Method and apparatus for optimizing antenna weight parameters

Abstract

The present invention provides an antenna weight parameter optimization method and device, the method comprising: obtaining state information and initial antenna weight parameters of a base station to be optimized; based on the state information and the initial antenna weight parameters, iteratively optimizing through a pre-constructed simulation environment model and optimization algorithm model to obtain the optimal antenna weight parameters; replacing the initial antenna weight parameters with the optimal antenna weight parameters. The antenna weight parameter optimization method and device provided by the present invention iteratively optimizes the initial antenna weight parameters through a simulation environment model and an optimization algorithm model, thereby quickly solving and obtaining the optimal antenna weight parameters. Compared with the traditional antenna weight parameter optimization method, the optimal antenna weight parameters can be obtained more efficiently and conveniently, and the optimization efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a method and device for optimizing antenna weight parameters. Background Art

[0002] Massive MIMO (Massive Multiple Input Multiple Output) is a key technology for improving network coverage and system capacity in 5G (5th Generation Mobile Communication Technology). In a Massive MIMO system, base stations are equipped with a large number of antennas. As the number of beam sectors and dynamic scanning dimensions increase and the cell environment becomes more complex, it becomes more difficult to optimize and adjust antenna weight parameters.

[0003] Traditional antenna weight parameter optimization methods have the problems of high computational complexity and long optimization time, which makes it difficult to meet the needs of actual applications.

[0004] Therefore, there is an urgent need for a more efficient and convenient antenna weight parameter optimization method to solve the above technical problems. Summary of the invention

[0005] The present invention provides an antenna weight parameter optimization method and device, which are used to solve the defects of high calculation complexity and long optimization time in the antenna weight parameter optimization method in the prior art, and realize more efficient and convenient antenna weight parameter optimization.

[0006] In a first aspect, the present invention provides a method for optimizing antenna weight parameters, the method comprising:

[0007] Obtaining status information and initial antenna weight parameters of the base station to be optimized;

[0008] Based on the state information and the initial antenna weight parameters, iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain optimal antenna weight parameters;

[0009] The initial antenna weight parameters are replaced with the optimal antenna weight parameters.

[0010] According to an antenna weight parameter optimization method provided by the present invention, based on the state information and the initial antenna weight parameters, an iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters, including:

[0011] Inputting the state information and the initial antenna weight parameters into a simulation environment model to obtain current communication performance data output by the simulation environment model;

[0012] Inputting the initial antenna weight parameter and the current communication performance data into an optimization algorithm model to obtain a predicted value of the antenna weight parameter output by the optimization algorithm model;

[0013] Inputting the state information and the predicted value of the antenna weight parameter into the simulation environment model, and obtaining corresponding communication performance data again;

[0014] The antenna weight parameter prediction value and the corresponding communication performance data are input into the optimization algorithm model again to perform the next antenna weight parameter prediction until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, thereby obtaining the optimal antenna weight parameter.

[0015] The present invention realizes iterative optimization of antenna weight parameters by cooperating with a simulation environment model and an optimization algorithm model. The communication performance data corresponding to the antenna weight parameters input into the model is obtained through the simulation environment model. Then, the antenna weight parameters and the communication performance data are input into the optimization algorithm model for fitting the objective function, and the next antenna weight parameter prediction value is determined. After multiple iterative optimization processes, the optimal antenna weight parameters are quickly determined.

[0016] According to an antenna weight parameter optimization method provided by the present invention, after inputting the state information and the initial antenna weight parameter into a simulation environment model and obtaining the current communication performance data output by the simulation environment model, the method further includes:

[0017] The state information, initial antenna weight parameters and current communication performance data are cached.

[0018] The setting of the cache function can improve the speed of acquiring simulation sample data. Since the state information, initial antenna weight parameters and corresponding communication performance data have been classified and saved, in the subsequent iterative optimization process, if the antenna weight parameters received by the simulation environment model are already stored in the storage data, then the relevant data can be directly retrieved from the cache, which reduces the amount of repeated calculations, improves the calculation efficiency of the simulation sample data, and thus reduces the optimization time of the entire optimization method.

[0019] According to an antenna weight parameter optimization method provided by the present invention, based on the state information and the initial antenna weight parameters, an iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters, including:

[0020] Inputting the state information and the initial antenna weight parameters into a simulation environment model to obtain current communication performance data output by the simulation environment model;

[0021] Inputting the initial antenna weight parameter and the current communication performance data into an optimization algorithm model to obtain a predicted value of the antenna weight parameter output by the optimization algorithm model;

[0022] Performing characterization learning on the predicted value of the antenna weight parameter to obtain discrete data of the antenna weight;

[0023] Inputting the antenna weight discrete data into the simulation environment model to obtain corresponding communication performance data again;

[0024] The antenna weight discrete data and the corresponding communication performance data are input into the optimization algorithm model again to perform the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, thereby obtaining the optimal antenna weight parameter.

[0025] Through an efficient representation learning mechanism, the difficulty of solving optimization problems can be reduced, and the conversion between low-dimensional and high-dimensional data and between continuous data and discrete data can be realized, thereby reducing the difficulty of implementing the entire optimization method.

[0026] According to an antenna weight parameter optimization method provided by the present invention, before inputting the initial antenna weight parameter and the current communication performance data into an optimization algorithm model and obtaining the antenna weight parameter prediction value output by the optimization algorithm model, the method further includes:

[0027] The optimization algorithm model is pre-trained using pre-acquired historical experience data.

[0028] The setting of the above-mentioned model pre-training link can effectively improve the data processing reliability and stability of the optimization algorithm model, thereby improving the optimization efficiency of the entire optimization method.

[0029] According to an antenna weight parameter optimization method provided by the present invention, the optimization algorithm model is pre-trained by using pre-acquired historical experience data, including:

[0030] Using pre-acquired historical experience data as observation data, a reward function is fitted; wherein the observation data includes antenna weight parameter samples and corresponding communication performance data samples, and the reward function is used to characterize the linear relationship between the communication performance data and the antenna weight parameter;

[0031] Based on the observation data and the reward function, a next antenna weight parameter to be trained is determined as a sample of the antenna weight parameter prediction value.

[0032] According to an antenna weight parameter optimization method provided by the present invention, the communication performance data includes signal strength, signal-to-noise ratio and signal coverage value;

[0033] The antenna weight parameters of the base station to be optimized include horizontal beam width, vertical beam width, direction angle and downtilt angle.

[0034] According to a method for optimizing antenna weight parameters provided by the present invention, the base station to be optimized is a 5G base station using a large-scale multiple-input multiple-output system.

[0035] In a second aspect, the present invention further provides an antenna weight parameter optimization device, the device comprising:

[0036] An acquisition module is used to obtain the status information and initial antenna weight parameters of the base station to be optimized;

[0037] An optimization module, configured to perform iterative optimization based on the state information and the initial antenna weight parameters through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters;

[0038] A configuration module is used to replace the initial antenna weight parameters with the optimal antenna weight parameters.

[0039] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the antenna weight parameter optimization method as described in any one of the above are implemented.

[0040] The antenna weight parameter optimization method and device provided by the present invention iteratively optimize the initial antenna weight parameters through a simulation environment model and an optimization algorithm model, thereby quickly solving for the optimal antenna weight parameters. Compared with the traditional antenna weight parameter optimization method, the optimal antenna weight parameters can be obtained more efficiently and conveniently, and the optimization efficiency is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a schematic diagram of the flow chart of the method for optimizing antenna weight parameters provided by the present invention;

[0043] Figure 2 It is a schematic diagram of the implementation principle of the antenna weight parameter optimization method provided by the present invention;

[0044] Figure 3It is a schematic diagram of the three-dimensional channel simulation effect of the 5G Massive MIMO three-dimensional channel simulation model;

[0045] Figure 4 It is a structural schematic diagram of the antenna weight parameter optimization device provided by the present invention;

[0046] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Since the antenna weight parameter optimization method and device provided in the embodiment of the present invention are mainly targeted at the application scenario of 5G base stations using Massive MIMO systems (large-scale multiple-input multiple-output systems), a large number of cheap antenna devices are used on the base station side to improve the data rate and link reliability. In a massive MIMO system using a large antenna array, the signal can be dynamically adjusted in the horizontal and vertical directions, so that the energy can be more accurately concentrated on a specific UE (User Equipment), thereby reducing interference between cells and supporting spectrum space multiplexing between multiple UEs.

[0049] The Massive MIMO system uses a large number of transceivers (TRX) and multiple antenna arrays to improve the broadcast mechanism of Long Term Evolution (LTE). It transmits multiple beam sectors through the beamforming mechanism, improves the power density of the transmitted electromagnetic waves, enhances the coverage range, and greatly improves the regional spectrum efficiency.

[0050] In a massive MIMO system, base stations are equipped with a large number of antennas, usually dozens, hundreds or even thousands, which is several times or even dozens of times the number of antennas in existing MIMO systems. Therefore, as the number of beam sectors and dynamic scanning dimensions increase, the number of combinations of beam sector antenna weights and wireless parameters in the massive MIMO system increases dramatically, coupled with the complex cell environment, making the optimization and adjustment of the massive MIMO system extremely complex.

[0051] At present, in the field of communication signal optimization of MIMO systems, how to adjust antenna parameters according to the environment and user distribution to improve communication status has become the focus of people's attention. Therefore, how to optimize the optimal mode and antenna weight parameters of the cell broadcast beam in the large-scale MIMO scenario to improve the network coverage quality and enhance the coverage effect of large-scale MIMO has become a problem that needs to be solved urgently.

[0052] The following is an explanation of the existing antenna weight parameter optimization methods, which are mainly divided into the following three categories:

[0053] The first category: artificial experience method

[0054] When deploying base stations, operators need to combine test data, site information, antenna files, electronic maps and other data to adjust antenna parameters to change the horizontal beamwidth, vertical beamwidth, azimuth angle, downtilt angle and other parameters of the broadcast beam. In actual business, manual experience and business data are often used to optimize antenna weight parameters. After calculation using empirical formulas, the optimal mode and RF parameter recommendations are set to improve network coverage and anti-interference capabilities through the set weight configuration plan.

[0055] However, the manual experience method relies on the engineering experience of business personnel from a large number of practices. The experience of business personnel determines the final optimization effect, so the reliability of the optimization process is difficult to guarantee. The beam characteristics of base stations in different scenarios, that is, the adjustable range, are different. Therefore, relying solely on manual experience to optimize broadcast channel coverage requires high skills of personnel, and the workload is large, which is time-consuming and labor-intensive. At the same time, this method has high investment costs. After the initial optimization, engineers need to continue to re-optimize parameter values ​​according to changes in scene requirements.

[0056] The second category: Classical optimization method

[0057] The classical optimization method uses the classical optimization algorithm to model the antenna weight parameter optimization problem, designs the optimization target according to the user and service distribution in the cell, uses the beamforming capability to determine the optimization parameters, and solves the optimal solution for the cell coverage and capacity corresponding to the Massive MIMO system. The classical optimization method mainly uses optimization methods such as particle swarm algorithm, genetic algorithm, simulated annealing algorithm, etc. to calculate the optimal mode and RF parameters of the cell broadcast beam in the Massive MIMO scenario, so as to improve the coverage quality of the 5G network and enhance the coverage effect of Massive MIMO.

[0058] However, since the large-scale MIMO optimization problem is generally a multivariate hybrid nonlinear problem, the multi-parameter joint optimization algorithm designed by the classical optimization method is relatively more complex. Due to the high parameter dimension, the computational complexity is high and the optimization time is long, so it is difficult to put into practical application. At the same time, the 5G antenna weight combination varies a lot, and more scene-by-scene research is needed to explore the maximum efficiency of the weights to meet the needs of users in more scenes. The traditional optimization algorithm needs to be re-optimized for new business scenarios, and the model does not have the ability to generalize.

[0059] The third category: intelligent learning method

[0060] Intelligent learning methods include deep neural network learning methods based on supervised learning and reinforcement learning methods based on reinforcement learning algorithms such as Q-Learning and DDPG (Deep Deterministic Policy Gradient). The deep neural network learning method uses the channel state information of a large number of beamforming matrices as training samples to maximize the system spectrum efficiency through neural network learning; the reinforcement learning method needs to model the MIMO optimization problem as a Markov decision process, define the state space, action space and reward function, and the intelligent agent learns the optimal antenna weights and parameters through continuous trial and error.

[0061] However, the learning efficiency of intelligent learning methods is low. Intelligent learning algorithms need to solve the optimal parameter values ​​from a large amount of data, and the learning cost is high.

[0062] Therefore, in order to solve the problem of high computational complexity and long optimization time of existing antenna weight parameter optimization methods, the embodiment of the present invention combines the classical optimization method with the intelligent learning method, and adopts Bayesian optimization and reinforcement learning technology to efficiently solve the optimal antenna weight parameters, thereby greatly improving the optimization efficiency and simulation calculation cost. Figure 1-Figure 3 The invention describes an antenna weight parameter optimization method and device provided by an embodiment of the invention.

[0063] Figure 1 The invention shows an optimization method for antenna weight parameters provided by an embodiment of the present invention, the method comprising:

[0064] Step 110: Acquire the status information and initial antenna weight parameters of the base station to be optimized.

[0065] In this embodiment, the base station to be optimized mainly refers to a 5G base station that adopts a large-scale multi-input multi-output system, wherein the status information may include information such as the location and power consumption of the base station, and the initial antenna weight parameters may include parameters such as the horizontal beamwidth, vertical beamwidth direction angle and downtilt angle.

[0066] Step 120: Based on the state information and the initial antenna weight parameters, an iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters.

[0067] In this embodiment, the specific implementation process of the above step 120 may include:

[0068] Firstly, the state information and initial antenna weight parameters are input into the simulation environment model to obtain the current communication performance data output by the simulation environment model;

[0069] Then, the initial antenna weight parameters and current communication performance data are input into the optimization algorithm model to obtain the antenna weight parameter prediction value output by the optimization algorithm model;

[0070] Next, the state information and the predicted values ​​of the antenna weight parameters are input into the simulation environment model to obtain the corresponding communication performance data again;

[0071] Finally, the antenna weight parameter prediction value and the corresponding communication performance data are input into the optimization algorithm model again for the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, and the optimal antenna weight parameter is obtained.

[0072] Preferably, after inputting the initial antenna weight parameters into the virtual communication environment and obtaining the current communication performance data, the following steps may also be performed:

[0073] The status information, initial antenna weight parameters and current communication performance data are cached.

[0074] Preferably, before inputting the predicted value of the antenna weight parameter into the simulation environment model and obtaining the corresponding communication performance data again, the following steps may also be included:

[0075] The predicted values ​​of antenna weight parameters are characterized and learned to obtain discrete data of antenna weights.

[0076] That is, based on the state information and the initial antenna weight parameters, the optimal antenna weight parameters are obtained by iteratively searching through the pre-built simulation environment model and optimization algorithm model, which may include:

[0077] Inputting the state information and initial antenna weight parameters into the simulation environment model to obtain current communication performance data output by the simulation environment model;

[0078] Inputting the initial antenna weight parameters and current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction value output by the optimization algorithm model;

[0079] Perform characterization learning on the predicted values ​​of antenna weight parameters to obtain discrete data of antenna weights;

[0080] Inputting the antenna weight discrete data into the simulation environment model, and obtaining the corresponding communication performance data again;

[0081] The antenna weight discrete data and the corresponding communication performance data are input into the optimization algorithm model again to predict the antenna weight parameters for the next time, until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, and the optimal antenna weight parameters are obtained.

[0082] Preferably, before inputting the initial antenna weight parameters and the current communication performance data into the optimization algorithm model and obtaining the antenna weight parameter prediction value output by the optimization algorithm model, the following may also be included:

[0083] The optimization algorithm model is pre-trained using pre-acquired historical experience data.

[0084] Specifically, the process of pre-training the optimization algorithm model using pre-acquired historical experience data may include:

[0085] Firstly, the reward function is fitted using the previously acquired historical experience data as the observation data; wherein the observation data includes antenna weight parameter samples and corresponding communication performance data samples, and the reward function is used to characterize the linear relationship between the communication performance data and the antenna weight parameters;

[0086] Then, based on the observed data and the reward function, the next antenna weight parameter to be trained is determined as a sample of the antenna weight parameter prediction value.

[0087] Step 130: Replace the initial antenna weight parameters with the optimal antenna weight parameters.

[0088] This step is the process of optimizing the configuration of the antenna of the base station to be optimized, that is, reconfiguring the antenna weight parameters of the base station to be optimized according to the optimal antenna weight parameters.

[0089] The following is combined with Figure 2 The optimization principle involved in the above step 120 in the embodiment of the present invention is described in detail.

[0090] Considering that the 5G Massive MIMO broadcast channel adopts a narrow beam polling transmission method, the antenna gain is greatly improved compared with the traditional antenna. At the same time, it supports multi-dimensional adjustment of the outer envelope mode of the broadcast channel. The horizontal beamwidth, vertical beamwidth, azimuth angle and downtilt angle of the broadcast beam can be changed by adjusting the antenna weight parameters. In a single Massive MIMO system, the active antenna processing unit (AAU) device supports up to tens of thousands of broadcast channel beam forms.

[0091] The large-scale MIMO optimization problem involves a large number of design decisions, and its optimization target is not only multi-peaked, non-convex, high-dimensional, and has huge decision space. In addition, the evaluation of network signals in different scenarios is costly. There is no clear mathematical expression for network coverage and interference in actual scenarios, so it takes a high cost to observe the return value of the optimization target. This is a typical black box optimization problem, and the embodiment of the present invention provides an effective global optimization method for this purpose. The optimal solution of the antenna weight parameters can be obtained after only a few objective function evaluations.

[0092] See attached Figure 2 This embodiment mainly optimizes the antenna weight parameters through the simulation environment model and the optimization algorithm model, and realizes efficient iterative optimization of the antenna weight parameters through the cooperation of the two.

[0093] Specifically, see Appendix Figure 2 The simulation environment model of this embodiment includes two parts: a data visualization model and an intelligent agent environment model. The data visualization model is generated by a 5G signal beam three-dimensional channel simulator. Figure 3 Figures a and b show the 3D channel simulation effects of the 5G Massive MIMO 3D channel simulation model under different antenna weight parameter configurations. Figure 3 The petal-shaped areas in Figures a and b represent the visualized network signals. The ultimate goal of this three-dimensional channel simulation model is to build a virtual communication environment on a computing device, which can quickly output the signal strength and signal-to-noise ratio of the user device under different base station locations, different antenna parameters, etc. After the intelligent agent side (i.e., the base station side) generates an action A (i.e., the antenna weight parameter configuration scheme) and sends it to the three-dimensional channel simulation model, it will receive the simulation results such as the signal coverage value returned by the three-dimensional channel simulation model, and the reward R corresponding to the action can be determined based on the returned data.

[0094] It is understandable that the three-dimensional channel simulation model, as an interface for obtaining states, actions, and rewards, needs to be further encapsulated in the simulation environment, that is, to build a complete intelligent agent environment.

[0095] The agent environment model includes two parts: result cache and efficient action representation. In order to speed up the acquisition of simulation results, this embodiment adds a result cache mechanism, which can save all states S (i.e., base station location, power consumption and other information), actions A (i.e., antenna weight parameter configuration information), and rewards R (i.e., communication performance data). For existing action parameters, the corresponding rewards are obtained directly from the result cache without the need to recalculate the simulation results through the three-dimensional channel simulation model. Therefore, the setting of the result cache mechanism can reduce the amount of repeated calculations and improve simulation efficiency.

[0096] In addition, in actual application, the result cache data can also be used as historical experience data to optimize the preheating of the algorithm model, that is, pre-training, to improve the learning efficiency of the algorithm.

[0097] Since the Massive MIMO optimization problem has many state (such as power consumption, position, etc.) and action (such as horizontal beam width, vertical beam width, direction angle and downtilt angle, etc.) parameters, in order to reduce the difficulty of solving the optimization problem, this embodiment efficiently represents and models the environmental state and action, so as to represent and learn the complex environmental characteristics around the base station and the high-dimensional action parameters.

[0098] To this end, this embodiment adopts multidimensional scaling (MDS) to characterize the learning mechanism, describes the similarity between variables according to the distance of high-dimensional variables, and thus uses low-dimensional space to represent high-dimensional multivariate data to achieve dimensionality reduction processing.

[0099] When the MDS algorithm runs for the first time, it starts with the initial action set A ini A random initial action A is selected and sent to the agent environment. The initial action here is the configuration parameter used in the actual business. The agent environment will return a reward R to the optimization algorithm model. The optimization algorithm model selects the next action A based on the current reward R. i ′, the action output by the optimization algorithm model is a three-dimensional low-dimensional continuous action, which will be converted into a discrete action A after being represented by the efficient action of the agent environment model. i , the actions after characterization learning will be applied to the three-dimensional channel simulation model to continue the simulation calculation to output the corresponding rewards and feed back to the optimization algorithm model.

[0100] After multiple cycles, the optimization algorithm model obtains actions and rewards from the simulation environment model and learns repeatedly, and finally learns the optimal antenna weight parameters.

[0101] Considering that the data-driven optimization algorithm model needs to interact massively with the simulation environment model during the iterative optimization training process, resulting in huge computational overhead, it is necessary to improve the utilization efficiency of simulation samples as much as possible.

[0102] To this end, the optimization algorithm used in the optimization algorithm model of the embodiment of the present invention combines Bayesian optimization learning and the Contextual Bandit method, namely the GP-based contextual bandits algorithm, which can achieve a more efficient and targeted action space exploration strategy, reduce the number of interactive simulations of three-dimensional channel simulation, accelerate algorithm training and optimization, and quickly solve the optimal antenna weight parameters.

[0103] The optimization algorithm model is based on the optimization principle of the GP-based contextual bandits algorithm. The main thing is that the intelligent agent selects the next action to obtain the maximum reward value based on the observed data until the optimal action parameters are obtained.

[0104] The optimization algorithm model in this embodiment mainly includes two parts: a probabilistic proxy model and an acquisition function. The probabilistic proxy model is used to learn the objective function (i.e., reward function) of the large-scale MIMO optimization problem from the observed data, set the initial assumption prior, obtain incremental information through iterative interaction between the probabilistic proxy model and the agent environment, and then correct the assumption prior to obtain a more accurate probabilistic proxy model.

[0105] Taking the probabilistic agent model of Gaussian process as an example, assuming that the finite action parameter points are {x1,…,x N}, the corresponding output objective function (i.e. reward function) is {f(x1),…,f(x N )}, which can form a data set Among them, the output value obeys Gaussian distribution, that is:

[0106]

[0107] Among them, μ represents the mean vector, that is, μ = m(x); K(x, x′) represents the covariance matrix, also known as the Gaussian process kernel function, and its expression is:

[0108]

[0109] Among them, f(x) and f(x′) both represent reward values, and m(x) and m(x′) both represent mean vectors.

[0110] Assume that the training set is D(X,y), X={x1,…,x N},y={f(x1),…,f(x N )}, given a new input action x * , get y * =f(x * ), which is a Gaussian distribution, input action x * Then the corresponding objective function prediction mean and variance are obtained.

[0111] The acquisition function in this embodiment can effectively balance the relationship between exploring the uncertainty area to obtain more unknown information and using the existing information to find the current optimal action parameters, thereby reducing unnecessary objective function evaluations. In other words, the acquisition function used in this embodiment has higher exploration efficiency.

[0112] The acquisition function is composed of the observed data set The obtained posterior distribution is constructed and maximized, that is, the maximum acquisition function value is used to guide the selection of the next action candidate point, that is, to obtain the predicted value of the antenna weight parameter. The process can be expressed as:

[0113]

[0114] Among them, x t+1 is the next action, i.e. the predicted value of antenna weight parameter, α t is the reward function, x is the current action, D 1:t It is a historical observation dataset.

[0115] Taking the Gaussian process confidence boundary strategy (GP-UCB) as an example, the acquisition function is expressed as:

[0116]

[0117] The parameter β balances the expectation μ(x) and the variance σ(x).

[0118] It can be understood that the acquisition function in this embodiment can also be implemented in other forms, such as PI, EI, UCBTS and other function instances, which can also be used to construct the acquisition function.

[0119] The optimization algorithm model based on the GP-based contextual bandits algorithm constructs a Gaussian Process probability proxy model as a priori probability to fit the true target function, and then uses the Bandit algorithm as the acquisition function to actively select the most "potential" evaluation point for evaluation based on the fitting results. The iterative process of the entire optimization algorithm mainly includes:

[0120] First, the next most promising action x is selected according to the maximized acquisition function t+1 ;

[0121] Then, depending on the action x chosen t+1 Evaluate the reward function value, that is:

[0122] y t+1 =f(x t+1 )+ε t+1 (5)

[0123] Among them, y t+1 Indicates action x t+1 The corresponding reward function value, f(x t+1 ) represents the predicted value of the reward function, ε t+1 Indicates the noise value, which can be reasonably defined according to the actual application scenario or set to 0.

[0124] Finally, the newly obtained action and reward value pair {x t+1 ,y t+1}Add to historical observation set D 1:t In the process, the probability proxy model is updated to prepare for the next iteration until the action with the largest reward value is obtained, that is, the antenna weight parameters with the best communication performance.

[0125] It can be seen that the antenna weight parameter optimization method provided by the embodiment of the present invention models the antenna weight optimization problem of the Massive MIMO system as a black box model, and realizes the construction and evaluation of the objective function to be optimized based on the Bayesian optimization and contextual bandit algorithm. The intelligent agent outputs an action, and the Massive MIMO system returns a result as a reward value. The intelligent agent continuously interacts with the environment, evaluates the objective function in multiple iterations, and finally learns the optimal action parameters.

[0126] The intelligent agent can select the optimal antenna weights and wireless parameter configuration combination of the Massive MIMO system to achieve the best network performance and improve the cell communication coverage. This method can quickly learn the optimal antenna parameters based on a small amount of data samples when the simulation cost is high, greatly improving the optimization efficiency and simulation calculation cost.

[0127] Subsequent test simulation verification shows that the antenna weight parameter optimization method based on the ContextualBandit optimization algorithm applied to Massive MIMO 5G single base station has the advantages of short calculation time and high optimization efficiency. It can learn the approximate optimal antenna weight parameters through fewer interactive learning times of the optimization algorithm and the three-dimensional channel simulator. Compared with the current weight configuration scheme selected by artificial experience method, the optimization effect is significantly improved.

[0128] The following is a description of the antenna weight parameter optimization device provided by the present invention. The antenna weight parameter optimization device described below and the antenna weight parameter optimization method described above can be referenced to each other.

[0129] Figure 4 An antenna weight parameter optimization device provided by an embodiment of the present invention is shown, and the device includes:

[0130] An acquisition module 410 is used to acquire the state information and initial antenna weight parameters of the base station to be optimized;

[0131] The optimization module 420 is used to iteratively search for the optimal antenna weight parameters through a pre-built simulation environment model and an optimization algorithm model based on the state information and the initial antenna weight parameters to obtain the optimal antenna weight parameters;

[0132] The configuration module 430 is used to replace the initial antenna weight parameters with the optimal antenna weight parameters.

[0133] In an exemplary embodiment, the above-mentioned optimization module 420 is specifically used to: input the state information and initial antenna weight parameters into the simulation environment model to obtain the current communication performance data output by the simulation environment model; input the initial antenna weight parameters and the current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction values ​​output by the optimization algorithm model; input the state information and the antenna weight parameter prediction values ​​into the simulation environment model to obtain the corresponding communication performance data again; input the antenna weight parameter prediction values ​​and the corresponding communication performance data into the optimization algorithm model again to perform the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction values ​​are optimal, and the optimal antenna weight parameters are obtained.

[0134] Preferably, the optimization module 420 inputs the state information and initial antenna weight parameters into the simulation environment model, obtains the current communication performance data output by the simulation environment model, and can also be used to cache the state information, initial antenna weight parameters and current communication performance data.

[0135] In an exemplary embodiment, the optimization module 420 inputs the predicted antenna weight parameter values ​​into the simulation environment model, and before obtaining the corresponding communication performance data again, it can also be used to: perform characterization learning on the predicted antenna weight parameter values ​​to obtain discrete antenna weight data.

[0136] That is to say, the above-mentioned optimization module 420 is specifically used to: input the state information and the initial antenna weight parameters into the simulation environment model to obtain the current communication performance data output by the simulation environment model; input the initial antenna weight parameters and the current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction values ​​output by the optimization algorithm model; perform characterization learning on the antenna weight parameter prediction values ​​to obtain antenna weight discrete data; input the antenna weight discrete data into the simulation environment model to obtain the corresponding communication performance data again; input the antenna weight discrete data and the corresponding communication performance data into the optimization algorithm model again to perform the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction values ​​are optimal, and the optimal antenna weight parameters are obtained.

[0137] Preferably, the above-mentioned optimization module 420, before inputting the initial antenna weight parameters and current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction value output by the optimization algorithm model, can also be used to pre-train the optimization algorithm model through pre-acquired historical experience data.

[0138] In an exemplary embodiment, the optimization module 420 implements the pre-training function in the following manner, including:

[0139] The reward function is fitted using the previously acquired historical experience data as observation data; wherein the observation data includes antenna weight parameter samples and corresponding communication performance data samples, and the reward function is used to characterize the linear relationship between the communication performance data and the antenna weight parameter;

[0140] Based on the observed data and the reward function, the next antenna weight parameter to be trained is determined as a sample of the antenna weight parameter prediction value.

[0141] Specifically, the communication performance data in this embodiment may include signal strength, signal-to-noise ratio, and signal coverage value.

[0142] In this embodiment, the antenna weight parameters of the base station to be optimized may include horizontal beam width, vertical beam width, direction angle, and downtilt angle.

[0143] Specifically, in this embodiment, the base station to be optimized may be a 5G base station that adopts a large-scale multiple-input multiple-output system.

[0144] It can be seen that the antenna weight parameter optimization device provided in the embodiment of the present invention can learn the optimal result only through a small amount of data samples after simulation interaction, and the optimization efficiency is greatly improved; the device can be applicable to base stations in different application scenarios, and only the initial state information needs to be adjusted for re-optimization; historical experience data can be integrated, and a preheating link can be set in the optimization algorithm model. By preheating the optimization algorithm model with historical business data as the initial finite action, the debugging time can be reduced; a result caching mechanism is designed to avoid repeated simulation calculations, and the entire optimization process is more efficient and convenient.

[0145] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the antenna weight parameter optimization method, which includes: obtaining the state information and the initial antenna weight parameters of the base station to be optimized; based on the state information and the initial antenna weight parameters, iteratively searching through the pre-built simulation environment model and the optimization algorithm model to obtain the optimal antenna weight parameters; replacing the initial antenna weight parameters with the optimal antenna weight parameters.

[0146] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0147] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the antenna weight parameter optimization method provided by the above methods, the method including: obtaining the state information and initial antenna weight parameters of the base station to be optimized; based on the state information and the initial antenna weight parameters, iteratively searching through a pre-constructed simulation environment model and optimization algorithm model to obtain the optimal antenna weight parameters; replacing the initial antenna weight parameters with the optimal antenna weight parameters.

[0148] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the antenna weight parameter optimization method provided by the above-mentioned methods, the method comprising: obtaining state information and initial antenna weight parameters of the base station to be optimized; based on the state information and the initial antenna weight parameters, iteratively searching for the optimal antenna weight parameters through a pre-constructed simulation environment model and optimization algorithm model; and replacing the initial antenna weight parameters with the optimal antenna weight parameters.

[0149] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0150] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing antenna weight parameters, characterized in that: include: Obtaining status information and initial antenna weight parameters of the base station to be optimized; Based on the state information and the initial antenna weight parameters, iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain optimal antenna weight parameters; Replacing the initial antenna weight parameters with the optimal antenna weight parameters; Based on the state information and the initial antenna weight parameters, an iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters, including: Inputting the state information and the initial antenna weight parameters into a simulation environment model to obtain current communication performance data output by the simulation environment model; Inputting the initial antenna weight parameter and the current communication performance data into an optimization algorithm model to obtain a predicted value of the antenna weight parameter output by the optimization algorithm model; Inputting the state information and the predicted value of the antenna weight parameter into the simulation environment model, and obtaining corresponding communication performance data again; The antenna weight parameter prediction value and the corresponding communication performance data are input into the optimization algorithm model again to perform the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, thereby obtaining the optimal antenna weight parameter; After inputting the state information and the initial antenna weight parameters into the simulation environment model to obtain the current communication performance data output by the simulation environment model, the method further includes: Cache the state information, initial antenna weight parameters and current communication performance data; Before inputting the initial antenna weight parameter and the current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction value output by the optimization algorithm model, the method further includes: Pre-training the optimization algorithm model using pre-acquired historical experience data; Pre-training the optimization algorithm model using pre-acquired historical experience data includes: Using pre-acquired historical experience data as observation data, a reward function is fitted; wherein the observation data includes antenna weight parameter samples and corresponding communication performance data samples, and the reward function is used to characterize the linear relationship between the communication performance data and the antenna weight parameter; Based on the observation data and the reward function, a next antenna weight parameter to be trained is determined as a sample of the antenna weight parameter prediction value.

2. A method for optimizing antenna weight parameters, characterized in that: include: Obtaining status information and initial antenna weight parameters of the base station to be optimized; Based on the state information and the initial antenna weight parameters, iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain optimal antenna weight parameters; Replacing the initial antenna weight parameters with the optimal antenna weight parameters; Based on the state information and the initial antenna weight parameters, an iterative optimization is performed through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters, including: Inputting the state information and the initial antenna weight parameters into a simulation environment model to obtain current communication performance data output by the simulation environment model; Inputting the initial antenna weight parameter and the current communication performance data into an optimization algorithm model to obtain a predicted value of the antenna weight parameter output by the optimization algorithm model; Performing characterization learning on the predicted value of the antenna weight parameter to obtain discrete data of the antenna weight; Inputting the antenna weight discrete data into the simulation environment model to obtain corresponding communication performance data again; The antenna weight discrete data and the corresponding communication performance data are input into the optimization algorithm model again to perform the next antenna weight parameter prediction, until the communication performance data corresponding to the antenna weight parameter prediction value is optimal, and the optimal antenna weight parameter is obtained; After inputting the state information and the initial antenna weight parameters into the simulation environment model to obtain the current communication performance data output by the simulation environment model, the method further includes: Cache the state information, initial antenna weight parameters and current communication performance data; Before inputting the initial antenna weight parameter and the current communication performance data into the optimization algorithm model to obtain the antenna weight parameter prediction value output by the optimization algorithm model, the method further includes: Pre-training the optimization algorithm model using pre-acquired historical experience data; Pre-training the optimization algorithm model using pre-acquired historical experience data includes: Using pre-acquired historical experience data as observation data, a reward function is fitted; wherein the observation data includes antenna weight parameter samples and corresponding communication performance data samples, and the reward function is used to characterize the linear relationship between the communication performance data and the antenna weight parameter; Based on the observation data and the reward function, a next antenna weight parameter to be trained is determined as a sample of the antenna weight parameter prediction value.

3. The method for optimizing antenna weight parameters according to claim 1 or 2, characterized in that: The communication performance data includes signal strength, signal-to-noise ratio and signal coverage value; The antenna weight parameters of the base station to be optimized include horizontal beam width, vertical beam width, direction angle and downtilt angle.

4. The method for optimizing antenna weight parameters according to claim 1 or 2, characterized in that: The base station to be optimized is a 5G base station that adopts a large-scale multiple-input multiple-output system.

5. An antenna weight parameter optimization device, characterized in that: The device is used to execute the antenna weight parameter optimization method according to claim 1 or 2; the device comprises: An acquisition module is used to obtain the status information and initial antenna weight parameters of the base station to be optimized; An optimization module, configured to perform iterative optimization based on the state information and the initial antenna weight parameters through a pre-built simulation environment model and an optimization algorithm model to obtain the optimal antenna weight parameters; A configuration module is used to replace the initial antenna weight parameters with the optimal antenna weight parameters.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the antenna weight parameter optimization method according to any one of claims 1 to 4 are implemented.

Images