WO2022148494A1 - 过热处理方法及装置、存储介质 - Google Patents

过热处理方法及装置、存储介质 Download PDF

Info

Publication number
WO2022148494A1
WO2022148494A1 PCT/CN2022/076562 CN2022076562W WO2022148494A1 WO 2022148494 A1 WO2022148494 A1 WO 2022148494A1 CN 2022076562 W CN2022076562 W CN 2022076562W WO 2022148494 A1 WO2022148494 A1 WO 2022148494A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
overheating
model
terminal
configuration
Prior art date
Application number
PCT/CN2022/076562
Other languages
English (en)
French (fr)
Inventor
陈晓宇
韩立锋
Original Assignee
展讯通信(上海)有限公司
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 展讯通信(上海)有限公司 filed Critical 展讯通信(上海)有限公司
Priority to US18/260,693 priority Critical patent/US20240061480A1/en
Publication of WO2022148494A1 publication Critical patent/WO2022148494A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F1/00Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
    • G06F1/16Constructional details or arrangements
    • G06F1/20Cooling means
    • G06F1/206Cooling means comprising thermal management
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F1/00Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
    • G06F1/16Constructional details or arrangements
    • G06F1/20Cooling means
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W76/00Connection management
    • H04W76/20Manipulation of established connections
    • H04W76/27Transitions between radio resource control [RRC] states

Definitions

  • Embodiments of the present invention relate to the field of communications, and in particular, to a method and device for overheating, and a storage medium.
  • AI artificial intelligence
  • the technical problem solved by the embodiments of the present invention is that when the terminal performs the training and inference of the AI model, it is easy to cause the terminal to overheat.
  • an embodiment of the present invention provides an overheating method, including: when it is detected that the terminal is overheated, obtaining configuration indication information, where the configuration indication information is used to indicate the working state configuration of the AI model; according to the configuration Instructions to adjust the working status of the corresponding AI model.
  • the configuration indication information is static information pre-configured on the network side, and when it is detected that the terminal is overheated, acquiring the configuration indication information includes: when it is detected that the terminal is overheated, locally acquiring the information from the network.
  • the static information pre-configured on the side the static information includes: configuration permission information of the AI model, where the configuration permission information of the AI model is used to indicate the configuration permission of the AI model.
  • the overheating method further includes: after adjusting the working state of the corresponding AI model, when it is detected that the terminal overheating is relieved or eliminated, readjusting the work of the AI model based on the configuration permission information of the AI model. state.
  • the obtaining configuration indication information when it is detected that the terminal is overheated includes: when it is detected that the terminal is overheated, obtaining dynamic information configured by the network side from the network side, where the dynamic information is used to indicate The working state configuration of the AI model when the terminal is overheated, and the configuration indication information includes the dynamic information.
  • the acquiring configuration indication information when it is detected that the terminal is overheated includes: when it is detected that the terminal is overheating, reporting the first overheating auxiliary information to the network side, so that the network side receives the After the first overheat auxiliary information, the dynamic information is delivered.
  • the first overheating assistance information includes at least one of the following information: first overheating indication information, AI model configuration suggestion information, and available computing resource information.
  • the first overheating indication information is used to indicate that the terminal is overheated, and the overheating of the terminal is related to the running of the AI model.
  • the AI model configuration recommendation information includes at least one of the following information: AI model information recommended to be retained, AI model information recommended to be closed, AI model closure time recommended, AI model information recommended to reduce precision, AI model recommended duration of reduced precision.
  • the AI model configuration suggestion information is obtained in the following manner: at least one of the configuration information of historical AI models, the computing resources occupied by each AI model when running, and the computing resources expected to be reduced for alleviating or eliminating overheating. One, determining the AI model configuration suggestion information.
  • the available computing power resource information includes at least one of the following information: available computing power resources at the current moment, and computing power resources expected to be reduced for alleviating or eliminating overheating.
  • the method for overheating further includes: when it is detected that the terminal is overheated or eliminated, acquiring additional dynamic information configured by the network side from the network side, where the additional dynamic information is used for Indicates the working state configuration of the AI model when the terminal is overheated or eliminated; readjusts the working state of the AI model according to the additional dynamic information.
  • the obtaining additional dynamic information configured by the network side from the network side when detecting that the terminal overheating is relieved or eliminated The second overheating assistance information is sent to the network side, so that the network side sends the additional dynamic information after receiving the second overheating assistance information.
  • the second overheating assistance information includes at least one of the following information: second overheating indication information, AI model configuration suggestion information, and available computing power resource information.
  • the second overheat indication information is used to instruct the terminal to relieve overheating or eliminate overheating.
  • the AI model configuration suggestion information includes at least one of the following information: AI model information that is recommended to be enabled, and AI model information that is recommended to improve accuracy.
  • the AI model configuration suggestion information is obtained in the following manner: according to at least one of the configuration information of historical AI models, the computing resources occupied by each AI model when running, and the computing resources available when overheating is not expected, Determine the AI model configuration suggestion information.
  • the available computing power resource information includes at least one of the following information: available computing power resources at the current moment, and available computing power resources when overheating is not expected.
  • the first overheating assistance information and the second overheating assistance information are reported to the network side through RRC signaling.
  • the overheating method further includes: when it is detected that the terminal is overheated, if it is currently in an idle state or an inactive state, establishing a connection with the network side.
  • the dynamic information further includes static information, and the static information includes configuration authority information of the AI model, and the configuration authority information of the AI model is used to indicate the configuration authority of the AI model.
  • An embodiment of the present invention also provides another method for overheating, including: delivering configuration indication information to a terminal, so that when the terminal is overheated, the working state of the corresponding AI model is adjusted according to the configuration indication information, and the configuration indication information is used for Indicates the working state configuration of the AI model.
  • the configuration indication information is preconfigured static information
  • the delivering the configuration indication information to the terminal includes: preconfiguring the static information and delivering it to the terminal
  • the static information includes: an AI model
  • the configuration permission information of the AI model is used to indicate the configuration permission of the AI model.
  • the static information is pre-configured.
  • the delivering configuration indication information to the terminal includes: configuration dynamic information, where the dynamic information is used to indicate the working state configuration of the AI model when the terminal is overheated, the configuration indication information includes the dynamic information, the The dynamic information is configured when the terminal is overheated; the dynamic information is delivered to the terminal.
  • the configuring dynamic information includes: configuring dynamic information when receiving first overheating assistance information, the first overheating assistance information being reported by the terminal when overheating is detected.
  • the first overheating assistance information includes at least one of the following information: first overheating indication information, AI model configuration suggestion information, and available computing resource information.
  • the dynamic information includes at least one of the following information: AI model information that is kept open, AI model information that is closed, AI model closing duration, AI model information that reduces the precision, and duration that the precision of the AI model is reduced.
  • the overheating method further includes: configuring additional dynamic information, where the additional dynamic information is used to indicate the configuration of the working state of the AI model when the terminal is overheated or eliminated; sending the additional dynamic information to the terminal, so that the terminal readjusts the working state of the AI model according to the additional dynamic information.
  • the configuring the additional dynamic information includes configuring the additional dynamic information when receiving second overheating assistance information, the second overheating assistance information being reported by the terminal when overheating is relieved or overheating is eliminated.
  • the second overheating assistance information includes at least one of the following information: second overheating indication information, AI model configuration suggestion information, and available computing power resource information.
  • the additional dynamic information includes at least one of the following information: AI model information that is enabled, and AI model information that improves accuracy.
  • the dynamic information further includes static information, and the static information includes configuration authority information of the AI model, and the configuration authority information of the AI model is used to indicate the configuration authority of the AI model.
  • An embodiment of the present invention further provides an overheating device, including: an acquisition unit, configured to acquire configuration indication information when it is detected that the terminal is overheated, where the configuration indication information is used to indicate the working state configuration of the AI model; the configuration unit is configured with According to the configuration instruction information, the working state of the corresponding AI model is adjusted.
  • An embodiment of the present invention further provides another overheating device, including: a sending unit configured to deliver configuration indication information to a terminal, so that when the terminal is overheated, the working state of the corresponding AI model is adjusted according to the configuration indication information, and the The configuration indication information is used to indicate the working state configuration of the AI model.
  • a sending unit configured to deliver configuration indication information to a terminal, so that when the terminal is overheated, the working state of the corresponding AI model is adjusted according to the configuration indication information, and the The configuration indication information is used to indicate the working state configuration of the AI model.
  • An embodiment of the present invention further provides a storage medium, where the storage medium is a non-volatile storage medium or a non-transitory storage medium, and a computer program is stored thereon, and the computer program executes any of the foregoing when the computer program is run by a processor The steps of the superheat treatment method.
  • An embodiment of the present invention further provides a device for overheating, including a memory and a processor, the memory stores a computer program that can run on the processor, and the processor executes any of the above when running the computer program The steps of a superheat treatment method.
  • the configuration instruction information is static information pre-configured on the network side
  • the static information includes the configuration permission information of the AI model, so that when the terminal is overheated, the terminal can obtain the static information locally from the terminal, and based on the static information of the AI model in the terminal.
  • the configuration authority of the AI model indicated by the configuration authority information can adjust the working status of the corresponding AI model, so that the problem of terminal overheating can be solved without interacting with the network side, and the efficiency of solving terminal overheating can be improved.
  • the working state of the AI model is re-adjusted based on the configuration permission information of the AI model, so as to realize the configuration of the working state of the AI model without interacting with the network side, Restore the working state of some AI models, such as re-opening AI models that were turned off when overheated or improving the accuracy of AI models whose accuracy was reduced when overheating, etc., to ensure the network communication performance of terminals and improve user experience.
  • the configuration indication information may include dynamic information.
  • the dynamic information configured by the network side is obtained from the network side, which helps the network side configure dynamic information for the terminal according to the overall situation of the network side.
  • the reported first overheating auxiliary information may include one or more of first overheating indication information, AI model configuration suggestion information, available computing power resource information, etc.
  • the terminal may indicate the terminal to the network side through the first overheating auxiliary information. Overheating due to the operation of the AI model. For some terminals with certain capabilities, the AI model configuration suggestion information and/or available computing resource information can even be used to give some suggestions to the network side or its own actual situation, which helps the network side to configure The terminal configures more appropriate dynamic information.
  • the additional dynamic information configured by the network side is obtained from the network side, and the working state of the AI model is re-adjusted according to the additional dynamic information, so that the terminal can restore the working state of some AI models, such as restarting.
  • the AI model that is turned off when it is overheated or the accuracy of the AI model that reduces the accuracy when it is overheating is improved, etc., to help ensure the network communication performance of the terminal and improve the user experience.
  • Fig. 1 is the flow chart of a kind of terminal overheating method in the embodiment of the present invention
  • FIG. 3 is a schematic structural diagram of a terminal overheating device in an embodiment of the present invention.
  • FIG. 4 is a schematic structural diagram of another terminal overheating device in an embodiment of the present invention.
  • performing AI model training and inference on a terminal may easily lead to overheating of the terminal. Even if the training and inference of the AI model are performed on the network side or in the cloud, and the terminal participates in this centralized AI training and inference process, the problem of terminal overheating still exists. When the terminal is overheated, the terminal is prone to frequency reduction and freezes, which affects the user experience.
  • the configuration indication information is obtained, and the working state of the AI model is adjusted according to the working state configuration of the AI model indicated by the configuration indication information. Adjustment of operating conditions to alleviate or eliminate terminal overheating problems is expected.
  • Embodiments of the present invention provide an overheating method.
  • the execution body of the overheating method may be the terminal itself, or may be a control chip, a processing chip, a baseband chip, or other appropriate components that can be used in the terminal.
  • the terminal may also be called user equipment (User Equipment, UE), and the terminal may be various appropriate terminals, such as a mobile phone, a computer, an Internet of Things device, a vehicle-mounted device, etc., but is not limited thereto.
  • FIG. 1 a flowchart of a method for overheating in an embodiment of the present invention is provided, and the method for overheating may specifically include the following steps:
  • Step S11 when it is detected that the terminal is overheated, obtain configuration indication information.
  • the acquired configuration indication information may be used to indicate the working state configuration of the AI model.
  • AI models usually requires corresponding computing resources.
  • the occupied graphics processor Graphics Processing Unit, GPU
  • the computing resources occupied by the AI model running are related to the terminal temperature. The greater the computing resources occupied when the AI model is running, the easier the terminal is to overheat.
  • the computing resources required by the AI model running in the terminal are related to the occupancy of the GPU in the terminal. Therefore, the running status of the AI model and whether the terminal is overheated can be inferred from the occupancy of the GPU.
  • whether the terminal is overheated can be determined according to the occupancy of the GPU in the terminal and the total capacity of the GPU.
  • the occupancy rate of the GPU is calculated according to the occupancy rate of the GPU and the total amount of the GPU. When the GPU occupancy rate is greater than the set threshold, it is determined that the terminal is overheated.
  • the overheating of the terminal may be the overheating of the entire terminal, or the overheating of some systems or components in the terminal.
  • the chip is overheated, the transmitter module is overheated, the central processing unit (CPU) is overheated, the GPU is overheated, and the baseband chip is overheated.
  • the CPU central processing unit
  • the GPU is overheated
  • the baseband chip is overheated.
  • the chip or other suitable components can receive the detection result of the terminal overheating by other coupled components, for example, the chip or other suitable components Some specified components in the terminal, the temperature of the system, etc. sent by the coupled temperature detection device can be received. It is also possible to configure a temperature detection device on the chip or other appropriate components themselves, and determine whether the terminal overheating problem occurs according to the temperature detected by the self-configured temperature detection device.
  • Step S12 according to the configuration instruction information, adjust the working state of the corresponding AI model.
  • the working state of the corresponding AI model may be adjusted according to the working state configuration of the AI model indicated by the configuration indication information.
  • adjusting the working state of the corresponding AI model may include at least one of the following: keeping some AI models turned on, turning off some AI models, reducing the accuracy of some AI models, and the like.
  • the configuration indication information is obtained, and the working state of the AI model is adjusted according to the working state configuration of the AI model indicated by the configuration indication information. Or eliminating terminal overheating issues is to be expected.
  • the configuration indication information acquired in step S11 may be static information preconfigured on the network side acquired locally.
  • static information preconfigured by the network side can be obtained locally.
  • the static information may include configuration permission information of the AI model, where the configuration permission information of the AI model is used to indicate the configuration permission for the AI model.
  • the terminal can adjust the working status of the AI model based on the configuration permission for the AI model indicated by the configuration permission information of the AI model. Solve the problem of terminal overheating by changing the working state of the AI model on the terminal side.
  • the objects with configuration permissions for the AI model are different.
  • the execution body of the overheating method is the terminal itself, the terminal has the right to configure the AI model.
  • the execution body of the overheating method is a chip or some other components, the chip or some other components have control authority over the AI model.
  • the working state of the AI model can be readjusted based on the configuration permission information of the AI model.
  • re-adjusting the working state of the AI model may include re-opening some AI models that are turned off when the terminal is overheated, improving the accuracy of some AI models whose accuracy is reduced when the terminal is overheating, increasing the amount of data uploaded by the terminal, etc.
  • the configuration of the AI model ensures the performance of the terminal's communication network and improves the user experience.
  • whether the terminal is overheated is relieved or overheated can be determined according to the duration. For example, if the working state of the corresponding AI model reaches the set duration after adjustment, it is determined that the terminal overheating is relieved or the overheating is eliminated.
  • the temperature of the terminal can be used to determine whether the overheating of the terminal is relieved or the overheating is eliminated. For example, when the terminal temperature is lower than the first temperature threshold, it is determined that the terminal overheating is relieved; when the terminal temperature is lower than the second temperature threshold, it is determined that the terminal overheating has been eliminated, and the first temperature threshold is higher than the second temperature threshold.
  • the configuration authority for the AI model may include at least one of the following: configuration authority for opening and closing the AI model, configuration authority for the shutdown duration of the AI model, and precision adjustment for the AI model. Configuration permissions, configuration permissions to adjust the accuracy level of AI models, etc.
  • the configuration permission to open and close the AI model means that for the AI model with the configuration permission to open and close, you can decide whether to stop using the AI model, or decide whether to open the AI model.
  • the AI model with the configuration permission to turn on and off can be turned on by itself.
  • the AI model can be timed after the AI model is closed, and the AI model can be automatically activated when the set maximum closing time is reached to ensure the communication performance of the terminal.
  • AI models that do not have the configuration permission to turn on and off it is necessary to negotiate with the network side to turn off or turn on these AI models, and the working status of these AI models can be adjusted based on the instructions of the network side.
  • the configuration authority for the precision adjustment of the AI model refers to the AI model with the configuration authority for the precision adjustment, which can decide the precision adjustment of the AI model by itself.
  • the precision of AI models with configuration permissions for precision adjustment can be reduced at its own discretion. Or, when the terminal overheating is alleviated or the overheating is eliminated, it can be automatically decided to improve the precision of the AI model with the configuration permission of precision adjustment.
  • the accuracy level of the AI model can be configured according to the configuration authority for adjusting the accuracy level indicated.
  • the accuracy level of the AI model can be reduced according to the configuration authority adjusted by the indicated accuracy level.
  • the accuracy level of the AI model can be increased according to the configuration permission adjusted by the indicated accuracy level.
  • the configuration authority information of the AI model included in the static information configured on the network side can be indicated to the terminal in various ways, as long as the network side and the terminal both know the specific representative of the configuration authority information of the AI model. Meaning can be.
  • indicators 0 and 1 are used to indicate the configuration authority of the terminal to open and close the AI model. If the configuration authority information of the AI model is the indicator 1, it means that the terminal has the configuration authority to open and close the AI model. If the configuration permission information of the AI model is the indicator 0, it means that the terminal does not have the configuration permission to turn on and off the AI model. When the terminal overheats, the terminal can stop using the AI model with indicator 1 on its own. For the AI model with an indicator of 0, even if the terminal is overheated, it needs to be retained, and the terminal cannot decide whether it can be turned off.
  • configuration permission information corresponding to different types of configuration permissions may use different identifiers, and other methods may also be used to indicate the configuration permission information of the AI model according to actual needs.
  • the network side may deliver static information when configuring the AI model for the terminal or updating the AI model.
  • the static information may be delivered to the terminal as attribute information of the AI model, may also be delivered to the terminal as instruction information, or may be delivered to the terminal as other types of information, which is not limited here.
  • the network side when the network side configures an AI model for the terminal, the network side can deliver static information corresponding to each AI model, that is, each AI model corresponds to a piece of static information, and the static information of each AI model is determined by the static information.
  • the configuration permission information for the included AI model applies to that AI model.
  • the static information of all AI models on the terminal needs to be obtained.
  • the network side when the network side configures the AI model for the terminal, the network side can also put the configuration permission information of all AI models in one static information in batches, that is, the static information includes all AI models in the terminal. Configure permission information.
  • the terminal When the terminal is overheated, the terminal only needs to obtain a static message to learn the configuration permissions for all AI models from the static message.
  • the static information corresponding to each AI model or the updated AI model may be delivered separately, that is, each AI model corresponds to a piece of static information, and each AI model corresponds to a piece of static information.
  • the configuration permission information of the AI model contained in the static information of the model is applicable to the AI model.
  • the network side when the network side updates the AI model for the terminal, the network side can also put the configuration permission information of all AI models in one static information in batches, that is, the static information includes all AI models in the terminal. Configure permission information.
  • the terminal When the terminal is overheated, the terminal only needs to obtain a static message to learn the configuration permissions for all AI models from the static message.
  • the updated static information can be delivered, and the updated static information includes the configuration authority information of the AI model after adjustment.
  • the terminal may cover the existing static information with the updated static information, and use the updated static information as the basis for adjusting the working state of the subsequent AI model.
  • the terminal When the terminal readjusts the working state of the AI model based on the configuration permission information of the AI model, the overheating problem of the terminal is usually alleviated or disappears. Permission information, readjust the working state of the AI model, and the terminal may still be overheated.
  • the terminal can adjust the working state of the AI model again according to the working state configuration of the AI model when the terminal is overheated indicated by the dynamic information.
  • the AI model in the terminal can be reconfigured to alleviate or eliminate the terminal overheating.
  • the terminal may be currently in an idle state or an inactive state.
  • a connection with the network side is established. After the connection with the network side is established, the first overheat assistance information is reported.
  • acquiring the configuration indication information in step S11 may be acquiring dynamic information configured by the network side from the network side.
  • the configuration indication information includes the dynamic information, and the dynamic information is used to indicate the working state configuration of the AI model when the terminal is overheated.
  • the terminal can adjust the working state of the AI model according to the working state configuration of the AI model during overheating indicated by the dynamic information. By adjusting the working state of the AI model, the AI model in the terminal can be reconfigured to alleviate or eliminate overheating.
  • the network side may configure dynamic information for the terminal in real time according to the first overheating assistance information and deliver it to the terminal. Therefore, the dynamic information configured for the terminal each time can conform to the current actual situation of the terminal.
  • the network side can also select a configuration from several preset configurations according to the preconfigured policy and mapping relationship according to the mapping relationship configured in the mapping table, and different configurations correspond to the work of the AI model.
  • the state configuration is different, dynamic information is generated based on the selected configuration, and the dynamic information is delivered to the terminal.
  • the configuration corresponding to the overheating problem caused by different AI models is different, or the configuration corresponding to different types of terminals is different. It is understandable that the corresponding configuration can also be determined based on other factors, and in subsequent use, based on the current actual Select the corresponding configuration as required.
  • the dynamic information may include at least one of the following information: AI model information that is kept on, AI model information that is turned off, AI model shutdown duration, AI model information with reduced precision, duration of AI model precision reduction, and the like.
  • the dynamic information may include identification information (ID) of the AI model, configuration identification information of the AI model, and the like.
  • ID identification information
  • the identification information of the AI model is used to identify the AI model and uniquely corresponds to the AI model.
  • the configuration identification information of the AI model is used to identify the working status configuration of the AI model. When the working status configuration of the AI model is different, the corresponding configuration identification information is different. Therefore, the working status configuration of each AI model can be determined according to the ID and configuration identification information of the AI model.
  • the network side may deliver the configuration indication information to the terminal through radio resource control (Radio Resource Control, RRC) signaling.
  • RRC Radio Resource Control
  • the first overheating auxiliary information is reported to the network side, so that the network side sends the dynamic information after receiving the first overheating auxiliary information.
  • the terminal may report the first overheating assistance information to the network side through RRC signaling.
  • the RRC signaling may be terminal assistance information (UE Assistance Information, UAI).
  • the first overheating assistance information is carried in an overheating assistance information element (OverheatingAssistance IE).
  • OverheatingAssistance IE Overheating assistance information element
  • the first overheating assistance information may include at least one of the following information: first overheating indication information, AI model configuration suggestion information, available computing power resource information, and the like.
  • the first overheat indication information is used to indicate that the terminal is overheated, and the overheating of the terminal is related to the operation of the AI model.
  • the terminal may indicate to the network side that the terminal is overheated through the first overheating indication information, and the reason for the overheating of the terminal is related to factors such as an AI model running on the terminal.
  • the AI model configuration suggestion information in the first overheating assistance information includes at least one of the following information: AI model information recommended to be retained, AI model information recommended to be closed, AI model shutdown duration recommended, and AI recommended to reduce precision. Model information, and the recommended time for reducing the precision of the AI model, etc.
  • the terminal may obtain the AI model configuration suggestion information in the following manner: according to the configuration information of historical AI models, the computing power resources occupied by each AI model when running, and the computing power expected to be reduced for alleviating or eliminating overheating At least one of the resources to determine the AI model configuration suggestion information.
  • the available computing power resource information includes at least one of the following information: available computing power resources at the current moment, computing power resources expected to be reduced for alleviating overheating or eliminating overheating, and the like.
  • the information contained in the first overheating assistance information is different. in particular:
  • the first overheat assist information may include only the first overheat indication information.
  • the network side receives the first overheating assistance information that only includes the first overheating indication information, it can know that the terminal is overheated due to the excessive number of running AI models or the excessive computing resources occupied by the running of the AI model. , the network side configures dynamic information for the terminal in response to the first overheating auxiliary information and delivers it to the terminal.
  • the terminal can give some reasonable suggestions to the network side when it is overheated.
  • the first overheat assistance information reported by the terminal may include AI model configuration suggestion information.
  • the terminal can use the AI model configuration recommendation information to indicate to the network the AI model information that the terminal recommends to keep, the AI model information that is recommended to be closed, the time to close the proposed AI model, the information about the AI model that recommends reducing the precision, and the length that the AI model is recommended to reduce the precision for, etc. one or more of.
  • the network side can configure dynamic information for the terminal considering the AI model configuration recommendation information, so as to improve the matching degree between the dynamic information configured for the terminal and the actual demand of the terminal, and help to better alleviate the terminal overheating.
  • the terminal can calculate the available computing resources at the current moment, the computing resources that are expected to be reduced to alleviate overheating or eliminate overheating, etc., and report
  • the first overheating assistance information includes one or more of available computing power resources at the current moment, computing power resources that are expected to be reduced for alleviating overheating or eliminating overheating, and the like.
  • the network side can accurately know that the terminal is overheating due to too many AI models running or participating in too many AI-related network functions, and can obtain information about the available computing power resources of the terminal and how to alleviate the overheating or Eliminate computing resources that are expected to be reduced by overheating, etc. Therefore, the network side can comprehensively consider one or more of the terminal's available computing resources, air interface, base station, neighbor cell reporting capability, and historical dynamic information previously configured for the terminal, and determine how to configure appropriate dynamic information for the terminal.
  • the reported first overheating auxiliary information may include two or all of the first overheating indication information, AI model configuration suggestion information, and available computing power resource information. No more examples here.
  • the terminal when the terminal detects overheating, the terminal may currently be in an idle state or an inactive state.
  • the terminal When the terminal is currently in an idle state or an inactive state, a connection with the network side is established. After the connection with the network side is established, the first overheat assistance information is reported.
  • the dynamic information may further include static information, and the static information includes configuration permission information of the AI model.
  • the configuration permission information of the AI model is used to indicate the configuration permission for the AI model.
  • the terminal may configure the working state of the AI model according to the working state configuration of the AI model when the terminal is overheated indicated by the dynamic information.
  • the static information can be stored locally. When the terminal overheats later, the terminal can decide to adjust the working status of the AI model based on the configuration permission information of the AI model in the static information.
  • the static information can also be used to readjust the working state of the AI model based on the configuration permission information of the AI model in the static information when the terminal is overheated or eliminated.
  • the network side can carry static information in the dynamic information when the terminal overheating problem occurs for the first time, or after the terminal overheating exceeds the set number of times, to indicate the configuration authority of the AI model. to the terminal.
  • the terminal overheats later, it does not need to interact with the network side, that is, it can decide how to adjust the working status of the AI model based on the configuration permissions of the AI model indicated by the static information, thereby improving the efficiency of overheating.
  • the overheating method may further include the following steps on the basis of the overheating method described in FIG. 1 S13 and step S14, specifically:
  • Step S13 when it is detected that the terminal overheating is relieved or the overheating is eliminated, additional dynamic information configured by the network side is acquired from the network side.
  • the additional dynamic information is used to indicate the working state configuration of the AI model when the terminal is overheated or eliminated.
  • the additional dynamic information may include at least one of the following information: enabled AI model information, AI model information with improved accuracy, and the like.
  • Step S14 readjust the working state of the AI model according to the additional dynamic information.
  • step S13 may include: when it is detected that the terminal overheating is relieved or the overheating is eliminated, reporting the second overheating auxiliary information to the network side, so that the network side sends the second overheating auxiliary information after receiving the second overheating auxiliary information. additional dynamic information.
  • the terminal may report the second overheating assistance information to the network side through RRC signaling.
  • the RRC signaling may be UAI.
  • the second overheating assistance information may be carried in the OverheatingAssistance IE. It can be understood that the RRC signaling may also be other RRC signaling.
  • the second overheating assistance information may include at least one of the following information: second overheating indication information, AI model configuration suggestion information, available computing power resource information, and the like.
  • the second overheat indication information is used to instruct the terminal to relieve or eliminate overheating. That is, the terminal may indicate to the network side the overheating mitigation or overheating elimination of the terminal through the second overheating indication information.
  • the AI model configuration suggestion information in the second overheating assistance information includes at least one of the following information: AI model information suggested to be enabled, AI model information suggested to improve accuracy, and the like.
  • the terminal may obtain the AI model configuration suggestion information in the following manner: according to the configuration information of historical AI models, the computing resources occupied by each AI model when running, and the available computing resources when overheating is not expected to occur. At least one, determining the AI model configuration suggestion information.
  • the available computing power resource information may include at least one of the following information: available computing power resources at the current moment, available computing power resources when overheating is not expected, and the like.
  • the information contained in the second overheating assistance information is different. in particular:
  • the second overheat assist information may include only the second overheat indication information.
  • the network side receives the second overheating assistance information including only the second overheating indication information, it can know that the terminal is overheated or eliminated, and can configure additional dynamic information for the terminal in response to the second overheating assistance information.
  • the terminal may provide advice to the network side through the AI model configuration advice information in the reported second overheating assistance information.
  • the terminal can indicate to the network side one or more of the AI model information that the terminal recommends to open, the AI model information that recommends improving the accuracy, and the like through the AI model configuration suggestion information.
  • the network side can configure additional dynamic information for the terminal considering the AI model configuration recommendation information, so as to improve the match between the additional dynamic information configured for the terminal and the actual needs of the terminal, and help ensure the communication performance of the terminal.
  • the terminal can calculate the available computing power resources at the current moment, the available computing power resources when overheating is not expected, etc., and report the second
  • the overheating auxiliary information includes one or more of available computing power resources at the current moment, available computing power resources when overheating is not expected to occur, and the like.
  • the network side can accurately learn the available computing power resource information of the terminal. Therefore, the network side can comprehensively consider one or more of the terminal's available computing resources, air interface, base station, neighbor cell reporting capabilities, and historical additional dynamic information that has been configured for the terminal, and determine how to configure appropriate additional dynamic information for the terminal. information.
  • the reported second overheating auxiliary information may include two or all of the second overheating indication information, AI model configuration suggestion information, and available computing power resource information. No more examples here.
  • the terminal when the terminal detects overheating mitigation or overheating elimination, if the terminal may be currently in an idle state or an inactive state, a connection with the network side is established. After the connection with the network side is established, the second overheating assistance information is reported.
  • the embodiment of the present invention also provides another overheating method, which can be used on the network side.
  • the overheating method may specifically include the following steps: delivering configuration instruction information to a terminal, so that when the terminal is overheated, adjust the corresponding The working state of the AI model, and the configuration indication information is used to indicate the working state configuration of the AI model. Therefore, when the terminal is overheated, the working state of the corresponding AI model can be adjusted according to the configuration instruction information. For example, it is expected to turn off the working status of some AI models, adjust the accuracy of some AI models, etc., so as to alleviate the problem of terminal overheating.
  • the configuration indication information is preconfigured static information. That is, the network side can preconfigure static information, and deliver the preconfigured static information to the terminal.
  • the static information may include configuration permission information of the AI model, and the configuration permission information of the AI model is used to indicate the configuration permission of the AI model.
  • the network side may pre-configure the static information when configuring the AI model for the terminal or updating the AI model.
  • the configuration indication information may include dynamic information, where the dynamic information is configured when the terminal is overheated.
  • the dynamic information is used to indicate the working state configuration of the AI model when the terminal is overheated, and the configuration indication information includes the dynamic information.
  • the dynamic information can be delivered to the terminal.
  • the network side may deliver configuration indication information including dynamic information to the terminal through RRC signaling.
  • the dynamic information may include at least one of the following information: AI model information that remains open, AI model information that is closed, AI model closing duration, AI model information that reduces the precision, duration that reduces the precision of the AI model, etc. .
  • the terminal reports the first overheat assistance information to the network side when overheating is detected.
  • the network side may configure dynamic information when receiving the first overheat assistance information.
  • the first overheating assistance information may include at least one of the following information: first overheating indication information, AI model configuration suggestion information, available computing power resource information, and the like.
  • first overheating auxiliary information For the first overheating auxiliary information, reference may be made to the relevant description of the first overheating auxiliary information in the foregoing embodiments of the present invention, and details are not repeated here.
  • the network side may configure dynamic information for the terminal in real time according to the first overheating assistance information and deliver it to the terminal. Therefore, the dynamic information configured for the terminal each time can conform to the current actual situation of the terminal.
  • the network side can also select a configuration from several preset configurations according to the preconfigured policy and mapping relationship according to the mapping relationship configured in the mapping table, and different configurations correspond to the work of the AI model.
  • the state configuration is different, dynamic information is generated based on the selected configuration, and the dynamic information is delivered to the terminal.
  • the configuration corresponding to the overheating problem caused by different AI models is different, or the configuration corresponding to different types of terminals is different. It is understandable that the corresponding configuration can also be determined based on other factors, and in subsequent use, based on the current actual Select the corresponding configuration as required.
  • the network side may configure additional dynamic information for the terminal when overheating of the terminal is alleviated or eliminated.
  • the additional dynamic information is used to indicate the working state configuration of the AI model when the terminal overheating is relieved or the overheating is eliminated.
  • the network side sends the additional dynamic information to the terminal, and the terminal readjusts the working state of the corresponding AI model according to the additional dynamic information.
  • the additional dynamic information may include at least one of the following information: AI model information that is enabled, and AI model information that improves accuracy.
  • the terminal when the terminal detects overheating mitigation or overheating elimination, the terminal may report the second overheating assistance information to the network side.
  • the network side configures additional dynamic information. Delivers the configured additional dynamic information to the terminal.
  • the network side may deliver the additional dynamic information to the terminal through RRC signaling.
  • the dynamic information may further include static information
  • the static information includes configuration permission information of the AI model.
  • the configuration permission information of the AI model is used to indicate the configuration permission for the AI model.
  • the terminal can configure the working state of the AI model according to the working state configuration of the AI model when the terminal is overheated indicated by the dynamic information.
  • the static information can be stored locally on the terminal. When the terminal is overheated later, the terminal can decide to adjust the working status of the AI model based on the configuration permission information of the AI model in the static information.
  • the static information can also be used to readjust the working state of the AI model based on the configuration permission information of the AI model in the static information when the terminal is overheated or eliminated.
  • the network side can carry static information in the dynamic information when the terminal overheating problem occurs for the first time, or after the terminal overheating exceeds the set number of times, to indicate the configuration authority of the AI model. to the terminal.
  • the terminal overheats later, it does not need to interact with the network side, that is, it can decide how to adjust the working status of the AI model based on the configuration permissions of the AI model indicated by the static information, thereby improving the efficiency of overheating.
  • AI model described in the above embodiments of the present invention may be a model with an artificial intelligence (AI-enabled) network function in the terminal, so as to promote the development of the communication network towards automation and intelligence.
  • AI-enabled artificial intelligence
  • Embodiments of the present invention further provide an overheating device, which may be used for terminals, and may also be used for some chips or other components.
  • an overheating device which may be used for terminals, and may also be used for some chips or other components.
  • FIG. 3 a schematic structural diagram of an overheating device in an embodiment of the present invention is given, and the overheating device 30 may include:
  • an obtaining unit 31 configured to obtain configuration indication information when it is detected that the terminal is overheated, where the configuration indication information is used to indicate the working state configuration of the AI model;
  • the configuration unit 32 is configured to adjust the working state of the corresponding AI model according to the configuration instruction information.
  • the above-mentioned overheating device 30 may correspond to a chip with an overheating function in a terminal (also referred to as user equipment); or a chip with a data processing function, such as a baseband chip; or a user equipment
  • each module/unit included in each device and product described in the above embodiments it may be a software module/unit, a hardware module/unit, or a part of a software module/unit, a part of which is a software module/unit. is a hardware module/unit.
  • each module/unit included therein may be implemented by hardware such as circuits, or at least some modules/units may be implemented by a software program.
  • Running on the processor integrated inside the chip the remaining (if any) part of the modules/units can be implemented by hardware such as circuits; for each device and product applied to or integrated in the chip module, the modules/units contained therein can be They are all implemented by hardware such as circuits, and different modules/units can be located in the same component of the chip module (such as chips, circuit modules, etc.) or in different components, or at least some of the modules/units can be implemented by software programs.
  • the software program runs on the processor integrated inside the chip module, and the remaining (if any) part of the modules/units can be implemented by hardware such as circuits; for each device and product applied to or integrated in the terminal, each module contained in it
  • the units/units may all be implemented in hardware such as circuits, and different modules/units may be located in the same component (eg, chip, circuit module, etc.) or in different components in the terminal, or at least some of the modules/units may be implemented by software programs Realization, the software program runs on the processor integrated inside the terminal, and the remaining (if any) part of the modules/units can be implemented in hardware such as circuits.
  • the embodiment of the present invention further provides another overheating device, which can be used on the network side.
  • the overheating device 40 may include: a sending unit 41 .
  • the sending unit 41 is configured to deliver configuration indication information to the terminal, so that when the terminal is overheated, the working state of the corresponding AI model is adjusted according to the configuration indication information, and the configuration indication information is used to indicate the working state configuration of the AI model.
  • the overheating device 40 may further include a receiving unit 42, and the receiving unit 42 may be configured to receive auxiliary information reported by the terminal.
  • the auxiliary information reported by the terminal may include the first overheating auxiliary information.
  • the first overheat auxiliary information is reported when the terminal is overheated.
  • the overheating device 40 may generate configuration indication information including dynamic information according to the first overheating assistance information.
  • the auxiliary information reported by the terminal may include second overheating auxiliary information.
  • the second overheating assistance information is reported when the terminal overheating is relieved or the overheating is eliminated.
  • the overheating device 40 may generate additional dynamic information according to the second overheating assistance information, and may send the second overheating assistance information to the terminal through the sending unit 41 .
  • An embodiment of the present invention further provides a storage medium, where the storage medium is a non-volatile storage medium or a non-transitory storage medium, and a computer program is stored thereon, and the computer program is provided by any of the foregoing embodiments when the processor runs The steps of the superheat treatment method.
  • An embodiment of the present invention further provides a device for overheating, including a memory and a processor, the memory stores a computer program that can run on the processor, and the processor executes any of the above when running the computer program
  • a device for overheating including a memory and a processor, the memory stores a computer program that can run on the processor, and the processor executes any of the above when running the computer program
  • the examples provide steps of the superheat treatment method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Debugging And Monitoring (AREA)

Abstract

本申请实施例公开了一种过热处理方法及装置、存储介质,所述过热处理方法,包括:当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;根据所述配置指示信息,调整相应AI模型的工作状态。上述方案通过对运行的AI模型的工作状态的调整,能够缓解或消除终端过热。

Description

过热处理方法及装置、存储介质 技术领域
本发明实施例涉及通信领域,尤其涉及一种过热处理方法及装置、存储介质。
背景技术
随人工智能(Artificial Intelligence,AI)技术在通信网络广泛应用,终端上应用AI技术也越来越普及。
然而,在终端执行AI模型的训练和推理,易导致终端过热。即使AI模型的训练和推理放在网络侧或云端执行,终端参与这种中心化AI训练和推理过程,依然存在终端过热问题。当终端过热时,易出现降频和卡顿,影响用户体验。
发明内容
本发明实施例解决的技术问题是在终端执行AI模型的训练和推理时,易导致终端过热。
为解决上述技术问题,本发明实施例提供一种过热处理方法,包括:当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;根据所述配置指示信息,调整相应AI模型的工作状态。
可选的,所述配置指示信息为网络侧预配置的静态信息,所述当检测到终端过热时,获取配置指示信息,包括:当检测到所述终端过热时,从本地获取由所述网络侧预配置的静态信息,所述静态信息包括:AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
可选的,所述过热处理方法,还包括:在调整相应AI模型的工作状态之后,当检测到终端过热缓解或者过热消除时,基于所述AI模型的配置权限信息,重新调整AI模型的工作状态。
可选的,所述当检测到终端过热时,获取配置指示信息,包括:当检测到所述终端过热时,从网络侧获取由所述网络侧配置的动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息。
可选的,所述当检测到终端过热时,获取配置指示信息,包括:当检测到所述终端过热时,上报第一过热辅助信息至所述网络侧,使得所述网络侧收到所述第一过热辅助信息后,下发所述动态信息。
可选的,所述第一过热辅助信息包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息。
可选的,所述第一过热指示信息用于指示终端过热,且终端过热与AI模型运行相关。
可选的,所述AI模型配置建议信息包括以下至少一种信息:建议保留的AI模型信息、建议关闭的AI模型信息、建议AI模型的关闭时长、建议降低精度的AI模型信息、建议AI模型的降低精度时长。
可选的,采用如下方式得到所述AI模型配置建议信息:根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及为缓解过热或消除过热预计减少的算力资源中的至 少一种,确定所述AI模型配置建议信息。
可选的,所述可用算力资源信息包括以下至少一种信息:当前时刻可用算力资源、为缓解过热或消除过热预计减少的算力资源。
可选的,所述过热处理方法,还包括:当检测到所述终端过热缓解或者过热消除时,从所述网络侧获取由所述网络侧配置的附加动态信息,所述附加动态信息用于指示所述终端过热缓解或者过热消除时AI模型的工作状态配置;根据所述附加动态信息,重新调整AI模型的工作状态。
可选的,所述当检测到所述终端过热缓解或者过热消除时,从所述网络侧获取由所述网络侧配置的附加动态信息,包括:当检测到终端过热缓解或者过热消除时,上报第二过热辅助信息至所述网络侧,使得所述网络侧收到所述第二过热辅助信息后,下发所述附加动态信息。
可选的,所述第二过热辅助信息包括以下至少一种信息:第二过热指示信息、AI模型配置建议信息、可用算力资源信息。
可选的,所述第二过热指示信息用于指示终端过热缓解或者过热消除。
可选的,所述AI模型配置建议信息包括以下至少一种信息:建议开启的AI模型信息、建议提升精度的AI模型信息。
可选的,采用如下方式得到所述AI模型配置建议信息:根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及预计不引起过热时可用算力资源中的至少一种,确定所述AI模型配置建议信息。
可选的,所述可用算力资源信息,包括以下至少一种信息:当前时刻可用算力资源、预计不引起过热时可用算力资源。
可选的,通过RRC信令上报所述第一过热辅助信息和所述第二过热辅助信息至所述网络侧。
可选的,所述过热处理方法,还包括:当检测到终端过热时,若当前处于空闲态或者非激活态,则建立与网络侧的连接。
可选的,所述动态信息还包括静态信息,所述静态信息包括AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
本发明实施例还提供另一种过热处理方法,包括:下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。
可选的,所述配置指示信息为预配置的静态信息,所述下发配置指示信息至终端,包括:预配置所述静态信息并下发至所述终端,所述静态信息包括:AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
可选的,为所述终端配置AI模型或者更新AI模型时,预配置所述静态信息。
可选的,所述下发配置指示信息至终端,包括:配置动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息,所述动态信息为当所述终端过热时所配置;将所述动态信息下发至所述终端。
可选的,所述配置动态信息,包括:当接收到第一过热辅助信息时,配置动态信息,所述第一过热辅助信息由所述终端在检测到过热时上报。
可选的,所述第一过热辅助信息包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息。
可选的,所述动态信息包括以下至少一种信息:保持开启的AI模型信息、关闭的AI模型信息、AI模型的关闭时长、降低精度的AI模型信息、AI模型的精度降低时长。
可选的,所述过热处理方法,还包括:配置附加动态信息,所述附加动态信息用于指示所述终端过热缓解或者过热消除时AI模型的工作状态配置;将所述附加动态信息下发至所述终端,使得所述终端根据所述附加动态信息重新调整AI模型的工作状态。
可选的,所述配置附加动态信息,包括:当接收到第二过热辅助信息时,配置所述附加动态信息,所述第二过热辅助信息由所述终端在过热缓解或者过热消除时上报。
可选的,所述第二过热辅助信息包括以下至少一种信息:第二过热指示信息、AI模型配置建议信息、可用算力资源信息。
可选的,所述附加动态信息包括以下至少一种信息:开启的AI模型信息、提高精度的AI模型信息。
可选的,所述动态信息还包括静态信息,所述静态信息包括AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
本发明实施例还提供一种过热处理装置,包括:获取单元,用于当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;配置单元,用于根据所述配置指示信息,调整相应AI模型的工作状态。
本发明实施例还提供另一种过热处理装置,包括:发送单元,用于下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。
本发明实施例还提供一种存储介质,所述存储介质为非易失性存储介质或非瞬态存储介质,其上存储有计算机程序,所述计算机程序被处理器运行时执行上述任一种过热处理方法的步骤。
本发明实施例还提供一种过热处理装置,包括存储器和处理器,所述存储器上存储有能够在所述处理器上运行的计算机程序,所述处理器运行所述计算机程序时执行上述任一种过热处理方法的步骤。
与现有技术相比,本发明实施例的技术方案具有以下有益效果:
当检测到终端过热时,获取配置指示信息,根据配置指示信息指示的AI模型的工作状态配置,调整AI模型的工作状态,通过对运行的AI模型的工作状态的调整,以缓解或消除终端过热问题是可以预期的。
进一步,配置指示信息为网络侧预配置的静态信息,静态信息包括AI模型的配置权限信息,从而当终端过热时,终端即可从终端本地获取静态信息,并基于静态信息终端中的AI模型的配置权限信息指示的对AI模型的配置权限来调整相应AI模型的工作状态,从而可以在不与网络侧交互的情况下,解决终端过热的问题,提高终端过热的解决效率。
进一步,终端过热缓解或者过热消除时,基于所述AI模型的配置权限信息,重新调整AI模型的工作状态,以实现在不与网络侧进行交互的情况下,对AI模型的工作状态进行配置,恢复一些AI模型的工作状态,如重新开启在过热时关闭的AI模型或者提升在过热时降低精度的AI模型的精度等,保障终端的网络通信性能,提升用户体验。
进一步,配置指示信息可以包括动态信息,当终端过热时,从网络侧获取由网络侧配置的动态信息,有助于网络侧根据网络侧的整体情况为终端配置动态信息。
进一步,上报的第一过热辅助信息可以包括第一过热指示信息、AI模型配置建议信息、可用算力资源信息等中的一种或多种,终端可以通过第一过热辅助信息向网络侧指示终端因AI模型的运行原因导致过热。对于一些具有一定能力的终端,甚至可以通过AI模型配置建议信息和/或可用算力资源信息来给网络侧一些建议或者自身的实际情况,有助于网络侧根据终端的实际需求或者实际情况为终端配置较为合适的动态信息。
进一步,当终端过热缓解或者过热消除时,从网络侧获取由网络侧配置的附加动态信息,根据附加动态信息重新调整AI模型的工作状态,可以使得终端恢复一些AI模型的工作状态,如重新开启在过热时关闭的AI模型或者提升在过热时降低精度的AI模型的精度等,有助于保障终端的网络通信性能,提升用户体验。
附图说明
图1是本发明实施例中的一种终端过热处理方法的流程图;
图2是本发明实施例中的另一种终端过热处理方法的流程图;
图3是本发明实施例中的一种终端过热处理装置的结构示意图;
图4是本发明实施例中的另一种终端过热处理装置的结构示意图。
具体实施方式
如背景技术中所述,在终端执行AI模型的训练和推理,易导致终端过热。即使AI模型的训练和推理放在网络侧或云端执行,终端参与这种中心化AI训练和推理过程,依然存在终端过热问题。当终端过热时,终端易出现降频和卡顿,从而影响用户体验。
为解决上述问题,在本发明实施例中,当检测到终端过热时,获取配置指示信息,根据配置指示信息指示的AI模型的工作状态配置,调整AI模型的工作状态,通过对运行的AI模型的工作状态的调整,以缓解或消除终端过热问题是可以预期的。
为使本发明实施例的上述目的、特征和有益效果能够更为明显易懂,下面结合附图对本发明的具体实施例做详细的说明。
本发明实施例提供一种过热处理方法,过热处理方法的执行主体可以为终端本身,也可以为能够用于终端的控制芯片、处理芯片、基带芯片等芯片或者其他各种恰当的元器件等。所述终端也可以称为用户设备(User Equipment,UE),所述终端可以是各种恰当的终端,例如手机、电脑、物联网设备、车载设备等,但并不限于此。
参见图1,给出了本发明实施例中的一种过热处理方法的流程图,过热处理方法具体可以包括如下步骤:
步骤S11,当检测到终端过热时,获取配置指示信息。
在具体实施中,所获取的配置指示信息可以用于指示AI模型的工作状态配置。
经研究发现,当终端执行AI模型的训练和推理时,需要消耗终端的算力资源,带来终端功耗提升。终端功耗的提升相应地会带来终端温度的升高。此外,即使AI模型的训练和推理放在网络侧或云端执行,终端参与这种中心化AI训练和推理过程,需要上传大量数据,导致终端的发射功耗增大。终端发射功耗增大时,易导致终端的温度升高。AI模型的运行 数目等与终端温度的相关,AI模型的运行数目越多,则终端越容易产生过热问题。
在一些实施例中,可以根据终端的温度判断是否出现终端过热。例如,当终端的温度高于设定温度阈值时,判定终端过热。
经研究发现,通常AI模型运行需要占用相应的算力资源,通常运行AI的模型数目较多或者运行的AI模型所需的算力资源较大时,所占用的图形处理器(Graphics Processing Unit,GPU)量也就较大。而AI模型运行时占用的算力资源与终端温度相关,AI模型运行时占用的算力资源越大,则终端越容易产生过热问题。而终端中运行的AI模型所需的算力资源与终端中的GPU的占用量相关,因此,可以通过GPU的占用量推断AI模型的运行情况以及终端是否出现过热问题。
在一些实施例中,可以根据终端中GPU的占用量以及GPU的总容量来判断终端是否过热。例如,根据GPU的占用量以及GPU的总量计算GPU的占用率,当GPU占用率大于设定阈值时,判定终端过热。
需要说明的是,上述举例仅为示意性说明,还可以存在其他的方式来判断终端是否过热,此处不再一一举例。
在具体实施中,终端过热可以为终端整体过热,也可以为终端中的部分系统或者元器件过热。如,芯片过热、发射模块过热、中央处理器(central processing unit,CPU)过热、GPU过热、基带芯片过热等。
在具体实施中,当由芯片或者其他恰当的元器件执行过热处理方法时,芯片或其他恰当的元器件可以接收其他耦接的部件对终端过热的检测结果,例如,芯片或者其他恰当的元器件可以接收耦接的温度检测装置发送的终端中的一些指定部件、系统的温度等。也可以在芯片或者其他恰当的元器件本身配置有温度检测装置,根据自身配置的温度检测装置所检测到的温度来判定是否出现终端过热问题。
步骤S12,根据所述配置指示信息,调整相应AI模型的工作状态。
在具体实施中,在获取到配置指示信息之后,可以根据配置指示信息指示的AI模型的工作状态配置,调整相应AI模型的工作状态。
在一些实施例中,调整相应AI模型的工作状态可以包括以下至少一种:保持开启一些AI模型、关闭一些AI模型、降低一些AI模型的精度等。
由上可知,当检测到终端过热时,获取配置指示信息,根据配置指示信息指示的AI模型的工作状态配置,调整AI模型的工作状态,通过对运行的AI模型的工作状态的调整,以缓解或消除终端过热问题是可以预期的。
在具体实施中,步骤S11中获取的配置指示信息可以为从本地获取的网络侧预配置的静态信息。
具体而言,在检测到终端过热时,可以在本地获取由网络侧预配置的静态信息。
静态信息可以包括AI模型的配置权限信息,其中,AI模型的配置权限信息用于指示对AI模型的配置权限。终端可以基于AI模型的配置权限信息指示的对AI模型的配置权限,调整AI模型的工作状态。通过更改终端侧的AI模型的工作状态,解决终端过热的问题。
根据过热处理方法的执行主体不同,具有对AI模型的配置权限的对象不同。当过热处理方法的执行主体为终端本身时,终端具有对AI模型的配置权限。当过热处理方法的执行主体为芯片或其他一些元器件时,芯片或其他一些元器件具有对AI模型的控制权限。
进一步地,在调整相应AI模型的工作状态之后,当检测到终端过热缓解或者过热消除时,可以基于AI模型的配置权限信息,重新调整AI模型的工作状态。例如,重新调整AI模型的工作状态可以包括重新开启在终端过热时关闭的一些AI模型、提升一些在终端过热时降低精度的AI模型的精度、增大终端上传的数据量等,通过调整终端的AI模型的配置,保障终端的通信网络的性能,提升用户体验。
在具体实施中,可以通过多种方式判断终端是否过热缓解或者过热消除。
在一些实施例中,可以通过时长来判断终端是否过热缓解或者过热消除。例如,若相应AI模型的工作状态调整后达到设定时长,则判定终端过热缓解或者过热消除。
由于过热缓解和过热消除对应的程度不同,具体可以通过配置设定时长的长短来区分是过热缓解还是过热消除。当相应AI模型的工作状态调整后达到第一设定时长时,判定终端过热缓解。当相应AI模型的工作状态调整后达到第二设定时长时,判定终端过热消除,第一设定时长小于第二设定时长。
在一些实施例中,可以通过终端的温度来判断终端是否过热缓解或者过热消除。例如,当终端温度低于第一温度阈值时,判定终端过热缓解;当终端温度低于第二温度阈值,判定终端过热消除,第一温度阈值高于第二温度阈值。
需要说明的是,还可以通过其他方式判定终端是否过热缓解或者过热消除,此处不做限定。
在一些非限制性实施例中,对AI模型的配置权限可以包括以下至少一种:对AI模型的开启及关闭的配置权限、对AI模型的关闭时长的配置权限、对AI模型的精度调整的配置权限、对AI模型的精度等级调整的配置权限等。
对AI模型的开启及关闭的配置权限,指对于具有开启及关闭的配置权限的AI模型,可以自行决定是否停止使用该AI模型,或者自行决定是否开启该AI模型。当终端过热时,可以自行决定停止具有开启及关闭的配置权限的AI模型。或者,在终端过热缓解或者过热消除时,可以自行决定开启具有开启及关闭的配置权限的AI模型。
进一步地,当具有对AI模型的关闭时长的配置权限时,在关闭时长的配置权限允许的最大关闭时长下,可以自行决定对AI模型的关闭时长。
进一步地,对于具有关闭时长的配置权限的AI模型,自将AI模型关闭之后可以计时,当达到设定的最大关闭时长之后可以自动启动该AI模型,以保障终端的通信性能。
对于不具有开启及关闭的配置权限的AI模型,不能自行决定是否停止使用该AI模型。当终端过热时,需保留不具有开启及关闭的配置权限的AI模型,维持不具有开启及关闭的配置权限的AI模型的运行,而不能自行决定是否可以关闭该AI模型。
进一步地,对于不具有开启及关闭的配置权限的AI模型,需要与网络侧协商这些AI模型的关闭或者开启,可以基于网络侧的指示来调整这些AI模型的工作状态。
对AI模型的精度调整的配置权限,指对于具有精度调整的配置权限的AI模型,可以自行决定对AI模型的精度调整。当终端过热时,可以自行决定降低具有精度调整的配置权限的AI模型的精度。或者,在终端过热缓解或者过热消除时,可以自动决定提升具有精度调整的配置权限的AI模型的精度。
进一步地,当具有AI模型的精度等级调整的配置权限时,可以根据指示的精度等级调整的配置权限对AI模型的精度等级进行配置。当终端过热时,可以根据指示的精度等级调 整的配置权限降低AI模型的精度等级。或者,在终端过热缓解或者过热消失时,可以根据指示的精度等级调整的配置权限提升AI模型的精度等级。
在具体实施中,网络侧配置的静态信息所包括的AI模型的配置权限信息可以通过多种方式指示给终端,只需满足网络侧与终端之间均知道AI模型的配置权限信息的具体代表的含义即可。
例如,针对终端对AI模型的开启及关闭的配置权限,采用指示符0和1来指示终端对AI模型的开启及关闭的配置权限。若AI模型的配置权限信息为指示符1时,表征终端具有该AI模型的开启及关闭的配置权限。若AI模型的配置权限信息为指示符0时,表征终端不具有该AI模型的开启及关闭的配置权限。在终端过热时,终端可以自行停止使用指示符为1的AI模型。而对于指示符为0的AI模型,即使终端过热,也需保留,终端不能自行决定是否可以关闭。
需要说明的是,不同类型的配置权限对应的配置权限信息可以采用不同的标识,根据实际需求还可以采用其他方式来指示AI模型的配置权限信息,此处不再一一举例。
在具体实施中,网络侧可以在为终端配置AI模型或者更新AI模型时,下发静态信息。静态信息可以作为AI模型的属性信息下发至终端,也可以作为指示信息下发至终端,也可以作为其他类型的信息下发至终端,此处不做限定。
在一些实施例中,网络侧为终端配置AI模型时,网络侧可以分别下发每个AI模型对应的静态信息,也即每个AI模型均对应一个静态信息,每个AI模型的静态信息所包含的AI模型的配置权限信息适用于该AI模型。当终端过热时需要获取终端上所有AI模型的静态信息。
在另一些实施例中,网络侧为终端配置AI模型时,网络侧也可以批量的将所有AI模型的配置权限信息放于一个静态信息中,也即该静态信息中包含终端中所有AI模型的配置权限信息。当终端过热时,终端仅需获取一个静态信息即可从静态信息中可以得知对所有AI模型的配置权限。
在又一些实施例中,网络侧为终端更新AI模型时,可以对分别下发每个AI模型或者更新的AI模型对应的静态信息,也即每个AI模型均对应一个静态信息,每个AI模型的静态信息中所包含的AI模型的配置权限信息适用于该AI模型,当终端过热时需要获取终端上所有AI模型的静态信息。
在又一些实施例中,网络侧为终端更新AI模型时,网络侧也可以批量的将所有AI模型的配置权限信息放于一个静态信息中,也即该静态信息中包含终端中所有AI模型的配置权限信息。当终端过热时,终端仅需获取一个静态信息即可从静态信息中可以得知对所有AI模型的配置权限。
在具体实施中,基于网络侧对整体配置调整需求,存在对AI模型的配置权限的调整的情况。当网络侧需要对AI模型的配置权限进行调整时,则可以下发更新的静态信息,更新的静态信息包括调整后的AI模型的配置权限信息。终端收到更新后的静态信息之后,可以将更新后的静态信息覆盖已有的所述静态信息,作为所述静态信息,用于后续的AI模型的工作状态的调整依据。
当终端基于AI模型的配置权限信息,重新调整AI模型的工作状态之后,通常终端的过热问题会缓解或者消失,但是若终端没有配置权限的AI模型运行数目较多时,即使终端基 于AI模型的配置权限信息,重新调整AI模型的工作状态,也可能依然存在终端过热的情况。
为了解决重新调整AI模型的工作状态之后,终端依然过热的问题,在本发明实施例中,若基于所述AI模型的配置权限信息,重新调整AI模型的工作状态之后,若检测到终端依然过热,则进一步从网络侧获取由所述网络侧配置的动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息。在获取到动态信息之后,终端可以根据动态信息指示的终端过热时AI模型的工作状态配置,再次调整AI模型的工作状态。通过再次调整AI模型的工作状态,实现对终端中AI模型的重新配置,以缓解或消除终端过热。
在一些实施例中,基于所述AI模型的配置权限信息,重新调整AI模型的工作状态之后,若检测到终端依然过热,终端依然过热时,终端当前可能处于空闲态或者非激活态。当终端当前处于空闲态或者非激活态时,建立与网络侧的连接。在建立与网络侧的连接之后,上报第一过热辅助信息。
在具体实施中,步骤S11中获取配置指示信息可以为从网络侧获取由所述网络侧配置的动态信息。所述配置指示信息包括所述动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置。在获取到动态信息之后,终端可以根据动态信息指示的过热时AI模型的工作状态配置,调整AI模型的工作状态。通过调整AI模型的工作状态,实现对终端中AI模型的重新配置,以缓解过热或消除过热。
在一些实施例中,网络侧可以根据第一过热辅助信息,实时为终端配置动态信息并下发至终端。从而每次为终端配置的动态信息能够符合终端当前的实际情况。
在另一些实施例中,网络侧也可以根据预配置的策略和映射关系,按照映射表中配置的映射关系从预设的几种配置中选择一种配置,不同的配置对应的AI模型的工作状态配置不同,基于选择的配置生成动态信息,并将动态信息下发至终端。如不同的AI模型引起的过热问题所对应的配置不同,或者不同类型的终端对应的配置不同,可以理解的是,还可以基于其他因素确定对应的配置,并在后续使用时,基于当前的实际需求选择对应的配置。
在具体实施中,动态信息可以包括以下至少一种信息:保持开启的AI模型信息、关闭的AI模型信息、AI模型的关闭时长、降低精度的AI模型信息、AI模型的精度降低时长等。
在一些实施例中,动态信息可以包括AI模型的标识信息(ID)以及AI模型的配置标识信息等。AI模型的标识信息用于标识AI模型,且与AI模型唯一对应。AI模型的配置标识信息用于标识AI模型的工作状态配置,AI模型的工作状态配置不同时对应的配置标识信息不同,如采用0标识AI模型关闭,采用1标识AI模型的保持开启等。从而可以根据AI模型的ID以及配置标识信息,来确定各AI模型的工作状态配置。
在一些实施例中,网络侧可以通过无线资源控制(Radio Resource Control,RRC)信令将配置指示信息下发至终端。
在具体实施中,当检测到所述终端过热时,上报第一过热辅助信息至所述网络侧,使得所述网络侧收到所述第一过热辅助信息后,下发所述动态信息。
在具体实施中,终端可以通过RRC信令上报第一过热辅助信息至网络侧。例如,RRC信令可以为终端辅助信息(UE Assistance Information,UAI)。此时,将第一过热辅助信息携带于过热辅助信息元素(OverheatingAssistance IE)中。可以理解的是,RRC信令也可以为其他RRC信令。
在具体实施中,第一过热辅助信息可以包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息等。
在具体实施中,第一过热指示信息用于指示终端过热,且终端过热与AI模型运行相关。终端可以通过第一过热指示信息可以向网络侧指示终端过热,且终端过热原因与终端运行的AI模型等因素相关。
在具体实施中,第一过热辅助信息中的AI模型配置建议信息包括以下至少一种信息:建议保留的AI模型信息、建议关闭的AI模型信息、建议AI模型的关闭时长、建议降低精度的AI模型信息、建议AI模型的降低精度时长等。
在一些实施例中,终端可以采用如下方式得到所述AI模型配置建议信息:根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及为缓解过热或消除过热预计减少的算力资源中的至少一种,确定所述AI模型配置建议信息。
在具体实施中,所述可用算力资源信息包括以下至少一种信息:当前时刻可用算力资源、为缓解过热或消除过热预计减少的算力资源等。
在具体实施中,根据终端的实际情况不同,第一过热辅助信息所包含的信息不同。具体而言:
在一些实施例中,第一过热辅助信息可以仅包括第一过热指示信息。当网络侧收到仅包括第一过热指示信息的第一过热辅助信息时,则可以得知终端因AI模型的运行数目过多或者AI模型的运行占用的算力资源过多等原因导致终端过热,网络侧响应第一过热辅助信息为终端配置动态信息并下发至终端。
在另一些实施例中,对于具有能力知道自身AI模型的运行情况的终端,终端可以在过热时,给网络侧一些合理的建议。具体而言,终端上报的第一过热辅助信息可以包括AI模型配置建议信息。终端可以通过AI模型配置建议信息向网络侧指示终端建议保留的AI模型信息、建议关闭的AI模型信息、建议AI模型的关闭时长、建议降低精度的AI模型信息、建议AI模型的降低精度时长等中的一种或多种。网络侧可以考虑AI模型配置建议信息为终端配置动态信息,以提高为终端所配置的动态信息与终端的实际需求的匹配度,有助于较好的缓解终端过热。
在又一些实施例中,对于能力受限,无法向网络侧给出合理建议的终端,终端可以计算当前时刻可用算力资源、为缓解过热或消除过热预计减少的算力资源等,并在上报的第一过热辅助信息中包括当前时刻可用算力资源、为缓解过热或消除过热预计减少的算力资源等中的一种或多种。以使得网络侧收到第一过热辅助信息之后,可以准确的获知终端因AI模型运行过多或者参与AI相关网络功能过多导致过热,并可以获取终端的可用算力资源信息以及为缓解过热或消除过热预计减少的算力资源等。从而网络侧可以综合考虑终端的可用算力资源、空口、基站、邻小区上报的能力以及曾经为终端配置的历史动态信息等中的一种或多种,确定如何为终端配置合适的动态信息。
需要说明的是,根据终端的实际需求以及终端的实际能力等,上报的第一过热辅助信息可以包括第一过热指示信息、AI模型配置建议信息、可用算力资源信息中两种或全部,此处不再一一举例。
在一些实施例中,当终端检测到过热时,终端当前可能处于空闲态或者非激活态。当终端当前处于空闲态或者非激活态时,建立与网络侧的连接。在建立与网络侧的连接之后, 上报第一过热辅助信息。
在一些非限制性实施例中,所述动态信息还可以包括静态信息,所述静态信息包括AI模型的配置权限信息。所述AI模型的配置权限信息用于指示对AI模型的配置权限。此时,终端可以根据动态信息指示的终端过热时的AI模型的工作状态配置,对AI模型的工作状态进行配置。并可以将静态信息存储于本地,后续终端过热时,终端可以基于静态信息中的AI模型的配置权限信息,自行决定调整AI模型的工作状态。静态信息还可以用于在终端过热缓解或者过热消除时,终端基于静态信息中的AI模型的配置权限信息,重新调整AI模型的工作状态。
其中,关于终端基于静态信息对AI模型的工作状态进行调整的具体实现方式可以参见本发明上述实施例中静态信息部分的相关描述,此处不再赘述。
在具体实施中,网络侧可以在终端首次出现终端过热问题时,或者在终端出现过热的次数超出设定次数之后等场景下,在动态信息中携带静态信息,以将对AI模型的配置权限指示给终端。当终端后续出现过热时,可以不用与网络侧进行交互,即可以基于静态信息指示的对AI模型的配置权限,自行决定如何调整AI模型的工作状态,从而可以提高过热处理效率。
需要说明的是,上述实施例中所举部分示例以终端作为执行主体进行说明,在实际应用中,对于执行主体为芯片或者其他一些元器件时,具体执行步骤相似,参照上述实施例中的相关描述,此处不再赘述。
参照图2,给出了本发明实施例中的另一种过热处理方法,当配置指示信息包括动态信息时,过热处理方法在图1所述的过热处理方法的基础上,还可以包括如下步骤S13及步骤S14,具体而言:
步骤S13,当检测到所述终端过热缓解或者过热消除时,从所述网络侧获取由所述网络侧配置的附加动态信息。
在具体实施中,所述附加动态信息用于指示所述终端过热缓解或者过热消除时AI模型的工作状态配置。
在具体实施中,所述附加动态信息可以包括以下至少一种信息:开启的AI模型信息、提高精度的AI模型信息等。
步骤S14,根据所述附加动态信息,重新调整AI模型的工作状态。
进一步地,步骤S13可以包括:当检测到终端过热缓解或者过热消除时,上报第二过热辅助信息至所述网络侧,使得所述网络侧收到所述第二过热辅助信息后,下发所述附加动态信息。
在具体实施中,终端可以通过RRC信令上报第二过热辅助信息至网络侧。例如,RRC信令可以为UAI。此时,可以将第二过热辅助信息携带于OverheatingAssistance IE中。可以理解的是,RRC信令也可以为其他RRC信令。
在具体实施中,所述第二过热辅助信息可以包括以下至少一种信息:第二过热指示信息、AI模型配置建议信息、可用算力资源信息等。
在具体实施中,所述第二过热指示信息用于指示终端过热缓解或者过热消除。也即终端可以通过第二过热指示信息可以向网络侧指示终端过热缓解或过热消除。
在具体实施中,第二过热辅助信息中的所述AI模型配置建议信息包括以下至少一种信 息:建议开启的AI模型信息、建议提升精度的AI模型信息等。
在一些实施例中,终端可以采用如下方式得到所述AI模型配置建议信息:根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及预计不引起过热时可用算力资源中的至少一种,确定所述AI模型配置建议信息。
在具体实施中,所述可用算力资源信息可以包括以下至少一种信息:当前时刻可用算力资源、预计不引起过热时可用算力资源等。
在具体实施中,根据终端的实际情况不同,第二过热辅助信息所包含的信息不同。具体而言:
在一些实施例中,第二过热辅助信息可以仅包括第二过热指示信息。当网络侧收到仅包括第二过热指示信息的第二过热辅助信息时,则可以得知终端过热缓解或过热消除,可以响应于第二过热辅助信息,为终端配置附加动态信息。
在另一些实施例中,对于具有能力知道自身运行的AI模型情况的终端,可以在终端过热缓解或过热消除时,给网络侧一些合理的建议。具体而言,终端可以通过上报的第二过热辅助信息中的AI模型配置建议信息,给网络侧建议。终端可以通过AI模型配置建议信息向网络侧指示终端建议开启的AI模型信息、建议提升精度的AI模型信息等中的一种或多种。网络侧可以考虑AI模型配置建议信息为终端配置附加动态信息,以提高为终端所配置的附加动态信息与终端实际需求的匹配性,有助于保障终端的通信性能。
在又一些实施例中,对于能力受限,无法向网络侧给出合理建议的终端,终端可以计算当前时刻可用算力资源、预计不引起过热时可用算力资源等,并在上报的第二过热辅助信息中包括当前时刻可用算力资源、预计不引起过热时可用算力资源等中的一种或多种。网络侧收到第二过热辅助信息之后,可以准确的获知终端的可用算力资源信息。从而网络侧可以综合考虑终端的可用算力资源、空口、基站、邻小区上报的能力以及曾经为终端配置的历史附加动态信息等中的一种或多种,确定如何为终端配置合适的附加动态信息。
需要说明的是,根据终端的实际需求以及终端的实际能力等,上报的第二过热辅助信息可以包括第二过热指示信息、AI模型配置建议信息、可用算力资源信息中两种或全部,此处不再一一举例。
在一些实施例中,当终端检测到过热缓解或者过热消除时,若终端当前可能处于空闲态或者非激活态,则建立与网络侧的连接。在建立与网络侧的连接之后,上报第二过热辅助信息。
本发明实施例还提供另一种过热处理方法,可以用于网络侧,过热处理方法具体可以包括如下步骤:下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。从而在终端过热时,可以根据配置指示信息,调整相应AI模型的工作状态。例如,关闭一些AI模型的工作状态,调整一些AI模型的精度等,从而缓解终端过热问题是可以预期的。
在具体实施中,配置指示信息为预配置的静态信息。也即网络侧可以预配置的静态信息,并将预配置的静态信息下发至终端。
静态信息可以包括AI模型的配置权限信息,AI模型的配置权限信息用于指示对AI模型的配置权限。
网络侧可以为终端配置AI模型或者更新AI模型时,预配置所述静态信息。
在具体实施中,关于静态信息部分的描述可以参见本发明上述实施例中的描述,此处不再赘述。
在具体实施中,配置指示信息可以包括动态信息,所述动态信息为当所述终端过热时所配置。所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息。
网络侧完成动态信息配置之后,可以将动态信息下发至终端。
在一些实施例中,网络侧可以通过RRC信令将包含动态信息的配置指示信息下发至终端。
在具体实施中,所述动态信息可以包括以下至少一种信息:保持开启的AI模型信息、关闭的AI模型信息、AI模型的关闭时长、降低精度的AI模型信息、AI模型的精度降低时长等。
在具体实施中,终端在检测到过热时向网络侧上报第一过热辅助信息。网络侧可以在收到第一过热辅助信息时配置动态信息。
在具体实施中,所述第一过热辅助信息可以包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息等。
其中,对于第一过热辅助信息可以参考本发明上述实施例部分对第一过热辅助信息的相关描述,此处不再赘述。
在一些实施例中网络侧可以根据第一过热辅助信息,实时为终端配置动态信息并下发至终端。从而每次为终端配置的动态信息能够符合终端当前的实际情况。
在另一些实施例中,网络侧也可以根据预配置的策略和映射关系,按照映射表中配置的映射关系从预设的几种配置中选择一种配置,不同的配置对应的AI模型的工作状态配置不同,基于选择的配置生成动态信息,并将动态信息下发至终端。如不同的AI模型引起的过热问题所对应的配置不同,或者不同类型的终端对应的配置不同,可以理解的是,还可以基于其他因素确定对应的配置,并在后续使用时,基于当前的实际需求选择对应的配置。
在具体实施中,在终端过热缓解或者过热消除时,网络侧可以为终端配置附加动态信息。附加动态信息用于指示终端过热缓解或者过热消除时AI模型的工作状态配置。网络侧将附加动态信息下发至终端,终端根据附加动态信息重新调整相应AI模型的工作状态。
附加动态信息可以包括以下至少一种信息:开启的AI模型信息、提高精度的AI模型信息。
在具体实施中,终端检测到过热缓解或者过热消除时,可以向网络侧上报第二过热辅助信息。网络侧收到第二过热辅助信息时,配置附加动态信息。将配置的附加动态信息下发至终端。
在具体实施中,网络侧可以通过RRC信令将附加动态信息下发至终端。
在具体实施中,关于用于网络侧的过热处理方法的具体工作流程、工作原理以及与终端之间的交互过程,可以参见上述实施例提供的用于终端的过热处理方法中的相关描述,此处不再赘述。
在一些非限制性实施例中,所述动态信息还可以包括静态信息,所述静态信息包括AI模型的配置权限信息。所述AI模型的配置权限信息用于指示对AI模型的配置权限。此时,终端可以根据动态信息指示的终端过热时的AI模型的工作状态配置,对AI模型的工作状态 进行配置。并可以将静态信息存储于终端本地,后续终端过热时,终端可以基于静态信息中的AI模型的配置权限信息,自行决定调整AI模型的工作状态。静态信息还可以用于在终端过热缓解或者过热消除时,终端基于静态信息中的AI模型的配置权限信息,重新调整AI模型的工作状态。
其中,关于终端基于静态信息对AI模型的工作状态进行调整的具体实现方式可以参见本发明上述实施例中静态信息部分的相关描述,此处不再赘述。
在具体实施中,网络侧可以在终端首次出现终端过热问题时,或者在终端出现过热的次数超出设定次数之后等场景下,在动态信息中携带静态信息,以将对AI模型的配置权限指示给终端。当终端后续出现过热时,可以不用与网络侧进行交互,即可以基于静态信息指示的对AI模型的配置权限,自行决定如何调整AI模型的工作状态,从而可以提高过热处理效率。
需要说明的是,本发明上述实施例中所述的AI模型可以为终端中具有人工智能的(AI-enabled)网络功能的模型,以促进通信网络向自动化和智能化方向发展。
本发明实施例还提供一种过热处理装置,所述过热处理装置可以用于终端,也可以用于一些芯片或者其他元器件。参照图3,给出了本发明实施例中的一种过热处理装置的结构示意图,过热处理装置30可以包括:
获取单元31,用于当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;
配置单元32,用于根据所述配置指示信息,调整相应AI模型的工作状态。
在具体实施中,过热处理装置30的工作原理及工作流程可以参见本发明上述实施例中提供的图1及图2所示意的过热处理方法中的描述,此处不再赘述。
在具体实施中,上述的过热处理装置30可以对应于终端(也可称为用户设备)中具有过热处理功能的芯片;或者对应于具有数据处理功能的芯片,如基带芯片;或者对应于用户设备中包括具有过热处理功能的芯片的芯片模组;或者对应于具有数据处理功能芯片的芯片模组,或者对应于用户设备。
在具体实施中,关于上述实施例中描述的各个装置、产品包含的各个模块/单元,其可以是软件模块/单元,也可以是硬件模块/单元,或者也可以部分是软件模块/单元,部分是硬件模块/单元。
例如,对于应用于或集成于芯片的各个装置、产品,其包含的各个模块/单元可以都采用电路等硬件的方式实现,或者,至少部分模块/单元可以采用软件程序的方式实现,该软件程序运行于芯片内部集成的处理器,剩余的(如果有)部分模块/单元可以采用电路等硬件方式实现;对于应用于或集成于芯片模组的各个装置、产品,其包含的各个模块/单元可以都采用电路等硬件的方式实现,不同的模块/单元可以位于芯片模组的同一组件(例如芯片、电路模块等)或者不同组件中,或者,至少部分模块/单元可以采用软件程序的方式实现,该软件程序运行于芯片模组内部集成的处理器,剩余的(如果有)部分模块/单元可以采用电路等硬件方式实现;对于应用于或集成于终端的各个装置、产品,其包含的各个模块/单元可以都采用电路等硬件的方式实现,不同的模块/单元可以位于终端内同一组件(例如,芯片、电路模块等)或者不同组件中,或者,至少部分模块/单元可以采用软件程序的方式实现,该软件程序运行于终端内部集成的处理器,剩余的(如果有)部分模块/单元可 以采用电路等硬件方式实现。
本发明实施例还提供另一种过热处理装置,过热处理装置可以用于网络侧。参照图4给出了本发明实施例中的一种过热处理装置的结构示意图。过热处理装置40可以包括:发送单元41。发送单元41用于下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。
在具体实施例中,过热处理装置40还可以包括接收单元42,接收单元42可以用于接收终端上报的辅助信息。
在一些实施例中,终端上报的辅助信息可以包括第一过热辅助信息。第一过热辅助信息为终端过热时上报。过热处理装置40可以根据第一过热辅助信息生成包括动态信息的配置指示信息。
在另一些实施例中,终端上报的辅助信息可以包括第二过热辅助信息。第二过热辅助信息为终端过热缓解或者过热消除时上报。过热处理装置40可以根据第二过热辅助信息生成附加动态信息,并可以通过发送单元41将第二过热辅助信息发送至终端。
在具体实施中,过热处理装置40的工作原理及工作流程可以参见本发明上述实施例中提供的用于网络侧的过热处理方法中的描述,此处不再赘述。
本发明实施例还提供一种存储介质,存储介质为非易失性存储介质或非瞬态存储介质,其上存储有计算机程序,所述计算机程序被处理器运行时执行上述任一实施例提供的过热处理方法的步骤。
本发明实施例还提供一种过热处理装置,包括存储器和处理器,所述存储器上存储有能够在所述处理器上运行的计算机程序,所述处理器运行所述计算机程序时执行上述任一实施例提供过热处理方法的步骤。
需要说明的是,上文已经描述了具体实施方案,但这些实施方案并非要限制本发明公开的范围,即使仅相对于特定特征描述单个实施方案的情况下也是如此。本发明公开中提供的特征示例意在进行例示,而非限制,除非做出不同表述。在具体实施中,可根据实际需求,在技术上可行的情况下,将一项或者多项从属权利要求的技术特征与独立权利要求的技术特征进行组合,并可通过任何适当的方式而不是仅通过权利要求书中所列举的特定组合来组合来自相应独立权利要求的技术特征。
本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成,该程序可以存储于任一计算机可读存储介质中,存储介质可以包括:ROM、RAM、磁盘或光盘等。
虽然本发明披露如上,但本发明并非限定于此。任何本领域技术人员,在不脱离本发明的精神和范围内,均可作各种更动与修改,因此本发明的保护范围应当以权利要求所限定的范围为准。

Claims (36)

  1. 一种过热处理方法,其特征在于,包括:
    当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;
    根据所述配置指示信息,调整相应AI模型的工作状态。
  2. 如权利要求1所述的过热处理方法,其特征在于,所述配置指示信息为网络侧预配置的静态信息,所述当检测到终端过热时,获取配置指示信息,包括:
    当检测到所述终端过热时,从本地获取由所述网络侧预配置的静态信息,所述静态信息包括:AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
  3. 如权利要求2所述的过热处理方法,其特征在于,还包括:
    在调整相应AI模型的工作状态之后,当检测到终端过热缓解或者过热消除时,基于所述AI模型的配置权限信息,重新调整AI模型的工作状态。
  4. 如权利要求1所述的过热处理方法,其特征在于,所述当检测到终端过热时,获取配置指示信息,包括:
    当检测到所述终端过热时,从网络侧获取由所述网络侧配置的动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息。
  5. 如权利要求4所述的过热处理方法,其特征在于,所述当检测到终端过热时,获取配置指示信息,包括:
    当检测到所述终端过热时,上报第一过热辅助信息至所述网络侧,使得所述网络侧收到所述第一过热辅助信息后,下发所述动态信息。
  6. 如权利要求5所述的过热处理方法,其特征在于,所述第一过热辅助信息包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息。
  7. 如权利要求6所述的过热处理方法,其特征在于,所述第一过热指示信息用于指示终端过热,且终端过热与AI模型运行相关。
  8. 如权利要求6所述的过热处理方法,其特征在于,所述AI模型配置建议信息包括以下至少一种信息:建议保留的AI模型信息、建议关闭的AI模型信息、建议AI模型的关闭时长、建议降低精度的AI模型信息、建议AI模型的降低精度时长。
  9. 如权利要求6所述的过热处理方法,其特征在于,采用如下方式得到所述AI模型配置建议信息:
    根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及为缓解过热或消除过热预计减少的算力资源中的至少一种,确定所述AI模型配置建议信息。
  10. 如权利要求6所述的过热处理方法,其特征在于,所述可用算力资源信息包括以下至少一种信息:当前时刻可用算力资源、为缓解过热或消除过热预计减少的算力资源。
  11. 如权利要求4至10任一项所述的过热处理方法,其特征在于,还包括:
    当检测到所述终端过热缓解或者过热消除时,从所述网络侧获取由所述网络侧配置的附加动态信息,所述附加动态信息用于指示所述终端过热缓解或者过热消除时AI模型的工作状态配置;
    根据所述附加动态信息,重新调整AI模型的工作状态。
  12. 如权利要求11所述的过热处理方法,其特征在于,所述当检测到所述终端过热缓解或者过热消除时,从所述网络侧获取由所述网络侧配置的附加动态信息,包括:
    当检测到终端过热缓解或者过热消除时,上报第二过热辅助信息至所述网络侧,使得所述网络侧收到所述第二过热辅助信息后,下发所述附加动态信息。
  13. 如权利要求12所述的过热处理方法,其特征在于,所述第二过热辅助信息包括以下至少一种信息:第二过热指示信息、AI模型配置建议信息、可用算力资源信息。
  14. 如权利要求13所述的过热处理方法,其特征在于,所述第二过热指示信息用于指示过热缓解或者过热消除。
  15. 如权利要求13所述的过热处理方法,其特征在于,所述AI模型配置建议信息包括以下至少一种信息:建议开启的AI模型信息、建议提升精度的AI模型信息。
  16. 如权利要求13所述的过热处理方法,其特征在于,采用如下方式得到所述AI模型配置建议信息:
    根据历史AI模型的配置信息、各AI模型运行时占用的算力资源以及预计不引起过热时可用算力资源中的至少一种,确定所述AI模型配置建议信息。
  17. 如权利要求13所述的过热处理方法,其特征在于,所述可用算力资源信息,包括以下至少一种信息:当前时刻可用算力资源、预计不引起过热时可用算力资源。
  18. 如权利要求5所述的过热处理方法,其特征在于,通过RRC信令上报所述第一过热辅助信息至所述网络侧。
  19. 如权利要求4所述的过热处理方法,其特征在于,还包括:
    当检测到终端过热时,若当前处于空闲态或者非激活态,则建立与网络侧的连接。
  20. 如权利要求4所述的过热处理方法,其特征在于,所述动态信息还包括静态信息,所述静态信息包括AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
  21. 一种过热处理方法,其特征在于,包括:
    下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。
  22. 如权利要求21所述的过热处理方法,其特征在于,所述配置指示信息为预配置的静态信息,所述下发配置指示信息至终端,包括:
    预配置所述静态信息并下发至所述终端,所述静态信息包括:AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
  23. 如权利要求22所述的过热处理方法,其特征在于,为所述终端配置AI模型或者更新AI模型时,预配置所述静态信息。
  24. 如权利要求21所述的过热处理方法,其特征在于,所述下发配置指示信息至终端,包括:
    配置动态信息,所述动态信息用于指示终端过热时AI模型的工作状态配置,所述配置指示信息包括所述动态信息,所述动态信息为当所述终端过热时所配置;
    将所述动态信息下发至所述终端。
  25. 如权利要求24所述的过热处理方法,其特征在于,所述配置动态信息,包括:
    当接收到第一过热辅助信息时,配置动态信息,所述第一过热辅助信息由所述终端在 检测到过热时上报。
  26. 如权利要求25所述的过热处理方法,其特征在于,所述第一过热辅助信息包括以下至少一种信息:第一过热指示信息、AI模型配置建议信息、可用算力资源信息。
  27. 如权利要求24所述的过热处理方法,其特征在于,所述动态信息包括以下至少一种信息:保持开启的AI模型信息、关闭的AI模型信息、AI模型的关闭时长、降低精度的AI模型信息、AI模型的精度降低时长。
  28. 如权利要求24至27任一项所述的过热处理方法,其特征在于,还包括:
    配置附加动态信息,所述附加动态信息用于指示所述终端过热缓解或者过热消除时AI模型的工作状态配置;
    将所述附加动态信息下发至所述终端,使得所述终端根据所述附加动态信息重新调整AI模型的工作状态。
  29. 如权利要求28所述的过热处理方法,其特征在于,所述配置附加动态信息,包括:
    当接收到第二过热辅助信息时,配置所述附加动态信息,所述第二过热辅助信息由所述终端在过热缓解或者过热消除时上报。
  30. 如权利要求29所述的过热处理方法,其特征在于,所述第二过热辅助信息包括以下至少一种信息:第二过热指示信息、AI模型配置建议信息、可用算力资源信息。
  31. 如权利要求28所述的过热处理方法,其特征在于,所述附加动态信息包括以下至少一种信息:开启的AI模型信息、提高精度的AI模型信息。
  32. 如权利要求24所述的过热处理方法,其特征在于,所述动态信息还包括静态信息,所述静态信息包括AI模型的配置权限信息,所述AI模型的配置权限信息用于指示对AI模型的配置权限。
  33. 一种过热处理装置,其特征在于,包括:
    获取单元,用于当检测到终端过热时,获取配置指示信息,所述配置指示信息用于指示AI模型的工作状态配置;
    配置单元,用于根据所述配置指示信息,调整相应AI模型的工作状态。
  34. 一种过热处理装置,其特征在于,包括:
    发送单元,用于下发配置指示信息至终端,使得所述终端过热时,根据配置指示信息,调整相应AI模型的工作状态,所述配置指示信息用于指示AI模型的工作状态配置。
  35. 一种存储介质,所述存储介质为非易失性存储介质或非瞬态存储介质,其上存储有计算机程序,其特征在于,所述计算机程序被处理器运行时执行权利要求1至32任一项所述的过热处理方法的步骤。
  36. 一种过热处理装置,包括存储器和处理器,所述存储器上存储有能够在所述处理器上运行的计算机程序,其特征在于,所述处理器运行所述计算机程序时执行权利要求1至32中任一项所述的过热处理方法的步骤。
PCT/CN2022/076562 2021-01-08 2022-02-17 过热处理方法及装置、存储介质 WO2022148494A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/260,693 US20240061480A1 (en) 2021-01-08 2022-02-17 Method and device for processing overheat and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110025138.7A CN114756104A (zh) 2021-01-08 2021-01-08 过热处理方法及装置、存储介质
CN202110025138.7 2021-01-08

Publications (1)

Publication Number Publication Date
WO2022148494A1 true WO2022148494A1 (zh) 2022-07-14

Family

ID=82325528

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/076562 WO2022148494A1 (zh) 2021-01-08 2022-02-17 过热处理方法及装置、存储介质

Country Status (3)

Country Link
US (1) US20240061480A1 (zh)
CN (1) CN114756104A (zh)
WO (1) WO2022148494A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2622831A (en) * 2022-09-29 2024-04-03 Nokia Technologies Oy AI/ML configuration feedback

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118214750A (zh) * 2022-12-15 2024-06-18 维沃移动通信有限公司 Ai算力上报方法、终端及网络侧设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909183A (zh) * 2015-12-23 2017-06-30 西安中兴新软件有限责任公司 一种温度控制方法及移动终端
CN107463478A (zh) * 2017-07-06 2017-12-12 北京小米移动软件有限公司 终端设备的控制方法及装置
CN108388333A (zh) * 2018-01-25 2018-08-10 福州瑞芯微电子股份有限公司 一种基于电量设置乘法器精度的功耗调整方法和装置
US20190294964A1 (en) * 2018-03-20 2019-09-26 National Institute Of Advanced Industrial Science And Technology Computing system

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909183A (zh) * 2015-12-23 2017-06-30 西安中兴新软件有限责任公司 一种温度控制方法及移动终端
CN107463478A (zh) * 2017-07-06 2017-12-12 北京小米移动软件有限公司 终端设备的控制方法及装置
CN108388333A (zh) * 2018-01-25 2018-08-10 福州瑞芯微电子股份有限公司 一种基于电量设置乘法器精度的功耗调整方法和装置
US20190294964A1 (en) * 2018-03-20 2019-09-26 National Institute Of Advanced Industrial Science And Technology Computing system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2622831A (en) * 2022-09-29 2024-04-03 Nokia Technologies Oy AI/ML configuration feedback

Also Published As

Publication number Publication date
CN114756104A (zh) 2022-07-15
US20240061480A1 (en) 2024-02-22

Similar Documents

Publication Publication Date Title
WO2022148494A1 (zh) 过热处理方法及装置、存储介质
US11044681B2 (en) Configuring dual connectivity maximum transmit power
EP2705699A1 (en) Temperature-driven airlink selection in a multi-mode wireless device
US20170201936A1 (en) Method and Device for Controlling Network Connection
WO2018040258A1 (zh) 基站信号发射功率的调整方法、装置、设备及存储介质
WO2016066121A1 (zh) 一种实现基站和终端间灵活harq定时的方法及装置
US9985449B2 (en) Communication method, power adapter and terminal
AU2013397749A1 (en) Timing value adjustment method and apparatus
EP3261390B1 (en) Transmission power adjusting method and apparatus in gsm
CN108733454B (zh) 一种虚拟机故障处理方法和装置
EP3209005A1 (en) Task processing device, intelligent apparatus, task processing method and baseband processor
CN103856977B (zh) SCell删除方法及装置
WO2018143881A1 (en) Deferring a transmission of a message
WO2016050137A1 (zh) 一种基于tm9的载波聚合方法和设备
MX2021014520A (es) Metodo y aparato de adaptacion de modo de puerto.
CN110876149B (zh) 一种带宽部分bwp的接入方法、终端及网络设备
CN108234266B (zh) 总线型通讯网络的数据传输方法和装置及存储介质
CN105813187B (zh) 一种信息处理方法及电子设备
WO2023231829A1 (zh) 载波聚合控制方法及装置、终端和网络侧设备
US10178564B2 (en) Technique for handling a communication failure in a spectrum sharing communication system
WO2017166234A1 (zh) 信号传输方法、信号传输控制方法、用户设备及基站
WO2017049721A1 (zh) 用于基站的功耗控制方法及系统
CN113630129B (zh) 终端通讯控制方法、装置、设备及计算机可读存储介质
EP2608470A1 (en) Method and apparatus for achieving energy saving of data switching device
CN103577270A (zh) 控制分体式移动终端的使用方法及分体式移动终端

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22736655

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18260693

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22736655

Country of ref document: EP

Kind code of ref document: A1