces a significant challenge of model training delays under large-scale mobile device training scenarios. The reason for the excessive delay is as follows: 1) the calculation force difference of the mobile equipment participating in the training in the federal learning task is large, and the calculation force is unbalanced; 2) the geographical position of the mobile equipment is widely distributed, and the mobile equipment has strong mobility; 3) the dynamic change of the network state of the mobile equipment can not ensure the reliability. In the face of these problems, there are two main solutions to federal learning currently in the mainstream: 1) the state of the equipment is not considered, and a part of equipment is selected redundantly to participate in the training to allow some equipment to be disconnected in the training process; 2) and selecting the devices with sufficient currently available computation and network resources to participate in each round of training by considering the state of the candidate device at the current moment. However, because the device is constantly in dynamic change, the two modes cannot know whether the device can successfully complete the training task in the future after being selected to participate in the training.
Therefore, in a federal learning scenario where participating devices have strong dynamics, a technical problem to be solved by a person skilled in the art is urgently needed to design a mobile device calculation force prediction method, improve the proportion of effective participating devices in the federal learning scenario, and reduce the cost of the federal learning task.
Disclosure of Invention
The embodiment of the application provides a method and a device for predicting the calculation force of mobile equipment in a federated learning scene, so that the proportion of the effective participating equipment in the federated learning scene is improved, and the cost of a federated learning task is reduced.
In view of this, a first aspect of the present application provides a method for predicting computational power of a mobile device in a federated learning scenario, where the method includes:
collecting data information on a mobile device, wherein the data information comprises position information, time information, date information, network information and resource state information;
clustering the position information to obtain the hot spot position of the mobile equipment;
recording the stay time of the mobile equipment at the hotspot position through the time information;
calculating the average network state and the average resource state of the mobile equipment at the hotspot positions on different dates according to the date information, the stay time, the network information and the resource state information;
preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data;
inputting the training data into a recurrent neural network model for prediction, obtaining network information and resource state information of the mobile equipment after a preset time period, and outputting the network information and the resource state information.
Preferably, the location information is GPS data.
Preferably, the clustering the location information to obtain the hotspot location of the mobile device specifically includes:
carrying out noise filtration and sampling rate adjustment on the GPS data;
and clustering the filtered and adjusted GPS data to obtain the hot spot position of the mobile equipment.
Preferably, the network information specifically includes an upload network bandwidth and a download network bandwidth.
Preferably, the resource state information is specifically a CPU load rate.
Preferably, the recurrent neural network model is a long-short term memory model.
Preferably, the inputting the training data into a recurrent neural network model for prediction to obtain network information and resource state information of the mobile device after a preset time period and outputting specifically includes:
inputting the training data into a long-short term memory model;
predicting the hot spot position and the residence time after a preset time period simultaneously by adopting a mode of double output layers;
and mapping the network information and the resource state information of the mobile equipment by combining the hot spot position and the prediction result of the stay time to obtain and output the network information and the resource state information of the mobile equipment after a preset time period.
A second aspect of the present application provides a device for predicting computational power of a mobile device in a federated learning scenario, including: a collection module and a state prediction module;
the collection module specifically includes:
the mobile equipment comprises a collecting unit, a processing unit and a processing unit, wherein the collecting unit is used for collecting data information on the mobile equipment, and the data information comprises position information, time information, date information, network information and resource state information;
the clustering unit is used for clustering the position information to obtain the hot spot position of the mobile equipment;
the recording unit is used for recording the stay time of the mobile equipment at the hotspot position through the time information;
a calculating unit, configured to calculate an average network state and an average resource state of the mobile device at the hotspot location on different dates according to the date information, the staying time, the network information, and the resource state information;
the state prediction module specifically includes:
a preprocessing unit, configured to preprocess the location information, the time information, the date information, the average network state, and the average resource state as training data;
and the prediction unit is used for inputting the training data into a recurrent neural network model for prediction to obtain and output network information and resource state information of the mobile equipment after a preset time period.
Preferably, the recurrent neural network model is a long-short term memory model.
Preferably, the prediction unit specifically includes:
the input subunit is used for inputting the training data into a long-term and short-term memory model;
the double-output prediction subunit is used for simultaneously predicting the hot spot position and the stay time after a preset time period by adopting a double-output layer mode;
and the output subunit is configured to map the network information and the resource state information of the mobile device according to the prediction result of the hotspot position and the retention time, obtain the network information and the resource state information of the mobile device after a preset time period, and output the network information and the resource state information.
According to the technical scheme, the embodiment of the application has the following advantages:
the embodiment of the application provides a computing power prediction method of mobile equipment in a federated learning scene, which comprises the steps of collecting data information on the mobile equipment, wherein the data information comprises position information, time information, date information, network information and resource state information; clustering the position information to obtain the hot spot position of the mobile equipment; recording the stay time of the mobile equipment at the hot spot position through the time information; calculating the average network state and the average resource state of the mobile equipment at the hot spot position on different dates according to the date information, the stay time, the network information and the resource state information; preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data; inputting the training data into a recurrent neural network model for prediction, obtaining network information and resource state information of the mobile equipment after a preset time period, and outputting the network information and the resource state information.
Based on the calculation prediction mode of the mobile equipment provided by the application, in a federal learning task scene, the federal learning parameter server can make a selection according to the current resource state of the equipment and the equipment resource state predicted in a future stage, so that the equipment with the selected parameters in each round of federal learning task can complete local training on time and upload training results in time, the proportion of effectively participating in the equipment in the federal learning scene is effectively improved, and the time cost of the federal learning task is reduced.
Compared with the existing mainstream redundancy selection mode, the method for predicting the computing power of the mobile equipment provided by the application ensures that the federal learning task does not need to select redundant candidate equipment to prevent the disconnection or overtime of the selected equipment any more, but deduces the available state of the equipment in the future stage through a prediction model, thereby effectively improving the probability of successfully completing the training task by the selected equipment, and reducing the resource waste and the extra training time caused by selecting the redundant equipment. Compared with a selection mode only considering the current state of the equipment, the method for predicting the computing power of the mobile equipment can simultaneously acquire the current resource state and the predicted future resource state of the equipment, prevent the equipment with good current state and poor future state from being selected in a dynamic scene of federal learning, and simultaneously improve the probability of selecting the equipment with poor current state and good future state which can finally and smoothly complete a task, wherein the two equipment states are common states in the dynamic scene, so the method can solve the limitation caused by the original selection mode only considering the current state of the equipment.
Drawings
FIG. 1 is a scenario diagram of a federated learning task in an embodiment of the present application;
FIG. 2 is a flowchart of a method for predicting computing power of a mobile device in a federated learning scenario in a first embodiment of the present application;
FIG. 3 is a flowchart of a method for predicting computing power of a mobile device in a federated learning scenario in a second embodiment of the present application;
fig. 4 is a schematic structural diagram of a mobile device calculation force prediction apparatus in a federal learning scenario in a third embodiment of the present application;
FIG. 5 is a schematic diagram of a collection module of a mobile device calculation force prediction apparatus in a federal learning scenario according to a third embodiment of the present application;
fig. 6 is a schematic diagram of a state prediction module of a mobile device computing power prediction apparatus in a federal learning scenario in a third embodiment of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be understood that, when the present application is applied to a federal learning task scenario, please refer to fig. 1, where fig. 1 is a view of the federal learning task scenario in an embodiment of the present application, as shown in fig. 1, and fig. 1 includes a parameter server and mobile devices (device 1, device 2, and device 3). During federal learning, the position of the mobile device changes, so that the problems of unstable network bandwidth and insufficient available computing resources are caused, and the selected mobile device may not upload training results on time, so that the computing power of the mobile device in a future period of time needs to be predicted before the mobile device is selected, wherein the computing power includes a network bandwidth state and a computing resource state.
The application designs a computing power prediction method of mobile equipment in a federated learning scene in a first aspect. For convenience of understanding, please refer to fig. 2, where fig. 2 is a flowchart of a method for predicting computing power of a mobile device in a federal learning scenario in the first embodiment of the present application, and as shown in fig. 2, the method specifically includes:
s101, collecting data information on the mobile equipment, wherein the data information comprises position information, time information, date information, network information and resource state information.
It will be appreciated that before the computing power of the mobile device is predicted, data information is first collected for the current and past states of the mobile device, including location information, time information, date information, network information, and resource status information. The position information is the position of the mobile equipment at different periods and time, and can be acquired through the GPS function of the mobile equipment, specifically the real-time longitude and latitude of the GPS; the date information comprises dates such as working days, weekends, holidays and the like, and is used for recording the common geographic positions of the mobile equipment under the dates; the network information is the bandwidth of the mobile equipment, and specifically comprises uploading network bandwidth and downloading network bandwidth; the resource status information is an operational resource of the mobile device, such as a CPU load rate.
And S102, clustering the position information to obtain the hot spot position of the mobile equipment.
It should be noted that the hot spot location refers to a location where the mobile device frequently stays. The position information of the mobile equipment in a period of time is clustered, so that the hotspot position of the mobile equipment is obtained.
And S103, recording the stay time of the mobile equipment at the hot spot position through the time information.
It is understood that after the hot spot location is obtained, the dwell time of the mobile device at the hot spot location is counted.
And S104, calculating the average network state and the average resource state of the mobile equipment at the hot spot position on different dates according to the date information, the stay time, the network information and the resource state information.
It should be noted that the average network status and the average resource status may be calculated by the stay time at the specific location, and since the hot spot locations of the mobile device on the dates such as weekdays, weekends, holidays, etc. may be different, the average network status and the average resource status need to be combined with the date information to calculate the average network status and the average resource status at the hot spot locations of the mobile device on different dates.
And S105, preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data.
It is understood that the information obtained above can be input as training data of the prediction model after being preprocessed.
And S106, inputting the training data into the recurrent neural network model for prediction, obtaining and outputting network information and resource state information of the mobile equipment after a preset time period.
It should be noted that the prediction model adopted in the present application is a Recurrent Neural Network (RNN) model. The recurrent neural network model is a recurrent neural network which takes sequence data as input, recurs in the evolution direction of the sequence and all nodes (recurrent units) are connected in a chain manner, has memorability, parameter sharing and complete graphic, and has certain advantages when learning the nonlinear characteristics of the sequence.
The preprocessed data is represented as a six-tuple: [ poi ] Poii,sti,dti,ρi,di,ui],
Wherein the subscript i denotes the corresponding device i; the poi represents the hotspot geographic position of the device i obtained through a clustering algorithm; st represents the time length of the stay of the device i at the hotspot position; dt represents date information (weekday, weekend, holiday, etc.); ρ represents the average resource status (average load rate of the CPU) at the hot spot location during the period of the device i; d and u represent the average resource status (download rate and upload rate) at the hotspot location over the period of time of the CPU of device i. The preprocessed data are used as sequence data for input data of the recurrent neural network.
The output data is represented as a six-tuple: [ ρ (t), u (t), d (t), ρ*(t),u*(t),d*(t)]iWhere ρ (t) represents the currently observed device CPU load rate of device i at time t, ρ*(t) predicting the average equipment CPU load rate of the equipment i in the current time period at the moment t; u (t) and d (t) represent the upload rate and download rate, respectively, currently observed by device i at time t, u (t)*(t) and d*(t) represents the average upload rate and download rate, respectively, of the device i in the predicted round time period at time t.
The computing power prediction method for the mobile equipment in the federated learning scene, provided by the embodiment of the application, comprises the steps of collecting data information on the mobile equipment, wherein the data information comprises position information, time information, date information, network information and resource state information; clustering the position information to obtain the hot spot position of the mobile equipment; recording the stay time of the mobile equipment at the hot spot position through the time information; calculating the average network state and the average resource state of the mobile equipment at the hot spot position on different dates according to the date information, the stay time, the network information and the resource state information; preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data; inputting the training data into a recurrent neural network model for prediction, obtaining network information and resource state information of the mobile equipment after a preset time period, and outputting the network information and the resource state information. Based on the calculation prediction mode of the mobile equipment provided by the application, in a federal learning task scene, the federal learning parameter server can make a selection according to the current resource state of the equipment and the equipment resource state predicted in a future stage, so that the equipment with the selected parameters in each round of federal learning task can complete local training on time and upload training results in time, the proportion of effectively participating in the equipment in the federal learning scene is effectively improved, and the time cost of the federal learning task is reduced.
Compared with the existing mainstream redundancy selection mode, the method for predicting the computing power of the mobile equipment provided by the application ensures that the federal learning task does not need to select redundant candidate equipment to prevent the disconnection or overtime of the selected equipment any more, but deduces the available state of the equipment in the future stage through a prediction model, thereby effectively improving the probability of successfully completing the training task by the selected equipment, and reducing the resource waste and the extra training time caused by selecting the redundant equipment. Compared with a selection mode only considering the current state of the equipment, the method for predicting the computing power of the mobile equipment can simultaneously acquire the current resource state and the predicted future resource state of the equipment, prevent the equipment with good current state and poor future state from being selected in a dynamic scene of federal learning, and simultaneously improve the probability of selecting the equipment with poor current state and good future state which can finally and smoothly complete a task, wherein the two equipment states are common states in the dynamic scene, so the method can solve the limitation caused by the original selection mode only considering the current state of the equipment.
Experimental results show that the method can effectively improve the probability of selecting the available equipment in the federal learning scene, particularly when the candidate equipment is in an unstable environment in a mobile state, the problems of equipment computing resource waste and network resource waste caused by a redundancy selection mode without considering the equipment state, a selection mode only considering the current state of the equipment and the like in the existing mainstream are solved, and the time cost of each training task in the federal learning can be effectively shortened by selecting the equipment with sufficient available resources to participate in the training.
Referring to fig. 3, a second embodiment of the present application provides a method for predicting computing power of a mobile device in a federated learning scenario, including:
s201, collecting data information on the mobile device, wherein the data information comprises GPS data, time information, date information, uploading network bandwidth, downloading network bandwidth and CPU load rate.
It should be noted that step S201 in this embodiment is the same as step S101 in the first embodiment, where the location information specifically is a GPS data rate, the network information specifically includes an upload network bandwidth and a download network bandwidth, and the resource state information specifically is a CPU load rate.
S2021, performing noise filtering and sampling rate adjustment on the GPS data.
It can be appreciated that because of the noise and drift phenomena associated with GPS location information, noise filtering and sample rate adjustment of raw GPS data is required.
S2022, clustering the filtered and adjusted GPS data to obtain the hot spot position of the mobile equipment.
It can be understood that the effectiveness of the data can be improved by clustering the original GPS data after noise filtering and sampling rate adjustment.
And S203, recording the stay time of the mobile equipment at the hot spot position through the time information.
It is understood that step S203 in this embodiment is identical to step S103 in the first embodiment, and is not described herein again.
And S204, calculating the average network state and the average resource state of the mobile equipment at the hot spot position on different dates according to the date information, the stay time, the network information and the resource state information.
It is understood that step S204 in this embodiment is identical to step S104 in the first embodiment, and is not described herein again.
And S205, preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data.
It is understood that step S205 in this embodiment is identical to step S105 in the first embodiment, and is not described herein again.
S2061, inputting the training data into the long-short term memory model.
It should be noted that the recurrent neural network model in this embodiment is specifically a long-short term memory model. Long Short-term Memory (LSTM) is a special RNN, and mainly aims to solve the problems of gradient elimination and gradient explosion in the Long sequence training process. LSTM can perform better in longer sequences than normal RNNs.
S2062, adopting a double-output layer mode to simultaneously predict the hot spot position and the stay time after a preset time period.
And S2063, mapping the network information and the resource state information of the mobile equipment by combining the prediction results of the hot spot position and the residence time, obtaining and outputting the network information and the resource state information of the mobile equipment after a preset time period.
It should be noted that the output result adopts dual output layers, and the hot spot geographic location and the location dwell time duration are predicted at the same time. And finally, calculating the average CPU load rate and the uploading/downloading rate of the equipment in the time period at the position by combining the geographical position of the hot spot and the stay time of the position with the equipment operation resource information and the network resource information, and expressing the average CPU load rate and the uploading/downloading rate in a six-tuple manner: represented as a six-tuple: [ ρ (t), u (t), d (t), ρ*(t),u*(t),d*(t)]iWhere ρ (t) represents the currently observed device CPU load rate of device i at time t, ρ*(t) predicting the average equipment CPU load rate of the equipment i in the current time period at the moment t; u (t) and d (t) represent the upload rate and download rate, respectively, currently observed by device i at time t, u (t)*(t) and d*(t) respectively representing the times when the device i predicts the current round at the time tAverage upload rate and download rate over a period of time.
A second aspect of the present application provides a device for predicting computing power of a mobile device in a federated learning scenario, please refer to fig. 4.
The third embodiment of the present application provides a device for predicting computing power of a mobile device in a federated learning scenario, which includes a collection module 301 and a state prediction module 302.
The collection module specifically includes:
a collecting unit 3011, configured to collect data information on the mobile device, where the data information includes location information, time information, date information, network information, and resource status information;
the clustering unit 3012 is configured to cluster the location information to obtain a hot spot location of the mobile device;
a recording unit 3013, configured to record a staying time of the mobile device at the hot spot location through the time information;
a calculating unit 3014, configured to calculate, according to the date information, the staying time, the network information, and the resource state information, an average network state and an average resource state of the mobile device at the hotspot location on different dates;
the state prediction module 302 specifically includes:
a preprocessing unit 3021 configured to preprocess the location information, the time information, the date information, the average network status, and the average resource status as training data;
the prediction unit 3022 is configured to input the training data into the recurrent neural network model for prediction, obtain network information and resource state information of the mobile device after a preset time period, and output the network information and the resource state information.
Preferably, the location information is a GPS data rate, the network information includes an upload network bandwidth and a download network bandwidth, and the resource status information is a CPU load rate.
Further, the clustering unit 3012 includes:
and the adjusting subunit 30121 is configured to perform noise filtering and sample rate adjustment on the GPS data.
And the clustering subunit 30122 is configured to cluster the filtered and adjusted GPS data to obtain a hot spot location of the mobile device.
Preferably, the recurrent neural network model is specifically a long-short term memory model.
Further, the prediction unit 3022 specifically includes:
an input subunit 30221 configured to input training data into the long-term and short-term memory model;
a dual-output prediction subunit 30222, configured to perform prediction of a hot spot position and a dwell time after a preset time period simultaneously in a dual-output layer mode;
and the output subunit 30223 is configured to map the network information and the resource status information of the mobile device according to the prediction result of the hotspot position and the retention time, obtain and output the network information and the resource status information of the mobile device after a preset time period.
The operation of the collection module 301 is shown in fig. 5, and the operation of the state prediction module 302 is shown in fig. 6.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. A method for predicting computing power of mobile equipment in a federated learning scene is characterized by comprising the following steps:
collecting data information on a mobile device, wherein the data information comprises position information, time information, date information, network information and resource state information;
clustering the position information to obtain the hot spot position of the mobile equipment;
recording the stay time of the mobile equipment at the hotspot position through the time information;
calculating the average network state and the average resource state of the mobile equipment at the hotspot positions on different dates according to the date information, the stay time, the network information and the resource state information;
preprocessing the position information, the time information, the date information, the average network state and the average resource state to be used as training data;
inputting the training data into a recurrent neural network model for prediction, obtaining network information and resource state information of the mobile equipment after a preset time period, and outputting the network information and the resource state information.
2. The method for forecasting calculation power of a mobile device under the federal learning scenario as claimed in claim 1, wherein the location information is GPS data.
3. The method for predicting computational power of a mobile device under a federated learning scenario of claim 2, wherein the clustering the location information to obtain the hotspot location of the mobile device specifically comprises:
carrying out noise filtration and sampling rate adjustment on the GPS data;
and clustering the filtered and adjusted GPS data to obtain the hot spot position of the mobile equipment.
4. The method for forecasting calculation power of mobile equipment in a federal learning scenario of claim 1, wherein the network information specifically includes an upload network bandwidth and a download network bandwidth.
5. The method for predicting computing power of a mobile device under a federal learning scenario of claim 1, wherein the resource status information is a CPU load rate.
6. The method for forecasting calculation power of mobile equipment in a federal learning scenario as claimed in claim 1, wherein the recurrent neural network model is specifically a long-short term memory model.
7. The method for predicting computational power of a mobile device under a federal learning scenario of claim 6, wherein the step of inputting the training data into a recurrent neural network model for prediction to obtain network information and resource status information of the mobile device after a preset time period and outputting the network information and the resource status information specifically comprises the steps of:
inputting the training data into a long-short term memory model;
predicting the hot spot position and the residence time after a preset time period simultaneously by adopting a mode of double output layers;
and mapping the network information and the resource state information of the mobile equipment by combining the hot spot position and the prediction result of the stay time to obtain and output the network information and the resource state information of the mobile equipment after a preset time period.
8. A mobile device calculation force prediction device under the federated learning scene is characterized by comprising: a collection module and a state prediction module;
the collection module specifically includes:
the mobile equipment comprises a collecting unit, a processing unit and a processing unit, wherein the collecting unit is used for collecting data information on the mobile equipment, and the data information comprises position information, time information, date information, network information and resource state information;
the clustering unit is used for clustering the position information to obtain the hot spot position of the mobile equipment;
the recording unit is used for recording the stay time of the mobile equipment at the hotspot position through the time information;
a calculating unit, configured to calculate an average network state and an average resource state of the mobile device at the hotspot location on different dates according to the date information, the staying time, the network information, and the resource state information;
the state prediction module specifically includes:
a preprocessing unit, configured to preprocess the location information, the time information, the date information, the average network state, and the average resource state as training data;
and the prediction unit is used for inputting the training data into a recurrent neural network model for prediction to obtain and output network information and resource state information of the mobile equipment after a preset time period.
9. The device for forecasting calculation power of mobile equipment in a federal learning scenario as claimed in claim 8, wherein the recurrent neural network model is specifically a long-short term memory model.
10. The device for predicting computational power of a mobile device under a federal learning scenario as claimed in claim 9, wherein the prediction unit specifically comprises:
the input subunit is used for inputting the training data into a long-term and short-term memory model;
the double-output prediction subunit is used for simultaneously predicting the hot spot position and the stay time after a preset time period by adopting a double-output layer mode;
and the output subunit is configured to map the network information and the resource state information of the mobile device according to the prediction result of the hotspot position and the retention time, obtain the network information and the resource state information of the mobile device after a preset time period, and output the network information and the resource state information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011418124.3A CN112418342A (en) | 2020-12-07 | 2020-12-07 | Mobile equipment calculation force prediction method and device under federated learning scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011418124.3A CN112418342A (en) | 2020-12-07 | 2020-12-07 | Mobile equipment calculation force prediction method and device under federated learning scene |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112418342A true CN112418342A (en) | 2021-02-26 |
Family
ID=74775897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011418124.3A Pending CN112418342A (en) | 2020-12-07 | 2020-12-07 | Mobile equipment calculation force prediction method and device under federated learning scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112418342A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
-
2020
- 2020-12-07 CN CN202011418124.3A patent/CN112418342A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
Non-Patent Citations (2)
| Title |
|---|
| HUAWEI HUANG ET AL.: "Prophet: Proactive Candidate-Selection for Federated Learning by Predicting the Qualities of Training and Reporting Phases", 《ARXIV.ORG》 * |
| HUAWEI HUANG ET AL: "Technique report", 《HTTPS://DRIVE.GOOGLE.COM/FILE/D/1NHVBZNACMRWIACTAX6D7F4A7O0KWO5MU/VIEW?USP=SHARING》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | Modeling temporal-spatial correlations for crime prediction | |
| CN109872535B (en) | Intelligent traffic passage prediction method, device and server | |
| US20170278382A1 (en) | Risk early warning method and apparatus | |
| CN103761309A (en) | Operation data processing method and system | |
| CN106055630A (en) | Log storage method and device | |
| CN111274340A (en) | People flow density monitoring processing method, equipment and storage medium | |
| CN111294812A (en) | Method and system for resource capacity expansion planning | |
| Dewasurendra et al. | Evidence filtering | |
| CN111311014B (en) | Service data processing method, device, computer equipment and storage medium | |
| CN111382334B (en) | Data processing method and device, computer and readable storage medium | |
| CN110262863B (en) | Method and device for displaying terminal main interface | |
| CN113382071B (en) | Link creation method and device based on hybrid cloud architecture | |
| US20180332449A1 (en) | Data usage recommendation generator | |
| CN112418342A (en) | Mobile equipment calculation force prediction method and device under federated learning scene | |
| CN113123955A (en) | Plunger pump abnormality detection method and device, storage medium and electronic device | |
| CN114785701A (en) | Method and device for predicting optical transport network traffic, electronic device and storage medium | |
| CN113873025B (en) | Data processing method and device, storage medium and electronic equipment | |
| CN115242717A (en) | Centralized storage system based on end-side calculation and control method | |
| CN112037023B (en) | Bank outlet management method and related device | |
| CN115454557A (en) | Dynamic display system for application program page | |
| CN113407491A (en) | Data processing method and device | |
| Chen | Human Habits Investigation: from mobility reconstruction to mobile traffic prediction | |
| CN115756875B (en) | Online service deployment method and system of machine learning model for streaming data | |
| CN114339796B (en) | Cell dormancy data processing method and device, electronic equipment and storage medium | |
| CN111523058B (en) | Service area analysis processing method, device, equipment and storage medium |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210226 |
|
| WD01 | Invention patent application deemed withdrawn after publication |