Access network selection method and device

Abstract

The invention discloses an access network selection method and device. The method comprises the following steps: obtaining performance data of a candidate network; inputting the obtained performance data into a pre-trained fuzzy neural network (FNN) to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes initial network parameters by using an Aperture Aperture Algorithm (AVOA), a position vector in the AVOA represents a mean value and a variance of a second layer of Gaussian membership function and a fifth layer of Gaussian membership function, and a fitness function in the AVOA is a mean square error function of a fuzzy neural network model; and for any user, determining a target network suitable for the user to access according to the performance score of the candidate network covering the user. The technical scheme provided by the embodiment of the invention is suitable for heterogeneous network access selection problems in various scenes such as satellite-ground fusion, and can make dynamic decisions according to one or more given standards, thereby ensuring seamless and efficient connection to optimize user experience.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of communication technology, and in particular, to a method and device for selecting an access network. Background Art

[0002] The access selection problem in heterogeneous wireless networks refers to determining the most appropriate access network for user devices in an environment where multiple wireless networks are available. Traditional wireless networks include cellular networks (such as 4G, 5G), Wi-Fi, Bluetooth, etc. With the widespread application of new types of networks such as satellite networks and the Internet of Things, the access capabilities of user devices have gradually diversified, making the access selection problem more and more complicated.

[0003] The main goal of the heterogeneous wireless network access selection problem is to achieve seamless and efficient connectivity through dynamic decision making. There are already a variety of decision mechanisms in the prior art, such as: signal strength-based methods prioritize networks according to signal strength and connect to the network with the strongest signal; quality of service-based methods consider factors such as latency, bandwidth, and reliability to select the network that can best support the service level required by a specific application; load-aware methods allocate users among available networks to maintain a balanced load, prevent network congestion, and improve overall performance; and so on.

[0004] Although these methods are relatively simple to implement, the decisions are made based on only partial considerations of some criteria and do not always guarantee the best user experience. Summary of the invention

[0005] In view of the above technical problems, the embodiments of the present application provide a method and device for selecting an access network, and the technical solution is as follows:

[0006] According to a first aspect of an embodiment of the present application, a method for selecting an access network is provided, the method comprising:

[0007] Obtaining performance data of candidate networks;

[0008] Input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model;

[0009] For any user, a target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user.

[0010] According to a second aspect of an embodiment of the present application, there is provided an access network selection device, the device comprising:

[0011] A performance data obtaining unit, used to obtain performance data of a candidate network;

[0012] A performance scoring unit is used to input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model;

[0013] The selection output unit is used to determine, for any user, a target network suitable for the user to access based on the performance scores of the candidate networks covering the user.

[0014] The technical solution provided in the embodiment of the present application implements network access selection based on AVOA (African Vulture Optimization Algorithm) and FNN (Fuzzy Neural Network). It is applicable to heterogeneous network access selection problems in various scenarios such as satellite-ground integration, and can make dynamic decisions based on one or more given criteria, thereby ensuring seamless and efficient connections to optimize user experience.

[0015] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and do not limit the embodiments of the present application.

[0016] In addition, any embodiment of the present application does not need to achieve all of the above-mentioned effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is a schematic diagram of an access network selection system according to an embodiment of the present application;

[0019] Figure 2 It is a schematic diagram of the FNN model structure of an embodiment of the present application;

[0020] Figure 3 It is a flowchart of the AVOA algorithm of the embodiment of the present application;

[0021] Figure 4is another flow chart of the AVOA algorithm of the embodiment of the present application;

[0022] Figure 5 This is a schematic diagram of a satellite-ground integrated access scenario in an embodiment of the present application;

[0023] Figure 6 is another schematic diagram of the access network selection system according to an embodiment of the present application;

[0024] Figure 7 is another schematic diagram of the FNN model structure of an embodiment of the present application;

[0025] Figure 8 is a flowchart of an access network selection method according to an embodiment of the present application;

[0026] Figures 9 to 12 2 is a schematic diagram of the simulation results of the satellite-ground fusion access scenario algorithm according to an embodiment of the present application;

[0027] Fig.13 It is a structural diagram of an access network selection device according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be described in detail below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field should belong to the scope of protection.

[0029] In view of the problems existing in the existing access network selection scheme, the present application provides a network access selection scheme based on AVOA (African Vulture Optimization Algorithm) and FNN (Fuzzy Neural Network). It can make dynamic decisions based on one or more given criteria, thereby ensuring seamless and efficient connection to optimize user experience.

[0030] Figure 1 The figure shows a schematic diagram of the access network selection system provided by the present application. The whole system is implemented based on FNN, where the input of FNN includes two categories:

[0031] 1) Performance indicators of candidate access networks:

[0032] Specifically, here we refer to the performance indicators that need to be considered when selecting an access network, which can be one or more, such as network bandwidth, delay, signal strength, data speed, etc. If there are M candidate networks and N performance indicators, then this part of the input is an M×N dimensional vector.

[0033] 2) User type:

[0034] Different types of users have different access preference weights for different networks, which are independent of the performance of the candidate networks and can cover factors such as brand preference and price preference.

[0035] The role of FNN is to perform an overall score based on the candidate network's own performance indicators, and then output the final access selection result based on user preferences.

[0036] It is understandable that without considering the "user type", that is, assuming that all users have the same preferences, then Figure 1 The "user type" submodule in the input module can be omitted, and the output module will only make access selection based on the performance score of each candidate network. The "network performance score" submodule can also be regarded as an output module.

[0037] Figure 2 As shown, it is a schematic diagram of the model of the FNN provided by the present application, which mainly includes a five-layer structure. The first layer is the input layer, which is composed of input language nodes, and each input variable corresponds to an input language node. The second layer is the fuzzification layer, which maps the input variables to fuzzy variables through Gaussian membership functions. The third and fourth layers are fuzzy reasoning layers, which apply fuzzy reasoning rules. The fifth layer is the defuzzification layer, which is used to convert the fuzzy reasoning output into a common output variable.

[0038] The forward propagation process of each layer of the FNN used in this application is specifically described below. For the convenience of description, the i-th node of the k-th layer (k = 1, 2, 3, 4, 5) of the FNN is used as follows: Represents its input, Represents its output.

[0039] 1) First layer - input layer:

[0040]

[0041] The number of nodes in the first layer corresponds to all performance indicators of all candidate networks. For example, if there are 3 candidate networks in total and each candidate network considers 2 candidate indicators, the corresponding number of nodes is 6.

[0042] exist Figure 3 In the specific implementation shown, i=1, 2, 3, 4, 5, 6, j=1, Represents the jth value of the i-th node in the first layer.

[0043] 2) The second layer - the fuzzy layer:

[0044] The fuzzification step uses Gaussian membership function to convert the precise quantities of the six input linguistic variables into fuzzy quantities and maps them into fuzzy sets. In this embodiment, the number of fuzzy sets for each linguistic variable is set to 3, represented by low (L), medium (M), and high (H), respectively. That is to say, the number of nodes in the second layer is 3 times the number of nodes in the second layer. The input / output of each node is as follows:

[0045]

[0046]

[0047] exist Figure 3 In the specific implementation shown, i=1, 2, ..., 18, j=1, represents the jth value of the i-th node in the second layer, and Represents the mean and variance of the Gaussian membership function.

[0048] 3) The third layer - fuzzy reasoning layer:

[0049] The basic format of fuzzy rules follows the IF-THEN structure, i.e. IF (a set of conditions are met) THEN (a set of conclusions are derived). The conditions are the fuzzy sets of input linguistic variables from the fuzzification layer, and the conclusions are the fuzzy sets of candidate network scores.

[0050] In this embodiment, for 3 fuzzy sets and 6 input linguistic variables, there are 3 6 = 729 fuzzy rules, so there are 729 nodes in the third layer, and each node uses the minimum value of the second layer output as the input value, that is:

[0051]

[0052]

[0053] Where i=1,2,...,729,j=1,2,3,4,5,6, Represents the j-th value of the i-th node in the third layer.

[0054] 4) The fourth layer - normalization layer:

[0055] Since the number of fuzzy sets is set to 3 in this embodiment, and the number of potential candidate networks output is 3, the number of nodes in this layer is 3×3=9, and the input / output of each node is:

[0056]

[0057]

[0058] where i = 1, 2, ..., 9, G i is the set of nodes in the third layer connected to node i in the fourth layer, Represents the j-th value of the i-th node in the fourth layer.

[0059] 5) Fifth layer - output layer:

[0060] The regional center method is used to score the candidate networks as the output of the fuzzy neural network model. Since there are 3 candidate networks in this embodiment, the fifth layer has 3 nodes, and the input / output of each node is:

[0061]

[0062]

[0063] Where i = 1, 2, 3, B i is the set of nodes in the fourth layer connected to node i in the fifth layer, represents the jth value of the i-th node in the fifth layer, and Represents the mean and variance of the Gaussian membership function.

[0064] It is understandable that in Figure 3 In the network model shown, the number of nodes in each layer is determined based on the specific settings of "3 candidate networks, 2 performance indicators, and 3 fuzzy sets", which should not be understood as limitations on the solution of the present application.

[0065] For the second layer (fuzzification layer) in the above FNN, in order to obtain the optimal initial values ​​of the mean and variance of the Gaussian membership function, the present application uses the AVOA algorithm to optimize the above parameters.

[0066] AVOA is a metaheuristic optimization algorithm inspired by the foraging behavior of African vultures. It simulates the search and foraging behavior of African vultures in the natural environment, including how they use vision and smell to find food, and how they interact with each other. The algorithm explores the solution space by simulating these behavior patterns to find the optimal or near-optimal solution.

[0067] This application uses AVOA to optimize the initial parameters of the FNN model. Traditional parameter search methods such as grid search or random search may be inefficient and time-consuming. As an efficient global search algorithm, AVOA can quickly locate a position close to the optimal solution by simulating the foraging behavior of African vultures, thereby accelerating the parameter optimization process.

[0068] The implementation of the AVOA basic algorithm can refer to the relevant prior art and will not be described in detail in this application. Relative to the basic AVOA algorithm, this application also provides an improved AVOA algorithm. First, the maximum number of iterations is set to I, the number of populations is N, and the task of each iteration is to select the best vulture position in the N populations.

[0069] See also Figure 3 As shown, for the nth population (n≤N), the specific processing method for the i-th iteration may include the following steps:

[0070] S101, initialization to determine the best vulture in any population;

[0071] The population of the vulture group is recorded as N, and it is divided into two groups according to the fitness function. In this application, the fitness function is the mean square error function of the fuzzy neural network model. Each group searches for food together with different abilities. The two vultures with the strongest foraging and navigation abilities are called Best Vulture 1 and Best Vulture 2 respectively. The rest of the vultures will try to follow the best vulture when they are hungry and lack energy.

[0072] For each group of vultures, at the i-th iteration, the best solution R(i) is selected from the two best vultures, as shown below:

[0073]

[0074] Wherein L1 and L2 are two probability values, and satisfy 0<L1<1, 0<L2<1 and L1+L2=1.

[0075] S102, calculating the hunger rate of the vulture, and determining the updating stage of the vulture position according to the hunger rate;

[0076] The foraging behavior of vultures depends on their hunger rate. When vultures are hungry, they will compete with stronger vultures for food in an aggressive manner. Otherwise, vultures will travel long distances to forage for food and avoid conflicts. The calculation formula of the hunger rate F of vultures is defined as follows:

[0077]

[0078]

[0079] Among them, i is the current iteration number, I is the maximum iteration number, 0<rand1<1, -1<z<1 and -2<h<2 are random numbers, and t is an intermediate variable.

[0080] According to the currently calculated hunger rate, the subsequent vulture position update phase is determined. If |F|≥1, the improved AVOA algorithm will enter the exploration phase S103, otherwise it will enter the utilization phase S104.

[0081] S103, exploration phase;

[0082] The initial position of the vulture is initialized by the piecewise linear chaos mapping method, as shown in the following formula:

[0083]

[0084] Where z(1) is a random number between 0 and 1. The initial position P(i) of the vulture is as follows:

[0085] P(i)=(ub-lb)×z(i)+lb

[0086] Where lb = 0 and ub = 1 are the lower and upper limits of the vulture's position. During the exploration phase, the vulture's position vector P(i) is updated according to the preset search strategy.

[0087] The first strategy is to look for food near the best vulture, as shown below:

[0088] P(i+1)=R(i)-D(i)×F

[0089] D(i)=|X×R(i)-P(i)|

[0090] Where P(i+1) is the updated vulture position vector in the next iteration, F is the hunger rate, R(i) is the best vulture selected, D(i) is the random distance, and X = 2 × rand, 0 < rand < 1 is a random number. The introduction of X is to simulate the behavior of the best vulture moving randomly to protect food and avoid being attacked by other vultures.

[0091] The second strategy is to search for food randomly throughout the area, as shown in the following equation:

[0092] P(i+1)=R(i)-F+rand2×((ub-lb)×rand3+lb)

[0093] Among them, 0<rand2<1 and 0<rand3<1 are random numbers.

[0094] The improved AVOA algorithm selects the corresponding strategy by introducing the parameter P1, as shown in the following formula:

[0095]

[0096] in is a random number.

[0097] S104, utilization stage;

[0098] In the improved AVOA algorithm provided in this application, the utilization phase can be further divided into two sub-phases. First, a threshold T needs to be set to distinguish different sub-phases according to the relationship between |F| and T. For the sake of generality, the T value is set to 0.5 here, but this specific value should not be understood as a limitation on the scheme of this application. The two sub-phases are described in detail below:

[0099] 1) When 0.5≤|F|<1, the improved AVOA algorithm enters the first sub-stage of the utilization phase, which contains two different strategies: fighting for food and circling flight.

[0100] Fighting for food: When |F|≥0.5, the vultures have enough energy. Stronger vultures will fight each other for food sources, while weaker vultures will wait patiently and then steal food from stronger vultures. In this case, the position vector is updated as follows:

[0101] P(i+1)=D(i)×(F+rand4)-d(t)

[0102] d(t)=R(i)-P(i)

[0103] Among them, D(i)=|X×R(i)-P(i)|, 0<rand4<1 is a random number.

[0104] Circling flight: Each vulture will form a companion flight with one of the two best vultures. The flight equation is as follows:

[0105]

[0106] P(i+1)=R(i)-(S1+S2)

[0107] Among them, 0<rand5<1 and 0<rand6<1 are random numbers.

[0108] The improved AVOA algorithm selects the corresponding strategy by introducing the parameter P2, as shown in the following formula:

[0109]

[0110] in is a random number.

[0111] 2) When |F| < 0.5, the improved AVOA algorithm enters the second sub-stage of the utilization phase. This sub-stage contains two different strategies: the gathering of vultures at the food source and the fierce competition of vultures for food.

[0112] Vultures gather around the food source: Under this strategy, all vultures gather around the food source, as shown below:

[0113]

[0114]

[0115]

[0116] Where BestVulture1(i) and BestVulture2(i) are the best vultures in the current iteration.

[0117] Fierce competition for food among vultures: When |F| < 0.5, fierce competition for food will occur between the best vulture and other vultures, as shown in the following formula:

[0118] P(i+1)=R(i)-|d(t)|×F×Levy(d)

[0119] Where d(t) = R(i) - P(i) is the distance between the vulture and one of the two best vultures, and Levy(d) is a random number generated by the Levy flight pattern, as shown below:

[0120]

[0121] in,

[0122]

[0123] The value of β is generally 1.5, and u and v are random variable values ​​that follow the standard Gaussian distribution.

[0124] The improved AVOA algorithm selects the corresponding strategy by introducing parameter P3, as shown in the following formula:

[0125]

[0126] in is a random number.

[0127] Through the above S101 to S104, N populations are traversed respectively, and after completing I rounds of iteration, the final optimal vulture position is obtained. The complete iterative process can be found in Figure 4 shown.

[0128] Compared with the basic AVOA algorithm, the improved AVOA algorithm provided in the present application does not randomly select the initial position of the vulture when initializing the initial position of the vulture, but initializes the position of the vulture through a piecewise linear chaotic mapping method. Initializing the position of the vulture through a piecewise linear chaotic mapping method can ensure the diversity of the initial positions of the vultures and improve the global search capability of the algorithm, thereby effectively accelerating the convergence speed of the algorithm and being able to obtain the optimized FNN network parameters more quickly.

[0129] The implementation principles of FNN and AVOA adopted in the present application are respectively introduced above. The access network selection method of the present application is described in detail below in combination with specific application scenarios.

[0130] The satellite-ground converged network aims to enhance the connectivity and coverage of future networks by combining satellite communication systems and ground communication systems to create a global information network with seamless coverage. Initially, satellite communications were mainly used to provide long-distance communications and wide-area coverage in areas where ground infrastructure was lacking. The convergence of satellite and ground networks has become critical to ensuring ubiquitous and reliable connectivity, especially in remote or underserved areas. Advances in low-orbit satellite and high-throughput satellite technology have significantly improved the service capabilities of satellite communications. These developments provide lower latency and higher bandwidth, making the satellite-ground converged network more practical and efficient. The satellite-ground converged network plays a vital role in providing global connectivity, including services such as broadband Internet, mobile communications, and emergency communications.

[0131] Figure 5 As shown in FIG. 1 , a schematic diagram of a satellite-ground integrated network access selection scenario is shown, where both the satellite network and the 5G network are candidate networks for providing network access services. The signal coverage of the satellite and 5G networks overlaps, the mathematical model of the access request of the mobile user is the Poisson process arrival model, and the mathematical model of the user's location is the random movement model within the coverage area. It is assumed here that the mobile user has the ability to access multiple connections to different wireless networks. Based on network performance indicators and user types, the satellite-ground integrated network access selection method determines the user's access network type. The network performance indicators are updated regularly by the user and used as inputs to the access selection model, mainly including network bandwidth (BW) and delay (D). Different user types are described by preference vectors for candidate networks. The main optimization goal of the network access selection problem is to select the candidate network with the highest user satisfaction. Other secondary goals include improving system throughput and reducing network request blocking rate.

[0132] Here we assume that the candidate networks include a satellite network and two 5G networks. Figure 1 The access network selection system provided can specify the parameters of the "input module", such as Figure 6 As shown, where:

[0133] There are 6 performance indicators for candidate access networks: The bandwidth and latency correspond to one satellite network and two 5G networks respectively;

[0134] User type is recorded as vector Represents the user's preference weight for different networks.

[0135] based on Figure 2 The FNN model shown in FIG. 2 concretizes the input of the first layer into the performance indicators of the above six candidate access networks, such as Figure 7 As shown, here we still use three fuzzy sets, L, M, and H, and there are still three candidate networks and two performance indicators. Therefore Figure 8 The number of nodes in each layer is shown as Figure 2 The above description is exactly the same as that shown in the figure, and no repetition is given here.

[0136] The AVOA algorithm is used to optimize the initial network parameters of the FNN. The AVOA algorithm here can be the basic AVOA algorithm or the improved AVOA algorithm mentioned above, where the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model.

[0137] See also Figure 8 As shown, based on the system provided above, the access network selection method provided in this application may include the following steps:

[0138] S201, obtaining performance data of a candidate network;

[0139] S202, input the obtained performance data into FNN, and after completing forward propagation in the fuzzy neural network, obtain the performance scores of different candidate networks. This process can be referred to the previous description of the input and output of each layer of FNN, which will not be repeated here.

[0140] S203: Determine a target network suitable for the current target user to access based on the type of the current user and the performance scores of the candidate networks.

[0141] The output module takes the user type and network score of the current user as input, where the user type is recorded as Represents the user's preference weight for each network, and the network score is recorded as Different situations are considered according to the user's location. When user i is only covered by the satellite network, the access selection result is O i =sat. When user i is covered by both the satellite network and the 5G1 (or 5G2) network, if (or ), user i is connected to the satellite network (i.e. i=sat). Otherwise, user i connects to the 5G network, i.e. i =5G1(or O i =5G2). When user i is covered by satellite and two 5G networks at the same time, user i is connected to The network corresponding to the maximum value in .

[0142] It should be noted that, considering that the performance of each candidate network is relatively stable as a whole, the above steps S201-202 do not need to be recalculated every time a selection is required. According to actual needs, the performance of the candidate network can be scored according to a certain period (e.g., every hour, every day, etc.).

[0143] It is understandable that in actual applications, simple rules and other methods can also be used to characterize the user's preference for different networks. Therefore, the above-mentioned method of characterizing using weights should not be understood as limiting the present application. In addition, as mentioned above, the user type here is used to characterize the personal preferences of different users for different networks. If it is believed that there is no difference in the personal preferences of users for networks, the "user type" factor can be ignored, and access selection can be made directly based on the performance score of the network covered by the current user.

[0144] In order to verify the effectiveness of the proposed network access selection method, this application is verified by simulation. In the simulation scenario, consider a circular simulation area with a radius of 500 meters, which can be covered by a single satellite beam, a coverage radius of 300 meters for a single 5G base station, 100 users, a moving speed of 3 meters per second, a satellite network bandwidth of 10-100MHz, a latency of 25-100 milliseconds, a 5G network bandwidth of 20-20MHz, a latency of 10-50 milliseconds, a user request rate of 1-10Mbps, a real-time user's request duration of 30 seconds, and a non-real-time user's request duration of 60 seconds. Real-time users prefer terrestrial 5G networks because of their lower latency. Non-real-time users prefer larger bandwidths and can sacrifice latency. To describe this difference, the weight vectors of real-time users and non-real-time users are set to (0.2, 0.4, 0.4) and (0.4, 0.3, 0.3), respectively. The hyperparameter settings of the improved AVOA algorithm are as follows, where N is 50, the maximum number of iterations is 1000, L1 is 0.8, L2 is 0.2, p1 is 0.6, p2 is 0.4, and p3 is 0.6.

[0145] Fig. 9 This is a comparison chart of the convergence speed of different optimization algorithms, in which the comparison object is the Dragonfly Algorithm (DA). The horizontal axis in the figure is the iteration round, and the vertical axis is the error of the optimization target. Compared with the traditional Dragonfly Optimization Algorithm (DA), the improved AVOA algorithm provided in this application has a faster convergence speed.

[0146] Fig.10 This is a comparison chart of the blocking rates of four access selection algorithms, where the comparison objects include: Fuzzy Logic method, unoptimized fuzzy neural network (FNN), and DA algorithm optimized fuzzy neural network (DA-FNN). In the figure, the horizontal axis is the arrival rate of user requests (number of requests / second), and the vertical axis is the blocking rate of user requests. The higher the blocking rate, the more user requests cannot be served. As the arrival rate of user requests increases, the blocking rate will also increase. It can be seen that the network selection method of improved AVOA optimized fuzzy neural network has a lower network request blocking rate.

[0147] Fig.11 This is a comparison chart of the system throughput of four access selection algorithms. The horizontal axis in the figure is the arrival rate of user requests (number of requests / second), and the vertical axis is the system throughput. The higher the system throughput, the more user requests the system serves. As the arrival rate of user requests increases, the system throughput will also increase. It can be seen that the network selection method of the improved AVOA optimized fuzzy neural network has a higher system throughput.

[0148] Fig.12 This is a user satisfaction comparison chart of four access selection algorithms. The horizontal axis in the figure is the arrival rate of user requests (number of requests / second), and the vertical axis is the user satisfaction rate. The higher the user satisfaction rate, the higher the proportion of users served by the system who can obtain network selection results that meet their preferences. When the user request arrival rate is relatively low, the user satisfaction rate can reach 100%, and all users can meet their network selection preferences. As the user request arrival rate increases, the user satisfaction rate will decrease. It can be seen that the improved AVOA optimized fuzzy neural network network selection method has a higher user satisfaction rate.

[0149] Corresponding to the above method embodiment, the present application embodiment also provides an access network selection device, see Fig.13 As shown, the device may include:

[0150] The performance data obtaining unit 110 is used to obtain the performance data of the candidate network;

[0151] The performance scoring unit 120 is used to input the obtained performance data into the pre-trained FNN to obtain the performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model; FNN is a fuzzy neural network, and AVOA is an African vulture optimization algorithm;

[0152] The selection output unit 130 is used to determine, for any user, a target network suitable for the user to access according to the performance scores of the candidate networks covering the user.

[0153] In a specific implementation of the present application, in the stage of optimizing the initial network parameters using AVOA, the initial position of the vulture is initialized by a piecewise linear chaotic mapping device, as shown in the following formula:

[0154]

[0155] Where z(1) is a random number between 0 and 5, and the initial position P(i) of the vulture is as follows:

[0156] P(i)=(ub-lb)×z(i)+lb

[0157] lb=0 and ub=1 are the lower and upper limits of the vulture position.

[0158] In a specific embodiment of the present application, the pre-trained FNN may include: an input layer, a fuzzification layer, a fuzzy reasoning layer, a normalization layer, and an output layer;

[0159] The number of nodes in the input layer is M×N, where M is the number of candidate networks and N is the network performance index used for decision making;

[0160] The number of nodes in the fuzzification layer is M×N×K, where K is the number of fuzzy sets used in the fuzzification layer;

[0161] The number of nodes in the fuzzy reasoning layer is K M×N ;

[0162] The number of nodes in the normalization layer is L×M.

[0163] The number of nodes in the output layer is M.

[0164] In a specific implementation of the present application, the selection output unit may be specifically used for:

[0165] Obtain user preference information for different candidate networks

[0166] A target network suitable for the user to access is determined based on the performance scores and preference information of the candidate networks covering the user.

[0167] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0168] The embodiment of the present application also provides a computer device, which at least includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aforementioned access network selection method when executing the program. The method at least includes:

[0169] Obtaining performance data of candidate networks;

[0170] Input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model; FNN is a fuzzy neural network, and AVOA is an African vulture optimization algorithm;

[0171] For any user, a target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user.

[0172] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the aforementioned access network selection method is implemented. The method at least includes:

[0173] Obtaining performance data of candidate networks;

[0174] Input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model; FNN is a fuzzy neural network, and AVOA is an African vulture optimization algorithm;

[0175] For any user, a target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user.

[0176] It can be known from the description of the above implementation mode that a person skilled in the art can clearly understand that the embodiment of the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the embodiment of the present application or some parts of the embodiment.

[0177] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts refer to the partial description of the method embodiment. The device embodiment described above is merely schematic, wherein the modules described as separate components may or may not be physically separated, and the functions of each module can be implemented in the same one or more software and / or hardware when implementing the embodiment scheme of the present application. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the embodiment scheme. Those of ordinary skill in the art can understand and implement it without paying creative work.

[0178] The above is only a specific implementation of the embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the embodiment of the present application. These improvements and modifications should also be regarded as the protection scope of the embodiment of the present application.

Claims

1. A method for selecting an access network, characterized in that: The method includes: Obtaining performance data of candidate networks; Input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model; FNN is a fuzzy neural network, and AVOA is an African vulture optimization algorithm; For any user, a target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user.

2. According to the method of claim 1, in the stage of optimizing the initial network parameters by using AVOA, the initial position of the vulture is initialized by a piecewise linear chaotic mapping method, as shown in the following formula: Where z(1) is a random number between 0 and 1, and the initial position P(i) of the vulture is as follows: P(i)=(ub-lb)×z(i)+lb lb=0 and ub=1 are the lower and upper limits of the vulture position.

3. The method according to claim 1, wherein the pre-trained FNN comprises: Input layer, fuzzification layer, fuzzy reasoning layer, normalization layer and output layer; The number of nodes in the input layer is M×N, where M is the number of candidate networks and N is the network performance index used for decision-making; The number of nodes in the fuzzification layer is M×N×K, where K is the number of fuzzy sets used in the fuzzification layer; The number of nodes in the fuzzy reasoning layer is K M×N ; The number of nodes in the normalization layer is L×M. The number of nodes in the output layer is M.

4. The method according to claim 1, wherein determining a target network suitable for the user to access based on the performance scores of the candidate networks covering the user comprises: Obtain user preference information for different candidate networks A target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user and the preference information.

5. An access network selection device, characterized in that: The device includes: A performance data obtaining unit, used to obtain performance data of a candidate network; A performance scoring unit is used to input the obtained performance data into a pre-trained FNN to obtain performance scores of different candidate networks; wherein the pre-trained FNN optimizes the initial network parameters using the AVOA algorithm: the position vector in the AVOA algorithm represents the mean and variance of the second and fifth layer Gaussian membership functions, and the fitness function in the AVOA algorithm is the mean square error function of the fuzzy neural network model; FNN is a fuzzy neural network, and AVOA is an African vulture optimization algorithm; The selection output unit is used to determine, for any user, a target network suitable for the user to access based on the performance scores of the candidate networks covering the user.

6. According to the device of claim 5, in the stage of optimizing the initial network parameters by using AVOA, the initial position of the vulture is initialized by a piecewise linear chaotic mapping device, as shown in the following formula: Where z(1) is a random number between 0 and 5, and the initial position P(i) of the vulture is as follows: P(i)=(ub-lb)×z(i)+lb lb=0 and ub=1 are the lower and upper limits of the vulture position.

7. The apparatus according to claim 5, wherein the pre-trained FNN comprises: Input layer, fuzzification layer, fuzzy reasoning layer, normalization layer and output layer; The number of nodes in the input layer is M×N, where M is the number of candidate networks and N is the network performance index used for decision-making; The number of nodes in the fuzzification layer is M×N×K, where K is the number of fuzzy sets used in the fuzzification layer; The number of nodes in the fuzzy reasoning layer is K M×N ; The number of nodes in the normalization layer is L×M. The number of nodes in the output layer is M.

8. The device according to claim 5, wherein the selection output unit is specifically configured to: Obtain user preference information for different candidate networks A target network suitable for the user to access is determined based on the performance scores of the candidate networks covering the user and the preference information.

9. A computer 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 method according to any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images