Network construction system and method, recording medium, program product and control system

Abstract

The present invention provides a recursive Bayesian network construction system, construction method, computer-readable recording medium, non-transitory computer program product, and wireless network control system. The system is executed by a processor to: establish an initial group and set the initial group as the current group; generate a network for each combination pattern in the current group and establish a corresponding recursive Bayesian network to obtain a set of recursive Bayesian networks corresponding to the current group; evolve the current group using an evolutionary algorithm and a fitness function to obtain a next group; determine whether a termination condition is satisfied based on the fitness function and the set of recursive Bayesian networks corresponding to the current group; and, if the termination condition is not satisfied, repeat the aforementioned steps. If the termination condition is satisfied, select a solution network in the current group as a task network based on the fitness function.

Description

Technical Field

[0001] The present invention relates to a genetic algorithm technology applied to a recursive Bayesian network, and more particularly to a recursive Bayesian network applicable to a wireless network control system. Background Art

[0002] Self-organizing network (SON) refers to the automation of mobile networks and the minimization of human intervention in cellular / wireless network management. This concept was introduced by 3GPP in Release 8. The main goals of self-organizing networks can be roughly divided into three points. First, to introduce intelligence and autonomous adaptability into cellular networks; second, to reduce capital and operating expenditures; and third, to improve network performance in terms of network capacity, coverage, services provided and experience. However, many operations of existing self-organizing network solutions on the market are still done manually (for example, network failures are usually repaired directly); and many operations are still done manually (for example, network failures are usually repaired directly by engineers) and some open challenges remain unresolved, such as the coordination of self-organizing network functions, or the proper resolution of the trade-offs between centralized and decentralized self-organizing network implementations.

[0003] Recurrent Bayesian networks (RBNs) have the advantage of leveraging prior knowledge to model causal relationships between problem features as dependencies between random variables, thus handling the inherent uncertainty in the problem. However, training a large RBN typically requires a large number of training samples to learn useful information from the data. Furthermore, the structure of the RBN to be trained is predetermined, which can lead to inefficient utilization of computing resources. Summary of the Invention

[0004] In view of this, some embodiments of the present invention provide a recursive Bayesian network construction system, a recursive Bayesian network construction method, a computer-readable recording medium, a non-transitory computer program product, and a wireless network control system to improve the existing technical problems.

[0005] One embodiment of the present invention provides a recursive Bayesian network construction system comprising at least one processor. The at least one processor is configured to perform the following steps to generate a task network: establishing an initial population, wherein the initial population includes a plurality of combination pattern generation networks, and setting the initial population as a current population; establishing a corresponding recursive Bayesian network for each combination pattern generation network in the current population to obtain a set of recursive Bayesian networks corresponding to the current population; evolving the current population using an evolutionary algorithm and a fitness function to obtain a next population, and setting the next population as the current population; determining whether a termination condition is satisfied based on the fitness function and the set of recursive Bayesian networks corresponding to the current population; and, in response to the termination condition not being satisfied, repeating the aforementioned steps. In response to the termination condition being satisfied, selecting a solution network in the current population as the task network based on the fitness function.

[0006] One embodiment of the present invention provides a recursive Bayesian network construction method, which is executed by a processor. The recursive Bayesian network construction method is used to generate a task network. The recursive Bayesian network construction method includes: establishing an initial group, wherein the initial group includes multiple combination pattern generation networks, and setting the initial group as the current group; establishing a corresponding recursive Bayesian network for each combination pattern generation network in the current group to obtain a recursive Bayesian network set corresponding to the current group; evolving the current group using an evolutionary algorithm and a fitness function to obtain a next group, and setting the next group as the current group; judging whether a termination condition is satisfied based on the fitness function and the recursive Bayesian network set corresponding to the current group; and repeating the aforementioned steps in response to the termination condition not being satisfied, and selecting a solution network in the current group as the task network based on the fitness function in response to the termination condition being satisfied.

[0007] Some embodiments of the present invention provide a computer-readable medium storing a program and a non-transitory computer program product. When a processor loads and executes the program, the aforementioned recursive Bayesian network construction method can be implemented.

[0008] One embodiment of the present invention provides a wireless network control system. The wireless network control system uses the aforementioned recursive Bayesian network to construct a task network generated by the system. The wireless network control system is coupled to a radio access network (RAN) and includes an optimizer and a control unit. The optimizer is configured to receive the network status and target policy of the radio access network and output an optimal configuration to the radio access network to configure and optimize the radio access network. The control unit includes the aforementioned task network. The optimizer inputs multiple configuration feature vectors of each solution to the control unit based on multiple solutions in a solution space. The task network evaluates at least one key performance indicator (KPI) of each configuration feature vector to obtain a KPI value for each configuration feature vector. The optimizer selects an optimal solution in the solution space based on the KPI value of each configuration feature vector. The optimizer then outputs an optimal configuration to the radio access network based on the optimal solution.

[0009] Based on the foregoing, some embodiments of the present invention provide a recursive Bayesian network construction system, construction method, computer-readable recording medium, and non-transitory computer program product. By evolving a combinatorial pattern to generate a population of networks, a suitable recursive Bayesian network can be efficiently constructed. Some embodiments of the present invention also provide a wireless network control system that automatically implements a self-organizing network solution by applying the task network generated by the recursive Bayesian network construction system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 FIG. 1 is a block diagram of a recursive Bayesian network construction system according to an embodiment of the present invention.

[0011] Figure 2 FIG. 1 is a schematic diagram of a network generated by a combination mode according to an embodiment of the present invention.

[0012] Figure 3 FIG. 4 is a schematic diagram of neural network encoding according to an embodiment of the present invention.

[0013] Figure 4 FIG. 4 is a schematic diagram illustrating a neural network mutation process according to an embodiment of the present invention.

[0014] Figure 5 FIG. 4 is a schematic diagram of a neural network exchange process according to an embodiment of the present invention.

[0015] Figure 6-1 FIG. 4 is a schematic diagram of a Bayesian network according to an embodiment of the present invention.

[0016] Figure 6-2FIG. 4 is a schematic diagram of a Bayesian network according to an embodiment of the present invention.

[0017] Figure 7 FIG. 4 is a schematic diagram of a recursive Bayesian network according to an embodiment of the present invention.

[0018] Figure 8 FIG. 1 is a schematic diagram illustrating a recursive Bayesian network construction according to an embodiment of the present invention.

[0019] Figure 9 FIG. 4 is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention.

[0020] Figure 10 FIG. 4 is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention.

[0021] Figure 11 FIG. 4 is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention.

[0022] Figure 12 It is a structural block diagram of an electronic device according to one embodiment of the present specification.

[0023] Figure 13 This is a block diagram of a wireless network control system that uses a recursive Bayesian network to construct a task network.

[0024] The description of the accompanying drawings is as follows:

[0025] 100: Recursive Bayesian Network Construction System

[0026] 101, 1201: Processor

[0027] 102: Memory

[0028] 205, 303, 805: Combined pattern generation network

[0029] 2051, 701, 704, 707, 710: input layer

[0030] 2052, 702, 705, 708, 711: hidden layers

[0031] 2053, 703, 706, 709, 712: output layer

[0032] w, w1, w2: output

[0033] 201, 202, 203, 204, 801, 802, 803, 804: Nodes

[0034] 301: Node Gene

[0035] 302, 401, 403, 501, 502, 505: Linking genes

[0036] 402, 404, 503, 504, 506: Neural Network

[0037] 5011, 5012, 5013, 5014, 5021, 5022, 5023, 5024, 5025, 5051, 5052, 5053, 5054, 5055, 5056: Connected genetic elements

[0038] 601-608: Nodes

[0039] 609: Probability Table

[0040] 610-612: Conditional Probability Table

[0041] 700: Recursive Bayesian Network

[0042] 7021, 7051, 7081, 7111, 800: Bayesian networks

[0043] 7022, 7052, 7082, 7112: recursive layers

[0044] 1200: Electronic equipment

[0045] 1202: Internal memory

[0046] 1203: Non-volatile memory

[0047] 1300: Wireless Network Control System

[0048] 1301: Wireless Access Network

[0049] 1302: Optimizer

[0050] 1303: Control Unit

[0051] 1304: Mission Network

[0052] P(.): probability

[0053] P(X1): probability of random variable X1

[0054] P(X2|X1), P(X3|X1), P(X4|X2,X3): conditional probability

[0055] t: time

[0056] X1, X2, X3, X4, C, S, R, W: random variables

[0057] T, F: elements of the set

[0058] M1, M2, M3: Node status

[0059] S901~S905, S1001~S1002, S1101: Steps DETAILED DESCRIPTION

[0060] The aforementioned and other technical contents, features and effects of the present invention will be clearly presented in the following detailed description of the embodiments with reference to the drawings. The proportions or sizes of the various elements in the drawings are expressed in an exaggerated, omitted or approximate manner for the understanding and reading of those familiar with this technology. They are not intended to limit the limiting conditions under which the present invention can be implemented, so they have no technical substantive significance. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed by the present invention without affecting the effects and purposes that can be achieved by the present invention. The same reference numerals will be used to represent the same or similar elements in all drawings. The term "coupling" or "connection" mentioned in the following embodiments may refer to any direct or indirect, wired or wireless connection means.

[0061] Figure 1 FIG is a block diagram of a recursive Bayesian network construction system according to an embodiment of the present invention. Figure 1 , the recursive Bayesian network construction system 100 includes a processor 101 and a memory 102. The memory 102 is configured to store intermediate results that need to be temporarily stored during the calculation process. It is worth noting that although Figure 1 Only a single processor 101 is shown, but the recursive Bayesian network construction system 100 may also be configured with multiple processors to increase the computing speed.

[0062] The following describes in detail the recursive Bayesian network construction method according to an embodiment of the present invention and how the modules of the recursive Bayesian network construction system 100 operate in coordination with each other with reference to the drawings.

[0063] Figure 2 FIG. 4 is a schematic diagram of a compositional pattern-producing network (CPPN) according to an embodiment of the present invention. Figure 3 FIG. 4 is a schematic diagram of neural network encoding according to an embodiment of the present invention. Figure 4 FIG. 1 is a schematic diagram illustrating a neural network mutation process according to an embodiment of the present invention. Figure 5 FIG. 1 is a schematic diagram illustrating a neural network crossover process according to an embodiment of the present invention. Figure 9 This is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention. Figures 2 to 5 as well as Figure 9 .

[0064] exist Figure 9 In the illustrated embodiment, the combined pattern generation network is a neural network. Figure 2 Taking the illustrated combined pattern generation network 205 as an example, the combined pattern generation network 205 includes an input layer 2051, a hidden layer 2052, and an output layer 2053. Each node in the hidden layer 2052 can use a different activation function, such as a trigonometric sine function, a Gaussian function, etc. It is worth noting that the combined pattern generation network 205 is merely illustrative of a general combined pattern generation network, and the present invention is not limited to the connection relationships of the combined pattern generation network 205.

[0065] Using a combined pattern generation network, the connection pattern of the neural network can be stored as a four-dimensional hypercube, where each point of the four-dimensional hypercube can be represented as (x1, y1, x2, y2), where x1, y1, x2, y2 are the coordinate components of the point of the four-dimensional hypercube. Each point encodes the connection between two nodes. The combined pattern generation network 205 can be viewed as a four-dimensional function CPPN (1):

[0066] w=CPPN(x1,y1,x2,y2) (1).

[0067] Wherein w is the output value of the combination pattern generation network 205. Figure 2 As shown in the figure, the connection weight between node (x1, y1) and node (x2, y2) can be obtained by inputting (x1, y1, x2, y2) into the four-dimensional function CPPN (1). The output w is the connection weight between node (x1, y1) and node (x2, y2). Figure 2 In the example shown, since the output of the four-dimensional function CPPN(1) corresponding to (-1, 1, 1, 1) is w1, the connection weight between node 201 (encoded as (-1, 1)) and node 202 (encoded as (1, 1)) is w1; since the output of the four-dimensional function CPPN(1) corresponding to (0, -1, 1, -1) is w2, the connection weight between node 203 (encoded as (0, -1)) and node 204 (encoded as (1, -1)) is w2. It is worth noting that in this embodiment, a minimum threshold w is first given. min , when the output w is less than the minimum threshold w min In the above manner, the combination of the pattern generation network 205 and the corresponding four-dimensional function CPPN (1) can establish a corresponding neural network topology.

[0068] See also Figure 3 , a general neural network can be represented by numbering its nodes and connections so that the neural network can perform evolutionary operations. Figure 3 For the example shown, by numbering the nodes of the combined pattern generation network 205 from 1 to 7, a numbered combined pattern generation network 303 can be obtained. The numbered combined pattern generation network 303 can be represented by a node gene 301 and a link gene 302. "Input" recorded in the node gene 301 indicates that this node is an input node; "Output" recorded in the node gene 301 indicates that this node is an output node; and "Hidden" recorded in the node gene 301 indicates that this node is a hidden node. "In" recorded in the link gene 302 indicates the starting node of the link; "Out" recorded in the link gene 302 indicates the ending node of the link; "Weight" recorded in the link gene 302 indicates the weight value of the link; and "Innovation" recorded in the link gene 302 indicates the innovation value of the link. The innovation value of the link is used to match the genome when performing the exchange process in the neural network evolution operation. The use of the innovation value of the link will be further described in the following embodiments.

[0069] In the Figure 3 After the neural network is represented as node genes 301 and link genes 302, the neural network can further perform evolutionary operations. Figure 4 As shown, neural network 402 has linker gene 401. Linker gene elements are randomly added to linker gene 401, such as linker gene 403, which has an "in" value of 2, an "out" value of 5, a "weight" of 0.1, and an "innovation" value of 6. Linker gene 403 corresponds to neural network 404. The process by which neural network 402 evolves into neural network 404 is called a neural network mutation process.

[0070] It is worth noting that the aforementioned random addition of elements linked to gene 401 can be achieved by using general software to simulate random functions (such as the random() function of the random module in Python) to randomly generate appropriate "in", "out", "weight" and "innovation".

[0071] See also Figure 5 , neural network 503 has link gene 501, which has link gene elements 5011, 5012, 5013, and 5014. Neural network 504 has link gene 502, which has link gene elements 5021, 5022, 5023, 5024, and 5025. To perform the exchange process in the neural network evolution operation on neural network 503 and neural network 504 to generate a new neural network, first, the link gene elements of the link genes of the two are matched based on the "innovation" number. Figure 5In the illustrated example, link gene element 5011 matches link gene element 5021, link gene element 5012 matches link gene element 5022, and link gene element 5013 matches link gene element 5023. Link gene element 5014 does not match the link gene element of link gene 502, and link gene elements 5024 and 5025 do not match the link gene element of link gene 501. Of these matching gene elements, one is randomly selected and added to the link gene of the new neural network.

[0072] In the aforementioned example, link gene element 5011 matches link gene element 5021, so one of link gene element 5011 and link gene element 5021 is randomly selected and placed into link gene 505 of neural network 506. Link gene element 5012 matches link gene element 5022, so one of link gene element 5012 and link gene element 5022 is randomly selected and placed into link gene 505 of neural network 506, and so on. Link gene elements that do not match, such as link gene elements 5014, 5024, and 5025, are unconditionally placed into link gene 505 of neural network 506. Ultimately, link gene 505 and the corresponding neural network 506 are obtained, where link gene 505 includes link gene elements 5051, 5052, 5053, 5054, 5055, and 5056. Neural network 503 and neural network 504 are referred to as parents of neural network 506 , and neural network 506 is referred to as offspring of neural network 503 and neural network 504 .

[0073] The foregoing Figures 3 to 5 The algorithm depicted for evolving the topology and connection weights of a neural network is called the NeuroEvolution of Augmenting Topologies method.

[0074] It is worth noting that the aforementioned process of randomly selecting one of the linked gene elements 5012 and 5022 can be achieved by simulating a random function using general software (e.g., the sample() function of the random module in Python) to randomly link the gene element 5012 and the linked gene element 5022.

[0075] exist Figure 9 In step S901, the processor 101 creates an initial group and sets the initial group as the current group. The initial group includes a plurality of combination pattern generation networks randomly generated by the processor 101 (for example, Figure 2The number of initial populations is a preset population number M, where M is a positive integer. The processor 101 sets the aforementioned initial population as the current population. The processor 101 also receives a fitness function from the outside to evolve the current population. It is worth noting that the processor 101 can use the random simulation functions provided by general software (such as the functions of the random module in Python) to randomly generate the values ​​of the node genes and the link genes (such as Figure 3 (shown) generates a network using a random generation combination pattern.

[0076] Figure 6-1 、 Figure 6-2 is a schematic diagram of a Bayesian network according to an embodiment of the present invention. Before explaining step S902, please refer to Figure 6-1 and Figure 6-2 The Bayesian network is a probabilistic graphical model that uses directed acyclic graphs (DAGs) to learn a set of random variables {X1, X2, ..., X n} and the properties of its n sets of conditional probability distributions (CPDs). For example, the Bayesian network can be used to represent the probability relationship between a disease and its related symptoms. When a certain symptom is known, the Bayesian network can be used to calculate the probability of various possible diseases. Generally speaking, the nodes of the Bayesian network represent random variables, which can be observable variables, latent variables, unknown parameters, etc. The arrow connecting two nodes represents whether the two random variables are causally related or non-conditionally independent. If there is no arrow connecting the two nodes, the random variables are said to be conditionally independent of each other. If two nodes are connected by a single arrow, it means that one node is a parent node and the other node is a child node. The child node will generate a conditional probability value for the parent node.

[0077] Figure 6-1A Bayesian network is shown, including node 601 (represented by random variable X1), node 602 (represented by random variable X2), node 603 (represented by random variable X3), and node 604 (represented by random variable X4). The arrow connecting node 601 (represented by random variable X1) and node 602 (represented by random variable X2) indicates that these two random variables X1 and X2 are causally related or unconditionally independent. Furthermore, node 601 (represented by random variable X1) is a parent node, and another node 602 (represented by random variable X2) is a child node. Node 601 (represented by random variable X1) and node 602 (represented by random variable X2) generate a conditional probability P(X2|X1), node 601 (represented by random variable X1) and node 603 (represented by random variable X3) generate a conditional probability P(X3|X1), and node 604 (represented by random variable X4) and node 602 (represented by random variable X2) and node 603 (represented by random variable X3) generate a conditional probability P(X4|X 2, P(X1) represents the probability of node 601 (represented by the random variable X1).

[0078] In one embodiment of the present invention, the Bayesian network is discrete in nature. In this embodiment, the conditional probability can be represented by a conditional probability table (CPT). The sum of the probabilities in any column of the CPT must be 1. Figure 6-2 Taking the Bayesian network shown as an example, node 605 represents the random variable C for cloudy weather, node 606 represents the random variable S for watering, node 607 represents the random variable R for raining, and node 608 represents the random variable W for wet grass. The random variables C, S, R, and W all take values ​​from the set {T, F}, where T and F are elements of the aforementioned sets, with T representing occurrence and F representing non-occurrence. P(C) is represented by probability table 609, P(S|C) by conditional probability table 610, P(R|C) by conditional probability table 611, and P(W|S,R) by conditional probability table 612.

[0079] Figure 7 FIG is a schematic diagram of a recurrent Bayesian network (Recurrent Neural Networks) according to an embodiment of the present invention. Figure 7 As shown, recursive Bayesian network 700 includes input layer 701, hidden layer 702, and output layer 703. Hidden layer 702 includes a Bayesian network 7021 and a recursive layer 7022. Recursive layer 7022 is used to record the state of the Bayesian network and add the recorded state at the next moment to the input at the next moment. Figure 7The following diagram illustrates the operation of the recurrent Bayesian network 700 as it is expanded over time, where the output layer 706 at time t-1, the output layer 709 at time t, and the output layer 712 at time t+1 correspond to the output layer 703; the input layer 704 at time t-1, the input layer 707 at time t, and the input layer 710 at time t+1 correspond to the input layer 701; the hidden layer 705 at time t-1, the hidden layer 708 at time t, and the hidden layer 711 at time t+1 correspond to the hidden layer 702; the Bayesian network 7051 and the recurrent layer 7052 of the hidden layer 705 at time t-1, the Bayesian network 7081 and the recurrent layer 7082 of the hidden layer 708 at time t, and the Bayesian network 7111 and the recurrent layer 7112 of the hidden layer 711 at time t+1 correspond to the Bayesian network 7021 and the recurrent layer 7022 of the hidden layer 702. At time t-1, hidden layer 705 stores node state M1 in recursive layer 7052. At time t, node state M1 is input to hidden layer 708. Similarly, at time t, hidden layer 705 stores node state M2 in recursive layer 7082. At time t+1, node state M2 is input to hidden layer 711. Similarly, at time t+1, hidden layer 711 stores node state M3 in recursive layer 7112, and so on.

[0080] Figure 8 This is a schematic diagram of a recursive Bayesian network construction according to an embodiment of the present invention. Figure 8 and Figure 9 In step S902, the processor 101 generates a network for each combination pattern in the current group and establishes a corresponding recursive Bayesian network. Figure 8 The above method of establishing the corresponding recursive Bayesian network is described as an example. For the combined pattern generation network 805, for all time t, the combined pattern generation network 805 takes all possible two nodes of the Bayesian network 800 of the recursive Bayesian network at time t as input (such as Figure 8The illustrated node 801 (represented by random variable X1) and node 802 (represented by random variable X2) obtain the conditional probability P(X2|X1). It is worth noting that if the two input nodes have no causal relationship or are conditionally independent, the output of the combined pattern generation network 805 is 0. In this way, a Bayesian network 800 of a recursive Bayesian network at time t can be established through the combined pattern generation network 805, wherein the Bayesian network 800 of the recursive Bayesian network at time t includes node 801 (represented by random variable X1), node 802 (represented by random variable X2), node 803 (represented by random variable X3) and node 804 (represented by random variable X4); and node 802 (represented by random variable X2) has a node relationship with its parent node. The conditional probability of node 801 (represented by random variable X1) is P(X2|X1), the conditional probability of node 803 (represented by random variable X3) for its parent node 801 (represented by random variable X1) is P(X3|X1), and the conditional probability of node 804 (represented by random variable X4) for its parent nodes 802 (represented by random variable X2) and 803 (represented by random variable X3) is P(X4|X2,X3). P(X1) represents the probability of node 801 (represented by random variable X1).

[0081] Through the above-mentioned establishment method, the processor 101 can establish a corresponding recursive Bayesian network for each combination pattern generation network in the current group. Therefore, the processor 101 can finally obtain a set of recursive Bayesian networks corresponding to the current group.

[0082] In step S903, the processor 101 uses an evolutionary algorithm and a fitness function to evolve the current group to obtain a next group, and sets the next group as the current group. In one embodiment of the present invention, the processor 101 uses the aforementioned fitness function to calculate the fitness value of each combination pattern generation network in the current group, and uses the fitness value of each combination pattern generation network in the current group to sort each combination pattern generation network in the current group. The processor 101 uses the sorting result to select N combination pattern generation networks in the current group with larger fitness, where N is a positive integer and N is less than the aforementioned number of groups M. The processor 101 then uses the aforementioned fitness function to calculate the fitness value of each combination pattern generation network in the current group, and uses the fitness value of each combination pattern generation network in the current group to sort each combination pattern generation network in the current group. The processor 101 uses the sorting result to select N combination pattern generation networks in the current group with larger fitness, where N is a positive integer and N is less than the aforementioned number of groups M. Figure 4 、 Figure 5 The illustrated neural network evolution operation performs the mutation and exchange processes on the selected N combination pattern generation networks to expand the N combination pattern generation networks to a population number M to obtain the next population. The processor 101 then sets the next population as the current population to proceed to the next step.

[0083] In one embodiment of the present invention, the processor 101 utilizes the Figure 4 、 Figure 5The illustrated process of performing the mutation and swapping processes on the selected N combination pattern generation networks in the neural network evolution operation includes processor 101 randomly performing the mutation and swapping processes on the N combination pattern generation networks according to a swapping probability. For example, if the swapping probability is 0.8, processor 101 has an 80% probability of performing the swapping process on the N combination pattern generation networks and an 80% probability of performing the mutation process on the N combination pattern generation networks.

[0084] It is worth noting that as long as the above Figures 3 to 5 The illustrated topology-enhancing neuroevolutionary algorithms for evolving the topology structure and connection weights of a neural network can all be applied to evolving the aforementioned current population, and the present invention is not limited to the aforementioned embodiments.

[0085] In step S904, processor 101 determines whether a termination condition is met based on the fitness function and the set of recursive Bayesian networks corresponding to the current population. In this embodiment, the termination condition is that the number of repetitions of steps S902 and S903 has reached a predetermined value, or that, based on the fitness function, the current population contains a candidate network whose fitness value is greater than a predetermined fitness value. If the termination condition is not met, processor 101 executes steps S902 and S903 again. If the termination condition is met, processor 101 executes step S905.

[0086] In step S905, since the current group already contains at least one candidate network whose fitness value is greater than a predetermined fitness value, the processor 101 selects a solution network in the current group whose fitness value is greater than the predetermined fitness value as the task network according to the fitness function.

[0087] Figure 10 FIG is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention. Figure 10 In the illustrated embodiment, the Bayesian network is discrete in nature. Step S902 includes steps S1001 and S1002. In step S1001, the processor 101 selects a current combination pattern generation network in the current group. In step S1002, the processor 101 uses the outputs of the current combination pattern generation network regarding the child nodes and the parent nodes corresponding to the child nodes in the corresponding recursive Bayesian network to establish a conditional probability table for the parent nodes corresponding to the child nodes in the corresponding recursive Bayesian network, thereby establishing the corresponding recursive Bayesian network.

[0088] Figure 11 FIG is a flow chart of a recursive Bayesian network construction method according to an embodiment of the present invention. Figure 11In the illustrated embodiment, step S901 further includes step S1101. In step S1101, processor 101 randomly generates an initial population based on a parameter. In this embodiment, the parameter is a value generated by processor 101 based on the time at which step S901 is started. Processor 101 then initializes a common software random function (e.g., the random() function in the Python random module) based on this value to randomly generate the initial population. Processor 101 then sets the initial population as the current population.

[0089] Figure 12 FIG is a block diagram of an electronic device according to an embodiment of the present specification. Figure 12 As shown, at the hardware level, electronic device 1200 includes a processor 1201, internal memory 1202, and non-volatile memory 1203. Internal memory 1202 is, for example, random-access memory (RAM). Non-volatile memory is, for example, at least one disk drive. Of course, electronic device 1200 may also include hardware required for other functions.

[0090] The internal memory 1202 and the non-volatile memory 1203 are used to store programs. The programs may include program codes, and the program codes include computer operating instructions. The internal memory 1202 and the non-volatile memory 1203 provide instructions and data to the processor 1201. The processor 1201 reads the corresponding computer program from the non-volatile memory 1203 into the internal memory 1202 and then runs it. The processor 1201 is specifically used to execute Figures 9 to 13 The steps described.

[0091] The processor 1201 may be an integrated circuit chip with signal processing capabilities. During implementation, the methods and steps disclosed in the aforementioned embodiments may be completed by hardware integrated logic circuits in the processor 801 or by software instructions. The processor 1201 may be a general-purpose processor, including a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, and may implement or execute the methods and steps disclosed in the aforementioned embodiments.

[0092] The embodiments of this specification also provide a computer-readable storage medium, which stores at least one instruction. When the at least one instruction is executed by the processor 1201 of the electronic device 1200, the processor 1201 of the electronic device 1200 can execute the various methods and steps disclosed in the aforementioned embodiments.

[0093] Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other internal memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmitting media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0094] Figure 13 This is a block diagram of a wireless network control system that uses the aforementioned recursive Bayesian network to construct a task network. Figure 13 The wireless network control system 1300 is externally coupled to a radio access network (RAN) 1301. The wireless network control system 1300 includes a recursive Bayesian network construction system 100, an optimizer 1302, and a control unit 1303. The optimizer 1302 is configured to receive the network status of the RAN 1301 and an external target policy. The optimizer 1302 is configured to output an optimal configuration to the RAN 1301 to configure and optimize the RAN 1301. The control unit 1303 is configured to receive and store a task network 1304 from the recursive Bayesian network construction system 100. The recursive Bayesian network construction system 100 generates the task network 1304 using pre-collected data related to the RAN 1301 and a fitness function associated with the RAN 1301. Task network 1304 supports both the second-level (intervention) and third-level (counterfactual) causal reasoning models. Therefore, task network 1304 can be used to answer questions such as, "Given a time series of radio access network configuration feature vectors, what key performance indicators will radio access network 1301 achieve?"

[0095] Based on multiple solutions in the solution space, optimizer 1302 inputs multiple configuration feature vectors for each solution to control unit 1303. Task network 1304 then evaluates the key performance indicators (KPIs) of each input configuration feature vector to obtain a KPI value for each configuration feature vector. Optimizer 1302 selects the optimal solution in the solution space based on the KPI value of each configuration feature vector. Optimizer 1302 then outputs the optimal configuration based on the optimal solution to radio access network 1301.

[0096] In some embodiments of the present invention, the components of each configuration feature vector include a topology of the wireless access network, a routing table, a traffic intensity matrix, and a power intensity matrix.

[0097] In some embodiments of the present invention, the aforementioned key performance indicator is selected from one of a delay indicator, a jitter indicator, and a loss indicator, or a combination thereof.

[0098] In some embodiments of the present invention, the radio access network 1301 is a 5G radio access network.

[0099] Based on the foregoing, some embodiments of the present invention provide a recursive Bayesian network construction system, construction method, computer-readable recording medium, and non-transitory computer program product, which can efficiently construct a suitable recursive Bayesian network by evolving a population of networks composed of combinatorial patterns.

[0100] Some embodiments of the present invention provide a wireless network control system that can automatically implement a self-organizing network solution by applying the task network generated by the aforementioned recursive Bayesian network construction system.

[0101] Although the technical contents of the present invention have been disclosed above in terms of preferred embodiments, they are not intended to limit the present invention. Any slight changes and modifications made by anyone skilled in the art without departing from the spirit of the present invention should be included in the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A recursive Bayesian network construction system, comprising a processor, wherein the processor is configured to perform the following steps to generate a task network: (a) setting a current group, wherein the current group includes a plurality of combined pattern generation networks; (b) establishing a corresponding recursive Bayesian network for each of the combined pattern generation networks in the current group to obtain a set of recursive Bayesian networks corresponding to the current group; (c) evolving the current population using an evolutionary algorithm and a fitness function to obtain a next population, and setting the next population as the current population; (d) determining whether a termination condition is satisfied based on the fitness function and the recursive Bayesian network set corresponding to the current population; and (e) In response to the termination condition not being met, repeating steps (b), (c) and (d), and in response to the termination condition being met, selecting a solution network in the current population as the task network based on the fitness function.

2. The recursive Bayesian network construction system of claim 1 , wherein step (b) comprises: Selecting a current combined pattern generation network in the current group; and An output generated by a child node and a parent node corresponding to the child node in the corresponding recursive Bayesian network is generated by the current combination mode generation network, and a conditional probability table of the parent node corresponding to the child node in the corresponding recursive Bayesian network is established. 3 . The recursive Bayesian network construction system as claimed in claim 2 , wherein the evolutionary algorithm is a topology-enhanced neural evolutionary algorithm.

4. The recursive Bayesian network construction system of claim 1 , wherein the termination condition is that the number of repetitions of steps (b), (c), and (d) reaches a predetermined value or, based on the fitness function, the current population includes a candidate network such that a fitness value of the candidate network is greater than a predetermined fitness value.

5. The recursive Bayesian network construction system of claim 1 , wherein step (a) comprises: randomly generating an initial group according to a parameter; and setting the initial group as the current group, wherein the initial group includes the combined pattern generation network.

6. A recursive Bayesian network construction method, executed by a processor to generate a task network, the recursive Bayesian network construction method comprising: (a) setting a current group, wherein the current group includes a plurality of combined pattern generation networks; (b) establishing a corresponding recursive Bayesian network for each of the combined pattern generation networks in the current group to obtain a set of recursive Bayesian networks corresponding to the current group; (c) evolving the current population using an evolutionary algorithm and a fitness function to obtain a next population, and setting the next population as the current population; (d) determining whether a termination condition is satisfied based on the fitness function and the recursive Bayesian network set corresponding to the current population; and (e) In response to the termination condition not being met, repeating steps (b), (c) and (d), and in response to the termination condition being met, selecting a solution network in the current population as the task network based on the fitness function.

7. The recursive Bayesian network construction method according to claim 6, wherein step (b) comprises: Selecting a current combined pattern generation network in the current group; and An output generated by a child node and a parent node corresponding to the child node in the corresponding recursive Bayesian network is generated by the current combination mode generation network, and a conditional probability table of the parent node corresponding to the child node in the corresponding recursive Bayesian network is established. 8 . The recursive Bayesian network construction method as claimed in claim 7 , wherein the evolutionary algorithm is a topology-enhanced neural evolutionary algorithm.

9. The recursive Bayesian network construction method of claim 6 , wherein the termination condition is that the number of repetitions of steps (b), (c), and (d) has reached a predetermined value or, based on the fitness function, the current population includes a candidate network such that a fitness value of the candidate network is greater than a predetermined fitness value.

10. The recursive Bayesian network construction method of claim 6, wherein step (a) comprises: randomly generating an initial group according to a parameter; and setting the initial group as the current group, wherein the initial group includes the combined pattern generation network.

11. A computer-readable recording medium storing a program, wherein when a processing unit loads and executes the stored program, the method according to any one of claims 6 to 10 is performed. 12 . A non-transitory computer program product storing at least one instruction, which, when executed by a processor, causes the processor to perform the method according to claim 6 .

13. A wireless network control system coupled to a wireless access network and comprising: A recursive Bayesian network construction system includes a processor, wherein the processor is configured to perform the following steps to generate a task network: (a) setting a current group, wherein the current group includes a plurality of combined pattern generation networks; (b) establishing a corresponding recursive Bayesian network for each of the combined pattern generation networks in the current group to obtain a set of recursive Bayesian networks corresponding to the current group; (c) evolving the current population using an evolutionary algorithm and a fitness function to obtain a next population, and setting the next population as the current population; (d) determining whether a termination condition is satisfied based on the fitness function and the recursive Bayesian network set corresponding to the current population; and (e) in response to the termination condition not being met, repeating steps (b), (c), and (d), and in response to the termination condition being met, selecting a solution network in the current population as the task network based on the fitness function; an optimizer configured to receive a network status and a target policy of the wireless access network and output an optimal configuration to the wireless access network to configure and optimize the wireless access network; and A control unit is configured to receive and store the mission network. The optimizer inputs a plurality of configuration feature vectors of each solution into the control unit based on a plurality of solutions in a solution space. The mission network evaluates at least one key performance indicator of each configuration feature vector to obtain at least one key performance indicator value of each configuration feature vector. The optimizer selects an optimal solution in the solution space based on the at least one key performance indicator value of each configuration feature vector, and outputs the optimal configuration to the wireless access network based on the optimal solution. 14 . The wireless network control system of claim 13 , wherein each component of the configuration feature vector comprises a topology structure of the wireless access network, a routing table, a traffic intensity matrix, and a power intensity matrix. 15 . The wireless network control system of claim 13 , wherein the at least one key performance indicator is selected from the group consisting of a free delay indicator, a jitter indicator, and a loss indicator.

16. The wireless network control system of claim 13, wherein step (b) comprises: Selecting a current combined pattern generation network in the current group; and An output generated by a child node and a parent node corresponding to the child node in the corresponding recursive Bayesian network is generated by the current combination mode generation network, and a conditional probability table of the parent node corresponding to the child node in the corresponding recursive Bayesian network is established. 17 . The wireless network control system as claimed in claim 16 , wherein the evolutionary algorithm is a topology-enhanced neuro-evolutionary algorithm.

18. The wireless network control system as described in claim 13, wherein the termination condition is that the number of times steps (b), (c) and (d) are repeated has reached a preset value or, based on the fitness function, the current group includes a candidate network such that a fitness value of the candidate network is greater than a preset fitness value.

19. The wireless network control system as claimed in claim 13, wherein step (a) comprises: randomly generating an initial group according to a parameter; and setting the initial group as the current group, wherein the initial group includes the combined pattern generation network.

Images