WO2023125855A1 - 模型更新方法及通信设备 - Google Patents

模型更新方法及通信设备 Download PDF

Info

Publication number
WO2023125855A1
WO2023125855A1 PCT/CN2022/143594 CN2022143594W WO2023125855A1 WO 2023125855 A1 WO2023125855 A1 WO 2023125855A1 CN 2022143594 W CN2022143594 W CN 2022143594W WO 2023125855 A1 WO2023125855 A1 WO 2023125855A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
communication device
information
prediction
time series
Prior art date
Application number
PCT/CN2022/143594
Other languages
English (en)
French (fr)
Inventor
贾承璐
杨昂
孙鹏
Original Assignee
维沃移动通信有限公司
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 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2023125855A1 publication Critical patent/WO2023125855A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B17/00Monitoring; Testing
    • H04B17/30Monitoring; Testing of propagation channels
    • H04B17/391Modelling the propagation channel
    • H04B17/3913Predictive models, e.g. based on neural network models
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B17/00Monitoring; Testing
    • H04B17/30Monitoring; Testing of propagation channels
    • H04B17/391Modelling the propagation channel
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W24/00Supervisory, monitoring or testing arrangements
    • H04W24/08Testing, supervising or monitoring using real traffic

Definitions

  • the present application belongs to the technical field of communication, and in particular relates to a model updating method and communication equipment.
  • NTN non-terrestrial network
  • high-speed rail high-speed rail
  • other high-mobility scenarios the rapid changes in the scattering environment around communication devices have severely shortened the coherence time of wireless channels. , that is to say, the detection result of the wireless channel by the communication device will quickly fail, so more frequent wireless channel detection must be performed to maintain good communication performance.
  • Time series prediction is a method for communication devices to predict future communication measurements based on observations of communication measurements in the past period of time. Therefore, time series prediction models are widely used in predicting key parameters related to time series in wireless communication networks.
  • Common time series forecasting models include Gaussian process (Gaussian process, GP). Gaussian process achieves good forecasting performance with lower model storage, calculation and parameter interaction overhead.
  • model parameters of the time series forecasting model may need to be updated, so how to update the time series forecasting model’s Model parameters become an urgent problem to be solved.
  • Embodiments of the present application provide a model updating method and a communication device, which can solve the problem of how to update model parameters of a time series prediction model.
  • a method for updating a model includes: a first communication device acquires model update information; the first communication device updates the time series prediction model used by the first communication device according to the model update information, and the time series prediction model For the execution of the prediction task; wherein, the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • a method for updating a model includes: the second communication device sends model update information to the first communication device, the model update information is used to update the time series prediction model used by the first communication device, and the time series prediction The model is used to predict the execution of the task; wherein, the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • a model update device in a third aspect, includes an acquisition module and an update module; the acquisition module is used to obtain model update information; the update module is used to update the time used by the first communication device according to the model update information Sequence prediction model, time series prediction model is used for the execution of prediction tasks; wherein, the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode .
  • a model update device in a fourth aspect, includes a sending module; the sending module is used to send model update information to the first communication device, and the model update information is used to update the time series prediction used by the first communication device Model, the time series prediction model is used to perform the prediction task; wherein, the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • a communication device in a fifth aspect, includes a processor and a memory, and the memory stores programs or instructions that can run on the processor, and when the program or instructions are executed by the processor, the first aspect or the second aspect is implemented. steps of the method.
  • a communication device including a processor and a communication interface, wherein, when the communication device is the first communication device, the communication interface is used to acquire model update information; the processor updates the first model update information according to the model update information.
  • the time series forecasting model, the time series forecasting model is used for the execution of the forecasting task, and the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • a communication system including: a terminal and a network-side device, the terminal is used to execute the steps of the model updating method in the first aspect, and the network-side device can be used to execute the steps of the model updating method in the second aspect ; or the network side device is used to execute the steps of the model updating method in the first aspect, and the terminal can be used to execute the steps of the model updating method in the second aspect.
  • the eighth aspect provides a readable storage medium, the readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the steps of the model updating method as described in the first aspect are implemented, or The steps for implementing the model updating method described in the second aspect.
  • a ninth aspect provides a chip, the chip includes a processor and a communication interface, the communication interface is coupled to the processor, the processor is used to run programs or instructions, and implement the model described in the first aspect update method, or implement the model update method described in the second aspect.
  • a computer program product is provided, the computer program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the method described in the first aspect or the second aspect The steps of the model update method.
  • the first communication device obtains model update information; the first communication device updates the time series prediction model used by the first communication device according to the model update information, and the time series prediction model is used for the execution of the prediction task; wherein,
  • the model update information includes at least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • the first communication device can update the information according to the model , updating the time series prediction model used by the first communication device, so that the time series prediction model can perform the prediction task more accurately, and thus enable the first communication device to obtain good communication performance.
  • FIG. 1 is a schematic diagram of a wireless communication system provided by an embodiment of the present application.
  • Fig. 2 is a schematic diagram of a neural network provided by an embodiment of the present application.
  • Fig. 3 is a schematic diagram of a neuron provided in an embodiment of the present application.
  • Fig. 4 is a schematic diagram of the variation of the normalized mean square error of different training window sizes with the prediction time unit provided by the embodiment of the present application;
  • Fig. 5 is a schematic diagram of a smooth-turn motion trajectory of a terminal's motion speed provided by an embodiment of the present application
  • Fig. 6 is a schematic diagram of an average error of prediction with different sampling intervals provided by the embodiment of the present application.
  • Fig. 7 is a schematic flow chart of a model update method provided by the embodiment of the present application.
  • Fig. 8 is a schematic structural diagram of a model updating device provided by an embodiment of the present application.
  • Fig. 9 is a schematic structural diagram of another model updating device provided by the embodiment of the present application.
  • FIG. 10 is a schematic structural diagram of a communication device provided by an embodiment of the present application.
  • Fig. 11 is a schematic hardware diagram of a communication device provided by an embodiment of the present application.
  • first, second and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific sequence or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application are capable of operation in sequences other than those illustrated or described herein and that "first" and “second” distinguish objects. It is usually one category, and the number of objects is not limited. For example, there may be one or more first objects.
  • “and/or” in the description and claims means at least one of the connected objects, and the character “/” generally means that the related objects are an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced LTE-Advanced
  • LTE-A Long Term Evolution-Advanced
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • system and “network” in the embodiments of the present application are often used interchangeably, and the described technology can be used for the above-mentioned system and radio technology, and can also be used for other systems and radio technologies.
  • the following description describes the New Radio (New Radio, NR) system for example purposes, and uses NR terms in most of the following descriptions, but these techniques can also be applied to applications other than NR system applications, such as the 6th generation (6- th Generation, 6G) communication system.
  • 6G 6th generation
  • Fig. 1 shows a block diagram of a wireless communication system to which the embodiment of the present application is applicable.
  • the wireless communication system includes a terminal 11 and a network side device 12 .
  • the terminal 11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer) or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a palmtop computer, a netbook, a super mobile personal computer (ultra-mobile personal computer, UMPC), mobile Internet device (Mobile Internet Device, MID), augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) equipment, robot, wearable device (Wearable Device) , Vehicle User Equipment (VUE), Pedestrian User Equipment (PUE), smart home (home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.), game consoles, personal computers (personal computer, PC), teller machine or self-service machine and other terminal side devices, wearable devices include: smart watches, smart bracelet
  • the network side device 12 may include an access network device or a core network device, where the access network device 12 may also be called a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function, or Wireless access network unit.
  • RAN Radio Access Network
  • RAN Radio Access Network
  • Wireless access network unit Wireless access network unit
  • the access network device 12 may include a base station, a wireless local area network (Wireless Local Area Network, WLAN) access point, a wireless fidelity (Wireless Fidelity, WiFi) node, etc., and the base station may be called a node B, an evolved node B (eNB), Access point, base transceiver station (Base Transceiver Station, BTS), radio base station, radio transceiver, basic service set (Basic Service Set, BSS), extended service set (Extended Service Set, ESS), home B node, home Evolved Node B, Transmission Reception Point (TRP) or some other appropriate term in the field, as long as the same technical effect is achieved, the base station is not limited to specific technical terms.
  • eNB evolved node B
  • BTS base transceiver station
  • BTS base transceiver station
  • BSS basic service set
  • Extended Service Set Extended Service Set
  • home B node home Evolved Node B
  • TRP Transmission Reception Point
  • Core network equipment may include but not limited to at least one of the following: core network nodes, core network functions, mobility management entities (Mobility Management Entity, MME), access mobility management functions (Access and Mobility Management Function, AMF), session management functions (Session Management Function, SMF), User Plane Function (UPF), Policy Control Function (Policy Control Function, PCF), Policy and Charging Rules Function (PCRF), edge application service Discovery function (Edge Application Server Discovery Function, EASDF), unified data management (Unified Data Management, UDM), unified data storage (Unified Data Repository, UDR), home subscriber server (Home Subscriber Server, HSS), centralized network configuration ( Centralized network configuration, CNC), network storage function (Network Repository Function, NRF), network exposure function (Network Exposure Function, NEF), local NEF (Local NEF, or L-NEF), binding
  • MME mobility management entities
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF User Plane Function
  • Policy Control Function Policy Control Function
  • Time series prediction is a method for communication equipment to predict future communication measurements based on the observation of communication measurements in the past period of time. It is essentially based on historical measurement data to mine the time correlation between the future and history.
  • AI artificial intelligence
  • neural network decision tree
  • support vector machine Bayesian classifier
  • a1, a2,..., aK is the input quantity
  • w is the weight (multiplicative coefficient)
  • b is the bias (additive coefficient)
  • ⁇ (.) is the activation function.
  • Common activation functions can include Sigmoid, tanh, linear rectification function (rectified linear unit, ReLU, also known as modified linear unit) and so on.
  • the parameters of the neural network can be optimized by gradient optimization algorithm.
  • Gradient optimization algorithm is a kind of algorithm that minimizes or maximizes the objective function (also called loss function), and the loss function is often a mathematical combination of model parameters and data. For example, given data x and its corresponding label Y, after constructing a neural network model f(), according to the input x, the predicted output f(x) can be obtained, and the gap between the predicted value and the real value can be calculated ( f(x)-Y), which is the loss function.
  • W and b the appropriate W and b to minimize the value of the loss function. The smaller the value of the loss function, the closer the model is to the real situation, that is, the higher the accuracy of the model.
  • BP error back propagation
  • the basic idea of the BP algorithm is: the learning process consists of two processes: the forward propagation of the signal and the back propagation of the error.
  • the input samples are passed in from the input layer, processed layer by layer by each hidden layer, and passed to the output layer. If the actual output of the output layer does not match the expected output, it will enter the error backpropagation stage.
  • Error backpropagation is to pass the output error back layer by layer through the hidden layer to the input layer in some form, and distribute the error to all units of each layer, so as to obtain the error signal of each layer unit, and this error signal is used as the correcting function of each unit.
  • This weight adjustment process of each layer of signal forward propagation and error back propagation is carried out repeatedly.
  • the process of continuously adjusting the weights is the learning and training process of the neural network. This process has been carried out until the error of the network output is reduced to an acceptable level, or until the preset number of learning times.
  • the optimization algorithm may include gradient descent (gradient descent), stochastic gradient descent (stochastic gradient descent, SGD), small batch gradient descent (mini-batch gradient descent), momentum method (momentum), stochastic gradient descent with momentum (Nesterov), adaptive gradient descent (ADAptive GRADient descent, Adagrad), Adadelta, root mean square error deceleration (root mean square prop, RMSprop), adaptive momentum estimation (adaptive moment estimation, Adam), etc.
  • gradient descent gradient descent
  • stochastic gradient descent stochastic gradient descent with momentum
  • NARPDient descent Adagrad
  • Adadelta root mean square error deceleration
  • adaptive momentum estimation adaptive moment estimation
  • GP is a commonly used time series forecasting model.
  • GP is a machine learning method developed based on statistical learning theory and Bayesian theory. It is suitable for dealing with complex regression and classification problems with high dimensions, small samples and nonlinearity, and has strong generalization ability.
  • GP has the advantages of easy implementation, small sample training, hyperparameter adaptive acquisition, flexible non-parametric inference, interpretable and probabilistic output, etc., and has the potential to solve complex time series prediction problems in future wireless communication systems. potential.
  • Fig. 4 shows the normalized mean squared error (NMSE) of different training window sizes as a function of prediction time units at a moving speed of 30 km/h, even when the training window size is only 4 In the case of 1 time unit, the CSI prediction for the first 2 time units still achieved relatively reliable results, and the prediction accuracy gradually decreased over time.
  • NMSE normalized mean squared error
  • Figure 6 shows that the position sampling interval is 10 time units, based on the history of 20 time units (such as the 0th, 10, 20, 30, ..., 190 time units) to predict the position of the next 5 time units (such as 200, 210, 220, 230, 240 time units), the predicted average position error varies with the position estimation error (The position of the historical 20 time units is estimated by the position reference signal PRS), even if the estimation error of the historical position information is 3m, the prediction error is still less than 0.5m.
  • 20 time units such as the 0th, 10, 20, 30, ..., 190 time units
  • an embodiment of the present application provides a model updating method, which can be applied to the wireless communication system shown in FIG. 1 , and the method can include the following steps 201 and 202 .
  • Step 201 the first communication device acquires model update information.
  • Step 202 the first communication device updates the time series prediction model used by the first communication device according to the model update information.
  • the above-mentioned time series prediction model can be used for the execution of the prediction task, and the above-mentioned model update information can include at least one of the following:
  • model update information may also include any other possible update information, which may be determined according to actual usage requirements, and is not limited in the embodiment of the present application.
  • the second A communication device may update the time series prediction model used by the first communication device according to the model update information, so that the time series prediction model can perform the prediction task more accurately. In this way, the first communication device can obtain good communication performance.
  • model update information for example, at least one of kernel function update information, model hyperparameter update information, prediction mode update information, or calculation mode update information
  • the above prediction mode may include a time window structure of the prediction task, and the time window structure may include at least one of the following:
  • the prediction task may include at least one of predicting a location and predicting a channel (for example, prediction of channel state information (CSI) of a channel).
  • CSI channel state information
  • the prediction task involved in the embodiment of the present application may also include any other possible tasks, which may be determined according to actual usage requirements, and are not limited in the embodiment of the present application.
  • the time unit may include but not limited to: reference signal period, prediction period, time slot, half time slot, symbol (such as an orthogonal frequency division multiplexing (orthogonal frequency division multiplex, OFDM) symbol ), subframe, wireless frame, millisecond, and second, which can be determined according to actual usage requirements.
  • reference signal period prediction period
  • time slot half time slot
  • symbol such as an orthogonal frequency division multiplexing (orthogonal frequency division multiplex, OFDM) symbol
  • subframe wireless frame
  • millisecond millisecond
  • second which can be determined according to actual usage requirements.
  • the length of the time unit of the training window of the above-mentioned time series prediction model refers to: the number of consecutive time units included in one training window.
  • the length of the time unit of the prediction window of the above time series prediction model refers to: the number of consecutive time units included in a prediction window.
  • each time unit has a piece of data input or output, that is, one time unit corresponds to one piece of data.
  • the above prediction task is the prediction of CSI, that is, predicting the CSI of M time units in the future according to the CSI of N time units in the past
  • the length of the time unit of the training window of the time series prediction model is N time units
  • the length of the time unit of the forecast window of the time series forecast model is M.
  • the interval between adjacent time units of the training window of the time series forecasting model is 2
  • the interval between adjacent time units of the prediction window of the time series forecasting model is 2N can be predicted according to the CSI of 0, 2, ..., 2N time units +2, 2N+4, ..., CSI of 2N+2M time units.
  • the model update information may also include at least one of the following:
  • the identification of the prediction task also called the target task to be updated
  • the identification of the model to be updated (also referred to as the target model, such as the time series prediction model in the embodiment of the present application);
  • the timestamp information of the model update to be updated.
  • the first communication device may be a terminal or a network side device, which may be determined according to actual usage requirements, and is not limited in the embodiment of the present application.
  • the network-side device mentioned in this embodiment of the application may be:
  • Core network nodes including network data analysis function (network data analytics function, NWDAF), location management function (location management function, LMF), or neural network processing nodes;
  • NWDAF network data analytics function
  • LMF location management function
  • neural network processing nodes including neural network processing nodes;
  • Access network nodes including base stations or newly defined neural network processing nodes
  • the first communication device may directly determine model update information according to information such as forecast tasks and/or forecast demands.
  • the first communication device may receive model update information from the second communication device.
  • step 202 may be specifically implemented through the following steps 203 and 202a.
  • Step 203 the second communication device sends model update information to the first communication device.
  • Step 202a the first communication device receives model update information from the second communication device.
  • the above-mentioned second communication device may be a network side device, or may be a terminal.
  • the second communication device in a case where the above-mentioned first communication device is a terminal, the second communication device may be a network side device; in a case where the first communication device is a network side device, the second communication device may be a terminal.
  • it may be determined according to actual usage requirements, and is not limited in this embodiment of the application.
  • the first communication device may store the same prediction model list as that of the second communication device, and the prediction model list may include an identifier of the prediction mode.
  • the first communication device may determine the prediction mode indicated by the update information of the prediction mode from the list of prediction models.
  • the second communication device may store the same prediction model list as that of the first communication device, and the prediction model list may include the identification of the prediction mode, so that the second communication device may configure the first communication device according to the model prediction list. Appropriate prediction model.
  • the second communication device may periodically send the above model update information to the first communication device, or may send the model update information to the first communication device aperiodically. It can be understood that the first communication device may update the time series prediction model periodically, or may update the time series prediction model aperiodically.
  • step 203 may be specifically implemented through the following step 203a.
  • Step 203a the second communication device sends model update information to the first communication device according to the fourth information.
  • the above fourth information may include at least one of the following:
  • the second communication device may send model update information to the first communication device.
  • the mobility information of the first communication device may include at least one of the moving speed of the first communication device, beam switching information of the first communication device, and cell switching information of the first communication device. item.
  • the second communication device may send model update information to the first communication device; and/or, when the beam switching frequency of the first communication device is greater than the first threshold When the preset threshold is reached, the second communication device may send model update information to the first communication device; and/or, when the first communication device completes cell switching, the second communication device may send model update information to the first communication device.
  • the foregoing prediction task may include at least one of predicting a position and predicting a channel.
  • the statistical information of the prediction results of the time series prediction model may include the mean value of the prediction results and/or the variance of the prediction results.
  • the second communication The device may send model update information to the first communications device.
  • the foregoing prediction performance requirements may include at least one of prediction accuracy and calculation delay.
  • the foregoing prediction task may include at least one of predicting a position and predicting a channel.
  • the predicted performance requirements and predicted tasks involved in the embodiments of the present application may also include any other possible performance requirements and tasks, which may be determined according to actual usage requirements, and are not limited in the embodiments of the present application.
  • the above-mentioned environmental perception information may include but not limited to: line of sight (line of sight, OS) environment, non-line of sight (non-line of sight, NLOS) environment, dense urban environment, rural The environment can be determined according to actual usage requirements.
  • the first communication device may send a model update request to the second communication device, so as to acquire model update information.
  • the model updating method provided by the embodiment of the present application may further include the following steps 204 and 205.
  • Step 204 the first communication device sends model update request information to the second communication device.
  • Step 205 the second communication device receives model update request information from the first communication device.
  • model update request information may be used to request to obtain the first update information;
  • model update information may include at least one of the following:
  • Second update information other than the first update information.
  • the second communication device may send the first update information obtained by the model update request information to the first communication device according to the model update request information, and/or , sending second update information except the first update information to the first communication device.
  • the second communication device may send the update information of the model hyperparameters to the first communication device, that is, the above model update information includes the first update information; and/or, the second communication device sends at least one update information of kernel function update information, prediction mode update information, and calculation mode update information to the first communication device, that is, the above-mentioned model update information includes The second update information other than the update information.
  • the above-mentioned model update request information may also indicate a condition or event that triggers an update of the time series prediction model.
  • the first communication device may periodically or aperiodically send a model update request to the second communication device, so as to obtain model update information.
  • the above step 204 may be specifically implemented through the following step 204a.
  • Step 204a the first communication device sends model update request information to the second communication device according to the first information.
  • the above-mentioned first information may include at least one of the following:
  • the first communication device may send model update request information to the second communication device; and/or, when the beam switching frequency of the first communication device is greater than When the second preset threshold is reached, the first communication device may send model update request information to the second communication device; and/or, when the first communication device completes cell handover, the first communication device may send a model update to the second communication device request information.
  • the first communication device may send model update request information to the second communication device.
  • step 204a the update of the time series prediction model is triggered by a condition.
  • step 204 may specifically be implemented through the following step 204b.
  • Step 204b When the first condition is satisfied, the first communication device sends model update request information to the second communication device.
  • the above-mentioned first condition may include at least one of the following:
  • the first communication device receives the model update instruction information sent by the second communication device
  • the above-mentioned model update indication information may be sent to the first communication device by the second communication device aperiodically (burstly), so that when the first communication device receives the model update After indicating the information, the first communication device may send model update request information to the second communication device, so that the aperiodic update of the time series prediction model may be implemented.
  • an event triggers an update of the time series prediction model.
  • model updating method provided in the embodiment of the present application may further include the following steps 206 and 207.
  • Step 206 the first communication device sends the first capability information to the second communication device.
  • Step 207 the second communication device receives first capability information from the first communication device.
  • the above first capability information may be used to indicate at least one of the following:
  • a model training configuration supported by the first communication device
  • Model prediction configurations supported by the first communications device are Model prediction configurations supported by the first communications device.
  • first capability information may be used to indicate the capability of the first communication device to update the configuration of the time series prediction model.
  • the above model optimizer configuration refers to the configuration of the optimizer used to update the time series prediction model.
  • the optimizer configuration may include at least one of the type of the optimizer and the state of the optimizer.
  • the optimizer configuration may also include other configuration information of the optimizer, which may be determined according to actual usage requirements, and is not limited in this embodiment of the present application.
  • the above-mentioned model training configuration may include the storage capability and computing capability of the first communication device, and the second communication device may configure corresponding model update information for the first communication device according to the model training configuration.
  • the first communication device may not support a relatively long training window size.
  • the second communication device may configure the time series prediction model to the first communication device, and after the first communication device receives configuration information of the time series prediction model, the first communication device may configure the time series prediction model to The second communication device feeds back whether the first communication device supports the time series prediction model.
  • model update method provided by the embodiment of the present application may further include the following steps 208-211.
  • Step 208 the second communication device sends the first model configuration to the first communication device.
  • Step 209 the first communication device receives the first model configuration from the second communication device.
  • Step 210 the first communication device sends the first feedback information to the second communication device.
  • Step 211 the second communication device receives first feedback information from the first communication device
  • the above first feedback information may be used to indicate whether the first communication device supports the time series prediction model indicated by the first model configuration.
  • the network-side device can first configure the time series prediction model to the terminal, and then the terminal reports to the network-side device whether it supports the network-side device Configured time series forecasting model.
  • the first communication device may first report the computing capability of the first communication device to the second communication device, and then the second communication device may configure time to the first communication device according to the computing capability Sequence prediction models.
  • model updating method provided by the embodiment of the present application may further include the following steps 212-215.
  • Step 212 the first communication device sends computing capability information to the second communication device.
  • the computing capability information may be used to indicate the computing capability of the first communication device.
  • Step 213 the second communication device receives computing capability information from the first communication device.
  • Step 214 the second communication device sends the second model configuration to the first communication device.
  • Step 215 the first communication device receives the second model configuration from the second communication device.
  • the time series prediction model indicated by the above second model configuration may be determined according to at least one of computing capability, processing delay, and prediction performance requirement of the first communication device.
  • the terminal may first report the computing capability of the terminal to the network-side device, and then after the network-side device receives the computing capability information, the network The side device can determine how to configure the time series prediction model to the terminal according to at least one of the terminal's computing capability, processing delay, and prediction performance requirements of the terminal's time series prediction model, so that the network side device can send the above-mentioned second Model configuration to configure the time series forecasting model to the terminal.
  • the processing delay may include but not limited to: model loading delay, data read-in and read-out delay, model calculation delay, etc., which may be determined according to actual usage requirements.
  • model updating method provided in the embodiment of the present application may further include the following steps 216 and 217.
  • Step 216 the first communication device sends the second information or update recommendation information of the time series prediction model to the second communication device.
  • Step 217 the second communication device receives the second information or update recommendation information of the time series prediction model from the first communication device.
  • the above-mentioned model update information is determined according to the above-mentioned second information.
  • the update recommendation information may include at least one of the following: kernel function recommendation information, model hyperparameter recommendation information, prediction mode recommendation information, and calculation mode recommendation information.
  • the second communication device may send corresponding model update information to the first communication device according to the update recommendation information.
  • the above update recommendation information may also include any other possible recommendation information, which may be determined according to actual usage requirements.
  • the second information may include at least one of the following:
  • the above statistical information of the estimation error of the time series prediction model may be the mean value and/or variance of the estimation error of the time series prediction model. Since GP predicts unknown information based on known information, the estimation error of the time series prediction model can be the estimation error of known information (historical information), such as channel state information (CSI) prediction, the past N time The CSI of a unit is estimated based on a pilot (sounding reference signal (SRS)), and there is an error in the estimation.
  • known information such as channel state information (CSI) prediction
  • SRS sounding reference signal
  • the statistical information of the model prediction error may be at least one of the mean value of the model prediction error and the variance of the model prediction error, which may be specifically determined according to actual usage requirements.
  • the statistical information of the noise may include a signal-to-noise ratio and a signal-to-interference ratio. Specifically, it may be determined according to actual usage requirements, and is not limited in this embodiment of the application.
  • the model update method provided in the embodiment of the present application may also include the following steps 218 and 219 .
  • Step 218 the first communication device sends the second capability information to the second communication device.
  • Step 219 the second communication device receives second capability information from the first communication device.
  • the above-mentioned second capability information may include at least one of the following:
  • the computing unit configuration information of the first communication device is not limited to
  • the foregoing second capability information may be used to indicate the hardware capability of the first communications device.
  • the above second capability information may further include at least one of the following:
  • the maximum amount of data that the first communication device supports caching such as the maximum number of time units of CSI that can be cached
  • the maximum calculation amount supported by the first communication device for example, how many time units of data are included in the training window of the time series prediction model
  • the maximum number of parallel computing threads supported by the first communication device for example, how many sub-models in the time series forecasting model are supported to be trained simultaneously.
  • the calculation mode may include the following calculation modes:
  • Serial computing mode that is, the training of the GP model (that is, the time series prediction model) is performed on the same computing unit, such as in a certain core of a CPU or a certain thread of a CPU;
  • Parallel computing mode that is, the training of the GP model (that is, the time series prediction model) is performed on multiple computing units at the same time, such as in multiple cores of a CPU or multiple threads of a CPU.
  • the model update method provided in the embodiment of the present application may further include the following steps 220 and 221 .
  • Step 220 the second communication device sends third information to the first communication device.
  • Step 221 the first communication device receives third information from the second communication device.
  • the above-mentioned third information may include at least one of the following:
  • the method of dividing the data set for example, dividing the data set into N parts, and training N sub-models of the time series prediction model at the same time, the length of the time unit of the training window of each sub-model becomes 1/N of the original. In this way, the computational complexity can be reduced.
  • the computational complexity of GP is O(n 3 ), and n is the number of samples in the training window. Therefore, when the number of samples in the data set is large, the computational complexity of GP is extremely high.
  • the The data set is divided into training multiple sub-models at the same time, which can significantly reduce the computational complexity);
  • the fusion method of the prediction results of multiple sub-models such as the weighted sum of the prediction results of multiple sub-models.
  • the data set in order to reduce the amount of calculation, can be divided into N parts, and N sub-models of the time series prediction model are trained.
  • the N sub-models are put together for training , have a common optimization objective or loss function. It can be understood that the process of optimizing the optimization objective is the process of training the N models at the same time.
  • the kernel function, model hyperparameters, prediction mode, and calculation mode are all parameter information of the time series prediction model
  • the first communication device obtains the model update information
  • the first communication device The time series forecasting model used by the first communication device can be updated according to the model update information, so that the time series forecasting model can perform the forecasting task more accurately, and thus enable the first communication device to obtain good communication performance.
  • the model update method provided in the embodiment of the present application may be executed by a model update device.
  • the model updating device performed by the model updating device is taken as an example to illustrate the model updating device provided in the embodiment of the present application.
  • the embodiment of the present application provides a model updating apparatus 300
  • the model updating apparatus 300 may include an acquiring module 301 and an updating module 302 .
  • the obtaining module 301 can be used to obtain model update information;
  • the update module 302 can be used to update the time series prediction model used by the first communication device according to the model update information, and the time series prediction model is used to perform the prediction task; wherein, the model
  • the updated information includes at least one of the following:
  • the acquiring module may include a receiving submodule, and the receiving submodule may be configured to receive model update information from the second communication device.
  • the model updating device may also include a sending module.
  • the sending module may be configured to send model update request information to the second communication device, where the model update request information is used to request to obtain first update information; wherein the model update information includes at least one of the following:
  • Second update information other than the first update information.
  • the sending module may specifically be configured to send model update request information to the second communication device according to the first information; wherein the first information includes at least one of the following:
  • the sending module may specifically be configured to send model update request information to the second communication device when the first condition is met;
  • the first condition includes at least one of the following:
  • the first communication device receives the model update instruction information sent by the second communication device
  • the forecast performance requirements of the time series forecast model configured by the first communication device are updated.
  • the prediction performance requirement includes at least one of prediction accuracy and calculation delay.
  • the model updating device may also include a sending module.
  • the sending module may be configured to send first capability information to the second communication device, where the first capability information is used to indicate at least one of the following:
  • a model training configuration supported by the first communication device
  • Model prediction configurations supported by the first communications device are Model prediction configurations supported by the first communications device.
  • the model updating device may also include a receiving module and a sending module.
  • the receiving module can be used to receive the first model configuration from the second communication device; the sending module can be used to send the first feedback information to the second communication device, and the first feedback information is used to indicate whether the first communication device supports the first model configuration.
  • the model updating device may also include a sending module and a receiving module.
  • the sending module can be used to send computing capability information to the second communication device, and the computing capability information is used to indicate the computing capability of the first communication device;
  • the receiving module can be used to receive the second model configuration from the second communication device, the second
  • the time series forecasting model indicated by the two model configurations is determined according to at least one of computing power, processing delay, and forecasting performance requirements of the time series forecasting model.
  • the model updating device may also include a sending module.
  • the sending module may be configured to send the second information or update recommendation information of the time series prediction model to the second communication device; wherein, the model update information is determined according to the second information.
  • the second information includes at least one of the following:
  • the prediction mode includes a time window structure of the prediction task, and the time window structure includes at least one of the following:
  • the first communication device stores the same prediction model list as that of the second communication device, and the prediction model list includes the identifier of the prediction mode.
  • the model update information includes update information of the calculation model;
  • the model update device may also include a sending module.
  • the sending module may be configured to send second capability information to the second communication device before the receiving submodule receives model update information from the second communication device, where the second capability information includes at least one of the following:
  • the computing unit configuration information of the first communication device is not limited to
  • the second capability information further includes at least one of the following:
  • the first communication device supports the maximum amount of cached data
  • the maximum number of parallel computing threads supported by the first communication device is the maximum number of parallel computing threads supported by the first communication device.
  • the update information of the calculation mode indicates that the calculation mode of the time series forecasting model is updated to a parallel calculation mode
  • the model update device may further include a receiving module.
  • the receiving module may be used to receive third information from the second communication device, where the third information includes at least one of the following:
  • model update information also includes at least one of the following:
  • the timestamp information of the model update to be updated.
  • the model update device since the kernel function, model hyperparameters, prediction mode, and calculation mode are all parameter information of the time series forecast model, after the model update device obtains the model update information, the model update device can according to The model update information updates the time series prediction model used by the first communication device, so that the time series prediction model can perform the prediction task more accurately, and thus enable the first communication device to obtain good communication performance.
  • the embodiment of the present application provides a model updating apparatus 400
  • the model updating apparatus 400 may include a sending module 401 .
  • the sending module 401 may be configured to send model update information to the first communication device, where the model update information is used to update the time series prediction model used by the first communication device, and the time series prediction model is used to perform the prediction task; wherein, the model update Information includes at least one of the following:
  • the sending module may specifically be configured to send model update information to the first communication device according to the fourth information; wherein the fourth information includes at least one of the following:
  • the model update information may also include a receiving module.
  • the receiving module may be configured to receive model update request information from the first communication device, where the model update request information is used to request to obtain first update information; wherein the model update information includes at least one of the following:
  • Second update information other than the first update information.
  • the model update information may also include a receiving module.
  • the receiving module may be configured to receive first capability information from the first communication device, where the first capability information is used to indicate at least one of the following:
  • a model training configuration supported by the first communication device
  • Model prediction configurations supported by the first communications device are Model prediction configurations supported by the first communications device.
  • the model update information may also include a receiving module.
  • the sending module can also be used to send the first model configuration to the first communication device;
  • the receiving module can be used to receive first feedback information from the first communication device, and the first feedback information is used to indicate whether the first communication device supports the first model configuration.
  • the model update information may also include a receiving module.
  • the receiving module can be used to receive computing capability information from the first communication device, and the computing capability information is used to indicate the computing capability of the first communication device; the sending module can also be used to send the second model configuration to the first communication device.
  • the time series forecasting model indicated by the two model configurations is determined according to at least one of computing power, processing delay, and forecasting performance requirements of the time series forecasting model.
  • the model update information may also include a receiving module.
  • the receiving module may be configured to receive second information or update recommendation information of the time series prediction model from the first communication device; wherein, the model update information is determined according to the second information.
  • the second information may include at least one of the following:
  • the prediction mode includes a time window structure of the prediction task, and the time window structure includes at least one of the following:
  • the second communication device stores the same prediction model list as that of the first communication device, and the prediction model list includes the identifier of the prediction mode.
  • the model update information includes update information of the calculation model; the model update information may also include a receiving module.
  • the receiving module may be configured to receive second capability information from the first communication device before the sending module sends model update information to the first communication device, where the second capability information includes at least one of the following:
  • the computing unit configuration information of the first communication device is not limited to
  • the second capability information further includes at least one of the following:
  • the first communication device supports the maximum amount of cached data
  • the maximum number of parallel computing threads supported by the first communication device is the maximum number of parallel computing threads supported by the first communication device.
  • the update information of the calculation mode indicates that the calculation mode of the time series prediction model is updated to a parallel calculation mode
  • the sending module is also used to send third information to the first communication device, the third information includes at least one of the following:
  • model update information also includes at least one of the following:
  • the timestamp information of the model update to be updated.
  • the model update device can send model update information to the first communication device, so that the first communication device
  • the device can update the time series prediction model used by the first communication device according to the model update information, so that the time series prediction model can perform the prediction task more accurately, and thus enable the first communication device to obtain good communication performance.
  • the model updating apparatus may be a communication device, such as a communication device with an operating system, or a component of the communication device, such as an integrated circuit or a chip.
  • the communication device may be a terminal, or other devices other than the terminal, such as a network side device.
  • the terminal may include but not limited to the types of terminal 11 listed above, and other devices may be core network devices or access network devices, servers, network attached storage (Network Attached Storage, NAS), etc., the embodiment of the present application Not specifically limited.
  • the model updating device provided in the embodiment of the present application can implement the various processes implemented in the above-mentioned model updating method embodiment, and achieve the same technical effect. To avoid repetition, details are not repeated here.
  • the embodiment of the present application further provides a communication device 500, including a processor 501 and a memory 502, and the memory 502 stores programs or instructions that can run on the processor 501, for example, the
  • the communication device 500 is a terminal
  • the program or instruction is executed by the processor 501
  • each step of the above-mentioned model updating method embodiment can be realized, and the same technical effect can be achieved.
  • the communication device 500 is a network-side device, when the program or instruction is executed by the processor 501, each step of the above-mentioned model updating method embodiment can be implemented, and the same technical effect can be achieved. To avoid repetition, details are not repeated here.
  • the embodiment of the present application also provides a communication device, including a processor and a communication interface.
  • the communication interface is used to obtain model update information; the processor updates the time series prediction model used by the first communication device according to the model update information; or, when the communication device is the second communication device
  • the communication interface is used to send model update information to the first communication device
  • the model update information is used to update the time series forecasting model used by the first communication device
  • the time series forecasting model is used to predict the execution of the task
  • the model update information includes At least one of the following: update information of kernel function; update information of model hyperparameters; update information of prediction mode; update information of calculation mode.
  • the communication device embodiment corresponds to the above method embodiment, and each implementation process and implementation manner of the above method embodiment can be applied to the communication device embodiment, and can achieve the same technical effect.
  • the communication device 600 includes: an antenna 61 , a radio frequency device 62 , a baseband device 63 , a processor 64 and a memory 65 .
  • the antenna 61 is connected to the radio frequency device 62 .
  • the radio frequency device 62 receives information through the antenna 61, and sends the received information to the baseband device 63 for processing.
  • the baseband device 63 processes the information to be sent and sends it to the radio frequency device 62
  • the radio frequency device 62 processes the received information and sends it out through the antenna 61 .
  • the methods performed by the communication devices (including the first communication device and the second communication device) in the above embodiments may be implemented in the baseband device 63, where the baseband device 63 includes a baseband processor.
  • the baseband device 63 can include at least one baseband board, for example, a plurality of chips are arranged on the baseband board, as shown in FIG.
  • the program executes the operations of the network side device shown in the above method embodiments.
  • the communication device may also include a network interface 66, such as a common public radio interface (common public radio interface, CPRI).
  • a network interface 66 such as a common public radio interface (common public radio interface, CPRI).
  • the communication device 600 of the embodiment of the present application further includes: instructions or programs stored in the memory 65 and operable on the processor 64, and the processor 64 calls the instructions or programs in the memory 65 to execute the instructions shown in FIG. 8 or FIG. 9 .
  • the methods executed by each module are shown to achieve the same technical effect. In order to avoid repetition, the details are not repeated here.
  • the embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, each process of the above-mentioned model updating method embodiment is realized, and the same technology can be achieved. Effect, in order to avoid repetition, will not repeat them here.
  • the processor is the processor in the terminal in the foregoing embodiment.
  • the readable storage medium includes a computer-readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and the like.
  • the embodiment of the present application further provides a chip, the chip includes a processor and a communication interface, the communication interface is coupled to the processor, and the processor is used to run programs or instructions to realize the various processes of the above-mentioned model update method embodiment, and can achieve the same Technical effects, in order to avoid repetition, will not be repeated here.
  • the chip mentioned in the embodiment of the present application may also be called a system-on-chip, a system-on-chip, a system-on-a-chip, or a system-on-a-chip.
  • the embodiment of the present application further provides a computer program/program product, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the various processes of the above-mentioned model updating method embodiment, And can achieve the same technical effect, in order to avoid repetition, no more details here.
  • the embodiment of the present application also provides a communication system, including: a terminal and a network side device, the terminal can be used to execute the steps of the above model updating method, and the network side device can be used to execute the steps of the above model updating method.
  • the term “comprising”, “comprising” or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase “comprising a " does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element.
  • the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions are performed, for example, the described methods may be performed in an order different from that described, and various steps may also be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
  • the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation.
  • the technical solution of the present application can be embodied in the form of computer software products, which are stored in a storage medium (such as ROM/RAM, magnetic disk, etc.) , CD-ROM), including several instructions to enable a terminal (which may be a mobile phone, computer, server, air conditioner, or network-side device, etc.) to execute the method of each embodiment of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Electromagnetism (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种模型更新方法及通信设备,属于通信技术领域,本申请实施例的模型更新方法包括:第一通信设备获取模型更新信息;第一通信设备根据模型更新信息,更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。

Description

模型更新方法及通信设备
相关申请的交叉引用
本申请主张在2021年12月30日在中国提交的中国专利申请号202111652530.0的优先权,其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域,具体涉及一种模型更新方法及通信设备。
背景技术
随着通信场景的移动性增强,例如非地面网络(non-terrestrial network,NTN)、高铁等多种高移动性场景,通信设备周围散射环境的快速变化,从而使得无线信道的相干时间被严重缩短,也就是说通信设备对无线信道的探测结果会迅速失效,所以必须进行更频繁的无线信道探测,以维持良好的通信性能。
时间序列预测是一种通信设备基于对过去一段时间通信测量量的观察,预测未来的通信测量量的方法,因此在预测无线通信网络中时间序列相关的关键参数中时间序列预测模型得到广泛应用。常见的时间序列预测模型包括高斯过程(Gaussian process,GP),高斯过程以较低模型存储、计算以及参数交互开销,实现良好的预测性能。
然而,在某些场景中(比如预测任务和/或预测需求发生变化),为保证时间序列预测模型的预测性能,时间序列预测模型的模型参数可能需要进行更新,从而如何更新时间序列预测模型的模型参数成为一个亟待解决的问题。
发明内容
本申请实施例提供一种模型更新方法及通信设备,能够解决如何更新时间序列预测模型的模型参数的问题。
第一方面,提供了一种模型更新方法,该方法包括:第一通信设备获取模型更新信息;第一通信设备根据模型更新信息,更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。
第二方面,提供了一种模型更新方法,该方法包括:第二通信设备向第一通信设备发送模型更新信息,模型更新信息用于更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。
第三方面,提供了一种模型更新装置,模型更新装置包括获取模块和更新模块;获取模块,用于获取模型更新信息;更新模块,用于根据模型更新信息,更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。
第四方面,提供了一种模型更新装置,该模型更新装置包括发送模块;发送模块,用于向第一通信设备发送模型更新信息,模型更新信息用于更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。
第五方面,提供了一种通信设备,该终端包括处理器和存储器,存储器存储可在处理器上运行的程序或指令,该程序或指令被处理器执行时实现如第一方面或第二方面的方法的步骤。
第六方面,提供了一种通信设备,包括处理器及通信接口,其中,在通信设备为第一通信设备的情况下,通信接口用于获取模型更新信息;处理器根据模型更新信息,更新第一通信设备使用的时间序列预测模型;或者,在通信设备为第二通信设备的情况下,通信接口用于向第一通信设备发送模型更新信 息,模型更新信息用于更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。
第七方面,提供了一种通信系统,包括:终端及网络侧设备,终端用于执行如第一方面的模型更新方法的步骤,网络侧设备可用于执行如第二方面的模型更新方法的步骤;或者网络侧设备用于执行如第一方面的模型更新方法的步骤,终端可用于执行如第二方面的模型更新方法的步骤。
第八方面,提供了一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如第一方面所述的模型更新方法的步骤,或者实现如第二方面所述的模型更新方法的步骤。
第九方面,提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如第一方面所述的模型更新方法,或实现如第二方面所述的模型更新方法。
第十方面,提供了一种计算机程序产品,所述计算机程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面或第二方面所述的模型更新方法的步骤。
在本申请实施例中,第一通信设备获取模型更新信息;第一通信设备根据模型更新信息,更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。通过该方案,由于核函数、模型超参数、预测模式和计算模式均为时间序列预测模型的参数信息,因此在第一通信设备获取到模型更新信息之后,第一通信设备可以根据该模型更新信息,更新第一通信设备使用的时间序列预测模型,从而使得时间序列预测模型能够更加准确地执行预测任务,进而使得第一通信设备获得良好的通信性能。
附图说明
图1是本申请实施例提供的一种无线通信系统的示意图;
图2是本申请实施例提供的一种神经网络的示意图;
图3是本申请实施例提供的一种神经元的示意图;
图4是本申请实施例提供的一种不同训练窗尺寸的归一化均方误差随预测时间单位的变化示意图;
图5是本申请实施例提供的一种终端的运动速度的smooth-turn运动轨迹示意图;
图6是本申请实施例提供的一种不同采样间隔预测的平均误差的示意图;
图7是本申请实施例提供的一种模型更新方法的流程示意图;
图8是本申请实施例提供的一种模型更新装置的结构示意图;
图9是本申请实施例提供的另一种模型更新装置的结构示意图;
图10是本申请实施例提供的通信设备的结构示意图;
图11是本申请实施例提供的通信设备的硬件示意图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员所获得的所有其他实施例,都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象,而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换,以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施,且“第一”、“第二”所区别的对象通常为一类,并不限定对象的个数,例如第一对象可以是一个,也可以是多个。此外,说明书以及权利要求中“和/或”表示所连接对象的至少其中之一,字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是,本申请实施例所描述的技术不限于长期演进型(Long Term Evolution,LTE)/LTE的演进(LTE-Advanced,LTE-A)系统,还可用于其他无线通信系统,诸如码分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division  Multiple Access,FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access,SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的系统和无线电技术,也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)系统,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR系统应用以外的应用,如第6代(6-th Generation,6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端11和网络侧设备12。其中,终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant,PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer,UMPC)、移动上网装置(Mobile Internet Device,MID)、增强现实(augmented reality,AR)/虚拟现实(virtual reality,VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(Vehicle User Equipment,VUE)、行人终端(Pedestrian User Equipment,PUE)、智能家居(具有无线通信功能的家居设备,如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer,PC)、柜员机或者自助机等终端侧设备,可穿戴式设备包括:智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是,在本申请实施例并不限定终端11的具体类型。网络侧设备12可以包括接入网设备或核心网设备,其中,接入网设备12也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备12可以包括基站、无线局域网(Wireless Local Area Network,WLAN)接入点、无线保真(Wireless Fidelity,WiFi)节点等,基站可被称为节点B、演进节点B(eNB)、接入点、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点、家用演进型B节点、发送接收点(Transmission Reception Point,TRP)或所述领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR系统中的基站为例进行介绍,并不限定基站的具体类型。核心网设备可以包含但不限于如下至少一项:核心网节点、核心网功能、移动管理实体(Mobility Management Entity,MME)、接入移动管理功能(Access and Mobility Management Function,AMF)、会话管理功能(Session Management Function,SMF)、用户平面功能(User Plane Function,UPF)、策略控制功能(Policy Control Function,PCF)、策略与计费规则功能单元(Policy and Charging Rules Function,PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function,EASDF)、统一数据管理(Unified Data Management,UDM),统一数据仓储(Unified Data Repository,UDR)、归属用户服务器(Home Subscriber Server,HSS)、集中式网络配置(Centralized network configuration,CNC)、网络存储功能(Network Repository Function,NRF),网络开放功能(Network Exposure Function,NEF)、本地NEF(Local NEF,或L-NEF)、绑定支持功能(Binding Support Function,BSF)、应用功能(Application Function,AF)等。需要说明的是,在本申请实施例中仅以NR系统中的核心网设备为例进行介绍,并不限定核心网设备的具体类型。
由于移动设备和移动流量的爆炸式增长以及海量应用场景的出现,未来的无线通信网络具有高速率、低时延以及移动性增强等多种指标需求。特别是对于移动性增强,像NTN、高铁等多种高移动性场景,由于周围散射环境的快速变化,无线信道的相干时间被严重缩短,使得通信设备(例如无线通信设备,比如终端、基站等)对无线信道的探测结果迅速失效,从而通信设备必须通过更加频繁的无线信道探测,维持良好的通信性能,这对网络的负载和能耗带来巨大的挑战。时间序列预测是一种通信设备基于对过去一段时间通信测量量的观察,预测未来的通信测量量的方法,本质上是基于历史测量量的数据,挖掘未来与历史之间的时间相关性。
目前,人工智能(artificial intelligence,AI)在各个领域得到了广泛的应用,将人工智能融入无线通信网络,显著提升吞吐量、时延以及用户容量等技术指标是未来的无线通信网络的重要任务。AI模块有多种实现方式,例如神经网络、决策树、支持向量机、贝叶斯分类器等。本申请以神经网络为例进 行说明,但是并不限定AI模块的具体类型。如图2所示,为一个神经网络的示意图。其中,神经网络由神经元组成,神经元的示意图如图3所示。其中a1,a2,…,aK为输入量,w为权值(乘性系数),b为偏置(加性系数),σ(.)为激活函数。常见的激活函数可以包括Sigmoid、tanh、线性整流函数(rectified linear unit,ReLU,也可以称为修正线性单元)等等。
神经网络的参数可以通过梯度优化算法进行优化。梯度优化算法是一类最小化或者最大化目标函数(也可以称为损失函数)的算法,而损失函数往往是模型参数和数据的数学组合。例如给定数据x和其对应的标签Y,在构建一个神经网络模型f()之后,根据输入x,可以得到预测输出f(x),并且可以计算出预测值和真实值之间的差距(f(x)-Y),即损失函数。在构建模型时,需要确定合适的W和b,使损失函数的值达到最小。损失函数的值越小,则模型越接近于真实情况,即模型的准确度越高。
优化算法大多是基于误差反向传播(error back propagation,BP)算法。BP算法的基本思想是:学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时,输入样本从输入层传入,经各隐层逐层处理后,传向输出层。若输出层的实际输出与期望的输出不符,则转入误差的反向传播阶段。误差反向传播是将输出误差以某种形式通过隐层向输入层逐层反传,并将误差分摊给各层的所有单元,从而获得各层单元的误差信号,此误差信号即作为修正各单元权值的依据。这种信号正向传播与误差反向传播的各层权值调整过程,是周而复始地进行的。权值不断调整的过程,也就是神经网络的学习训练过程。此过程一直进行到网络输出的误差减少到可接受的程度,或进行到预先设定的学习次数为止。
示例性地,优化算法可以包括梯度下降(gradient descent)、随机梯度下降(stochastic gradient descent,SGD)、小批量梯度下降(mini-batch gradient descent)、动量法(momentum)、带动量的随机梯度下降(Nesterov)、自适应梯度下降(ADAptive GRADient descent,Adagrad)、Adadelta、均方根误差降速(root mean square prop,RMSprop)、自适应动量估计(adaptive moment estimation,Adam)等。这些优化算法在误差反向传播时,都是根据损失函数得到的误差/损失,对当前神经元求导数/偏导,加上学习速率、之前的梯度/导数/偏导等影响,得到梯度,将梯度传给上一层。
GP是常用的一种时间序列预测模型。GP是基于统计学习理论和贝叶斯理论发展起来的一种机器学习方法,适用于处理高维度、小样本和非线性的复杂回归和分类问题,具有较强的泛化能力。与神经网络等方法相比,GP具有容易实现、小样本训练、超参数自适应获取、灵活非参数推断、可解释以及概率意义的输出等优点,具有解决未来无线通信系统复杂时间序列预测问题的潜力。
仿真结果表明,对于CSI、位置等时间相关性较强的变量,GP具有良好的预测效果。对于CSI预测,图4示出了30km/h的运动速度下,不同训练窗尺寸的归一化均方误差(normalized mean squared error,NMSE)随预测时间单位的变化,即使在训练窗尺寸只有4个时间单位的情况下,对起初2个时间单位的CSI预测仍取得了比较可靠的结果,预测精度随着时间推移逐渐降低。对于位置预测,在图5所示的15km/h的终端运动速度的smooth-turn运动轨迹下,图6示出了位置采样间隔为10个时间单位、基于历史20个时间单位(如第0,10,20,30,…,190时间单位)的位置预测未来5个时间单位(如第200,210,220,230,240时间单位)的位置,预测的平均位置误差随位置估计误差的变化(历史20个时间单位的位置通过位置参考信号PRS估计得到),即使历史位置信息的估计误差为3m的情况下,预测误差仍小于0.5m。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的模型更新方法进行详细地说明。
如图7所示,本申请实施例提供一种模型更新方法,该模型更新方法可以应用于如图1所示的无线通信系统,该方法可以包括下述的步骤201和步骤202。
步骤201、第一通信设备获取模型更新信息。
步骤202、第一通信设备根据模型更新信息,更新第一通信设备使用的时间序列预测模型。
其中,上述时间序列预测模型可以用于预测任务的执行,上述模型更新信息可以包括以下至少一项:
核函数的更新信息;
模型超参数的更新信息;
预测模式的更新信息;
计算模式的更新信息。
需要说明的是,本申请实施例中,上述模型更新信息还可以包括其它任意可能的更新信息,具体可以根据实际使用需求确定,本申请实施例不作限定。
本申请实施例中,在第一通信设备获取模型更新信息(例如核函数的更新信息、模型超参数的更新信息、预测模式的更新信息或计算模式的更新信息中的至少一项)之后,第一通信设备可以根据该模型更新信息,更新第一通信设备使用的时间序列预测模型,从而使得时间序列预测模型更加准确地执行预测任务。如此,可以使得第一通信设备获得良好的通信性能。
可选地,本申请实施例中,上述预测模式可以包括预测任务的时间窗的结构,该时间窗的结构可以包括以下至少一项:
时间序列预测模型的训练窗的时间单位的长度;
时间序列预测模型的预测窗的时间单位的长度;
时间序列预测模型的训练窗的相邻时间单位的间隔;
时间序列预测模型的预测窗的相邻时间单位的间隔。
可选地,本申请实施例中,上述预测任务可以包括预测位置和预测信道(例如信道的信道状态信息(channel state information,CSI)的预测)中的至少一项。
当然,实际实现时,本申请实施例涉及的预测任务还可以包括其它任意可能的任务,具体可以根据实际使用需求确定,本申请实施例不作限定。
可选地,本申请实施例中,时间单位可以包括但不限于:参考信号周期、预测周期、时隙、半时隙、符号(例如正交频分复用(orthogonal frequency division multiplex,OFDM)符号)、子帧、无线帧、毫秒、秒,具体可以根据实际使用需求确定。
本申请实施例中,上述时间序列预测模型的训练窗的时间单位的长度是指:一个训练窗内包含的连续的时间单位的数量。相应的,上述时间序列预测模型的预测窗的时间单位的长度是指:一个预测窗内包含的连续的时间单位的数量。其中,每个时间单位上均有一条数据输入或输出,即一个时间单位对应一条数据。
示例性地,假设上述预测任务为CSI的预测,即根据过去N个时间单位的CSI预测未来M个时间单位的CSI,则时间序列预测模型的训练窗的时间单位的长度为N个时间单位,时间序列预测模型的预测窗的时间单位的长度为M。另外,时间序列预测模型的训练窗相邻时间单位间隔为2,时间序列预测模型预测窗的相邻时间单位间隔为2,那么可以根据0,2,……,2N时间单位的CSI,预测2N+2,2N+4,……,2N+2M时间单位的CSI。
可选地,本申请实施例中,模型更新信息还可以包括以下至少一项:
待更新预测任务(也可以称为目标任务)的标识;
待更新模型(也可以称为目标模型,例如本申请实施例中的时间序列预测模型)的标识;
待更新模型更新的时间戳信息。
需要说明的是,本申请实施例中,第一通信设备可以为终端,也可以为网络侧设备,具体可以根据实际使用需求确定,本申请实施例不作限定。
可选地,本申请实施例提及的网络侧设备可以为:
a)核心网节点,包括网络数据分析功能(network data analytics function,NWDAF)、位置管理功能(location management function,LMF),或神经网络处理节点;
b)接入网节点,包括基站或新定义的神经网络处理节点;
c)以上多个节点的组合。
在一些实施例中,第一通信设备可以根据预测任务和/或预测需求等信息,直接确定模型更新信息。
在另一些实施例中,第一通信设备可以从第二通信设备接收模型更新信息。
对于上述另一些实施例,上述步骤202具体可以通过下述的步骤203和步骤202a实现。
步骤203、第二通信设备向第一通信设备发送模型更新信息。
步骤202a、第一通信设备从第二通信设备接收模型更新信息。
可选地,本申请实施例中,上述第二通信设备可以为网络侧设备,也可以为终端。示例性地,在上述第一通信设备为终端的情况下,第二通信设备可以为网络侧设备;在第一通信设备为网络侧设备的情况下,第二通信设备可以为终端。具体可以根据实际使用需求确定,本申请实施例不作限定。
可选地,本申请实施例中,第一通信设备可以存储有与第二通信设备相同的预测模型列表,该预测模型列表可以包括预测模式的标识。如此,在上述模型更新信息包括预测模式的更新信息的情况下,第一通信设备可以从该预测模型列表中确定该预测模式的更新信息指示的预测模式。
相应的,第二通信设备可以存储有与第一通信设备相同的预测模型列表,该预测模型列表可以包括预测模式的标识,从而使得第二通信设备可以根据该模型预测列表为第一通信设备配置合适的预测模式。
本申请实施例中,第二通信设备可以周期性地向第一通信设备发送上述模型更新信息,也可以非周期性地向第一通信设备发送模型更新信息。可以理解,第一通信设备可以周期性地更新时间序列预测模型,也可以非周期性地更新时间序列预测模型。
可选地,本申请实施例中,基于第二通信设备周期性地向第一通信设备发送模型更新信息,上述步骤203具体可以通过下述的步骤203a实现。
步骤203a、第二通信设备根据第四信息,向第一通信设备发送模型更新信息。
其中,上述第四信息可以包括以下至少一项:
时间序列预测模型的预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
时间序列预测模型的预测结果的统计信息;
时间序列预测模型的预测任务的更新信息;
时间序列预测模型的预测性能需求的更新信息;
环境感知信息。
可以理解,本申请实施例中,当第一通信设备使用的时间序列预测模型的预测任务和/或预测性能需求更新时,第二通信设备可以向第一通信设备发送模型更新信息。
可选地,本申请实施例中,上述第一通信设备的移动性信息可以包括第一通信设备的移动速度、第一通信设备的波束切换信息、第一通信设备的小区切换信息中的至少一项。
示例性地,当第一通信设备的移动速度大于第一预设阈值时,第二通信设备可以向第一通信设备发送模型更新信息;和/或,当第一通信设备的波束切换频率大于第二预设阈值时,第二通信设备可以向第一通信设备发送模型更新信息;和/或,当第一通信设备完成小区切换时,第二通信设备可以向第一通信设备发送模型更新信息。
可选地,本申请实施例中,上述预测任务可以包括预测位置和预测信道中的至少一项。
可选地,本申请实施例中,上述时间序列预测模型的预测结果的统计信息可以包括预测结果的均值和/或预测结果的方差。
示例性地,假设上述预测任务为预测位置,预测结果的统计信息为预测到的位置的方差,那么在第一通信设备预测到的位置的方差大于第三预设阈值的情况下,第二通信设备可以向第一通信设备发送模型更新信息。
可选地,本申请实施例中,上述预测性能需求可以包括预测精度和计算时延中的至少一项。上述预测任务可以包括预测位置和预测信道中的至少一项。
当然,实际实现时,本申请实施例涉及的预测性能需求和预测任务还可以包括其它任意可能的性能需求和任务,具体可以根据实际使用需求确定,本申请实施例不作限定。
可选地,本申请实施例中,上述环境感知信息可以包括但不限于:视距(line of sight,OS)环境、非视距(non-line of sight,NLOS)环境,密集城区环境、乡村环境,具体可以根据实际使用需求确定。
可选地,本申请实施例中,第一通信设备可以向第二通信设备发送模型更新请求,从而获取模型更新信息。基于此,在上述步骤203之前,本申请实施例提供的模型更新方法还可以包括下述的步骤204 和步骤205。
步骤204、第一通信设备向第二通信设备发送模型更新请求信息。
步骤205、第二通信设备从第一通信设备接收模型更新请求信息。
其中,上述模型更新请求信息可以用于请求获取第一更新信息;上述模型更新信息可以包括以下至少一项:
第一更新信息;
除第一更新信息之外的第二更新信息。
可以理解,在第二通信设备接收到上述模型更新请求信息之后,第二通信设备可以根据该模型更新请求信息,向第一通信设备发送模型更新请求信息请求获取的第一更新信息,和/或,向第一通信设备发送除第一更新信息之外的第二更新信息。
示例性地,假设模型更新请求信息指示的第一更新信息为模型超参数的更新信息,那么第二通信设备可以向第一通信设备发送模型超参数的更新信息,即上述模型更新信息包括第一更新信息;和/或,第二通信设备向第一通信设备发送核函数的更新信息、预测模式的更新信息、计算模式的更新信息中的至少一项更新信息,即上述模型更新信息包括除第一更新信息之外的第二更新信息。
可选地,本申请实施例中,上述模型更新请求信息还可以指示触发时间序列预测模型更新的条件或事件。
可选地,本申请实施例中,第一通信设备可以周期性或非周期性地向第二通信设备发送模型更新请求,从而获取模型更新信息。
在一些实施例中,上述步骤204具体可以通过下述的步骤204a实现。
步骤204a、第一通信设备根据第一信息,向第二通信设备发送模型更新请求信息。
其中,上述第一信息可以包括以下至少一项:
时间序列预测模型的预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
时间序列预测模型的预测结果的统计信息;
环境感知信息。
需要说明的是,本申请实施例中,对于第一通信设备的移动性信息、预测任务、预测结果的统计信息,以及环境感知信息的解释说明,具体可以参见上述实施例中的详细描述。为避免重复,此处不再赘述。
示例性地,当第一通信设备的移动速度大于第一预设阈值时,第一通信设备可以向第二通信设备发送模型更新请求信息;和/或,当第一通信设备的波束切换频率大于第二预设阈值时,第一通信设备可以向第二通信设备发送模型更新请求信息;和/或,当第一通信设备完成小区切换时,第一通信设备可以向第二通信设备发送模型更新请求信息。
示例性地,假设上述预测任务为预测位置,预测结果的统计信息为预测到的位置的方差,那么在第一通信设备预测到的位置的方差大于第三预设阈值的情况下,第一通信设备可以向第二通信设备发送模型更新请求信息。
可以理解,对于上述步骤204a,为条件触发时间序列预测模型的更新。
在另一些实施例中,上述步骤204具体可以通过下述的步骤204b实现。
步骤204b、在满足第一条件的情况下,第一通信设备向第二通信设备发送模型更新请求信息。
其中,上述第一条件可以包括以下至少一项:
第一通信设备接收到第二通信设备发送的模型更新指示信息;
时间序列预测模型的预测任务更新;
时间序列预测模型的预测性能需求更新。
需要说明的是,本申请实施例中,上述模型更新指示信息可以为第二通信设备非周期性(突发性)地发送给第一通信设备的,如此在第一通信设备接收到该模型更新指示信息之后,第一通信设备可以向 第二通信设备发送模型更新请求信息,从而可以实现时间序列预测模型的非周期性更新。
可以理解,对于上述步骤204b,为事件触发时间序列预测模型的更新。
可选地,本申请实施例提供的模型更新方法还可以包括下述的步骤206和步骤207。
步骤206、第一通信设备向第二通信设备发送第一能力信息。
步骤207、第二通信设备从第一通信设备接收第一能力信息。
其中,上述第一能力信息可以用于指示以下至少一项:
第一通信设备支持的核函数;
第一通信设备是否具有预测任务;
第一通信设备支持的模型优化器配置;
第一通信设备支持的模型训练配置;
第一通信设备支持的模型预测配置。
可以理解,上述第一能力信息可以用于指示第一通信设备更新时间序列预测模型的配置的能力。
本申请实施例中,上述模型优化器配置是指用于更新时间序列预测模型的优化器的配置。可选地,优化器配置可以包括优化器的种类、优化器的状态中的至少一项。当然,实际实现时,优化器配置还可以包括优化器的其它配置信息,具体可以根据实际使用需求确定,本申请实施例不作限定。
本申请实施例中,上述模型训练配置可以包括第一通信设备的存储能力、计算能力,第二通信设备可以根据该模型训练配置为第一通信设备配置相应的模型更新信息。示例性地,假设第一通信设备的存储能力和/计算能力较弱,那么第一通信设备可能不支持较长的训练窗尺寸。
可选地,在一种实施方式中,第二通信设备可以向第一通信设备配置时间序列预测模型,在第一通信设备接收到该时间序列预测模型的配置信息之后,第一通信设备可以向第二通信设备反馈第一通信设备是否支持该时间序列预测模型。
基于此,本申请实施例提供的模型更新方法还可以包括下述的步骤208-步骤211。
步骤208、第二通信设备向第一通信设备发送第一模型配置。
步骤209、第一通信设备从第二通信设备接收第一模型配置。
步骤210、第一通信设备向第二通信设备发送第一反馈信息。
步骤211、第二通信设备从第一通信设备接收第一反馈信息
其中,上述第一反馈信息可以用于指示第一通信设备是否支持第一模型配置指示的时间序列预测模型。
示例性地,假设上述第一通信设备为终端,第二通信设备为网络侧设备,那么网络侧设备可以先将时间序列预测模型配置给终端,然后终端再向网络侧设备上报是否支持网络侧设备配置的时间序列预测模型。
可选地,在另一种实施方式中,第一通信设备可以先向第二通信设备上报第一通信设备的计算能力,然后第二通信设备可以根据该计算能力,向第一通信设备配置时间序列预测模型。
基于此,本申请实施例提供的模型更新方法还可以包括下述的步骤212-步骤215。
步骤212、第一通信设备向第二通信设备发送计算能力信息。
其中,该计算能力信息可以用于指示第一通信设备的计算能力。
步骤213、第二通信设备从第一通信设备接收计算能力信息。
步骤214、第二通信设备向第一通信设备发送第二模型配置。
步骤215、第一通信设备从第二通信设备接收第二模型配置。
其中,上述第二模型配置指示的时间序列预测模型可以根据第一通信设备的计算能力、处理时延、第一通信设备的预测性能需求中的至少一项确定。
示例性地,假设上述第一通信设备为终端,第二通信设备为网络侧设备,那么终端可以先向网络侧设备上报终端的计算能力,然后在网络侧设备接收到上述计算能力信息之后,网络侧设备可以根据终端的计算能力、处理时延和终端的时间序列预测模型的预测性能需求中的至少一项,确定如何向终端配置时间序列预测模型,从而网络侧设备可以向终端发送上述第二模型配置,以向终端配置时间序列预测模 型。
可选地,本申请实施例中,处理时延可以包括但不限于:模型加载的时延、数据读入读出的时延、模型计算的时延等,具体可以根据实际使用需求确定。
可选地,本申请实施例提供的模型更新方法还可以包括下述的步骤216和步骤217。
步骤216、第一通信设备向第二通信设备发送第二信息或时间序列预测模型的更新推荐信息。
步骤217、第二通信设备从第一通信设备接收第二信息或时间序列预测模型的更新推荐信息。
其中,上述模型更新信息是根据上述第二信息确定的。
可选地,本申请实施例中,上述更新推荐信息可以包括以下至少一项:核函数的推荐信息、模型超参数的推荐信息、预测模式的推荐信息、计算模式的推荐信息。如此,在第二通信设备接收到该更新推荐信息之后,第二通信设备可以根据该更新推荐信息,向第一通信设备发送相应的模型更新信息。
当然,实际实现时,上述更新推荐信息还可以包括其它任意可能的推荐信息,具体可以根据实际使用需求确定。
可选地,本申请实施例中,第二信息可以包括以下至少一项:
时间序列预测模型的预测结果的统计信息;
时间序列预测模型的估计误差;
所述时间序列预测模型的估计误差的统计信息;
时间序列预测模型的模型预测误差;
所述时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
噪声的统计信息;
时间序列预测模型的预测性能需求。
本申请实施例中,上述时间序列预测模型的估计误差的统计信息可以为时间序列预测模型的估计误差的均值和/或方差。由于GP是基于已知的信息预测未知信息,因此时间序列预测模型的估计误差可以为已知信息(历史信息)的估计误差,如信道状态信息(channel state information,CSI)预测,过去N个时间单位的CSI是基于导频(探测参考信号(sounding reference signal,SRS))估计的,估计是存在误差的。
本申请实施例中,模型预测误差的统计信息可以为模型预测误差的均值、模型预测误差的方差中的至少一项,具体可以根据实际使用需求确定。
可选地,本申请实施例中,上述噪声的统计信息可以包括信噪比、信干燥比。具体可以根据实际使用需求确定,本申请实施例不作限定。
可选地,本申请实施例中,在模型更新信息包括计算模式的更新信息的情况下,在上述步骤203之前,本申请实施例提供的模型更新方法还可以包括下述的步骤218和步骤219。
步骤218、第一通信设备向第二通信设备发送第二能力信息。
步骤219、第二通信设备从第一通信设备接收第二能力信息。
其中,上述第二能力信息可以包括以下至少一项:
第一通信设备的计算能力信息;
第一通信设备的存储能力信息;
第一通信设备的计算单元配置信息。
可以理解,上述第二能力信息可以用于指示第一通信设备的硬件能力。
可选地,本申请实施例中,上述第二能力信息还可以包括以下至少一项:
第一通信设备支持缓存的最大数据量,例如最大可以缓存多少个时间单位的CSI;
第一通信设备支持计算的最大计算量,例如时间序列预测模型的训练窗包含多少个时间单位的数据;
第一通信设备支持的最大并行计算线程数,例如支持时间序列预测模型中的多少个子模型同时进行训练。
本申请实施例中,计算模式可以包括以下计算模式:
a)串行计算模式,即GP模型(即时间序列预测模型)训练在同一计算单元进行,如在CPU的某一核或CPU的某一线程内进行;
b)并行计算模式,即GP模型(即时间序列预测模型)训练在多个计算单元同时进行,如在CPU多个核或CPU的多个线程内进行。
可选地,在上述模型更新信息指示将时间序列预测模型的计算模式更新为并行计算模式的情况下,本申请实施例提供的模型更新方法还可以包括下述的步骤220和步骤221。
步骤220、第二通信设备向第一通信设备发送第三信息。
步骤221、第一通信设备从第二通信设备接收第三信息。
其中,上述第三信息可以包括以下至少一项:
数据集的划分方式,例如将数据集划分为N份,同时训练时间序列预测模型的N个子模型,每个子模型的训练窗的时间单位的长度变为原来的1/N。如此,可以降低计算复杂度,GP的计算复杂度为O(n 3),n为训练窗的样本数,因此当数据集的样本数很多时,GP的计算复杂度极高,通过并行计算将数据集划分为同时训练多个子模型,能够显著降低计算复杂度);
优化目标的选择,例如多个模型边缘函数的加权和;
模型优化器的选择;
模型优化器的初始状态;
多个子模型预测结果的融合方式,例如多个子模型预测结果的加权和。
需要说明的是,本申请实施例中,为了降低计算量,可以将数据集划分为N份,训练时间序列预测模型的N个子模型,在训练过程中,该N个子模型是放在一起的训练的,具有一个公共的优化目标或者是损失函数。可以理解,优化该优化目标的过程就是该N个模型同时训练的过程。
本申请实施例提供的模型更新方法,由于核函数、模型超参数、预测模式和计算模式均为时间序列预测模型的参数信息,因此在第一通信设备获取到模型更新信息之后,第一通信设备可以根据该模型更新信息,更新第一通信设备使用的时间序列预测模型,从而使得时间序列预测模型能够更加准确地执行预测任务,进而使得第一通信设备获得良好的通信性能。
本申请实施例提供的模型更新方法,执行主体可以为模型更新装置。本申请实施例中以模型更新装置执行模型更新方法为例,说明本申请实施例提供的模型更新装置。
如图8所示,本申请实施例提供一种模型更新装置300,该模型更新装置300可以包括获取模块301和更新模块302。获取模块301,可以用于获取模型更新信息;更新模块302,可以用于根据模型更新信息,更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:
核函数的更新信息;
模型超参数的更新信息;
预测模式的更新信息;
计算模式的更新信息。
可选地,获取模块可以包括接收子模块,接收子模块,可以用于从第二通信设备接收模型更新信息。
可选地,模型更新装置还可以包括发送模块。发送模块,可以用于向第二通信设备发送模型更新请求信息,模型更新请求信息用于请求获取第一更新信息;其中,模型更新信息包括以下至少一项:
第一更新信息;
除第一更新信息之外的第二更新信息。
可选地,发送模块,具体可以用于根据第一信息,向第二通信设备发送模型更新请求信息;其中,第一信息包括以下至少一项:
时间序列预测模型的预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
时间序列预测模型的预测结果的统计信息;
环境感知信息。
可选地,发送模块,具体可以用于在满足第一条件的情况下,向第二通信设备发送模型更新请求信息;
其中,第一条件包括以下至少一项:
第一通信设备接收到第二通信设备发送的模型更新指示信息;
时间序列预测模型的预测任务更新;
第一通信设备配置的时间序列预测模型的预测性能需求更新。
可选地,预测性能需求包括预测精度和计算时延中的至少一项。
可选地,模型更新装置还可以包括发送模块。该发送模块,可以用于向第二通信设备发送第一能力信息,第一能力信息用于指示以下至少一项:
第一通信设备支持的核函数;
第一通信设备是否具有预测任务;
第一通信设备支持的模型优化器配置;
第一通信设备支持的模型训练配置;
第一通信设备支持的模型预测配置。
可选地,模型更新装置还可以包括接收模块和发送模块。该接收模块,可以用于从第二通信设备接收第一模型配置;该发送模块,可以用于向第二通信设备发送第一反馈信息,第一反馈信息用于指示第一通信设备是否支持第一模型配置指示的时间序列预测模型。
可选地,模型更新装置还可以包括发送模块和接收模块。该发送模块,可以用于向第二通信设备发送计算能力信息,计算能力信息用于指示第一通信设备的计算能力;该接收模块,可以用于从第二通信设备接收第二模型配置,第二模型配置指示的时间序列预测模型根据计算能力、处理时延、时间序列预测模型的预测性能需求中的至少一项确定。
可选地,模型更新装置还可以包括发送模块。该发送模块,可以用于向第二通信设备发送第二信息或时间序列预测模型的更新推荐信息;其中,模型更新信息是根据第二信息确定的。
可选地,第二信息包括以下至少一项:
时间序列预测模型的预测结果的统计信息;
时间序列预测模型的估计误差;
时间序列预测模型的估计误差的统计信息;
时间序列预测模型的模型预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
噪声的统计信息;
时间序列预测模型的预测性能需求。
可选地,预测模式包括预测任务的时间窗的结构,时间窗的结构包括以下至少一项:
时间序列预测模型的训练窗的时间单位的长度;
时间序列预测模型的预测窗的时间单位的长度;
时间序列预测模型的训练窗的相邻时间单位的间隔;
时间序列预测模型的预测窗的相邻时间单位的间隔。
可选地,第一通信设备存储有与第二通信设备相同的预测模型列表,预测模型列表包括预测模式的标识。
可选地,模型更新信息包括计算模式的更新信息;模型更新装置还可以包括发送模块。该发送模块,可以用于在接收子模块从第二通信设备接收模型更新信息之前,向第二通信设备发送第二能力信息,第二能力信息包括以下至少一项:
第一通信设备的计算能力信息;
第一通信设备的存储能力信息;
第一通信设备的计算单元配置信息。
可选地,第二能力信息还包括以下至少一项:
第一通信设备支持缓存的最大数据量;
第一通信设备支持计算的最大计算量;
第一通信设备支持的最大并行计算线程数。
可选地,计算模式的更新信息指示将时间序列预测模型的计算模式更新为并行计算模式,可选地,模型更新装置还可以包括接收模块。该接收模块,可以用于从第二通信设备接收第三信息,第三信息包括以下至少一项:
数据集的划分方式;
优化目标的选择;
模型优化器的选择;
模型优化器的初始状态;
多个子模型预测结果的融合方式。
可选地,模型更新信息还包括以下至少一项:
待更新预测任务的标识;
待更新模型的标识;
待更新模型更新的时间戳信息。
本申请实施例提供的模型更新装置,由于核函数、模型超参数、预测模式和计算模式均为时间序列预测模型的参数信息,因此在模型更新装置获取到模型更新信息之后,模型更新装置可以根据该模型更新信息,更新第一通信设备使用的时间序列预测模型,从而使得时间序列预测模型能够更加准确地执行预测任务,进而使得第一通信设备获得良好的通信性能。
如图9所示,本申请实施例提供一种模型更新装置400,该模型更新装置400可以包括发送模块401。该发送模块401,可以用于向第一通信设备发送模型更新信息,模型更新信息用于更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行;其中,模型更新信息包括以下至少一项:
核函数的更新信息;
模型超参数的更新信息;
预测模式的更新信息;
计算模式的更新信息。
可选地,发送模块,具体可以用于根据第四信息,向第一通信设备发送模型更新信息;其中,第四信息包括以下至少一项:
时间序列预测模型的预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
时间序列预测模型的预测结果的统计信息;
时间序列预测模型的预测任务的更新信息;
时间序列预测模型的预测性能需求的更新信息;
环境感知信息。
可选地,模型更新信息还可以包括接收模块。该接收模块,可以用于从第一通信设备接收模型更新请求信息,模型更新请求信息用于请求获取第一更新信息;其中,模型更新信息包括以下至少一项:
第一更新信息;
除第一更新信息之外的第二更新信息。
可选地,模型更新信息还可以包括接收模块。该接收模块,可以用于从第一通信设备接收第一能力信息,第一能力信息用于指示以下至少一项:
第一通信设备支持的核函数;
第一通信设备是否具有预测任务;
第一通信设备支持的模型优化器配置;
第一通信设备支持的模型训练配置;
第一通信设备支持的模型预测配置。
可选地,模型更新信息还可以包括接收模块。发送模块,还可以用于向第一通信设备发送第一模型配置;该接收模块,可以用于从第一通信设备接收第一反馈信息,第一反馈信息用于指示第一通信设备是否支持第一模型配置指示的时间序列预测模型。
可选地,模型更新信息还可以包括接收模块。该接收模块,可以用于从第一通信设备接收计算能力信息,计算能力信息用于指示第一通信设备的计算能力;发送模块,还可以用于向第一通信设备发送第二模型配置,第二模型配置指示的时间序列预测模型根据计算能力、处理时延、时间序列预测模型的预测性能需求中的至少一项确定。
可选地,模型更新信息还可以包括接收模块。该接收模块,可以用于从第一通信设备接收第二信息或时间序列预测模型的更新推荐信息;其中,模型更新信息是根据第二信息确定的。
可选地,第二信息可以包括以下至少一项:
时间序列预测模型的预测结果的统计信息;
时间序列预测模型的估计误差;
时间序列预测模型的估计误差的统计信息;
时间序列预测模型的模型预测误差;
时间序列预测模型的模型预测误差的统计信息;
第一通信设备的移动性信息;
噪声的统计信息;
时间序列预测模型的预测性能需求。
可选地,预测模式包括预测任务的时间窗的结构,时间窗的结构包括以下至少一项:
时间序列预测模型的训练窗的时间单位的长度;
时间序列预测模型的预测窗的时间单位的长度;
时间序列预测模型的训练窗的相邻时间单位的间隔;
时间序列预测模型的预测窗的相邻时间单位的间隔。
可选地,第二通信设备存储有与第一通信设备相同的预测模型列表,预测模型列表包括预测模式的标识。
可选地,模型更新信息包括计算模式的更新信息;模型更新信息还可以包括接收模块。该接收模块,可以用于在发送模块向第一通信设备发送模型更新信息之前,从第一通信设备接收第二能力信息,第二能力信息包括以下至少一项:
第一通信设备的计算能力信息;
第一通信设备的存储能力信息;
第一通信设备的计算单元配置信息。
可选地,第二能力信息还包括以下至少一项:
第一通信设备支持缓存的最大数据量;
第一通信设备支持计算的最大计算量;
第一通信设备支持的最大并行计算线程数。
可选地,计算模式的更新信息指示将时间序列预测模型的计算模式更新为并行计算模式,发送模块,还用于向第一通信设备发送第三信息,第三信息包括以下至少一项:
数据集的划分方式;
优化目标的选择;
模型优化器的选择;
模型优化器的初始状态;
多个子模型预测结果的融合方式。
可选地,模型更新信息还包括以下至少一项:
待更新预测任务的标识;
待更新模型的标识;
待更新模型更新的时间戳信息。
本申请实施例中,由于核函数、模型超参数、预测模式和计算模式均为时间序列预测模型的参数信息,因此在模型更新装置可以通过向第一通信设备发送模型更新信息,使得第一通信设备可以根据该模型更新信息,更新第一通信设备使用的时间序列预测模型,从而使得时间序列预测模型能够更加准确地执行预测任务,进而使得第一通信设备获得良好的通信性能。
本申请实施例中,模型更新装置可以是通信设备,例如具有操作系统的通信设备,也可以是通信设备中的部件,例如集成电路或芯片。该通信设备可以是终端,也可以为除终端之外的其他设备,例如网络侧设备。示例性地,终端可以包括但不限于上述所列举的终端11的类型,其他设备可以为核心网设备或接入网设备、服务器、网络附属存储器(Network Attached Storage,NAS)等,本申请实施例不作具体限定。
本申请实施例提供的模型更新装置能够实现上述模型更新方法的实施例实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
可选地,如图10所示,本申请实施例还提供一种通信设备500,包括处理器501和存储器502,存储器502上存储有可在处理器501上运行的程序或指令,例如,该通信设备500为终端时,该程序或指令被处理器501执行时实现上述模型更新方法实施例的各个步骤,且能达到相同的技术效果。该通信设备500为网络侧设备时,该程序或指令被处理器501执行时实现上述模型更新方法实施例的各个步骤,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供一种通信设备,包括处理器和通信接口。在通信设备为第一通信设备的情况下,通信接口用于获取模型更新信息;处理器根据模型更新信息,更新第一通信设备使用的时间序列预测模型;或者,在通信设备为第二通信设备的情况下,通信接口用于向第一通信设备发送模型更新信息,模型更新信息用于更新第一通信设备使用的时间序列预测模型,时间序列预测模型用于预测任务的执行,模型更新信息包括以下至少一项:核函数的更新信息;模型超参数的更新信息;预测模式的更新信息;计算模式的更新信息。该通信设备实施例与上述方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该通信设备实施例中,且能达到相同的技术效果。
本申请实施例还提供了一种通信设备。如图11所示,该通信设备600包括:天线61、射频装置62、基带装置63、处理器64和存储器65。天线61与射频装置62连接。在上行方向上,射频装置62通过天线61接收信息,将接收的信息发送给基带装置63进行处理。在下行方向上,基带装置63对要发送的信息进行处理,并发送给射频装置62,射频装置62对收到的信息进行处理后经过天线61发送出去。
以上实施例中通信设备(包括第一通信设备和第二通信设备)执行的方法可以在基带装置63中实现,该基带装置63包括基带处理器。
基带装置63例如可以包括至少一个基带板,该基带板上设置有多个芯片,如图11所示,其中一个芯片例如为基带处理器,通过总线接口与存储器65连接,以调用存储器65中的程序,执行以上方法实施例中所示的网络侧设备操作。
该通信设备还可以包括网络接口66,该接口例如为通用公共无线接口(common public radio interface,CPRI)。
具体地,本申请实施例的通信设备600还包括:存储在存储器65上并可在处理器64上运行的指令或程序,处理器64调用存储器65中的指令或程序执行图8或图9所示各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
本申请实施例还提供一种可读存储介质,可读存储介质上存储有程序或指令,该程序或指令被处理器执行时实现上述模型更新方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再 赘述。
其中,处理器为上述实施例中的终端中的处理器。可读存储介质,包括计算机可读存储介质,如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片,芯片包括处理器和通信接口,通信接口和处理器耦合,处理器用于运行程序或指令,实现上述模型更新方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为系统级芯片,系统芯片,芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品,计算机程序/程序产品被存储在存储介质中,计算机程序/程序产品被至少一个处理器执行以实现上述模型更新方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供了一种通信系统,包括:终端及网络侧设备,终端可用于执行如上的模型更新方法的步骤,网络侧设备可用于执行如上的模型更新方法的步骤。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去、或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器,空调器,或者网络侧设备等)执行本申请各个实施例的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,均属于本申请的保护之内。

Claims (35)

  1. 一种模型更新方法,包括:
    第一通信设备获取模型更新信息;
    所述第一通信设备根据所述模型更新信息,更新所述第一通信设备使用的时间序列预测模型,所述时间序列预测模型用于预测任务的执行;
    其中,所述模型更新信息包括以下至少一项:
    核函数的更新信息;
    模型超参数的更新信息;
    预测模式的更新信息;
    计算模式的更新信息。
  2. 根据权利要求1所述的方法,其中,所述第一通信设备获取模型更新信息,包括:
    所述第一通信设备从第二通信设备接收所述模型更新信息。
  3. 根据权利要求2所述的方法,其中,所述第一通信设备从第二通信设备接收所述模型更新信息之前,所述方法还包括:
    所述第一通信设备向所述第二通信设备发送模型更新请求信息,所述模型更新请求信息用于请求获取第一更新信息;
    其中,所述模型更新信息包括以下至少一项:
    所述第一更新信息;
    除所述第一更新信息之外的第二更新信息。
  4. 根据权利要求3所述的方法,其中,所述第一通信设备向所述第二通信设备发送模型更新请求信息,包括:
    所述第一通信设备根据第一信息,向所述第二通信设备发送所述模型更新请求信息;
    其中,所述第一信息包括以下至少一项:
    所述时间序列预测模型的预测误差;
    所述时间序列预测模型的预测误差的统计信息;
    所述第一通信设备的移动性信息;
    所述时间序列预测模型的预测结果的统计信息;
    环境感知信息。
  5. 根据权利要求3所述的方法,其中,所述第一通信设备向所述第二通信设备发送模型更新请求信息,包括:
    在满足第一条件的情况下,所述第一通信设备向所述第二通信设备发送所述模型更新请求信息;
    其中,所述第一条件包括以下至少一项:
    所述第一通信设备接收到所述第二通信设备发送的模型更新指示信息;
    所述时间序列预测模型的预测任务更新;
    所述时间序列预测模型的预测性能需求更新。
  6. 根据权利要求5所述的方法,其中,所述预测性能需求包括预测精度和计算时延中的至少一项。
  7. 根据权利要求2所述的方法,其中,所述方法还包括:
    所述第一通信设备向所述第二通信设备发送第一能力信息,所述第一能力信息用于指示以下至少一项:
    所述第一通信设备支持的核函数;
    所述第一通信设备是否具有预测任务;
    所述第一通信设备支持的模型优化器配置;
    所述第一通信设备支持的模型训练配置;
    所述第一通信设备支持的模型预测配置。
  8. 根据权利要求2所述的方法,其中,所述方法还包括:
    所述第一通信设备从所述第二通信设备接收第一模型配置;
    所述第一通信设备向所述第二通信设备发送第一反馈信息,所述第一反馈信息用于指示所述第一通信设备是否支持所述第一模型配置指示的时间序列预测模型。
  9. 根据权利要求2所述的方法,其中,所述方法还包括:
    所述第一通信设备向所述第二通信设备发送计算能力信息,所述计算能力信息用于指示所述第一通信设备的计算能力;
    所述第一通信设备从所述第二通信设备接收第二模型配置,所述第二模型配置指示的时间序列预测模型根据所述计算能力、处理时延、所述时间序列预测模型的预测性能需求中的至少一项确定。
  10. 根据权利要求2所述的方法,其中,所述方法还包括:
    所述第一通信设备向所述第二通信设备发送第二信息或所述时间序列预测模型的更新推荐信息;
    其中,所述模型更新信息是根据所述第二信息确定的。
  11. 根据权利要求10所述的方法,其中,所述第二信息包括以下至少一项:
    所述时间序列预测模型的预测结果的统计信息;
    所述时间序列预测模型的估计误差;
    所述时间序列预测模型的估计误差的统计信息;
    所述时间序列预测模型的模型预测误差;
    所述时间序列预测模型的模型预测误差的统计信息;
    所述第一通信设备的移动性信息;
    噪声的统计信息;
    所述时间序列预测模型的预测性能需求。
  12. 根据权利要求1所述的方法,其中,所述预测模式包括预测任务的时间窗的结构,所述时间窗的结构包括以下至少一项:
    所述时间序列预测模型的训练窗的时间单位的长度;
    所述时间序列预测模型的预测窗的时间单位的长度;
    所述时间序列预测模型的训练窗的相邻时间单位的间隔;
    所述时间序列预测模型的预测窗的相邻时间单位的间隔。
  13. 根据权利要求2所述的方法,其中,所述第一通信设备存储有与所述第二通信设备相同的预测模型列表,所述预测模型列表包括预测模式的标识。
  14. 根据权利要求2所述的方法,其中,所述模型更新信息包括计算模式的更新信息;
    所述第一通信设备从第二通信设备接收所述模型更新信息之前,所述方法还包括:
    所述第一通信设备向所述第二通信设备发送第二能力信息,所述第二能力信息包括以下至少一项:
    所述第一通信设备的计算能力信息;
    所述第一通信设备的存储能力信息;
    所述第一通信设备的计算单元配置信息。
  15. 根据权利要求14所述的方法,其中,所述第二能力信息还包括以下至少一项:
    所述第一通信设备支持缓存的最大数据量;
    所述第一通信设备支持计算的最大计算量;
    所述第一通信设备支持的最大并行计算线程数。
  16. 根据权利要求1所述的方法,其中,所述计算模式的更新信息指示将所述时间序列预测模型的计算模式更新为并行计算模式,所述方法还包括:
    所述第一通信设备从第二通信设备接收第三信息,所述第三信息包括以下至少一项:
    数据集的划分方式;
    优化目标的选择;
    模型优化器的选择;
    模型优化器的初始状态;
    多个子模型预测结果的融合方式。
  17. 根据权利要求1所述的方法,其中,所述模型更新信息还包括以下至少一项:
    待更新预测任务的标识;
    待更新模型的标识;
    待更新模型更新的时间戳信息。
  18. 一种模型更新方法,包括:
    第二通信设备向第一通信设备发送模型更新信息,所述模型更新信息用于更新所述第一通信设备使用的时间序列预测模型,所述时间序列预测模型用于预测任务的执行;
    其中,所述模型更新信息包括以下至少一项:
    核函数的更新信息;
    模型超参数的更新信息;
    预测模式的更新信息;
    计算模式的更新信息。
  19. 根据权利要求18所述的方法,其中,所述第二通信设备向第一通信设备发送模型更新信息,包括:
    所述第二通信设备根据第四信息,向所述第一通信设备发送所述模型更新信息;
    其中,所述第四信息包括以下至少一项:
    所述时间序列预测模型的预测误差;
    所述时间序列预测模型的模型预测误差的统计信息;
    所述第一通信设备的移动性信息;
    所述时间序列预测模型的预测结果的统计信息;
    所述时间序列预测模型的预测任务的更新信息;
    所述时间序列预测模型的预测性能需求的更新信息;
    环境感知信息。
  20. 根据权利要求18所述的方法,其中,所述第二通信设备向第二通信设备发送模型更新信息之前,所述方法还包括:
    所述第二通信设备从所述第一通信设备接收模型更新请求信息,所述模型更新请求信息用于请求获取第一更新信息;
    其中,所述模型更新信息包括以下至少一项:
    所述第一更新信息;
    除所述第一更新信息之外的第二更新信息。
  21. 根据权利要求18所述的方法,其中,所述方法还包括:
    所述第二通信设备从所述第一通信设备接收第一能力信息,所述第一能力信息用于指示以下至少一项:
    所述第一通信设备支持的核函数;
    所述第一通信设备是否具有预测任务;
    所述第一通信设备支持的模型优化器配置;
    所述第一通信设备支持的模型训练配置;
    所述第一通信设备支持的模型预测配置。
  22. 根据权利要求18所述的方法,其中,所述方法还包括:
    所述第二通信设备向所述第一通信设备发送第一模型配置;
    所述第二通信设备从所述第一通信设备接收第一反馈信息,所述第一反馈信息用于指示所述第一通信设备是否支持所述第一模型配置指示的时间序列预测模型。
  23. 根据权利要求18所述的方法,其中,所述方法还包括:
    所述第二通信设备从所述第一通信设备接收计算能力信息,所述计算能力信息用于指示所述第一通 信设备的计算能力;
    所述第二通信设备向所述第一通信设备发送第二模型配置,所述第二模型配置指示的时间序列预测模型根据计算能力、处理时延、所述时间序列预测模型的预测性能需求中的至少一项确定。
  24. 根据权利要求18所述的方法,其中于,所述方法还包括:
    所述第二通信设备从所述第一通信设备接收第二信息或所述时间序列预测模型的更新推荐信息;
    其中,所述模型更新信息是根据所述第二信息确定的。
  25. 根据权利要求24所述的方法,其中,所述第二信息包括以下至少一项:
    所述时间序列预测模型的预测结果的统计信息;
    所述时间序列预测模型的估计误差;
    所述时间序列预测模型的估计误差的统计信息;
    所述时间序列预测模型的模型预测误差;
    所述时间序列预测模型的模型预测误差的统计信息;
    所述第一通信设备的移动性信息;
    噪声的统计信息;
    所述时间序列预测模型的预测性能需求。
  26. 根据权利要求18所述的方法,其中,所述预测模式包括预测任务的时间窗的结构,所述时间窗的结构包括以下至少一项:
    所述时间序列预测模型的训练窗的时间单位的长度;
    所述时间序列预测模型的预测窗的时间单位的长度;
    所述时间序列预测模型的训练窗的相邻时间单位的间隔;
    所述时间序列预测模型的预测窗的相邻时间单位的间隔。
  27. 根据权利要求18所述的方法,其中,所述第二通信设备存储有与所述第一通信设备相同的预测模型列表,所述预测模型列表包括预测模式的标识。
  28. 根据权利要求18所述的方法,其中,所述模型更新信息包括计算模式的更新信息;
    所述第二通信设备向第一通信设备发送模型更新信息之前,所述方法还包括:
    所述第二通信设备从所述第一通信设备接收第二能力信息,所述第二能力信息包括以下至少一项:
    所述第一通信设备的计算能力信息;
    所述第一通信设备的存储能力信息;
    所述第一通信设备的计算单元配置信息。
  29. 根据权利要求28所述的方法,其中,所述第二能力信息还包括以下至少一项:
    所述第一通信设备支持缓存的最大数据量;
    所述第一通信设备支持计算的最大计算量;
    所述第一通信设备支持的最大并行计算线程数。
  30. 根据权利要求18所述的方法,其中,所述计算模式的更新信息指示将所述时间序列预测模型的计算模式更新为并行计算模式,所述方法还包括:
    所述第二通信设备向所述第一通信设备发送第三信息,所述第三信息包括以下至少一项:
    数据集的划分方式;
    优化目标的选择;
    模型优化器的选择;
    模型优化器的初始状态;
    多个子模型预测结果的融合方式。
  31. 根据权利要求18所述的方法,其中,所述模型更新信息还包括以下至少一项:
    待更新预测任务的标识;
    待更新模型的标识;
    待更新模型更新的时间戳信息。
  32. 一种模型更新装置,所述模型更新装置包括获取模块和更新模块;
    所述获取模块,用于获取模型更新信息;
    所述更新模块,用于根据所述模型更新信息,更新第一通信设备使用的时间序列预测模型,所述时间序列预测模型用于预测任务的执行;
    其中,所述模型更新信息包括以下至少一项:
    核函数的更新信息;
    模型超参数的更新信息;
    预测模式的更新信息;
    计算模式的更新信息。
  33. 一种模型更新装置,所述模型更新装置包括发送模块;
    所述发送模块,用于向第一通信设备发送模型更新信息,所述模型更新信息用于更新第一通信设备使用的时间序列预测模型,所述时间序列预测模型用于预测任务的执行;
    其中,所述模型更新信息包括以下至少一项:
    核函数的更新信息;
    模型超参数的更新信息;
    预测模式的更新信息;
    计算模式的更新信息。
  34. 一种通信设备,包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如权利要求1至31中任一项所述的模型更新方法的步骤。
  35. 一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如权利要求1-31中任一项所述的模型更新方法的步骤。
PCT/CN2022/143594 2021-12-30 2022-12-29 模型更新方法及通信设备 WO2023125855A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111652530.0A CN116418432A (zh) 2021-12-30 2021-12-30 模型更新方法及通信设备
CN202111652530.0 2021-12-30

Publications (1)

Publication Number Publication Date
WO2023125855A1 true WO2023125855A1 (zh) 2023-07-06

Family

ID=86998074

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/143594 WO2023125855A1 (zh) 2021-12-30 2022-12-29 模型更新方法及通信设备

Country Status (2)

Country Link
CN (1) CN116418432A (zh)
WO (1) WO2023125855A1 (zh)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3591586A1 (en) * 2018-07-06 2020-01-08 Capital One Services, LLC Data model generation using generative adversarial networks and fully automated machine learning system which generates and optimizes solutions given a dataset and a desired outcome
CN110909884A (zh) * 2019-11-04 2020-03-24 深圳力维智联技术有限公司 一种用于更新参数模型的方法、装置、产品及介质
CN111797870A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 算法模型的优化方法、装置、存储介质及电子设备
US20210194733A1 (en) * 2018-09-10 2021-06-24 Huawei Technologies Co., Ltd. Channel Prediction Method and Related Device
CN113052388A (zh) * 2021-04-01 2021-06-29 北京工商大学 一种时间序列预测方法及装置
CN113407820A (zh) * 2021-05-29 2021-09-17 华为技术有限公司 模型训练方法及相关系统、存储介质
CN113570062A (zh) * 2020-04-28 2021-10-29 大唐移动通信设备有限公司 机器学习模型参数传递方法及装置
WO2021238397A1 (zh) * 2020-05-24 2021-12-02 中兴通讯股份有限公司 网络优化方法、服务器、客户端设备、网络侧设备、网络设备、系统和介质

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3591586A1 (en) * 2018-07-06 2020-01-08 Capital One Services, LLC Data model generation using generative adversarial networks and fully automated machine learning system which generates and optimizes solutions given a dataset and a desired outcome
US20210194733A1 (en) * 2018-09-10 2021-06-24 Huawei Technologies Co., Ltd. Channel Prediction Method and Related Device
CN111797870A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 算法模型的优化方法、装置、存储介质及电子设备
CN110909884A (zh) * 2019-11-04 2020-03-24 深圳力维智联技术有限公司 一种用于更新参数模型的方法、装置、产品及介质
CN113570062A (zh) * 2020-04-28 2021-10-29 大唐移动通信设备有限公司 机器学习模型参数传递方法及装置
WO2021238397A1 (zh) * 2020-05-24 2021-12-02 中兴通讯股份有限公司 网络优化方法、服务器、客户端设备、网络侧设备、网络设备、系统和介质
CN113052388A (zh) * 2021-04-01 2021-06-29 北京工商大学 一种时间序列预测方法及装置
CN113407820A (zh) * 2021-05-29 2021-09-17 华为技术有限公司 模型训练方法及相关系统、存储介质

Also Published As

Publication number Publication date
CN116418432A (zh) 2023-07-11

Similar Documents

Publication Publication Date Title
EP4156631A1 (en) Reinforcement learning (rl) and graph neural network (gnn)-based resource management for wireless access networks
US20240097993A1 (en) Machine Learning Model Distribution
CN113038501A (zh) 网络链路优化方法、装置及电子设备
Xue et al. Deep learning based channel prediction for massive MIMO systems in high-speed railway scenarios
US11128391B1 (en) System and method for predicting wireless channel path loss
WO2023125855A1 (zh) 模型更新方法及通信设备
WO2023125880A1 (zh) 模型的构建方法、装置及通信设备
WO2023143572A1 (zh) 基于人工智能ai模型的定位方法及通信设备
Jiewu et al. User traffic collection and prediction in cellular networks: Architecture, platform and case study
Yin et al. Channel prediction with liquid time-constant networks: an online and adaptive approach
WO2023103911A1 (zh) 测量方法、装置、设备及存储介质
WO2024008111A1 (zh) 数据采集方法及装置
CN116668309A (zh) Ai模型的在线学习方法、装置、通信设备及可读存储介质
US12022496B2 (en) Systems and methods for dynamic selection of a physical uplink control channel format
WO2023098662A1 (zh) 定位方法及通信设备
WO2024093997A1 (zh) 确定模型适用性的方法、装置及通信设备
WO2023098661A1 (zh) 定位方法及通信设备
WO2024083004A1 (zh) Ai模型配置方法、终端及网络侧设备
WO2024067281A1 (zh) Ai模型的处理方法、装置及通信设备
CN118042399A (zh) 模型监督方法、终端及网络侧设备
WO2024125391A2 (zh) Ai模型监测方法、ai模型性能的测量方法、装置及设备
WO2023179540A1 (zh) 信道预测方法、装置及无线通信设备
CN117641433A (zh) Csi预测处理方法、装置、通信设备及可读存储介质
Yuliana Comparative Analysis of Machine Learning Algorithms for 5G Coverage Prediction: Identification of Dominant Feature Parameters and Prediction Accuracy
CN118175048A (zh) 模型监督触发方法、装置、ue、网络侧设备、可读存储介质及通信系统

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: 22915104

Country of ref document: EP

Kind code of ref document: A1