Power distribution method and device, equipment and storage medium

Abstract

The invention discloses a power distribution method and device, equipment and a storage medium. The method is applied to an MIMO base station, and comprises the following steps: in a cellular large-scale MIMO network scene, obtaining position information data and channel quality information data of user equipment in a preset interval time period, performing data cleaning and disorder processing, establishing an initial data set, and taking an optimal power distribution result corresponding to the user equipment as a label; dividing the initial data set into a training set and a test set according to a preset proportion; inputting the training set into a preset power prediction model to obtain a trained power prediction model; and inputting to-be-detected data into the trained power prediction model, and predicting and outputting a power distribution result of each user terminal. According to the invention, a power allocation mechanism with optimal total power allocated to the user terminals of the same base station can be realized, and the purpose of saving the energy consumption of the base station is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of mobile communication technology, and in particular to a power allocation method, device, equipment and storage medium. BACKGROUND

[0002] In the network of the field of mobile communication, the base station is one of the devices consuming the most energy, so the energy saving of the base station is very important for reducing energy consumption, reducing operating costs and reducing the impact on the environment. Power allocation is a key link in the management of the base station, which involves the power allocation method of the base station at different times or in different areas. By reasonably allocating the transmission power of the base station, the system capacity can be improved to realize the access of more user equipment. For example, in a cellular network, by dynamically allocating transmission power and communication resources according to the needs of user equipment and the system channel quality, the best system capacity and coverage can be achieved, the power consumption is reduced under the premise of ensuring communication quality, and the purpose of energy saving is realized. In order to ensure transmission efficiency, the transmission power needs to be reasonably allocated, so as to ensure that the signal strength is within a proper range, reduce energy consumption cost under the premise of improving transmission efficiency, and also ensure that the signal can effectively resist noise and interference in the transmission process, and improve the signal-to-noise ratio and user experience.

[0003] The Multiple Input Multiple Output (MIMO) system can improve the communication quality and reliability by transmitting and receiving data on multiple antennas at the same time. Centralized large-scale MIMO further increases the number of antennas, so that the system can achieve higher system capacity in more users.

[0004] The existing power allocation method optimizes the design target (such as maximum spectrum efficiency or maximum throughput) of interest for power allocation through mathematical optimization model and program, and obtains the power allocation value through single network model training under the premise of limiting other parts, such as maximum rate or minimum delay. Initial sample data is usually used as a baseline method for comparison of new power allocation methods. The disadvantage of the existing method is that it depends heavily on channel information and has high requirements for its integrity. The existing method does not fully consider this sensitive time-varying characteristic of the wireless channel, which limits the improvement of the power allocation result on the performance of the communication system and affects the accuracy of the calculation of the optimal power allocation value. SUMMARY

[0005] In order to solve the above technical problems, the present application provides a power allocation method, device, equipment and storage medium, which can support larger throughput and system capacity with lower power consumption, achieve higher signal-to-noise ratio, better adapt to different network environments and user needs, and have higher flexibility and robustness.

[0006] In order to achieve the above object, the technical scheme of the present application is as follows:

[0007] A power allocation method applied to a MIMO base station, the method comprising the following steps:

[0008] Obtaining position information data and channel quality information data of user equipment in a preset interval period, and performing data cleaning and disorder processing, to form an initial data set, taking the optimal power allocation result corresponding to the user equipment as a label; dividing the initial data set into a training set and a test set according to a preset proportion;

[0009] Inputting the training set into a preset power prediction model, the power prediction model comprising a first neural network and a second neural network, inputting the channel quality information in a first time period into the first neural network for predicting the channel quality information in a second time period, and inputting the channel quality information in the second time period and the position information of the user equipment in the first time period into the second neural network, calculating a first loss value through a first loss function, judging whether the first loss value meets a preset requirement, if not, adjusting the network parameters for continuous training until the first loss value meets the preset requirement; if yes, inputting the training result into a second loss function to calculate a second loss value, judging whether the second loss value meets the preset requirement, if not, performing adaptive data augmentation, fusing the augmented data with the training set to obtain a fusion data set, inputting the fusion data set into the preset power prediction model for continuous training; if yes, outputting an initial power prediction model; inputting the test set into the initial power prediction model for optimization test, to obtain a trained power prediction model;

[0010] Inputting the real-time obtained position information and channel quality information of the user equipment in the current time period into the trained power prediction model, outputting the power allocation result of each user terminal and sending the downlink signal according to the power allocation result of each user terminal.

[0011] Preferably, the label acquisition specifically comprises the following steps:

[0012] Taking the position information of the user equipment as the input of a traditional power allocation optimization mathematical model, and estimating the channel between the user and the base station associated with the user;

[0013] Taking the maximization of the minimum user throughput as the target, establishing a mathematical model of the optimal power allocation result, and solving the established mathematical model of the optimal power allocation result by using a classical convex optimization algorithm, to obtain the optimal power allocation result.

[0014] Preferably, the position information comprises longitude and latitude information of the user terminal, and the channel quality information comprises quality of a wireless channel, channel parameters, multipath delay, Doppler frequency offset, rank of a MIMO channel and beamforming vectors.

[0015] Preferably, the second neural network adopts a multi-layer perception deep neural network, comprising an input layer, a hidden layer and an output layer; the input layer has 4 nodes, the hidden layer has 8 nodes and the output layer has 1 node.

[0016] Preferably, the first loss function L(w) is as follows:

[0017]

[0018] wherein x i is position information of the i-th sample point; y i is a power label value of the i-th position point;

[0019] w is a parameter of the second neural network; f(x i ,w) is a power value calculated by the neural network model;

[0020] l(f(x i ,w),y i ) represents a norm of the difference between f(x i ,w) and y i ; ||·||1 represents a one-norm of a vector; represents a summation of values from 1 to N in the brackets; is a summation of values of l(f(x i ,w),y i ) from i=1 to i=N.

[0021] Preferably, the second loss function R(w) is as follows:

[0022]

[0023] wherein, is position information of the i-th sample point after data enhancement; is a power label value of the i-th position point after data enhancement; a hyperparameter λ satisfies λ≥0, is a power value calculated by the neural network model using the enhanced sample data; represents a norm of the difference between and .

[0024] Preferably, the adaptive data augmentation is as follows:

[0025]

[0026] adaptive_α=σε+μ (4)

[0027] wherein (x i ,y i )∈D,(x j ,y j )∈D are any two groups of samples in the data set, ε ~ N(0,1) is a random variable subject to normal distribution, and the mean value μ and the standard deviation σ are obtained by calculation in the iterative training process of the second neural network, and adaptive_α is a coefficient value.

[0028] Based on the above, the application further discloses a power allocation device, which is applied to a MIMO base station and comprises an acquisition module, a model generation module and a power prediction module, wherein,

[0029] The acquisition module is used for acquiring position information data and channel quality information data of user equipment in a preset interval period, performing data cleaning and disorder processing, establishing an initial data set, taking an optimal power allocation result corresponding to the user equipment as a label, and dividing the initial data set into a training set and a test set according to a preset proportion.

[0030] The model generation module is used for inputting the training set into a preset power prediction model, the power prediction model comprising a first neural network and a second neural network, inputting channel quality information in a first time period into the first neural network for predicting channel quality information in a second time period, inputting the channel quality information in the second time period and position information of the user equipment in the first time period into the second neural network, calculating a first loss value through a first loss function, judging whether the first loss value meets a preset requirement, adjusting network parameters for continuous training until the first loss value meets the preset requirement if the first loss value does not meet the preset requirement, calculating a second loss value through a second loss function if the first loss value meets the preset requirement, judging whether the second loss value meets the preset requirement, performing adaptive data augmentation if the second loss value does not meet the preset requirement, fusing the augmented data with the training set to obtain a fusion data set, inputting the fusion data set into the preset power prediction model for continuous training, and outputting an initial power prediction model if the second loss value meets the preset requirement.

[0031] The power prediction module is used for inputting real-time acquired position information and channel quality information of user equipment in a current time period into the trained power prediction model, outputting power allocation results of each user terminal and performing downlink signal transmission according to the power allocation results of each user terminal.

[0032] Based on the above, the present invention also discloses an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a power distribution method as described above.

[0033] Based on the foregoing, the present invention also discloses a computer-readable storage medium having a computer program stored thereon, characterized in that the program is executed by a processor to implement a power distribution method as described above.

[0034] Based on the above technical solution, the beneficial effects of the present invention are:

[0035] 1) This invention proposes a two-stage cyclic power prediction model for massive MIMO scenarios in non-cellular networks to obtain the optimal power allocation scheme for downlink transmission of base stations; a more reasonable power allocation method can reduce the power consumption of base stations and achieve the purpose of energy conservation and emission reduction.

[0036] 2) The adaptive data augmentation technique proposed in this invention can improve the generalization ability of the model and the robustness of the algorithm;

[0037] 3) Experiments have verified that the algorithm proposed in this patent has lower error and better generalization performance. Attached Figure Description

[0038] Figure 1 This is a flowchart of a power allocation method in one embodiment;

[0039] Figure 2 This is a schematic diagram of the structure of a cellular massive MIMO network in one embodiment;

[0040] Figure 3 This is a schematic diagram of power allocation prediction time segmentation in one embodiment;

[0041] Figure 4 This is a schematic diagram of the structure of the second neural network in a power allocation method of one embodiment;

[0042] Figure 5 This is a schematic block diagram of a two-level loop of an MLP neural network model in one embodiment;

[0043] Figure 6 This is a schematic diagram illustrating the experimental comparison results in one embodiment;

[0044] Figure 7 This is a structural block diagram of a power distribution device in one embodiment;

[0045] Figure 8 This is an internal structural diagram of an electronic device in one embodiment. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described in connection with the drawings in the embodiments of the present application.

[0047] In one embodiment, as shown in Figure 1 A power allocation method is provided, which can be applied to a MIMO base station, and it can be understood that the MIMO base station corresponds to a MIMO-assisted cellular massive MIMO network communication system. A two-level cyclic power prediction model is proposed to obtain an optimal power allocation scheme for downlink transmission of the MIMO base station. In this embodiment, a convolutional neural network (CNN) is used as the first neural network, and a multi-layer perceptron (MLP) is used as the second neural network, but it is not limited thereto. Those skilled in the art can adjust the specific selection of the neural network according to actual needs. The power allocation method specifically includes the following steps:

[0048] In step 110, the position information and channel quality information of the user equipment in a preset time period are obtained, cleaned and out-of-order processed, an initial data set is established, and the optimal power allocation result corresponding to the user equipment is used as a label. The initial data set is divided into a training set and a test set according to a preset ratio.

[0049] In this embodiment, the MIMO base station obtains the position information and channel quality information of the user terminal as the model input. The user terminal position information includes the latitude and longitude information of the user terminal, which can be obtained by installing a road test software in the test terminal and an external GPS positioning device. The channel quality information includes the quality of the wireless channel, the channel parameters, the multipath delay, the Doppler frequency offset, the rank of the MIMO channel, and the beamforming vector, which can be obtained by capturing the content in the test terminal test log. The initial input values of the model include the initial input sample data parameters (position parameter X_t0 and wireless channel quality parameter Q_t0) and the initialized neural network model parameters (network model parameter ω0, initial hyperparameter λ0, initial mean μ0, and initial variance σ0).

[0050] Then, the MIMO base station calculates the training label value for the power prediction model. The label value refers to the target variable value output by the prediction model. In supervised learning, labeled training data is used to train the model, enabling it to learn the relationship between input features and labels. This embodiment uses a traditional power allocation model to calculate the label value. More specifically, the location information of the user equipment is used as the input to the traditional power allocation optimization mathematical model. The channel between the user and its associated base station is estimated. With the goal of maximizing the minimum user throughput, a mathematical model for the optimal power allocation result is established. The classic convex optimization algorithm is used to solve the established mathematical model for the optimal power allocation result, and the obtained optimal power allocation result is used as the label. Detailed explanation is as follows:

[0051] Sub-step 111: Channel estimation. The collected data undergoes data cleaning and other data processing to obtain the user equipment's location information, which is then used as input to the classical power allocation model. Channel characteristics between the user and its associated base stations are estimated using a channel algorithm.

[0052] Suppose Figure 2 The network shown is a cellular massive MIMO network containing L cells. Each cell p includes a base station with M antennas and K user equipment. The channel between user i and base station j in cell p is... And the channel satisfies the following distribution:

[0053]

[0054] in, Let be the known spatial correlation matrix of the base station. Assume that the base station and the user communicate using time-division duplex technology, and that the communication network uses τ. p =κ pilots, and user i in each cell uses the same pilots. Then, base station j uses the minimum mean square error estimation technique to obtain The estimate is:

[0055]

[0056] Where, ρ tr For the total uplink pilot power of the user, I M It is an M-dimensional identity matrix, n li ~N C (0,I M ) represents noise, ρ represents the service power of the base station, and δ represents the noise level. 2 Q is the noise power. 1i The autocorrelation matrix of the normalized received signal, Let be the covariance matrix of the channel. For the position is The users, its large-scale decline for:

[0057]

[0058] in, Let θ be the average channel gain from the antenna at base station j to user i in cell l, θ = -148 dB represent the channel gain at a reference distance of 1 km, and α be the path loss coefficient, where α = 2.83. Taking the example of large-scale fading between user i and base station j in a cell. The unit is dB.

[0059] Sub-step 112: Calculate the label value: Calculate the label value by solving the classical optimization model;

[0060] The formula for calculating the signal-to-noise ratio in a cellular massive MIMO network communication system is as follows:

[0061]

[0062] in Let ρ represent the signal-to-noise ratio (SNR) of the k-th user terminal within the j-th cell range of the network. jk This represents the power allocated to the k-th user terminal within the service range of the j-th CPU. This represents the precoding vector of the i-th UE associated with the l-th AP. Let E represent the channel estimation matrix of the k-th user terminal associated with the j-th base station, and let E denote the mean value. The optimal power allocation value is calculated using the maximum signal-to-noise ratio product criterion based on traditional optimization theory as the label of the multilayer sensing deep neural network. The mathematical expression of the optimization theory is as follows:

[0063]

[0064] satisfy

[0065] in Represents the maximum downlink transmission power, ρ jk Represents ρ jk This represents the power allocated to the k-th user terminal within the j-th cell range.

[0066] Specifically, a randomized method is used to shuffle the data to obtain the dataset. The acquired dataset is divided into a training set and a test set. The training set is used to train the model parameters, and the test set is used to evaluate the model's performance. In this embodiment, 80% of the dataset is used as the training set, and 2% is used for testing. The specific division ratio can be adjusted according to specific circumstances.

[0067] Step 120: Input the training set into a preset power prediction model. The preset power prediction model includes a first neural network and a second neural network. Channel quality information within a first time period is input into the first neural network to predict channel quality information within a second time period. Channel quality information within the second time period and location information of user equipment within the first time period are input into the second neural network. A first loss value is calculated using a first loss function. It is determined whether the first loss value meets the preset requirements. If not, the network parameters are adjusted and training continues until the first loss value meets the preset requirements. If yes, the training result is input into a second loss function to calculate a second loss value. It is determined whether the second loss value meets the preset requirements. If not, adaptive data augmentation is performed. The augmented data is fused with the training set to obtain a fused dataset. The fused dataset is input into the preset power prediction model for further training. If yes, the initial power prediction model is output. The test set is input into the initial power prediction model for optimization testing to obtain the trained power prediction model.

[0068] In this embodiment, unsupervised learning is used to train the first neural network, i.e., a CNN network, to predict the channel quality information Q_t during the second time period. The training process for the second neural network, i.e., an MLP neural network, employs supervised learning, using the first neural network's prediction of the channel quality information during the second time period as input data. (See [link]). Figure 3 .

[0069] Specifically, the structure of an MLP neural network is as follows: Figure 4 As shown, the input parameters of the MLP neural network include two categories: user terminal location information and channel quality information. Each category of parameters contains multiple feature components. The number of nodes in the input layer is selected as 4 nodes. The output parameters in this embodiment include the amount of power allocated to the base station. The number of nodes in the output layer is selected as 1 node. The number of nodes in the hidden layer can be two or three times the number of nodes in the input layer according to empirical methods. In order to avoid overfitting, it is usually not advisable to select too many nodes. The hidden layer is selected as 8 nodes and the output layer as 1 node.

[0070] Determine the activation function used for each node in the MLP neural network. Each hidden layer and output layer typically requires an activation function. The hidden layers use the ReLU activation function, mathematically expressed as follows:

[0071]

[0072] The output layer uses the Tanh activation function, the mathematical expression of which is as follows:

[0073]

[0074] To prevent overfitting, the loss function uses the L1 regularization function, the mathematical expression of which is as follows:

[0075]

[0076] And the preset threshold Th used to determine the stop of the training process. loss =0.1;

[0077] During model training, input samples are processed by a weight matrix and bias terms, and the output of each neuron is calculated using an activation function. The network parameters are then adjusted based on the loss function. In this embodiment, the weight parameters and bias terms are initialized using a random method.

[0078] The initial parameters and initial sample training set data determined in the above steps are input into the MLP neural network. During training, the network parameters are adjusted according to the selected loss function to minimize the loss.

[0079] The power prediction model employs a two-level loop structure within a large cycle T. The first-level loop determines whether to stop the model training process based on the output of the first loss function. The output of the model training result obtained after the first-level loop stops serves as the input of the second loss function. The judgment result of the second loss function determines whether to stop data updates. In the second-level loop, data updates are performed based on the output of the first-level loop. The updated data serves as the input sample data for the start of the next large cycle of the first-level loop.

[0080] Based on the training results and label values, a first loss value is calculated using a first loss function. It is then determined whether the first loss value meets a preset requirement. If not, the network parameters are adjusted, and training continues until the first loss value meets the preset requirement. If yes, the training results are input into a second loss function to calculate a second loss value. When the second loss value is less than or equal to a preset threshold Th, the loop stopping condition is met, and the model training loop is stopped. In this embodiment, the preset threshold Th = 0.2, but it can be adjusted by those skilled in the art based on specific experimental results. When the second loss value is greater than the preset threshold Th, the loop stopping condition is not met, and adaptive data augmentation is performed. The augmented data is then fused with the training set to obtain a fused dataset, which is input into the preset power prediction model for further training. The details are as follows:

[0081] The first loss function L(w) in the first-level loop is as follows:

[0082]

[0083] In the formula, x i It is the location information of the i-th sample point; y i It is the power label value at the i-th location; w is the neural network model parameter; f(x) i (f(x)) is the power value calculated by the neural network model; l(f(x)) i,w),y i ) represents the expression for f(x) i w) and y i Find the first norm of the difference; ||·||1 denotes the first norm of a vector; This represents the summation of values ​​from 1 to N over the positions indicated by the brackets. It is for l(f(x) i ,w),y i Sum the values ​​from i=1 to i=N.

[0084] The loss function value used in the second-level loop of the MLP neural network model is calculated. To achieve model generalization, a second loss function R(w) is defined, as follows:

[0085]

[0086] In the formula, It is the location information of the i-th sample point after data augmentation; It is the power label value of the i-th location point after data augmentation; the hyperparameter λ satisfies λ≥0. The power value is calculated using a neural network model with enhanced sample data. Indicates to and Find the norm of the difference.

[0087] The adaptive data augmentation method is as follows:

[0088]

[0089] α=σε+μ (4)

[0090] Where (x) i ,y i )∈D,(x j ,y j )∈D represents any two sets of samples in the dataset, ε~N(0,1) represents random variables that follow a normal distribution, and the mean μ and standard deviation σ are calculated during the iterative training of the MLP neural network, with adaptive_α being the coefficient value.

[0091] See Figure 5 The second neural network, namely the MLP neural network, employs a two-stage recurrent structure. The input is processed by the MLP neural network to obtain f(x,w), and simultaneously generates parameters μ and σ. 2 and λ. Then use μ,σ 2 The reparameterization process of formula (4) generates the random variable adaptive_α. Then, the fused input is generated using formula (3). The data is then fed into the neural network to calculate the output, and finally the loss function corresponding to the original sample and the fused sample is calculated. Each second-level loop cycle obtains an updated input dataset through the above process.

[0092] Step 130 involves inputting the real-time location information and channel quality information of user equipment within the current time period into the trained power prediction model, outputting the power allocation result for each user terminal, and transmitting downlink data based on the power allocation result for each user terminal.

[0093] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0094] experiment

[0095] Existing power allocation methods optimize the design objectives (such as maximizing spectral efficiency or throughput) of interest through mathematical optimization models and programs. Under constraints on other aspects, such as maximizing rate or minimizing latency, power allocation values ​​are obtained by training a single network model using initial sample data. These methods are typically used as baselines for comparing new algorithms. See also Figure 6 The figure shows a comparison between the training curves obtained by the baseline method and the training curves obtained by the method of this patent. The vertical axis represents the loss function value and the horizontal axis represents the number of training cycles. The smaller the loss function value, the better the model training result and the faster the model converges. The comparative experiment shows that the method of this patent has a faster convergence speed and is closer to the ideal optimal value.

[0096] In one embodiment, such as Figure 7 As shown, a power allocation device 600 is provided. This power allocation device can be applied to a MIMO base station, wherein the MIMO base station can be a very large-scale MIMO base station. It is understood that a very large-scale MIMO base station can form a hybrid field communication environment of very large-scale MIMO. The device may include: an acquisition module 610, a model generation module 620, and a power prediction module 630, wherein,

[0097] The acquisition module 610 is used to acquire location information data and channel quality information data of user equipment within a preset time interval, and perform data cleaning and disordering to build an initial dataset, using the optimal power allocation result corresponding to the user equipment as the label; and divide the initial dataset into a training set and a test set according to a preset ratio.

[0098] The model generation module 620 is used to input the training set into a preset power prediction model. The power prediction model includes a first neural network and a second neural network. Channel quality information within a first time period is input into the first neural network to predict channel quality information within a second time period. The channel quality information within the second time period and the location information of user equipment within the first time period are input into the second neural network. A first loss value is calculated using a first loss function. It is then determined whether the first loss value meets a preset requirement. If not, the network parameters are adjusted and training continues until the first loss value meets the preset requirement. If yes, the training result is input into a second loss function to calculate a second loss value. It is then determined whether the second loss value meets the preset requirement. If not, adaptive data augmentation is performed, and the augmented data is fused with the training set to obtain a fused dataset. The fused dataset is then input into the preset power prediction model for further training. If yes, an initial power prediction model is output. The test set is input into the initial power prediction model for optimization testing to obtain a trained power prediction model.

[0099] The power prediction module 630 is used to input the location information and channel quality information of user equipment acquired in real time within the current time period into the trained power prediction model, output the power allocation result of each user terminal, and transmit downlink signals according to the power allocation result of each user terminal.

[0100] It should be noted that the foregoing explanation of the power allocation method embodiments also applies to the power allocation device of this embodiment, and will not be repeated here. The power allocation device proposed according to the embodiments of this application can obtain the optimal power allocation scheme for downlink transmission of a MIMO base station.

[0101] like Figure 8 As shown, in one embodiment, an electronic device is provided, comprising a memory 701, a processor 702, and a computer program stored on the memory 701 and executable on the processor 702.

[0102] When the processor 702 executes the program, it implements a power allocation method provided in the above embodiments.

[0103] Furthermore, electronic devices also include:

[0104] Communication interface 703 is used for communication between memory 701 and processor 702.

[0105] The memory 701 is used to store computer programs that can run on the processor 702.

[0106] The memory 701 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0107] If the memory 701, processor 702, and communication interface 703 are implemented independently, then the communication interface 703, memory 701, and processor 702 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0108] Optionally, in a specific implementation, if the memory 701, processor 702, and communication interface 703 are integrated on a single chip, then the memory 701, processor 702, and communication interface 703 can communicate with each other through an internal interface.

[0109] The processor 702 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0110] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the line power allocation method described above.

[0111] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0112] Furthermore, in the embodiments of this application, the terms "first," "second," etc., are used to distinguish similar objects, and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, and the number of objects is not limited; for example, the first object can be one or more.

[0113] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0114] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0115] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0116] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0117] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0118] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A power distribution method, characterized by, The method is applied to a MIMO base station, and the method comprises the following steps: Obtaining position information data and channel quality information data of user equipment in a preset interval time period, and performing data cleaning and disorder processing, and establishing an initial data set, taking the optimal power allocation result corresponding to the user equipment as a label; the initial data set is divided into a training set and a test set according to a preset proportion; The training set is input into a preset power prediction model, the power prediction model comprises a first neural network and a second neural network, the channel quality information in the first time period is input into the first neural network for predicting the channel quality information in the second time period, and the channel quality information in the second time period and the position information of the user equipment in the first time period are input into the second neural network, a first loss value is calculated through a first loss function, whether the first loss value meets a preset requirement is judged, if not, the network parameters are adjusted for continuous training until the first loss value meets the preset requirement; if yes, the training result is input into a second loss function to calculate a second loss value, whether the second loss value meets the preset requirement is judged, if not, adaptive data augmentation is performed, the augmented data is fused with the training set to obtain a fusion data set, and the fusion data set is input into the preset power prediction model for continuous training; If yes, an initial power prediction model is output; the test set is input into the initial power prediction model for optimization test, and a trained power prediction model is obtained; The position information and the channel quality information of the user equipment in the current time period are input into the trained power prediction model in real time, the power allocation result of each user terminal is output, and the downlink signal is sent according to the power allocation result of each user terminal.

2. The power distribution method of claim 1, wherein, The label is obtained, and the following steps are specifically included: Taking the position information of the user equipment as the input of a traditional power allocation optimization mathematical model, and estimating the channel between the user and the base station associated with the user; A mathematical model of the optimal power allocation result is established by taking the maximum minimum user throughput as the target, and a classical convex optimization algorithm is used to solve the established mathematical model of the optimal power allocation result, and the optimal power allocation result is obtained.

3. The method of claim 1, wherein, The position information comprises longitude and latitude information of the user terminal, and the channel quality information comprises quality of a wireless channel, channel parameters, multipath delay, Doppler frequency offset, rank of a MIMO channel and a beamforming vector.

4. The method of claim 1, wherein, The second neural network adopts a multi-layer perception deep neural network, and comprises an input layer, a hidden layer and an output layer; the input layer has 4 nodes, the hidden layer has 8 nodes, and the output layer has 1 node.

5. A power distribution method as claimed in claim 4, characterized in that, The first loss function L(w) is as follows: In the formula, x i is position information of the i-th sample point; y i is a power label value of the i-th position point. w is a parameter of the second neural network; f(x i w) is a power value calculated by the neural network model l(f(x i ,w),y i ) represents a norm of the difference of f(x i ,w) and y i ; ||•||1 represents a one-norm of a vector; represents a summation over the values of the bracketed position from 1 to N; is a summation over the values of l(f(x i ,w),y i ) from i = 1 to i = N.

6. A power distribution method as claimed in claim 5, characterized in that, The second loss function (w) is as follows: In the formula, is the position information of the i-th sample point after data enhancement; is the power label value of the i-th position point after data enhancement; the hyperparameter λ satisfies λ≥0, is the power value calculated by the neural network model using the enhanced sample data; represents a norm of the difference between and .

7. A power distribution method as claimed in claim 6, characterized in that, The adaptive data augmentation is as follows: adaptive_α=σε+μ (4) wherein (x i ,y i )∈D,(x j ,y j )∈D are any two groups of samples in the data set, ε ~ N(0, 1) is a random variable subject to normal distribution, and the mean μ and the standard deviation σ are obtained by calculation in the iterative training process of the second neural network, and adaptive_α is a coefficient value.

8. A power distribution apparatus, characterized by, The device is applied to a MIMO base station, and the device comprises an acquisition module, a model generation module and a power prediction module, wherein, The acquisition module is configured to acquire position information data and channel quality information data of user equipment in a preset interval period, perform data cleaning and disordering processing, and establish an initial data set, with an optimal power allocation result corresponding to the user equipment as a label; and divide the initial data set into a training set and a test set according to a preset proportion; The model generation module is configured to input the training set into a preset power prediction model, the power prediction model including a first neural network and a second neural network, input channel quality information in a first time period into the first neural network for predicting channel quality information in a second time period, and input the channel quality information in the second time period and position information of the user equipment in the first time period into the second neural network, calculate a first loss value through a first loss function, determine whether the first loss value meets a preset requirement, if not, adjust network parameters for continuous training until the first loss value meets the preset requirement, if yes, input a training result into a second loss function to calculate a second loss value, determine whether the second loss value meets the preset requirement, if not, perform adaptive data augmentation, fuse the augmented data with the training set to obtain a fused data set, input the fused data set into the preset power prediction model for continuous training, and if yes, output an initial power prediction model; input the test set into the initial power prediction model for optimization testing to obtain a trained power prediction model; The power prediction module is configured to input real-time acquired position information and channel quality information of user equipment in a current time period into the trained power prediction model, output a power allocation result of each user terminal, and perform downlink signal transmission according to the power allocation result of each user terminal.

9. An electronic device, comprising: Comprise: A memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the program to implement a power allocation method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement a power allocation method according to any one of claims 1-7.

Images