Model training method, prediction method, device, apparatus and medium

Abstract

The model training method, the prediction method, the device, the equipment and the medium provided by the embodiments of the present application relate to the technical field of communication. The method comprises the following steps: receiving a plurality of generator parameters sent by a plurality of terminal devices, the plurality of generator parameters correspond to the plurality of terminal devices one by one; generating a training data set based on each generator parameter in the plurality of generator parameters by using a generator; and training a basic prediction model based on a plurality of training data sets to obtain a target prediction model, the target prediction model being used for predicting the communication characteristics of the terminal device based on the communication measurement data of the terminal device. The present application can reduce the communication overhead when the network device trains the model.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a model training method, prediction method, device, equipment and medium. Background Art

[0002] In communication systems, network equipment deploys AI / ML (Artificial Intelligence / Machine Learning), which makes predictions based on communication measurement data uploaded by terminal devices.

[0003] Currently, network devices train AI / ML by receiving large amounts of training data sent by terminal devices. This method of training AI / ML results in a large amount of communication overhead. Summary of the Invention

[0004] The embodiments of the present application provide a model training method, prediction method, apparatus, device, and medium, which are applied to the field of communication technology and can reduce communication overhead when training a model.

[0005] In a first aspect, an embodiment of the present application proposes a model training method, which is applied to a network device. The method includes:

[0006] receiving a plurality of generator parameters sent by a plurality of terminal devices, wherein the plurality of generator parameters correspond one to one with the plurality of terminal devices;

[0007] Using the generator, generating a training data set based on each of a plurality of generator parameters;

[0008] A basic prediction model is trained based on multiple training data sets to obtain a target prediction model, which is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device.

[0009] Optionally, also include:

[0010] Receive calibration data sent by multiple terminal devices;

[0011] Calibrate the target prediction model based on the calibration data.

[0012] Optionally, the calibration data includes first input data and first true value data.

[0013] Optionally, calibrate the target prediction model based on the calibration data, including:

[0014] Inputting the first input data into the target prediction model for processing to obtain first prediction data;

[0015] Adjust model parameters of the target prediction model according to the loss values ​​of the first prediction data and the first true value data.

[0016] Optionally, the base prediction model comprises a base prediction unit, and the target prediction model comprises a target prediction unit, the target prediction unit being obtained after training of the base prediction unit.

[0017] Optionally, the target prediction model further comprises at least one of a plurality of perturbation calibration units, a plurality of weight calibration units, or a plurality of offset calibration units, the plurality of perturbation calibration units corresponding to the plurality of terminal devices one by one, the plurality of weight calibration units corresponding to the plurality of terminal devices one by one, and the plurality of offset calibration units corresponding to the plurality of terminal devices one by one.

[0018] Optionally, the target prediction model further comprises a plurality of perturbation calibration units, and the processing of the first input data by the target prediction model to obtain the first prediction data comprises:

[0019] For each terminal device in the plurality of terminal devices, the first input data corresponding to the terminal device is processed according to the perturbation calibration unit corresponding to the terminal device to obtain perturbation data;

[0020] The perturbation data is input into the target prediction unit for processing to obtain the first prediction data.

[0021] Optionally, the model parameters of the target prediction model are adjusted according to the loss value of the first prediction data and the first true value data, and the adjusting comprises:

[0022] The parameters of the perturbation calibration unit corresponding to the terminal device are adjusted according to the loss value of the first prediction data and the first true value data.

[0023] Optionally, the target prediction model further comprises a plurality of weight calibration units and / or a plurality of offset calibration units, and the processing of the first input data by the target prediction model to obtain the first prediction data comprises:

[0024] For each terminal device in the plurality of terminal devices, the first input data corresponding to the terminal device is input into the target prediction unit for processing to obtain second prediction data;

[0025] The second prediction data is processed according to the weight calibration unit and / or the offset calibration unit corresponding to the terminal device to obtain the first prediction data.

[0026] Optionally, the model parameters of the target prediction model are adjusted according to the loss value of the first prediction data and the first true value data, and the adjusting comprises:

[0027] The parameters of the weight calibration unit and / or the parameters of the offset calibration unit corresponding to the terminal device are adjusted according to the loss value of the first prediction data and the first true value data.

[0028] Optionally, the target prediction model further includes: a plurality of disturbance calibration units, a plurality of weight calibration units, and a plurality of offset calibration units. The first input data is input into the target prediction model for processing to obtain the first prediction data, including:

[0029] For each terminal device among the plurality of terminal devices, performing disturbance processing on first input data corresponding to the terminal device according to a disturbance calibration unit corresponding to the terminal device to obtain disturbance data;

[0030] Inputting the disturbance data into the target prediction unit for processing to obtain third prediction data;

[0031] According to the weight calibration unit and the offset calibration unit corresponding to the terminal device, the third prediction data is adjusted to obtain the first prediction data.

[0032] Optionally, adjusting model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data includes:

[0033] According to the loss values ​​of the first predicted data and the first true value data, the parameters of the disturbance calibration unit, the parameters of the weight calibration unit and the parameters of the offset calibration unit corresponding to the terminal device are adjusted.

[0034] Optionally, the training data set includes: multiple training data, and the training data includes: second input data and second true value data.

[0035] Optionally, the second input data represents channel state information CSI, and the second true value data represents channel characteristics based on CSI.

[0036] Optionally, the second input data represents beam measurement information, and the second true value data represents beam characteristics predicted based on the beam measurement information.

[0037] In a second aspect, an embodiment of the present application proposes a prediction method, which is applied to a network device. The method includes:

[0038] receiving communication measurement data sent by a terminal device;

[0039] The communication measurement data is input into the target prediction model for processing to obtain communication features. The target prediction model is trained according to the model training method of any one of the first aspects.

[0040] In a third aspect, an embodiment of the present application proposes a model training method, which is applied to a terminal device. The method includes:

[0041] Train the generator and obtain the generator parameters of the trained generator;

[0042] Sends generator parameters to the network device.

[0043] Optionally, train the generator to obtain the generator parameters of the trained generator, including:

[0044] Identify random variables;

[0045] Input random variables into the generator to generate simulated data sets;

[0046] Get a real dataset;

[0047] A judger is used to determine the probability values ​​of the simulated data in the simulated data set and the real data in the real data set belonging to the real data;

[0048] Adjusting generator parameters of the generator and decision parameters of the decision maker according to the probability value;

[0049] The above steps are executed repeatedly to obtain the trained generator parameters.

[0050] Optionally, also include:

[0051] Get calibration data;

[0052] Send calibration data to network devices.

[0053] In a fourth aspect, an embodiment of the present application provides a prediction method, which is applied to a terminal device. The method includes:

[0054] Obtain communication measurement data;

[0055] The communication measurement data is sent to the network device, and the network device is used to predict the communication characteristics of the terminal device based on the communication measurement data.

[0056] In a fifth aspect, an embodiment of the present application provides a model training device, which is applied to a network device, and the device includes:

[0057] A receiving module, configured to receive a plurality of generator parameters sent by a plurality of terminal devices, wherein the plurality of generator parameters correspond one to one with the plurality of terminal devices;

[0058] A generation module, configured to employ the generator to generate a training data set based on each generator parameter of a plurality of generator parameters;

[0059] The training module is used to train a basic prediction model based on multiple training data sets to obtain a target prediction model, which is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device.

[0060] In a sixth aspect, an embodiment of the present application provides a prediction device, which is applied to a network device, and the device includes:

[0061] A receiving module, configured to receive communication measurement data sent by a terminal device;

[0062] The processing module is used to input the communication measurement data into the target prediction model for processing to obtain communication characteristics. The target prediction model is trained according to the model training method of any one of the first aspects.

[0063] In a seventh aspect, an embodiment of the present application provides a model training device, which is applied to a terminal device, and the device includes:

[0064] The training module is used to train the generator and obtain the generator parameters of the trained generator;

[0065] The sending module is used to send the generator parameters to the network device.

[0066] In an eighth aspect, an embodiment of the present application provides a prediction device, which is applied to a terminal device, and the device includes:

[0067] An acquisition module, used for acquiring communication measurement data;

[0068] The sending module is used to send communication measurement data to the network device, and the network device is used to predict the communication characteristics of the terminal device based on the communication measurement data.

[0069] In the ninth aspect, an embodiment of the present application provides a communication device, which includes: a processor, a transceiver, and a memory; the processor is communicatively connected to the transceiver and the memory respectively; the memory stores computer-executable instructions; the transceiver communicates and interacts with an external device; the processor executes the computer-executable instructions stored in the memory to implement the method described in any possible implementation method of the first to fourth aspects.

[0070] In the tenth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program or instruction is stored. When the computer program or instruction is run on a computer, the computer executes the method described in any possible implementation of the first to fourth aspects.

[0071] In the eleventh aspect, an embodiment of the present application provides a computer program product including a computer program, which, when the computer program is run on a computer, enables the computer to execute the method described in any possible implementation of the first to fourth aspects.

[0072] In a twelfth aspect, the present application provides a chip or chip system, comprising at least one processor and a communication interface, wherein the communication interface and the at least one processor are interconnected by a line, and the at least one processor is configured to run a computer program or instruction to perform the method described in any possible implementation of aspects 1 to 4. The communication interface in the chip may be an input / output interface, a pin, or a circuit, etc.

[0073] In one possible implementation, the chip or chip system described above in this application further includes at least one memory, wherein instructions are stored in the at least one memory. The memory may be a storage unit within the chip, such as a register or cache, or a storage unit of the chip (such as a read-only memory or random access memory).

[0074] It should be understood that the fifth to thirteenth aspects of the present application correspond to the technical solutions of the first to fourth aspects of the present application, and the beneficial effects achieved by each aspect and the corresponding feasible implementation methods are similar and will not be repeated.

[0075] The model training method, prediction method, apparatus, device and medium provided in the present application receive multiple generator parameters sent by multiple terminal devices, and the multiple generator parameters correspond one-to-one to the multiple terminal devices; a generator is used to generate a training data set based on each generator parameter in the multiple generator parameters; a basic prediction model is trained according to the multiple training data sets to obtain a target prediction model, and the target prediction model is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device, thereby reducing the communication overhead when the network device trains the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A flowchart of the steps of a model training method provided by the prior art;

[0077] Figure 2 A flowchart of a model training method provided in an embodiment of the present application;

[0078] Figure 3 A schematic diagram of determining a training data set provided in an embodiment of the present application;

[0079] Figure 4 A schematic diagram of a model training process provided in an embodiment of the present application;

[0080] Figure 5 A flowchart of another model training method provided in an embodiment of the present application;

[0081] Figure 6 A schematic diagram of a flow chart of a model calibration method provided in an embodiment of the present application;

[0082] Figure 7 A flowchart of another model training method provided in an embodiment of the present application;

[0083] Figure 8 A schematic diagram of a training generator provided in an embodiment of the present application;

[0084] Figure 9 A flowchart of a prediction method provided in an embodiment of the present application;

[0085] Figure 10 A schematic diagram of the structure of a model training device provided in an embodiment of the present application;

[0086] Figure 11 A schematic diagram of the structure of a prediction device provided in an embodiment of the present application;

[0087] Figure 12 A schematic diagram of the structure of another model training device provided in an embodiment of the present application;

[0088] Figure 13 A schematic diagram of the structure of another prediction device provided in an embodiment of the present application;

[0089] Figure 14 A schematic structural diagram of a communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0090] To facilitate a clear description of the technical solutions of the embodiments of this application, the words "exemplary" or "for example" are used in the embodiments of this application to indicate examples, illustrations, or explanations. Any embodiment or design described in this application as "exemplary" or "for example" should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0091] In the embodiments of this application, terms such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. For example, the terms "first chip" and "second chip" are used solely to distinguish between different chips and do not define their order. Those skilled in the art will understand that terms such as "first" and "second" do not define the quantity or execution order, and do not necessarily define differences.

[0092] It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0093] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, a--c, bc, or abc, where a, b, c can be single or plural.

[0094] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or plural.

[0095] It should be noted that the "at..." in the embodiments of the present application can be the instant when a certain situation occurs, or it can be a period of time after the situation occurs, and the embodiments of the present application do not specifically limit this. In addition, the display interface provided in the embodiments of the present application is only an example, and the display interface can also include more or less content.

[0096] The technical solutions of the embodiments of the present application can be applied to various communication systems, such as: long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD), universal mobile telecommunication system (UMTS), world-wide interoperability for microwave access (WiMAX) communication system, fifth generation (5G) system or new radio (NR), new systems that may appear in the future, such as sixth generation (6G) system, etc.

[0097] To facilitate understanding of the embodiments of the present application, a communication system applicable to the embodiments of the present application is first described in detail.

[0098] The network device and the terminal device in this application can communicate via a wireless link. When the network device acts as a communication transmitter, the terminal device can act as a communication receiver; when the network device acts as a communication receiver, the terminal device can act as a communication transmitter. The embodiment of this application does not limit the number of network devices and terminal devices included in the communication system. In addition, it should be understood that Figure 1 This is just a schematic diagram. The communication system may also include other network devices, which are not limited in this application. Figure 1 Not drawn in the middle.

[0099] The terminal device in the embodiments of the present application may also be referred to as: user equipment (UE), mobile station (MS), mobile terminal (MT), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication equipment, user agent or user device, etc.

[0100] The terminal device may be a device that provides voice / data connectivity to users, such as a handheld device or vehicle-mounted device with wireless connection function. Currently, some examples of terminal devices include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, in-vehicle devices, wearable devices, and terminal devices in 5G networks or future evolved public land mobile communication networks. The present application does not limit the terminal equipment in the network (PLMN).

[0101] By way of example and not limitation, in this application, the terminal device can be a terminal device in an internet of things (IoT) system. The internet of things is an important part of future information technology development, and its main technical feature is to connect objects through communication technology and network, so as to realize the intelligent network of man-machine interconnection and object-object interconnection. Illustratively, the terminal device in the embodiments of the present application can be a wearable device. The wearable device can also be called a wearable smart device, which is a general term for wearable devices that are designed and developed by applying wearable technology to daily wear, such as glasses, gloves, watches, clothing, and shoes. The wearable device is a portable device that can be directly worn on the body or integrated into the user's clothes or accessories. The wearable device is not only a hardware device, but also can realize powerful functions through software support and data interaction, cloud interaction. The general wearable smart device includes a full function, large size, and can realize complete or partial functions without relying on a smart phone, such as a smart watch or smart glasses, and only focuses on a certain application function, and needs to cooperate with other devices such as a smart phone, such as various smart wristbands, smart jewelry, and other devices for monitoring vital signs.

[0102] By way of example and not limitation, in the embodiments of the present application, the terminal device can also be a terminal device in machine type communication (MTC). In addition, the terminal device can also be a vehicle-mounted module, a vehicle-mounted module, a vehicle-mounted component, a vehicle-mounted chip, or a vehicle-mounted unit, etc. built-in as one or more components or units in a vehicle. The vehicle can implement the method provided in the present application by built-in vehicle-mounted module, vehicle-mounted module, vehicle-mounted component, vehicle-mounted chip or vehicle-mounted unit, etc. Therefore, the embodiments of the present application can also be applied to the Internet of Vehicles, such as vehicle to everything (V2X), long term evolution-vehicle (LTE-V), vehicle-to-vehicle (V2V) technology, etc.

[0103] The network device involved in this application can be a device that communicates with a terminal device. The network device can also be called an access network device or a wireless access network device. It can be a transmission reception point (TRP), an evolved NodeB (eNB or eNodeB) in an LTE system, a home base station (for example, home evolved NodeB, or home Node B, HNB), a base band unit (BBU), or a wireless controller in a cloud radio access network (CRAN) scenario, or the network device can be a relay station, an access point, a vehicle-mounted device, a wearable device, a network device in a 5G network, or a network device in a future evolved PLMN network, etc. It can also be an access point (AP) in a WLAN, or a gNB in ​​an NR system. The above-mentioned network devices can also be urban base stations, micro base stations, micro-micro base stations, femto-micro base stations, etc., and this application does not limit this.

[0104] The network in which the network device resides has strong computing capabilities. This computing capability can be provided by computing nodes included in the network or possessed by the network device itself. When this computing capability can be provided by computing nodes included in the network, the network device can connect to one or more computing nodes in the network and distribute task data received from terminal devices to the computing nodes so that the computing nodes process the task data. Examples of such computing nodes include edge computing servers (MECs), distributed cloud nodes, quantum computing nodes, and computing hosts. Within a computing node, one or more computing units can be included to enable concurrent processing of task data. Examples of such computing units include central processing units (CPUs) and graphics processing units (GPUs).

[0105] In one network structure, the network device may include a centralized unit (CU) node, a distributed unit (DU) node, a RAN device including a CU node and a DU node, or a RAN device including a control plane CU node (CU-CP node), a user plane CU node (CU-UP node), and a DU node.

[0106] Network equipment provides services for cells, and terminal devices communicate with the cells through transmission resources (for example, frequency domain resources, or spectrum resources) allocated by the network equipment. The cell can belong to a macro base station (for example, a macro eNB or macro gNB) or a base station corresponding to a small cell. Small cells here can include: metro cells, micro cells, pico cells, femto cells, etc. These small cells have the characteristics of small coverage and low transmission power, and are suitable for providing high-speed data transmission services.

[0107] Alternatively, the aforementioned device and computing node that communicate with the terminal device can be regarded as a whole, as the network device involved in this application, used to receive the generator parameters sent by the terminal device to train the target prediction model.

[0108] Currently, the training process of the network device side model is as follows: Figure 1 As shown, the specific steps include:

[0109] S101, the terminal device sends a model data capability collection signaling to the network device.

[0110] S102: The network device sends model data collection configuration information to the terminal device.

[0111] S103: The network device sends a reference signal to the terminal device.

[0112] S104: The terminal device sends training data to the network device.

[0113] S105: The network device performs model training based on the training data.

[0114] The inventors discovered that based on Figure 1 The model training method shown requires the terminal device to send a large amount of training data to the network device, resulting in high communication overhead. In addition, using the training data of a single terminal device to train the model will result in poor generalization of the model.

[0115] Based on this, the present application provides a model training method in which a terminal device only needs to send generator parameters to a network device, without having to send a large amount of training data, thereby reducing communication overhead. Furthermore, the network device generates a training dataset based on multiple generator parameters sent by multiple terminal devices to train a target prediction model, thereby achieving higher generalization of the target prediction model.

[0116] The model training method of this application is described in detail below with reference to the accompanying drawings. The execution subject of the embodiment shown in this application is a network device, and the specific form and quantity of each device shown are only examples and should not constitute any limitation on the implementation of the method provided in this application.

[0117] The terminal device in the embodiment of the present application can be the terminal device itself, or it can be a chip, chip system or processor that supports the terminal device to implement the model training method, or it can be a logic module or software that can implement all or part of the terminal device functions. The network device in the embodiment of the present application can be the network device itself, or it can be a chip, chip system or processor that supports the network device to implement the model training method, or it can be a logic module or software that can implement all or part of the network device functions. This application does not make specific restrictions on this.

[0118] Figure 2 A flow chart of a model training method provided in an embodiment of the present application. Figure 2 As shown, the method may include:

[0119] S201, the terminal device sends generator parameters to the network device.

[0120] Among them, multiple generator parameters correspond one-to-one to multiple terminal devices.

[0121] Reference Figure 2 , the number of multiple terminal devices is K, K is an integer greater than 1, the generator parameters can be expressed as , where k ranges from 1 to K, Represents the generator parameters sent by the k-th terminal device.

[0122] In the embodiment of the present application, the generator parameters It is the parameter obtained by the terminal device through training the generator.

[0123] Among them, the generator is trained using the Generative Adversarial Network (GAN) algorithm.

[0124] S202, using a generator to generate a training data set based on each generator parameter in a plurality of generator parameters.

[0125] In the embodiment of the present application, for each generator parameter Can generate a training data set , then K generator parameters can generate K training data sets accordingly.

[0126] It is understandable that the generator is pre-deployed on the network device side, and based on the generator parameters The generator uses express.

[0127] Specifically, a random variable Z1 is generated and input into the generator to obtain a training data set. Where Z1 is a Gaussian white noise random variable, which can be expressed by the following expression (1):

[0128]

[0129] In expression (1), is a component of the random variable Z1, which is an independent and identically distributed Gaussian random variable Satisfies normal distribution ,in The 0 in represents the center of the normal distribution. in Represents the variance of the normal distribution.

[0130] In some embodiments, the training data set includes: a plurality of training data, and the training data includes: second input data and second true value data.

[0131] It can be understood that the training data set corresponding to the terminal device k can be expressed by the following expression (2):

[0132]

[0133] Among them, in expression (2), Represents the training dataset The second input data in i, Represents the training dataset The i-th second true value data in , i ranges from 1 to M, and M is a positive integer.

[0134] It can be understood that the training data sets corresponding to multiple terminal devices k can be used express. Used to train the basic prediction model and obtain the target prediction model.

[0135] For example, referring to Figure 3 , input the random variable Z1 into each generator Get the training data set , each training data set Combine to get the training data set .

[0136] Furthermore, The data dimension is , The data dimension is .

[0137] In some embodiments, , ,in, represents the number of receiving antenna ports of terminal device k, Represents the number of transmitting antenna ports of terminal device k.

[0138] In some embodiments, wherein Can be a vector representing communication measurement data, for example, Used to represent the vector converted from the channel state information of terminal device k. It is also a vector used to represent the corresponding communication characteristics, that is, The corresponding truth value, for example, Represents one of the channels predicted based on the channel characteristics.

[0139] In some embodiments, the second input data represents CSI (Channel State Information), and the second true value data represents channel characteristics predicted based on the channel state information.

[0140] CSI describes channel characteristics and includes channel parameters such as amplitude attenuation, phase offset, and delay spread. Channel features are features predicted based on CSI, such as channel gain, channel phase, and optimal channel.

[0141] It can be understood that the present application can train a target prediction model for CSI prediction.

[0142] In some embodiments, the second input data represents beam measurement information, and the second true value data represents beam characteristics predicted based on the beam measurement information.

[0143] The beam measurement information includes at least one of the following: reference signal received power, reference signal received quality, signal to interference plus noise ratio, beam index, beam angle information, beam delay information, etc. Beam characteristics can be beam direction, beam width, beam gain, etc. predicted based on the beam measurement information.

[0144] It is understood that the present application can be used in beam management scenarios, and specific application scenarios include beam measurement, beam reporting, beam prediction, beam deployment, etc. Based on the above-mentioned present application, a target prediction model for beam prediction can be trained.

[0145] In summary, the second input data of the present application may also represent other measurable communication measurement data of the terminal device, and the second true value data may also represent communication features that can be predicted based on the communication measurement data. This application does not impose any limitation on this.

[0146] S203: Train a basic prediction model based on multiple training data sets to obtain a target prediction model.

[0147] The target prediction model is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device.

[0148] In the embodiment of the present application, multiple training data sets can be represented as .

[0149] Further, using Train the basic prediction model, where the basic prediction model can be expressed as ,in, Represents the trained model parameters.

[0150] In the embodiment of the present application, after the basic prediction model is trained, the target prediction model is obtained. The target prediction model has the same architecture as the basic prediction model, but different model parameters. The training process is to train the model parameters. , for example, the model parameters of the underlying prediction model , the model parameters of the target prediction model , N is a positive integer, it can be understood that the model parameters of the basic prediction model After N times of training and adjustment, we get .

[0151] The basic prediction model may be an LSTM (Long Short-Term Memory) network. In addition, the basic prediction model may also be other neural networks, which is not limited.

[0152] In some embodiments, referring to Figure 4 The basic prediction model includes a basic prediction unit, and the target prediction model includes a target prediction unit. The target prediction unit is obtained after the basic prediction unit is trained.

[0153] Further, based on Figure 4 The training process is: the second input data Input the basic prediction unit for processing to obtain the predicted value, and then use the preset loss function to calculate the predicted value and the second true value data The loss value is used to adjust the parameters of the basic prediction unit. After training the basic prediction unit with each training data in , the target prediction unit is obtained.

[0154] The basic prediction unit may be an LSTM or other neural network, without limitation.

[0155] In some embodiments, further, referring to Figure 5The target prediction model also includes: at least one of multiple disturbance calibration units, multiple weight calibration units or multiple offset calibration units, the multiple disturbance calibration units correspond one-to-one to the multiple terminal devices, the multiple weight calibration units correspond one-to-one to the multiple terminal devices, and the multiple offset calibration units correspond one-to-one to the multiple terminal devices.

[0156] Among them, reference Figure 5 , the terminal device k has a one-to-one correspondence with the disturbance calibration unit a1, the weight calibration unit b1 and the offset calibration unit c1. Figure 5 The training process is: the second input data The disturbance calibration unit ak is input to obtain disturbance data, and the disturbance data is input to the basic prediction unit for prediction to obtain prediction data. The prediction data is input to the weight calibration unit bk for processing, and the obtained data is input to the offset calibration unit ck for processing to obtain the predicted value dk. Then, the preset loss function is used to calculate the predicted value dk and the second true value data. The loss value is used to adjust the parameters of the perturbation calibration unit ak, the basic prediction unit, the weight calibration unit bk and the offset calibration unit ck. After training the basic prediction model with each training data, the target prediction model is obtained.

[0157] In the embodiment of the present application, the training data set is used Training basic prediction models , the trained model parameters are , the preset loss function can be expressed as , the training objective can be expressed as follows:

[0158]

[0159] Among them, expression (3) means that the training data set Training basic prediction models In the process, the preset loss function is used Calculate the loss value and use it to adjust the model parameters , to determine the minimum loss value are the model parameters of the training target.

[0160] Furthermore, the present application provides an expression (4) of a preset loss function as follows:

[0161]

[0162] In expression (4) Represents the training dataset The amount of training data in . Indicates that the second input data input model parameters are a prediction value output by the base prediction model, a second true value data corresponding to the second input data.

[0163] In the embodiments of the present application, the target prediction model can be trained in various ways, which are not limited herein.

[0164] To sum up, in the present application, the terminal device only needs to send the generator parameters to the network device, and does not need to send a large amount of training data, thereby reducing the communication overhead. In addition, the target prediction model does not need to be trained by a real training data set, so it is not limited by the size of the real training data set, thereby improving the training effect. In addition, the network device trains the target prediction model according to the training data set generated by the multiple generator parameters sent by the multiple terminal devices, so that the generalization of the target prediction model obtained is higher.

[0165] In some embodiments, referring to Figure 6 , the model training method further includes the following steps:

[0166] S601, the network device receives calibration data sent by multiple terminal devices.

[0167] Wherein, after the target prediction model is trained by the above method, the calibration data is used to calibrate the target prediction model.

[0168] Referring to Figure 6 , each terminal device in the multiple terminal devices sends a small amount of calibration data to the network device, and uses the calibration data to calibrate the target prediction model, which can improve the prediction accuracy of the target prediction model.

[0169] In some embodiments, the calibration data includes first input data and first true value data.

[0170] Wherein, the definition form of the calibration data is the same as that of the training data, but is independent of the training data. The calibration data is actually collected by the terminal device. The training data is generated by the network device based on the generator. For example, the first input data represents CSI, and the first true value data represents the channel characteristics predicted based on the channel state information. Or the first input data represents beam measurement information, and the first true value data represents the beam characteristics predicted based on the beam measurement information.

[0171] In the embodiments of the present application, the first input data can be collected by the terminal device. The first true value data can be collected by the terminal device through the air interface.

[0172] For example, the calibration data set corresponding to the terminal device k can be expressed as the following expression (5):

[0173]

[0174] wherein in expression (5), denotes the jth first input data in the calibration data set denotes the jth first true value data in the calibration data set denotes the jth first input data in the calibration data set denotes the jth first true value data in the calibration data set, j takes value 1 to P, and P is a positive integer.

[0175] Further, further, the data dimension of , the data dimension of .

[0176] In the embodiments of the present application, the terminal device only needs to send a small amount of calibration data to the network device, thereby reducing the communication overhead between the terminal device and the network device.

[0177] S602, the network device calibrates the target prediction model according to the calibration data.

[0178] In some embodiments, calibrating the target prediction model according to the calibration data comprises: inputting the first input data into the target prediction model for processing to obtain first prediction data; and adjusting model parameters of the target prediction model according to a loss value of the first prediction data and the first true value data.

[0179] In some embodiments, based on the target prediction model shown in Figure 4 , inputting the first input data into the target prediction model for processing to obtain first prediction data comprises: inputting the first input data into a target prediction unit for processing to obtain the first prediction data; and adjusting model parameters of the target prediction model according to a loss value of the first prediction data and the first true value data comprises: adjusting parameters of the target prediction unit according to the loss value of the first prediction data and the first true value data.

[0180] It can be understood that, based on Figure 4 , the model parameters of the target prediction model comprise: parameters of the target prediction unit.

[0181] In some embodiments, the target prediction model further comprises: a plurality of perturbation calibration units, the plurality of perturbation calibration units correspond one-to-one to the plurality of terminal devices, and based on Figure 5, the terminal device k corresponds to the disturbance calibration unit ak. If the target prediction model only includes multiple disturbance calibration units and target prediction units, the first input data is input into the target prediction model for processing to obtain first prediction data, including: for each terminal device among the multiple terminal devices, according to the disturbance calibration unit corresponding to the terminal device, the first input data corresponding to the terminal device is disturbed to obtain disturbance data; and the disturbance data is input into the target prediction unit for processing to obtain first prediction data.

[0182] Furthermore, according to the loss value of the first prediction data and the first true value data, the model parameters of the target prediction model are adjusted, including: according to the loss value of the first prediction data and the first true value data, the parameters of the disturbance calibration unit are adjusted.

[0183] For example, the first input data is calibrated by the disturbance calibration unit ak Perform disturbance processing to obtain disturbance data , the perturbation data The target prediction unit is input for processing to obtain first prediction data. A preset loss function is then used to calculate the loss value between the first prediction data and the first true value data, and the parameter ak of the disturbance calibration unit is adjusted according to the loss value.

[0184] In some embodiments, the target prediction model also includes: multiple weight calibration units and / or multiple offset calibration units, wherein multiple terminal devices correspond one-to-one to the multiple weight calibration units and the multiple offset calibration units, and the first input data is input into the target prediction model for processing to obtain the first prediction data, including: for each terminal device among the multiple terminal devices, the first input data corresponding to the terminal device is input into the target prediction unit for processing to obtain the second prediction data; according to the weight calibration unit and / or offset calibration unit corresponding to the terminal device, the second prediction data is processed to obtain the first prediction data.

[0185] Furthermore, according to the loss value of the first prediction data and the first true value data, the model parameters of the target prediction model are adjusted, including: according to the loss value of the first prediction data and the first true value data, the parameters of the weight calibration unit and / or the parameters of the offset calibration unit are adjusted.

[0186] For example, the first input data The target prediction unit is input for processing to obtain second prediction data; and the second prediction data is processed according to the weight calibration unit bk and / or the offset calibration unit ck corresponding to the terminal device to obtain the first prediction data.

[0187] In some embodiments, the target prediction model further includes: multiple disturbance calibration units, multiple weight calibration units, and multiple offset calibration units. The first input data is input into the target prediction model for processing to obtain first prediction data, including: for each terminal device among the multiple terminal devices, according to the disturbance calibration unit corresponding to the terminal device, the first input data corresponding to the terminal device is disturbed to obtain disturbance data; the disturbance data is input into the target prediction unit for processing to obtain third prediction data; according to the weight calibration unit and the offset calibration unit corresponding to the terminal device, the third prediction data is adjusted to obtain the first prediction data.

[0188] Furthermore, the model parameters of the target prediction model are adjusted according to the loss values ​​of the first prediction data and the first true value data, including: adjusting the parameters of the disturbance calibration unit, the parameters of the weight calibration unit, and the parameters of the offset calibration unit according to the loss values ​​of the first prediction data and the first true value data.

[0189] For example, the first input data is calibrated by the disturbance calibration unit ak Perform disturbance processing to obtain disturbance data , the perturbation data The target prediction unit processes the data to obtain third prediction data, which is then processed by the corresponding weight calibration unit bk and offset calibration unit ck to obtain first prediction data. A loss value between the first prediction data and the first true value data is then calculated using a preset loss function. The parameters ak of the disturbance calibration unit, the weight calibration unit bk, and / or the offset calibration unit ck are adjusted based on this loss value.

[0190] In the embodiment of the present application, the parameter of the disturbance calibration unit ak is the disturbance calibration parameter , the dimension of the disturbance calibration parameter is . First input data After being processed by the disturbance calibration unit ak, the disturbance data is obtained The expression (6) is as follows:

[0191]

[0192] Furthermore, the perturbation data Input the target prediction model obtained by the above training After processing, the third prediction data obtained can be expressed as , the third prediction data The data dimension is .

[0193] Among them, the third forecast data The processing process of the weight calibration unit bk and the offset calibration unit ck refers to the following expression (7):

[0194]

[0195] In expression (7) represents the weight calibration parameters of the weight calibration unit bk, Represents the offset calibration parameters of the offset calibration unit ck. Indicates the first prediction data.

[0196] In the embodiment of the present application, the disturbance calibration parameter , weight calibration parameters and offset calibration parameters The adjustment target is as follows (8):

[0197]

[0198] Among them, expression (8) can be understood as, When the L2 norm of is minimized, 、 and At least one of them is the calibration target of the target prediction model. Indicates the loss value between the first prediction data and the first true value data. In addition, the parameters of the target prediction unit can also be adjusted according to the loss value, which is not limited.

[0199] In the embodiment of the present application, the loss values ​​of the first predicted data and the first true value data may also be determined in other ways, which are not limited here.

[0200] In the embodiment of the present application, the model parameters of the target prediction model include: parameters of the target prediction unit, 、 and It can be understood that the present application provides a small amount of calibration data for each terminal device, and then the calibration data is perturbed and input into the trained target prediction unit for processing. The obtained third prediction data is then adjusted by the weight calibration unit and the offset calibration unit to obtain the first prediction data. Then the loss value between the first prediction data and the first true value data is calculated, and based on the loss value, 、 and Make adjustments to achieve calibration of the target prediction model and avoid repeated training of the target prediction model.

[0201] In the embodiment of the present application, method (1) is to use the Figure 4 The target prediction model is trained in the manner shown in FIG. 2 , and only the target prediction unit is calibrated in the calibration phase. Method (2) is used in the training phase. Figure 4The target prediction model is trained in the manner shown in FIG. 1 , and at least one of the disturbance calibration unit, weight calibration unit, or offset calibration unit is calibrated in the calibration phase. Method (3) is used in the training phase. Figure 5 The target prediction model is trained in the manner shown, and at least one of the simple calibration disturbance calibration unit, the weight calibration unit or the offset calibration unit is calibrated.

[0202] In the embodiment of the present application, the calibrated target prediction model can effectively extend the life cycle of the target prediction model, and when the target prediction model is deployed on a network device, it is not necessary to repeatedly train or fine-tune the target prediction unit, and only calibration data optimization is required. 、 and , the target prediction model can be calibrated, the accuracy of the calibrated target prediction model is improved in a lightweight way, the deployment overhead of the target prediction model is reduced, and the deployment efficiency of the target prediction model is improved.

[0203] Reference Figure 7 , shows a flowchart of the steps of another model training method provided by the present application, which is applied to a terminal device and specifically includes the following steps:

[0204] S701, the terminal device trains the generator and obtains the generator parameters of the trained generator.

[0205] Each of the K terminal devices is trained on the generator, and the generator parameters corresponding to the terminal device are obtained. For example, the generator parameters corresponding to terminal device k are .

[0206] In some embodiments, training a generator to obtain generator parameters of the trained generator includes: determining random variables; inputting the random variables into the generator to generate a simulated data set; obtaining a real data set; using a judge to determine the probability values ​​that the simulated data in the simulated data set and the real data in the real data set belong to the real data; adjusting the generator parameters of the generator and the judger parameters of the judge according to the probability values; and looping through the above steps to obtain the generator parameters.

[0207] The random variable Z2 determined by the terminal device can refer to the random variable Z1 in expression (1), which will not be described in detail here.

[0208] Further, refer to Figure 8 , the random variable Z2 is input into the generator to generate a simulated data set, where the expression (9) of the simulated data set is as follows:

[0209]

[0210] In expression (9), represents the simulation data set corresponding to the terminal device k, which includes multiple simulation data, including simulation input data and the corresponding simulated true data In expression (9), i takes a value from 1 to Q, where Q is a positive integer.

[0211] The real data set is the actual data collected by the terminal device. The expression (10) of the real data set is as follows:

[0212]

[0213] In expression (10), Represents the real data set corresponding to the terminal device k, which includes multiple real data, including real input data and the corresponding true ground truth data In expression (10), i ranges from 1 to R, where R is a positive integer.

[0214] Furthermore, the decision maker can be expressed as , both the simulated data set and the real data set are input into the decision maker, and the probability values ​​output by the decision maker include: and ,in, Indicates the probability value that the simulated data output by the judge belongs to the real data, It represents the probability value that the real data output by the judgement device belongs to the real data. Represents the decision maker parameters.

[0215] Then, the generator parameters of the generator and the judgement parameters of the judge are adjusted according to the probability value, referring to the following expression (11):

[0216]

[0217] In expression (11), E represents the mean value, express and The adjustment target is When the maximum value is taken and It can be understood that the goal of adjusting the generator parameters of the generator and the judgement parameters of the judge is to make The value of increases.

[0218] in, Figure 8 The loss function in can be express.

[0219] It is understood that this application The optimization target of the decision maker is to improve the ability to distinguish real data and non-real data. is to make the generator improve the ability to simulate real data.

[0220] S702, the terminal device sends the generator parameter to the network device.

[0221] In the embodiment of the application, after the terminal device obtains the generator parameter in the above-mentioned manner , it only needs to send the generator parameter to the network device to make the network device generate a training data set according to the generator parameter and further train the target prediction model.

[0222] In some embodiments, it further includes: obtaining calibration data; and sending the calibration data to the network device.

[0223] It can be understood that the calibration data is real data collected by the network device, and the specific description of the calibration data can refer to the calibration data on the network device side described above, which will not be repeated here.

[0224] In the embodiment of the application, the terminal device sends the calibration data to the network device, and the network device calibrates the target prediction model using the calibration data. The specific calibration process is described above and will not be repeated here.

[0225] In the application, the generator is trained by using the generative adversarial algorithm on the side of the plurality of terminal devices to obtain the generator parameter, and then the generator parameter is sent to the network device, so that the network device can generate a training data set corresponding to each terminal device by itself, avoiding the communication overhead caused by the transmission of a large amount of data sets over the air.

[0226] In summary, in the application, the terminal device only needs to send the generator parameter to the network device and does not need to send a large amount of training data, so as to reduce the communication overhead. Further, the terminal device only needs to send a small amount of calibration data to the network device to calibrate the target prediction model, so as to improve the prediction accuracy of the target prediction model.

[0227] Referring to Figure 9 , a step flowchart of a prediction method provided by the application is provided, and the specific steps are as follows:

[0228] S901, the terminal device obtains communication measurement data.

[0229] The terminal device can obtain the communication measurement data by collecting, and the communication measurement data can be CSI or beam measurement information.

[0230] S902, the terminal device sends the communication measurement data to the network device.

[0231] The network device is used to predict the communication characteristics of the terminal device based on the communication measurement data.

[0232] S903: The network device inputs the communication measurement data into the target prediction model for processing to obtain communication features.

[0233] The target prediction model is obtained by training using any of the above model training methods.

[0234] In an embodiment of the present application, if the communication measurement data is CSI, the communication characteristics may be channel characteristics predicted based on the CSI. If the communication measurement data is beam measurement information, the communication characteristics may be beam characteristics predicted based on the beam measurement information.

[0235] It can be understood that the present application can be applied to prediction in CSI scenarios, and can also be applied to beam management scenarios, such as beam measurement, beam reporting, beam prediction, or beam deployment. Among them, the terminal device can perform beam measurement to obtain beam measurement information, and send the beam measurement information to the network device. The network device processes the beam measurement information based on the target prediction model, and obtains beam characteristics such as the quality of the received signal of each beam. Furthermore, the network device can select a beam with a possible better signal quality according to the instructions of the received signal of each beam, determine the beam deployment information, and then provide it to the terminal device. The terminal device performs corresponding beam selection or beam switching according to the beam deployment information of the network device, thereby realizing resource allocation or beam management.

[0236] In some embodiments, based on Figure 4 ,If the target prediction model includes a target prediction unit, the communication measurement data is input into the target prediction unit for processing to obtain the communication features.

[0237] In some embodiments, based on Figure 5 If the target prediction model includes: target prediction unit, disturbance calibration unit, weight calibration unit and offset calibration unit. Then the communication measurement data can be After being processed by the disturbance calibration unit, the disturbance measurement data is obtained Then the perturbation measurement data Input the target measurement unit for processing to obtain predicted feature data . The predicted feature data After being processed by the weight calibration unit and the offset calibration unit, the communication characteristics are obtained .

[0238] In an embodiment of the present application, the calibrated target prediction model can be deployed on a network device, and the communication measurement data provided by the terminal device can be processed by the target prediction model to obtain predicted communication characteristics.

[0239] Figure 10 A structural schematic diagram of a model training apparatus is provided for an embodiment of the present application. It can be understood that the model training apparatus can correspond to the operation or steps of the corresponding network device in each of the foregoing method embodiments. The model training apparatus can be a network device or can be a component, such as a chip, a chip module, etc., configurable to the network device. As shown, the model training apparatus 100 can include: Figure 10

[0240] The receiving module 1001 is configured to receive a plurality of generator parameters sent by a plurality of terminal devices, the plurality of generator parameters corresponding to the plurality of terminal devices one by one.

[0241] The generating module 1002 is configured to generate a training data set based on each of the plurality of generator parameters using a generator.

[0242] The training module 1003 is configured to train a basic prediction model according to a plurality of training data sets to obtain a target prediction model, the target prediction model being used to predict a communication feature of a terminal device based on communication measurement data of the terminal device.

[0243] Optionally, the receiving module 1001 is further configured to receive calibration data sent by the plurality of terminal devices.

[0244] The model training apparatus 100 further includes a calibration module (not shown) configured to calibrate the target prediction model according to the calibration data.

[0245] Optionally, the calibration data includes first input data and first true value data.

[0246] Optionally, the calibration module is specifically configured to:

[0247] input the first input data into the target prediction model for processing to obtain first prediction data;

[0248] adjust model parameters of the target prediction model according to a loss value of the first prediction data and the first true value data.

[0249] Optionally, the basic prediction model includes a basic prediction unit, and the target prediction model includes a target prediction unit, the target prediction unit being obtained after training of the basic prediction unit.

[0250] Optionally, the target prediction model further includes at least one of a plurality of perturbation calibration units, a plurality of weight calibration units, or a plurality of offset calibration units, the plurality of perturbation calibration units corresponding to the plurality of terminal devices one by one, the plurality of weight calibration units corresponding to the plurality of terminal devices one by one, and the plurality of offset calibration units corresponding to the plurality of terminal devices one by one.

[0251] ​Optionally, the target prediction model further includes: a plurality of disturbance calibration units, wherein the calibration module is specifically configured to:

[0252] For each terminal device among the plurality of terminal devices, performing disturbance processing on first input data corresponding to the terminal device according to a disturbance calibration unit corresponding to the terminal device to obtain disturbance data;

[0253] The disturbance data is input into the target prediction unit for processing to obtain first prediction data.

[0254] Optionally, when the calibration module adjusts the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data, it is specifically used to:

[0255] According to the loss values ​​of the first predicted data and the first true value data, the parameters of the disturbance calibration unit corresponding to the terminal device are adjusted.

[0256] Optionally, the target prediction model further includes: a plurality of weight calibration units and / or a plurality of offset calibration units. When the calibration module inputs the first input data into the target prediction model for processing to obtain the first prediction data, the calibration module is specifically configured to:

[0257] For each terminal device among the plurality of terminal devices, inputting first input data corresponding to the terminal device into a target prediction unit for processing to obtain second prediction data;

[0258] According to the weight calibration unit and / or offset calibration unit corresponding to the terminal device, the second prediction data is processed to obtain the first prediction data.

[0259] Optionally, when the calibration module adjusts the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data, it is specifically used to:

[0260] According to the loss value of the first predicted data and the first true value data, the parameters of the weight calibration unit and / or the parameters of the offset calibration unit corresponding to the terminal device are adjusted.

[0261] Optionally, the target prediction model further includes: a plurality of disturbance calibration units, a plurality of weight calibration units, and a plurality of offset calibration units. When the calibration module inputs the first input data into the target prediction model for processing to obtain the first prediction data, the calibration module is specifically configured to:

[0262] For each terminal device among the plurality of terminal devices, performing disturbance processing on first input data corresponding to the terminal device according to a disturbance calibration unit corresponding to the terminal device to obtain disturbance data;

[0263] Inputting the disturbance data into the target prediction unit for processing to obtain third prediction data;

[0264] According to the weight calibration unit and the offset calibration unit corresponding to the terminal device, the third prediction data is adjusted to obtain first prediction data.

[0265] Optionally, the calibration module, when adjusting the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data, is specifically configured to:

[0266] According to the loss value of the first prediction data and the first true value data, the parameters of the perturbation calibration unit, the parameters of the weight calibration unit, and the parameters of the offset calibration unit corresponding to the terminal device are adjusted.

[0267] Optionally, the training data set includes a plurality of training data, and the training data includes second input data and second true value data.

[0268] Optionally, the second input data represents channel state information (CSI), and the second true value data represents channel characteristics predicted based on the CSI.

[0269] Optionally, the second input data represents beam measurement information, and the second true value data represents beam characteristics predicted based on the beam measurement information.

[0270] The model training apparatus provided in the embodiment can perform the model training method performed by the network device in the foregoing method embodiments, and has similar implementation principles and technical effects, which will not be described here.

[0271] Figure 11 A structural diagram of a prediction apparatus provided in the embodiment is shown. It can be understood that the prediction apparatus can correspond to the operations or steps of the corresponding network device in the foregoing various method embodiments. The prediction apparatus can be a network device or a component (for example, a chip, a chip module, etc.) that can be configured to the network device. As shown in the figure, the prediction apparatus 110 can include: Figure 11

[0272] The receiving module 1101 is configured to receive the communication measurement data sent by the terminal device.

[0273] The processing module 1102 is configured to input the communication measurement data into the target prediction model for processing to obtain communication characteristics, and the target prediction model is obtained by training according to the model training method of any one of the foregoing method embodiments.

[0274] The prediction apparatus provided in the embodiment can perform the prediction method performed by the network device in the foregoing method embodiments, and has similar implementation principles and technical effects, which will not be described here.

[0275] Figure 12 ​This is a structural diagram of a model training device provided in an embodiment of the present application. It is understandable that the model training device can implement the operations or steps of the corresponding terminal devices in the aforementioned various method embodiments. The model training device can be a terminal device or a component that can be configured in a terminal device, such as a chip, a chip module, etc. Figure 12 As shown, the model training device 120 may include:

[0276] A training module 1201 is used to train the generator and obtain the generator parameters of the trained generator;

[0277] The sending module 1202 is configured to send generator parameters to the network device.

[0278] Optionally, the training module 1201 is specifically configured to:

[0279] Identify random variables;

[0280] Input random variables into the generator to generate simulated data sets;

[0281] Get a real dataset;

[0282] A judger is used to determine the probability values ​​of the simulated data in the simulated data set and the real data in the real data set belonging to the real data;

[0283] Adjusting generator parameters of the generator and decision parameters of the decision maker according to the probability value;

[0284] Loop through the above steps to get the generator parameters.

[0285] Optionally, it further includes: an acquisition module (not shown) for acquiring calibration data;

[0286] The sending module 1202 is further configured to send calibration data to the network device.

[0287] The model training device provided in this embodiment can execute the model training method executed by the terminal device in the aforementioned method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0288] Figure 13 This is a schematic diagram of the structure of a prediction device provided in an embodiment of the present application. It is understood that the prediction device can implement the operations or steps of the corresponding terminal devices in the aforementioned various method embodiments. The prediction device can be a terminal device or a component that can be configured in the terminal device, such as a chip, a chip module, etc. Figure 13 As shown, the prediction device 130 may include:

[0289] Acquisition module 1301, used to acquire communication measurement data;

[0290] The sending module 1302 is configured to send communication measurement data to a network device, and the network device is configured to predict the communication characteristics of the terminal device based on the communication measurement data.

[0291] The prediction device provided in this embodiment can execute the prediction method executed by the terminal device in the aforementioned method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0292] Figure 14 This is a schematic diagram of the structure of a communication device provided in an embodiment of the present application. Figure 14 As shown, the communication device 140 may include: at least one processor 1401, a memory 1402, and a transceiver 1403. The processor 1401, the transceiver 1403, and the memory 1402 communicate with each other via an internal connection path. The memory 1402 is used to store instructions, and the processor 1401 is used to execute the instructions stored in the memory 802 to control the transceiver 1403 to send and / or receive indication information.

[0293] Optionally, the above-mentioned communication device may also include at least one storage module, which may include data and / or instructions. Other modules in the communication device (such as a receiving module, a sending module, a processing module, etc.) can read the data and / or instructions in the storage module to implement the corresponding method.

[0294] It should be noted that it should be understood that in each of the above embodiments, the sending module can be a transmitter when actually implemented, and the receiving module can be a receiver when actually implemented, or the sending module and the receiving module can be implemented through a transceiver, or the sending module and the receiving module can be implemented through a communication port. The processing module can be implemented in the form of software called by a processing element; it can also be implemented in the form of hardware. For example, the processing module can be at least one separately established processing element, or it can be integrated into a chip of the above-mentioned device for implementation. In addition, it can also be stored in the memory of the above-mentioned device in the form of program code, and called by a processing element of the above-mentioned device to perform the functions of the above-mentioned processing module. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element mentioned here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0295] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by a processing element invoking program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of invoking program code. For another example, the modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0296] It should be understood that the communication device may correspond to the terminal device in the above-mentioned method embodiment, or may correspond to the network device in the above-mentioned method embodiment. And it may be used to execute the various steps and / or processes performed by the terminal device or the network device in the above-mentioned method embodiment. Optionally, the memory 1402 may include a read-only memory and a random access memory, and provide instructions and data to the processor 1401. A portion of the memory 1402 may also include a non-volatile random access memory. The memory 1402 may be a separate device or may be integrated into the processor 1401. The processor 1401 may be used to execute the instructions stored in the memory 1402, and when the processor 1401 executes the instructions stored in the memory, the processor 1401 is used to execute the various steps and / or processes of the above-mentioned method embodiment.

[0297] The transceiver 1403 may include a transmitter and a receiver. The transceiver 1403 may further include an antenna, which may be one or more. The processor 1401, memory 1402, and transceiver 803 may be integrated on different chips. For example, the processor 1401 and memory 1402 may be integrated in a baseband chip, and the transceiver 803 may be integrated in a radio frequency chip. The processor 1401, memory 1402, and transceiver 1403 may also be integrated on the same chip. This application does not limit this.

[0298] Optionally, the communication device is a component configured in a terminal device or a network device, such as a chip, a chip system, etc.

[0299] The transceiver 1403 may also be a communication interface, such as an input interface and / or output interface, circuit, etc. The transceiver 803, the processor 1401 and the memory 1402 may be integrated into the same chip, such as a baseband chip.

[0300] It should be understood that the above-mentioned communication device may be one or more chips. For example, the communication device may be a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0301] During implementation, each step of the above method can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0302] It should be noted that the processor in the embodiments of the present application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above-described method embodiments can be completed by hardware integrated logic circuits in the processor or by software instructions. The above-described processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above-described method.

[0303] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM). It should be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0304] An embodiment of the present application also provides a model processing system, which includes the terminal device in the aforementioned embodiment and a network device.

[0305] The present application also provides a chip having a computer program stored thereon. When the computer program is executed by the chip, the method in the above embodiment is implemented.

[0306] The present application also provides a computer-readable storage medium, which may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media that can store program codes. Specifically, the computer-readable storage medium stores program instructions, and when the program instructions are executed, the method in the above embodiment is implemented.

[0307] The present application also provides a computer program product, which includes execution instructions stored in a readable storage medium. At least one processor of a communication device can read the execution instructions from the readable storage medium, and at least one processor executes the execution instructions so that the communication device implements the model training methods provided in the various embodiments described above.

[0308] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A model training method, characterized in that: Applied to a network device, the method includes: Receiving multiple generator parameters sent by multiple terminal devices, where the multiple generator parameters correspond one-to-one to the multiple terminal devices, and each generator parameter is obtained by the terminal device by training a generator of a generative adversarial network; Using a generator, based on each generator parameter of the multiple generator parameters, to generate a training data set, the training data set including multiple training data, each training data including second input data and second true data, the second input data representing channel state information CSI, the second true data representing a channel characteristic predicted based on the CSI, or the second input data representing beam measurement information, the second true data representing a beam characteristic predicted based on the beam measurement information; A basic prediction model is trained according to a plurality of training data sets to obtain a target prediction model, wherein the target prediction model is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device.

2. The model training method according to claim 1, characterized in that Also includes: receiving calibration data sent by the plurality of terminal devices; The target prediction model is calibrated according to the calibration data.

3. The model training method according to claim 2, characterized in that The calibration data includes first input data and first true value data.

4. The model training method according to claim 3, characterized in that The step of calibrating the target prediction model according to the calibration data includes: Inputting the first input data into the target prediction model for processing to obtain first prediction data; Adjust the model parameters of the target prediction model according to the loss values ​​of the first prediction data and the first true value data.

5. The model training method according to claim 4, characterized in that The basic prediction model includes a basic prediction unit, and the target prediction model includes a target prediction unit. The target prediction unit is obtained after the basic prediction unit is trained.

6. The model training method according to claim 5, characterized in that The target prediction model also includes: at least one of: multiple disturbance calibration units, multiple weight calibration units or multiple offset calibration units, the multiple disturbance calibration units correspond one-to-one to the multiple terminal devices, the multiple weight calibration units correspond one-to-one to the multiple terminal devices, and the multiple offset calibration units correspond one-to-one to the multiple terminal devices.

7. The model training method according to claim 5 or 6, characterized in that: The target prediction model further includes: a plurality of disturbance calibration units, wherein the first input data is input into the target prediction model for processing to obtain first prediction data, including: For each terminal device among the multiple terminal devices, performing disturbance processing on the first input data corresponding to the terminal device according to the disturbance calibration unit corresponding to the terminal device to obtain disturbance data; The disturbance data is input into the target prediction unit for processing to obtain the first prediction data.

8. The model training method according to claim 7, characterized in that: The adjusting the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data includes: According to the loss values ​​of the first predicted data and the first true value data, the parameters of the disturbance calibration unit corresponding to the terminal device are adjusted.

9. The model training method according to claim 5 or 6, characterized in that: The target prediction model further includes: a plurality of weight calibration units and / or a plurality of offset calibration units. Inputting the first input data into the target prediction model for processing to obtain first prediction data includes: For each terminal device among the plurality of terminal devices, inputting the first input data corresponding to the terminal device into the target prediction unit for processing to obtain second prediction data; According to the weight calibration unit and / or offset calibration unit corresponding to the terminal device, the second prediction data is processed to obtain the first prediction data.

10. The model training method according to claim 9, characterized in that: The adjusting the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data includes: According to the loss values ​​of the first predicted data and the first true value data, the parameters of the weight calibration unit and / or the parameters of the offset calibration unit corresponding to the terminal device are adjusted.

11. The model training method according to claim 5 or 6, characterized in that: The target prediction model further includes: a plurality of disturbance calibration units, a plurality of weight calibration units, and a plurality of offset calibration units. Inputting the first input data into the target prediction model for processing to obtain first prediction data includes: For each terminal device among the multiple terminal devices, performing disturbance processing on the first input data corresponding to the terminal device according to the disturbance calibration unit corresponding to the terminal device to obtain disturbance data; Inputting the disturbance data into the target prediction unit for processing to obtain third prediction data; According to the weight calibration unit and the offset calibration unit corresponding to the terminal device, the third prediction data is adjusted to obtain the first prediction data.

12. The model training method according to claim 11, characterized in that: The adjusting the model parameters of the target prediction model according to the loss value of the first prediction data and the first true value data includes: According to the loss values ​​of the first predicted data and the first true value data, the parameters of the disturbance calibration unit, the parameters of the weight calibration unit and the parameters of the offset calibration unit corresponding to the terminal device are adjusted.

13. A prediction method, characterized in that: Applied to a network device, the method includes: receiving communication measurement data sent by a terminal device; The communication measurement data is input into a target prediction model for processing to obtain communication features, wherein the target prediction model is trained by the model training method according to any one of claims 1 to 12.

14. A model training method, characterized in that: Applied to each terminal device among a plurality of terminal devices, the method includes: Train the generator to obtain the generator parameters of the trained generator. Each generator parameter is obtained by the terminal device through training the generator of the generative adversarial network; The generator parameters are sent to a network device, wherein the multiple generator parameters sent by the multiple terminal devices correspond one-to-one to the multiple terminal devices, and each generator parameter in the multiple generator parameters is used by the network device to generate a training data set, and the training data set includes multiple training data, and each training data includes second input data and second true value data, the second input data represents channel state information CSI, and the second true value data represents channel characteristics predicted based on the CSI, or the second input data represents beam measurement information, and the second true value data represents beam characteristics predicted based on the beam measurement information.

15. The model training method according to claim 14, characterized in that: The training generator obtains the generator parameters of the trained generator, including: Identify random variables; Inputting the random variables into the generator to generate a simulated data set; Get a real dataset; Using a judger to determine the probability values ​​of the simulated data in the simulated data set and the real data in the real data set belonging to the real data; adjusting a generator parameter of the generator and a decision parameter of the decision device according to the probability value; The above steps are executed repeatedly to obtain the trained generator parameters.

16. The model training method according to claim 14 or 15, characterized in that: Also includes: Get calibration data; The calibration data is sent to the network device.

17. A prediction method, characterized in that: Applied to a terminal device, the method includes: Obtain communication measurement data; The communication measurement data is sent to a network device, and the network device is used to input the communication measurement data into a target prediction model to predict the communication characteristics of the terminal device, and the target prediction model is trained according to the model training method according to any one of claims 1 to 12.

18. A model training device, characterized in that: Applied to network equipment, the device includes: A receiving module, configured to receive multiple generator parameters sent by multiple terminal devices, wherein the multiple generator parameters correspond one-to-one to the multiple terminal devices, and each generator parameter is obtained by the terminal device by training a generator of a generative adversarial network; a generating module, configured to use a generator to generate a training data set based on each generator parameter of the multiple generator parameters, where the training data set includes multiple training data, and each training data includes second input data and second true value data, where the second input data represents channel state information CSI, and the second true value data represents a channel characteristic predicted based on the CSI, or the second input data represents beam measurement information, and the second true value data represents a beam characteristic predicted based on the beam measurement information; The training module is used to train a basic prediction model based on multiple training data sets to obtain a target prediction model, where the target prediction model is used to predict the communication characteristics of the terminal device based on the communication measurement data of the terminal device.

19. A prediction device, characterized in that: Applied to network equipment, the device includes: A receiving module, configured to receive communication measurement data sent by a terminal device; A processing module is used to input the communication measurement data into a target prediction model for processing to obtain communication characteristics, and the target prediction model is trained according to the model training method according to any one of claims 1 to 12.

20. A model training device, characterized in that: Applied to each terminal device among a plurality of terminal devices, the apparatus comprises: A training module is used to train the generator and obtain the generator parameters of the trained generator. Each generator parameter is obtained by the terminal device through training the generator of the generative adversarial network; A sending module is used to send the generator parameters to the network device, wherein the multiple generator parameters sent by the multiple terminal devices correspond one-to-one to the multiple terminal devices, and each generator parameter of the multiple generator parameters is used by the network device to generate a training data set, and the training data set includes multiple training data, each training data includes second input data and second true value data, the second input data represents channel state information CSI, and the second true value data represents channel characteristics predicted based on the CSI, or the second input data represents beam measurement information, and the second true value data represents beam characteristics predicted based on the beam measurement information.

21. A prediction device, characterized in that: Applied to a terminal device, the device includes: An acquisition module, used for acquiring communication measurement data; A sending module is used to send the communication measurement data to a network device, and the network device is used to input the communication measurement data into a target prediction model to predict the communication characteristics of the terminal device. The target prediction model is trained according to the model training method according to any one of claims 1 to 12.

22. A communication device, characterized in that: The communication device includes: a processor, a transceiver, and a memory; the processor is communicatively connected to the transceiver and the memory respectively; The memory stores computer-executable instructions; The transceiver communicates and interacts with an external device; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 17.

23. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 17 is implemented.

24. A chip system, characterized in that: The system comprises at least one processor and a communication interface, wherein the communication interface and the at least one processor are interconnected via a line, and the at least one processor is configured to run a computer program or instruction to execute the method according to any one of claims 1 to 17.

25. A computer program product, characterized in that The method comprises a computer program which, when being executed, causes a computer to execute the method according to any one of claims 1 to 17.

Images