Robust network optimization method, related device and medium

Abstract

The embodiment of the invention provides a robust network optimization method, a related device and a medium. The method comprises the following steps: acquiring network sample data of a target network and antenna parameters of each cell in the target network; performing gradient estimation on the second optimization objective function through a zero-order estimation algorithm based on network sample data and antenna parameters to obtain an estimated gradient value; updating antenna parameters through gradient descent according to the estimated gradient value; respectively calculating a first value of a first optimization objective function and a second value of a second optimization objective function according to the updated antenna parameters; jumping to perform gradient estimation on the second optimization objective function through a zero-order estimation algorithm based on the network sample data and the antenna parameters to obtain an estimated gradient value until a preset condition is met so as to obtain a plurality of first values; and taking the antenna parameter corresponding to the minimum first value in the plurality of first values as a target antenna parameter. According to the embodiment of the invention, the efficiency and robustness of network optimization can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of communications, and in particular to a robust network optimization method, related devices and media. Background Art

[0002] With the rapid development of fifth-generation (5G) wireless cellular communications, network optimization plays a key role in improving 5G network performance. There are a large number of adjustable network parameters in 5G networks, such as antenna parameters, beam parameters, transmit power, etc. These network parameters need to be properly adjusted to align with the actual environment to maximize network performance.

[0003] In related technologies, a model of a real-world wireless network is constructed, and then simulation and optimization of network parameters are performed based on the constructed model, such as a digital twin network. However, the current network optimization algorithm does not fully utilize the randomness of the antenna network and has poor robustness. How to improve the efficiency and robustness of network optimization is an issue that needs to be discussed urgently. Summary of the invention

[0004] The embodiments of the present disclosure provide a robust network optimization method, related devices and media, aiming to improve the efficiency and robustness of network optimization.

[0005] In a first aspect, an embodiment of the present disclosure provides a robust network optimization method, the method comprising:

[0006] Acquire network sample data of a target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments;

[0007] Based on the network sample data and the antenna parameters, a second optimization objective function is gradient estimated by a zero-order estimation algorithm to obtain an estimated gradient value; wherein the first optimization objective function is constructed by a first quantile and a network efficiency optimization expectation, the first quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent an optimized mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters;

[0008] According to the estimated gradient value, updating the antenna parameters by gradient descent;

[0009] According to the updated antenna parameters, respectively calculating a first value of the first optimization objective function and a second value of the second optimization objective function, wherein the second optimization objective function is constructed by a second quantile and the network efficiency optimization expectation, and the second quantile is used to select a grid to be optimized;

[0010] Jumping to the step of performing gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value until a preset condition is met to obtain a plurality of the first values;

[0011] The antenna parameter corresponding to the smallest first value among the multiple first values ​​is used as the target antenna parameter.

[0012] In a second aspect, an embodiment of the present disclosure provides a robust network optimization device, including:

[0013] A parameter acquisition unit, configured to acquire network sample data of a target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments;

[0014] A gradient estimation unit, configured to perform gradient estimation on a second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value; wherein the first optimization objective function is constructed by a first quantile and a network efficiency optimization expectation, the first quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent an optimized mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters;

[0015] A gradient updating unit, configured to update the antenna parameters by gradient descent according to the estimated gradient value;

[0016] An optimization calculation unit, used to calculate a first value of the first optimization objective function and a second value of the second optimization objective function according to the updated antenna parameters, wherein the second optimization objective function is constructed by a second quantile and the network efficiency optimization expectation, and the second quantile is used to select a grid to be optimized;

[0017] an iterative jump unit, configured to jump to the step of performing gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value until a preset condition is met, so as to obtain a plurality of the first values;

[0018] A parameter determination unit is used to use the antenna parameter corresponding to the smallest first value among multiple first values ​​as the target antenna parameter.

[0019] Optionally, the robust network optimization device further includes a first quantile determination unit, wherein the first quantile determination unit is used to:

[0020] Obtaining a preset first loss function, wherein the first loss function is a non-smooth function;

[0021] Constructing a first quantile estimation function according to the first loss function, wherein the first quantile estimation function is used to output a first optimization variable that minimizes the mathematical expectation corresponding to the first loss function;

[0022] Inputting the network sample data into the first quantile estimation function to obtain the first optimization variable;

[0023] The first optimization variable is taken as the first quantile.

[0024] Optionally, the robust network optimization device further includes a second quantile determination unit, wherein the second quantile determination unit is used to:

[0025] Obtain a preset second loss function, wherein the second loss function is an upper bound proxy function corresponding to the first loss function;

[0026] According to the second loss function, construct a second quantile estimation function, wherein the second quantile estimation function is used to output a second optimization variable that minimizes the mathematical expectation corresponding to the second loss function;

[0027] Inputting the network sample data into the first quantile estimation function to obtain the second optimization variable;

[0028] The second optimization variable is used as the second quantile.

[0029] Optionally, the robust network optimization device further includes a network performance calculation unit, and the network performance calculation is used to:

[0030] The spectrum efficiency performance of the target network is determined according to an average of the spectrum efficiency performance of a first number of grids of the target network at a second number of moments.

[0031] Optionally, the first optimization objective function is as follows:

[0032]

[0033] in, For the first optimization objective function, Q τ is the first quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0034] Optionally, the second optimization objective function is as follows:

[0035]

[0036] in, For the second optimization objective function, is the second quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0037] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the robust network optimization method in the first aspect when executing the computer program.

[0038] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer executes the computer program, the robust network optimization method as in the first aspect is implemented.

[0039] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, comprising a computer program or computer instructions, wherein the computer program or the computer instructions are stored in a computer-readable storage medium, a processor of a computer device reads the computer program or the computer instructions from the computer-readable storage medium, and the processor executes the computer program or the computer instructions, so that the computer device performs the robust network optimization method as in the first aspect.

[0040] The present disclosure collects network sample data of a target network with two dimensions, space (grid) and time (moment). On the basis of the two-dimensional network sample data, an optimization objective function constructed by quantiles and network efficiency optimization expectations is set. The network efficiency optimization expectation is used to target the statistical characteristics of spectrum efficiency performance in time, thereby reducing the fluctuation of network performance in time; the quantile is used to target the statistical characteristics of spectrum efficiency performance in space, and the grids with lower spectrum efficiency performance in the first number of grids are optimized, thereby ensuring that most grids in space can achieve better spectrum efficiency performance. By iteratively optimizing the second optimization objective function set with a second quantile, and then determining the target antenna parameters based on the result of the second optimization objective function through the first optimization objective function with a first quantile, the target antenna parameters finally optimized are more in line with the actual situation of the target network, thereby improving the optimization efficiency. The robust network optimization method disclosed in the present disclosure can not only reduce time fluctuations, but also ensure that satisfactory spectrum efficiency performance is achieved in most geographic grids, and can improve the efficiency and robustness of network optimization.

[0041] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or understood by practicing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A system architecture diagram for a robust network optimization method according to an embodiment of the present disclosure;

[0043] Figure 2 A schematic diagram of a scenario of 5G network optimization in a certain area provided as an example of the present disclosure;

[0044] Figure 3 is a flowchart of a robust network optimization method according to an embodiment of the present disclosure;

[0045] Figure 4 A schematic diagram of a network optimization simulation result provided for an example of the present disclosure;

[0046] Figure 5 A schematic diagram of a network optimization simulation result provided for another example of the present disclosure;

[0047] Figure 6 is a module diagram of a robust network optimization device according to an embodiment of the present disclosure;

[0048] Figure 7 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical methods and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present disclosure and are not used to limit the present disclosure.

[0050] It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0051] In the description of the embodiments of the present disclosure, unless otherwise clearly defined, words such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above words in the embodiments of the present disclosure in combination with the specific content of the technical solution. In the embodiments of the present disclosure, words such as "further", "exemplarily" or "optionally" are used to indicate examples, illustrations or explanations, and should not be interpreted as being more preferred or more advantageous than other embodiments or design solutions. The use of words such as "further", "exemplarily" or "optionally" is intended to present related concepts in a specific way.

[0052] With the rapid development of fifth-generation (5G) wireless cellular communications, network optimization plays a key role in improving the performance of 5G networks. There are a large number of adjustable network parameters in 5G networks, such as antenna parameters, beam parameters, and transmit power. These network parameters need to be properly adjusted to align with the actual environment to maximize network performance. Since inappropriate network parameters may significantly degrade network performance and cause serious network outages, network optimization requires an offline approach. Traditionally, offline optimization of these network parameters relies on expertise and engineering experience, mainly through trial-and-error drive tests. Network experts usually try multiple network parameter configurations and conduct multiple rounds of drive test measurement activities during the network optimization process, which is time-consuming and labor-intensive and cannot achieve optimal network performance. Therefore, network optimization requires automated and intelligent strategies without human intervention. A promising approach is to build a model of a real-world wireless network and then perform simulation and optimization of network parameters based on the constructed model, such as a digital twin network. Inspired by this approach, current offline optimization methods for wireless networks are based on some mathematical models and advanced optimization algorithms.

[0053] In the related technology, some advanced algorithms that have been applied in network optimization include the following: Method 1, adjusting the downtilt angle and horizontal angle of the antenna to optimize coverage, by first making the objective function differentiable and then using a gradient-based method for optimization; Method 2, adjusting the horizontal angle of the antenna to optimize coverage, by defining a pseudo-metric horizontal angle solution space, and then using a meta-heuristic evolutionary algorithm to simulate natural evolution to find a suitable solution; Method 3, adjusting the downtilt angle of the antenna to optimize coverage and capacity, and optimizing the network by using a reinforcement learning algorithm, deep deterministic policy gradient (deepdeterministic policygradient).

[0054] However, the above methods do not fully utilize the randomness of wireless networks, do not consider robust optimization, and have poor robustness. How to improve the efficiency and robustness of network optimization is an issue that needs to be discussed urgently.

[0055] Based on this, the embodiment of the present disclosure provides a robust network optimization method, related devices and media. The embodiment of the present disclosure collects network sample data of a target network with two dimensions of space (grid) and time (moment). On the basis of the two-dimensional network sample data, an optimization objective function constructed by quantiles and network efficiency optimization expectations is set, and the statistical characteristics of the spectrum efficiency performance in time are targeted by the network efficiency optimization expectation, thereby reducing the fluctuation of network performance in time; the grids with lower spectrum efficiency performance in the first number of grids are optimized by quantiles for the statistical characteristics of the spectrum efficiency performance in space, thereby ensuring that most grids in space can achieve better spectrum efficiency performance. By iteratively optimizing the second optimization objective function with a second quantile, and then determining the target antenna parameters based on the result of the second optimization objective function by the first optimization objective function with a first quantile, the target antenna parameters finally optimized are more in line with the actual situation of the target network, thereby improving the optimization efficiency. The robust network optimization method disclosed in the present disclosure can not only reduce time fluctuations, but also ensure that satisfactory spectrum efficiency performance is achieved in most geographic grids, and can improve the efficiency and robustness of network optimization.

[0056] System architecture and scenario description of the application of the disclosed embodiments

[0057] Figure 1 1 is a system architecture diagram for the robust network optimization method according to an embodiment of the present disclosure, which includes a server 110, the Internet 120, and an object terminal 130, etc.

[0058] The server 110 refers to a computer system that can provide network optimization processing services. For example, it can obtain and output the optimized target antenna parameters based on the robust network optimization method provided in each embodiment of the present disclosure according to the network sample data from the target network and the initial antenna parameters of each cell in the target network, so that each cell in the target network can adjust its own antenna according to the target antenna parameters. Compared with the target terminal 130, the server 110 has higher requirements in terms of stability, security, performance, etc. The server 110 can be a high-performance computer in a network platform, a cluster of multiple high-performance computers, a part of a high-performance computer (such as a virtual machine), a combination of parts of multiple high-performance computers (such as virtual machines), etc. The server 110 can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0059] The object terminal 130 is a device that displays network optimization results, sample selection / input interface of network optimization, etc. to the object for the object to select the target network, set sampling conditions, etc. It includes various forms such as desktop computers, laptops, PDAs (personal digital assistants), mobile phones, car terminals, home theater terminals, and dedicated terminals. In addition, it can be a single device or a collection of multiple devices. For example, multiple devices are connected through a local area network and share a display device to work together, forming a terminal. The terminal can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0060] The embodiments of the present disclosure can be applied in various scenarios, such as Figure 2 A schematic diagram of a 5G network optimization scenario in a certain area is shown.

[0061] For example, Figure 2 As shown, the 5G network in the area is taken as the target network, assuming that the target network consists of Q cells 210, and the target network coverage area is divided into L grids.

[0062] The optimization process for the target network is as follows

[0063] Step 1: Obtain the spectrum efficiency performance of L grids in the target network at S times as network sample data, that is, the number of network sample data at this time is U=L*S. And obtain the initial antenna parameters of Q cells in the target network.

[0064] Step 2: Based on U network sample data and Q antenna parameters, a gradient estimation is performed on the second optimization objective function through a zero-order estimation algorithm to obtain an estimated gradient value; wherein the second optimization objective function is constructed through a second quantile and a network efficiency optimization expectation, the second quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent the optimization mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters;

[0065] Step 3: Update the antenna parameters by gradient descent according to the estimated gradient value;

[0066] Step 4: Calculate the first value of the first optimization objective function and the second value of the second optimization objective function according to the updated antenna parameters, wherein the first optimization objective function is constructed by the first quantile and the network efficiency optimization expectation, and the first quantile is used to select the grid to be optimized;

[0067] Step 5: Store the obtained first value and second value; wherein the second value is used to participate in the next gradient descent iterative calculation;

[0068] Step 6: loop through steps 2 to 5 for a preset number of times T to obtain multiple first values; use the antenna parameter corresponding to the smallest first value among the multiple first values ​​as the target antenna parameter;

[0069] Step 7: Based on the target antenna parameters, adjust the antennas of the Q cells in the target network to maximize the 5G network performance in the area.

[0070] It should be noted that steps 1-6 may be performed by the server 220. Step 7 may be that the server 220 sends the target antenna parameters to the corresponding cell so that the base station of the cell adjusts its own parameters; or the operation and maintenance personnel adjust the antennas of the base stations of the Q cells according to the target antenna parameters.

[0071] In the above example, network sample data of the target network with two dimensions of space (grid) and time (moment) are collected. On the basis of the two-dimensional network sample data, the optimization objective function constructed by the quantile and the network efficiency optimization expectation is set, and the network efficiency optimization expectation is used to target the statistical characteristics of the spectrum efficiency performance in time, thereby reducing the fluctuation of network performance in time; the quantile is used to target the statistical characteristics of the spectrum efficiency performance in space, and the grids with lower spectrum efficiency performance from the first number of grids are optimized, thereby ensuring that most of the grids in space can achieve better spectrum efficiency performance. By iteratively optimizing the second optimization objective function set with the second quantile, and then determining the target antenna parameters based on the result of the second optimization objective function through the first optimization objective function with the first quantile, the target antenna parameters finally optimized are more in line with the actual situation of the target network, thereby improving the optimization efficiency. The robust network optimization method disclosed in the present invention can not only reduce time fluctuations, but also ensure that satisfactory spectrum efficiency performance is achieved in most geographic grids, and can improve the efficiency and robustness of network optimization.

[0072] It should be understood that the above content only illustrates some application scenarios of the present disclosure. The business scenarios to which the present disclosure can be applied may include but are not limited to the specific embodiments listed above.

[0073] General description of the disclosed embodiments

[0074] It should be emphasized that the embodiments of the present disclosure can be applied to a variety of application scenarios, such as 5G network optimization, regional network optimization and other scenarios. In the related art, the current network optimization algorithm does not fully utilize the randomness of the antenna network and has poor robustness. Some embodiments of the present disclosure provide a robust network optimization method, related devices and media, which are intended to improve the efficiency and robustness of network optimization.

[0075] The robust network optimization method of the embodiment of the present disclosure may be executed on a server, may be executed on a terminal, or may be executed partially on a server and partially on a terminal.

[0076] like Figure 3 As shown, according to one embodiment of the present disclosure, the robust network optimization method includes at least but is not limited to:

[0077] Step 310: Obtain network sample data of the target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments;

[0078] Step 320: Based on the network sample data and antenna parameters, the second optimization objective function is estimated by a zero-order estimation algorithm to obtain an estimated gradient value; wherein the second optimization objective function is constructed by a second quantile and a network efficiency optimization expectation, the second quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent an optimized mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters;

[0079] Step 330: Update the antenna parameters by gradient descent according to the estimated gradient value;

[0080] Step 340: Calculate a first value of a first optimization objective function and a second value of a second optimization objective function according to the updated antenna parameters, wherein the first optimization objective function is constructed by a first quantile and a network efficiency optimization expectation, and the first quantile is used to select a grid to be optimized;

[0081] Step 350, jump to performing gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value (step 320), until a preset condition is met to obtain a plurality of first values;

[0082] Step 360: Use the antenna parameter corresponding to the smallest first value among the multiple first values ​​as the target antenna parameter.

[0083] Steps 310 to 360 are briefly described below.

[0084] In step 310, the target network refers to the network covered by the area where network optimization is required. A grid refers to a plurality of geographic grids obtained by geographically dividing the area where network optimization is required. A cell refers to each base station in the target network. In the present disclosure, a cell can be understood as a base station in the cell. Antenna parameters include the downtilt angle of the antenna and the horizontal angle of the antenna. Spectral efficiency performance, namely spectral efficiency (SE), is one of the important indicators for measuring the performance of a wireless communication system. It is defined as the transmission rate within a unit bandwidth, that is, how many bits of data can be transmitted per Hertz (Hz) of bandwidth.

[0085] For example, it is assumed that η represents the spectrum efficiency performance and ξ represents the antenna parameter. In the present disclosure, the task of network optimization is to optimize the spectrum efficiency performance of the target network by adjusting the antenna parameter ξ. The antenna parameter ξ of the target network can be expressed as:

[0086]

[0087] in, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, is the cell set of the target network,

[0088] In step 320, the zero-order estimation algorithm is an optimization method that does not require gradient information and is suitable for black-box optimization scenarios. That is, the objective function is a black box and its gradient information cannot be obtained. In the embodiment of the present disclosure, the zero-order estimation algorithm is used to obtain the estimated gradient value through function evaluation, so as to update the antenna parameters.

[0089] The second quantile refers to the grid of the target network, that is, the parameter of the spatial dimension of the network sample data. By setting the second quantile, the grids with lower spectral efficiency performance from the first number of grids of the target network are tuned, rather than directly tuning all the grids together, so that the overall spectral efficiency performance of the target network is better and more uniform, thereby improving the overall optimization efficiency. It should be noted that the specific definition of the second optimization objective function is reflected in the following embodiments and will not be repeated here.

[0090] Exemplarily, the network efficiency optimization expectation is expressed by the following formula:

[0091]

[0092] in, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number; in order to avoid antenna collision and overcoverage, δ1 and δ2 are set to the preset downtilt angle and horizontal angle adjustment thresholds respectively.

[0093] Formula (2) is mainly optimized by maximizing the mean of samples, where ε() is defined as the mean value, specifically:

[0094]

[0095] Wherein, U represents the total number of network sample data, U=L*S; Formula (3) is the average of U spectrum efficiency performances.

[0096] However, even if the average spectrum efficiency performance is very high, there will be some geographic grids at the edge of the cell with poor spectrum efficiency performance. In order to consider the spectrum efficiency performance of these poor grids, the quantile is set to incorporate spatial information (the impact of the grid's geographical location on the spectrum efficiency performance) into the optimization. Compared with the network efficiency optimization expectation, the quantile can provide more information about the distribution of the spectrum efficiency performance η, especially the tail behavior of the distribution of the random variable (spectrum efficiency performance η). The quantile can be expressed as the inverse function of the cumulative probability density function of the random variable (spectrum efficiency performance η).

[0097] For the spectrum efficiency performance η, its cumulative probability density function can be expressed as:

[0098]

[0099] in, is the cumulative probability density function of the spectrum efficiency performance η, Pr() represents the operation of taking probability, is the sampling value at a certain moment corresponding to the spectrum efficiency performance η.

[0100] Without loss of generality, combined with the cumulative probability density function of formula (4), for any 0<τ<1, the τ-quantile can be expressed as:

[0101]

[0102] For example, when τ=0.1, the τ-quantile represents a value. Among all network sample data, 10% of the samples are smaller than the τ-quantile value, and 90% of the samples are larger than the τ-quantile value.

[0103] In step 330, the estimated gradient value refers to the gradient value obtained by performing gradient estimation on the second optimization objective function through a zero-order estimation algorithm.

[0104] In step 340, the first value refers to the objective function value calculated by the first optimization objective function based on the updated antenna parameters. The second value refers to the objective function value calculated by the second optimization objective function based on the updated antenna parameters.

[0105] The difference between the second optimization objective function and the first optimization objective function is that the quantiles of the two are different, namely the first quantile and the second quantile. The first quantile and the second quantile can refer to the definition and explanation of quantiles in the above example, and the specific difference between the two is described in detail in the subsequent embodiments, which will not be repeated here.

[0106] In step 350, the preset condition refers to a condition that makes the antenna parameter optimization iteration converge. Exemplarily, the preset condition can be that the loop ends after the loop is executed T times, where T is a positive integer set according to the requirements. The preset condition can also be that the loop ends when the first value of the first optimization objective function is less than a preset threshold.

[0107] In step 360, based on the setting of the first optimization objective function / second optimization objective function disclosed in the present invention, the larger the value of the optimization objective function, the larger the spectrum efficiency performance corresponding to the target network. Therefore, the smallest first value among the first values ​​is the value with the largest spectrum efficiency performance corresponding to multiple loop iterations, so that the antenna parameters corresponding to the smallest first value are used as the optimal solution for the current optimization and as the target antenna parameters. After obtaining the target antenna parameters, the antennas of each cell in the target network can be set according to the target antenna parameters, so that the spectrum efficiency performance of the target network is optimized.

[0108] In the above-mentioned embodiments of steps 310 to 360, network sample data of the target network with two dimensions of space (grid) and time (moment) are collected. On the basis of the two-dimensional network sample data, the optimization objective function constructed by the quantile and the network efficiency optimization expectation is set, and the network efficiency optimization expectation is used to target the statistical characteristics of the spectrum efficiency performance in time, thereby reducing the fluctuation of network performance in time; the grids with lower spectrum efficiency performance in the first number of grids are optimized by the quantile for the statistical characteristics of the spectrum efficiency performance in space, thereby ensuring that most of the grids in space can achieve better spectrum efficiency performance. By iteratively optimizing the second optimization objective function with the second quantile, and then determining the target antenna parameters based on the result of the second optimization objective function by the first optimization objective function with the first quantile, the target antenna parameters finally optimized are more in line with the actual situation of the target network, thereby improving the optimization efficiency. The network optimization method disclosed in the present invention can not only reduce time fluctuations, but also ensure that satisfactory spectrum efficiency performance is achieved in most geographic grids, and can improve the efficiency and robustness of network optimization.

[0109] The above is a general description of step 310 to step 360. The specific implementation process of step 310 to step 360 will be described in detail below through various embodiments.

[0110] In one embodiment, the robust network optimization method further includes:

[0111] The spectrum efficiency performance of the target network is determined according to the average of the spectrum efficiency performance of the first number of grids of the target network at the second number of moments. Exemplarily, the spectrum efficiency performance of the target network may be obtained by using formula (3) provided in the above example.

[0112] In one embodiment, the first quantile is obtained by the following steps:

[0113] Obtaining a preset first loss function, wherein the first loss function is a non-smooth function;

[0114] According to the first loss function, construct a first quantile estimation function, wherein the first quantile estimation function is used to output a first optimization variable that minimizes the mathematical expectation corresponding to the first loss function;

[0115] Input the network sample data into the first quantile estimation function to obtain the first optimization variable;

[0116] The first optimized variable is taken as the first quantile.

[0117] In this embodiment, since the communication environment of the real 5G network is very complex, the cumulative density function of the spectrum efficiency performance in formula (4) cannot be obtained explicitly, so the value of the first quantile cannot be directly calculated by formula (5). In order to calculate the value of the first quantile that is more in line with the actual situation, we first define the first loss function ρ τ (), specifically:

[0118]

[0119] Among them, the first loss function ρ τ () is a non-smooth function, c is the independent variable, that is, the input of formula (6).

[0120] Furthermore, according to the first loss function ρ τ () Construct the first quantile estimation function Q η (τ) to calculate the value of the first quantile, specifically:

[0121]

[0122] Among them, z is the optimization variable.

[0123] It can be proved that by taking the derivative of h(z) in formula (7) and setting the partial derivative to zero, we can get τ-quantile as one of the solutions, so we can deduce:

[0124]

[0125] Finally, formula (8) is equivalent to:

[0126]

[0127] Based on formula (8) and formula (9), it can be determined that formula (7) can calculate the quantile defined by formula (5), that is, the optimization variable z obtained by formula (7) can be used as the first quantile.

[0128] After obtaining the first quantile estimation function, that is, formula (7), the network sample data Substituting into formula (7), where u∈U and U is the number of network sample data, the estimated value of the first quantile corresponding to the target network to be optimized can be calculated as follows:

[0129]

[0130] In one embodiment, based on the operation of calculating the mean of network sample data in formula (3) and the operation of calculating the first quantile in formula (10) provided in the above embodiment, a double-layer first optimization objective function based on expectation and quantile is designed, which is specifically as follows:

[0131]

[0132] in, is the first optimization objective function, Q τ is the first quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0133] In formula (11), the inner layer is to find the network efficiency optimization expectation, and the outer layer is to find the first quantile. The inner layer expectation is mainly aimed at the statistical characteristics of spectrum efficiency in time, and the outer layer quantile is mainly aimed at the statistical characteristics of spectrum efficiency in space. This robust design can not only mitigate time fluctuations, but also ensure satisfactory spectrum efficiency performance in most grids.

[0134] In one embodiment, the second quantile is obtained by the following steps:

[0135] Obtain a preset second loss function, wherein the second loss function is an upper bound proxy function corresponding to the first loss function;

[0136] According to the second loss function, construct a second quantile estimation function, wherein the second quantile estimation function is used to output a second optimization variable that minimizes the mathematical expectation corresponding to the second loss function;

[0137] Input the network sample data into the first quantile estimation function to obtain the second optimization variable;

[0138] The second optimization variable is taken as the second quantile.

[0139] In this embodiment, since the quantile in formula (10) is not smooth, this will affect the network optimization performance. Therefore, a smooth second loss function is set in the embodiment of the present disclosure, and the first optimization objective function can be rewritten into a smooth second optimization objective function by obtaining the second quantile estimation function of the smoothed quantile.

[0140] First, based on the first loss function (Formula (6)), the upper bound proxy function of Formula (6), i.e., the second loss function, is set to make Formula (6) smooth. Specifically, the second loss function S τ,γ ()for:

[0141]

[0142] Where γ is a non-negative smoothing factor. The larger the γ, the greater the degree of smoothness introduced. The upper bound proxy function defined by formula (12) is greater than or equal to the first loss function defined by formula (6). When γ approaches zero, the upper bound proxy function defined by formula (12) approaches the first loss function defined by formula (6). Furthermore, a smooth proxy function of the τ-quantile can be defined, that is, the smoothed second quantile estimation function

[0143]

[0144] in, is the network sample data, z is the optimization variable. From formula (10) and formula (13), it can be seen that formula (13) can improve the smoothness of network optimization, thereby improving the optimization performance.

[0145] In one embodiment, the operation of calculating the mean of network sample data in formula (3) and the operation of calculating the smoothed quantile in formula (13) provided in the above embodiment can change the first optimization objective function into the second optimization objective function as shown below:

[0146]

[0147] in, is the second optimization objective function, is the second quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0148] In one embodiment, in order to solve formula (14), since there is a black box mapping between antenna parameters and spectral efficiency performance, a zero-order estimation algorithm is used to estimate the gradient, as shown in the following formula (15):

[0149]

[0150] where μ is the estimated radius and μ>0.

[0151] Based on formula (15), the antenna parameters are updated using gradient descent, as shown in the following formula (16):

[0152]

[0153] Among them, Ω is the feasible domain set that satisfies the constraint conditions, and the constraint conditions are that the downtilt angle and horizontal angle of the antenna need to be within the corresponding adjustment threshold range.

[0154] For example, Figure 4 and Figure 5 As shown, four situations are compared, namely, before optimization, expectation optimization, joint expectation and quantile optimization (without smoothing), and joint expectation and quantile optimization (after smoothing). First of all, it can be seen that compared with before optimization, the three optimized situations, namely, expectation optimization, joint expectation and quantile optimization (without smoothing), and joint expectation and quantile optimization (after smoothing), can greatly reduce the geographic grid with poor spectrum efficiency performance. The curve of the robust network optimization method based on the expectation and quantile double-layer design of the embodiment of the present disclosure after smoothing is below all the curves, indicating that the robust network optimization method based on the expectation and quantile double-layer design of the embodiment of the present disclosure after smoothing has the best optimization performance. Moreover, both parameter settings of τ=0.1 and τ=0.3 can meet the robust optimization requirements of 5G networks in practice.

[0155] Description of the apparatus and device of the present disclosure

[0156] It is to be understood that, although the steps in the above-mentioned flowcharts are sequentially displayed according to the characterization of arrows, these steps are not necessarily executed in sequence according to the order of arrow characterization. Unless there is a clear description in the present embodiment, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0157] Figure 6A schematic diagram of the structure of a robust network optimization device provided in an embodiment of the present disclosure. The robust network optimization device 600 includes:

[0158] A parameter acquisition unit 610 is used to acquire network sample data of a target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments;

[0159] A gradient estimation unit 620 is used to perform gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters through a zero-order estimation algorithm to obtain an estimated gradient value; wherein the first optimization objective function is constructed through the first quantile and the network efficiency optimization expectation, the first quantile is used to select the grid to be optimized, and the network efficiency optimization expectation is used to represent the optimization mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters;

[0160] A gradient updating unit 630, configured to update antenna parameters by gradient descent according to the estimated gradient value;

[0161] The optimization calculation unit 640 is used to calculate the first value of the first optimization objective function and the second value of the second optimization objective function according to the updated antenna parameters, wherein the second optimization objective function is constructed by the second quantile and the network efficiency optimization expectation, and the second quantile is used to select the grid to be optimized;

[0162] An iterative jump unit 650 is used to jump to perform gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value until a preset condition is met to obtain a plurality of first values;

[0163] The parameter determination unit 660 is configured to use the antenna parameter corresponding to the smallest first value among the multiple first values ​​as the target antenna parameter.

[0164] Optionally, the robust network optimization device 600 further includes a first quantile determination unit (not shown in the figure), and the first quantile determination unit is used to:

[0165] Obtaining a preset first loss function, wherein the first loss function is a non-smooth function;

[0166] According to the first loss function, construct a first quantile estimation function, wherein the first quantile estimation function is used to output a first optimization variable that minimizes the mathematical expectation corresponding to the first loss function;

[0167] Input the network sample data into the first quantile estimation function to obtain the first optimization variable;

[0168] The first optimized variable is taken as the first quantile.

[0169] Optionally, the robust network optimization device 600 further includes a second quantile determination unit (not shown in the figure), and the second quantile determination unit is used to:

[0170] Obtain a preset second loss function, wherein the second loss function is an upper bound proxy function corresponding to the first loss function;

[0171] According to the second loss function, construct a second quantile estimation function, wherein the second quantile estimation function is used to output a second optimization variable that minimizes the mathematical expectation corresponding to the second loss function;

[0172] Input the network sample data into the first quantile estimation function to obtain the second optimization variable;

[0173] The second optimization variable is taken as the second quantile.

[0174] Optionally, the robust network optimization device 600 further includes a network performance calculation unit (not shown in the figure), and the network performance calculation is used to:

[0175] The spectrum efficiency performance of the target network is determined according to an average of the spectrum efficiency performance of the first number of grids of the target network at the second number of moments.

[0176] Optionally, the first optimization objective function is as follows:

[0177]

[0178] in, is the first optimization objective function, Q τ is the first quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0179] Optionally, the second optimization objective function is as follows:

[0180]

[0181] in, is the second optimization objective function, is the second quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, Q is the cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

[0182] Figure 6 Schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. Figure 6 As shown, the electronic device 2000 includes a memory 2100 and a processor 2200. The number of the memory 2100 and the processor 2200 may be one or more. Figure 6 In the example, a memory 2101 and a processor 2201 are taken; the memory 2101 and the processor 2201 in the network device can be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0183] The memory 2101 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the method provided in any embodiment of the present disclosure. The processor 2201 implements the robust network optimization method provided in any of the above embodiments by running the software programs, instructions and modules stored in the memory 2101.

[0184] The memory 2101 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function. In addition, the memory 2101 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device or other non-volatile solid-state storage device. In some instances, the memory 2101 further includes a memory remotely arranged relative to the processor 2201, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0185] An embodiment of the present disclosure further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the robust network optimization method provided by any embodiment of the present disclosure.

[0186] An embodiment of the present disclosure also provides a computer program product, including a computer program or computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer program or computer instructions from the computer-readable storage medium, and the processor executes the computer program or computer instructions, so that the computer device executes the robust network optimization method provided in any embodiment of the present disclosure.

[0187] The system architecture and application scenarios described in the embodiments of the present disclosure are intended to more clearly illustrate the technical solutions of the embodiments of the present disclosure, and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Those skilled in the art will appreciate that with the evolution of the system architecture and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present disclosure are equally applicable to similar technical problems.

[0188] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0189] In hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0190] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. By way of illustration, both applications running on a computing device and a computing device can be components. One or more components may reside in a process or an execution thread, and a component may be located on a computer or distributed between two or more computers. In addition, these components may be executed from various computer-readable media having various data structures stored thereon. Components may communicate, for example, through local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems through signals).

[0191] The above describes some embodiments of the present disclosure with reference to the accompanying drawings, but does not limit the scope of the present disclosure. Any modification, equivalent substitution and improvement made by those skilled in the art without departing from the scope and essence of the present disclosure shall be within the scope of the present disclosure.

Claims

1. A robust network optimization method, characterized in that: The method comprises: Acquire network sample data of a target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments; Based on the network sample data and the antenna parameters, a second optimization objective function is gradient estimated by a zero-order estimation algorithm to obtain an estimated gradient value; wherein the second optimization objective function is constructed by a second quantile and a network efficiency optimization expectation, the second quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent an optimized mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters; According to the estimated gradient value, updating the antenna parameters by gradient descent; According to the updated antenna parameters, respectively calculating a first value of a first optimization objective function and a second value of the second optimization objective function, wherein the first optimization objective function is constructed by a first quantile and the network efficiency optimization expectation, and the first quantile is used to select a grid to be optimized; Jumping to the step of performing gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value until a preset condition is met to obtain a plurality of the first values; The antenna parameter corresponding to the smallest first value among the multiple first values ​​is used as the target antenna parameter.

2. The method according to claim 1, characterized in that The first quantile is obtained by the following steps: Obtaining a preset first loss function, wherein the first loss function is a non-smooth function; Constructing a first quantile estimation function according to the first loss function, wherein the first quantile estimation function is used to output a first optimization variable that minimizes the mathematical expectation corresponding to the first loss function; Inputting the network sample data into the first quantile estimation function to obtain the first optimization variable; The first optimization variable is taken as the first quantile.

3. The method according to claim 2, characterized in that The second quantile is obtained by the following steps: Obtain a preset second loss function, wherein the second loss function is an upper bound proxy function corresponding to the first loss function; According to the second loss function, construct a second quantile estimation function, wherein the second quantile estimation function is used to output a second optimization variable that minimizes the mathematical expectation corresponding to the second loss function; Inputting the network sample data into the first quantile estimation function to obtain the second optimization variable; The second optimization variable is used as the second quantile.

4. The method according to claim 1, characterized in that: The method further comprises: The spectrum efficiency performance of the target network is determined according to an average of the spectrum efficiency performance of a first number of grids of the target network at a second number of moments.

5. The method according to claim 2, characterized in that: The first optimization objective function is as follows: in, For the first optimization objective function, is the first quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, is a cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

6. The method according to claim 3, characterized in that The second optimization objective function is as follows: in, For the second optimization objective function, is the second quantile estimation function, and are the antenna downtilt angle and horizontal angle of the qth cell respectively, is a cell set of the target network, represents the spectrum efficiency performance measured at the lth grid at the sth moment, L is the first number, S is the second number, ε() is used to calculate the average, δ1 and δ2 are the preset adjustment thresholds of the downtilt angle and the horizontal angle, respectively.

7. A robust network optimization device, characterized in that: include: A parameter acquisition unit, configured to acquire network sample data of a target network and antenna parameters of each cell in the target network, wherein the network sample data includes spectrum efficiency performance of a first number of grids of the target network at a second number of moments; A gradient estimation unit, configured to perform gradient estimation on a second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value; wherein the first optimization objective function is constructed by a first quantile and a network efficiency optimization expectation, the first quantile is used to select a grid to be optimized, and the network efficiency optimization expectation is used to represent an optimized mapping relationship between the spectrum efficiency performance of the target network and the antenna parameters; A gradient updating unit, configured to update the antenna parameters by gradient descent according to the estimated gradient value; An optimization calculation unit, used to calculate a first value of the first optimization objective function and a second value of the second optimization objective function according to the updated antenna parameters, wherein the second optimization objective function is constructed by a second quantile and the network efficiency optimization expectation, and the second quantile is used to select a grid to be optimized; an iterative jump unit, configured to jump to the step of performing gradient estimation on the second optimization objective function based on the network sample data and the antenna parameters by a zero-order estimation algorithm to obtain an estimated gradient value until a preset condition is met, so as to obtain a plurality of the first values; A parameter determination unit is used to use the antenna parameter corresponding to the smallest first value among multiple first values ​​as the target antenna parameter.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the robust network optimization method according to any one of claims 1 to 7 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer can implement the robust network optimization method according to any one of claims 1 to 7 when executing the computer program.

10. A computer program product comprising a computer program or computer instructions, characterized in that The computer program or the computer instructions are stored in a computer-readable storage medium, and the processor of a computer device reads the computer program or the computer instructions from the computer-readable storage medium, and the processor executes the computer program or the computer instructions, so that the computer device executes the robust network optimization method as described in any one of claims 1 to 7.

Images