5G base station cluster KPI prediction method and system based on multi-reservoir fuzzy cognitive map

Abstract

The present invention belongs to the field of communication technology, and provides a 5G base station cluster KPI prediction method and system based on a multi-reservoir fuzzy cognitive graph. The method comprises the following steps: obtaining KPI original sequence data of all base stations in a 5G base station cluster; preprocessing the KPI original sequence data to obtain a KPI dynamic time series of a base station network; based on a fuzzy cognitive graph model, corresponding the dynamic time series to a concept node in a fuzzy cognitive graph; adding fuzzy feedback on the basis of an original reasoning structure with fuzzy feedback, replacing each concept node in the fuzzy cognitive graph with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and applying multi-reservoir learning to obtain the state characteristics of the KPI of each base station; based on the state characteristics of the KPI, obtaining the KPI of the next moment by adopting iterative reasoning of fuzzy causal dynamics.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular to a 5G base station cluster KPI prediction method and system based on a multi-reservoir fuzzy cognitive map. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Compared with 2G, 3G, and 4G networks, 5G networks have higher frequency bands and suffer great losses during propagation. Therefore, more dense base stations need to be built. The KPIs of the base station network and the exploration of the rules behind these KPIs are of great significance for assisting decision-making. By predicting the KPIs of the base station network in a certain area, we can, to a certain extent, know in advance the changing trends of key indicators in that area, and thus assist operators in establishing reasonable emergency strategies.

[0004] Problems with existing technologies:

[0005] Commonly used time series data prediction models include DNN, LSTM, GRU and their variants. These models have many parameters and take a long time to train. Their black box characteristics obscure the interpretability of the models. Therefore, how to achieve a model that maintains interpretability and has high performance prediction to predict the KPI of 5G base station clusters is a problem. Summary of the invention

[0006] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a bearer network traffic prediction method and system based on the Grammar angle field. On the basis of the original reasoning structure with fuzzy feedback, each concept node in the graph is replaced by an ESN module, multi-library computing is applied, the state characteristics of the indicators are learned, and the prediction of indicators such as user experience rate is realized by iterative reasoning using fuzzy causal dynamics.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The first aspect of the present invention provides a 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map, comprising the following steps:

[0009] Obtain the original KPI sequence data of all base stations in the 5G base station cluster;

[0010] Preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network;

[0011] Based on the fuzzy cognitive graph model, the dynamic time series is mapped to the concept nodes in the fuzzy cognitive graph;

[0012] On the basis of the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive graph is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and multi-repository learning is applied to obtain the state characteristics of the KPI of each base station;

[0013] Based on the state characteristics of KPI, the KPI of the next moment is obtained by adopting iterative reasoning of fuzzy causal dynamics.

[0014] A second aspect of the present invention provides a 5G base station cluster KPI prediction system based on a multi-reservoir fuzzy cognitive map, comprising:

[0015] The data acquisition module is configured to: acquire KPI original sequence data of all base stations in the 5G base station cluster;

[0016] The data preprocessing module is configured to: preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network;

[0017] The state feature extraction module is configured to: correspond the dynamic time series to the concept nodes in the fuzzy cognitive map based on the fuzzy cognitive map model;

[0018] On the basis of the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive graph is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and multi-repository learning is applied to obtain the state characteristics of the KPI of each base station;

[0019] The KPI prediction module of the base station is configured to obtain the KPI of the base station at the next moment by adopting iterative reasoning of fuzzy causal dynamics based on the state characteristics of the KPI.

[0020] A third aspect of the present invention provides a computer-readable storage medium.

[0021] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map as described above.

[0022] A fourth aspect of the present invention provides a computer device.

[0023] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map as described above are implemented.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] On the basis of the original reasoning structure with fuzzy feedback, each concept node in the graph is replaced by an ESN module, and multi-library computing is applied to learn the state characteristics of a KPI in the 5G network. The ability of ESN to process large-scale non-stationary time series and the advantages of FCM in reasoning and interpretation are fully utilized for the prediction of KPIs in the 5G base station network, and the addition of ESN makes the proposed model have higher prediction accuracy. The 5G base station cluster in a certain area can be regarded as a graph structure, which is very similar to the structure of FCM. Through the above technology, high-precision interpretable prediction of the KPI of the 5G base station cluster can be achieved. In the present invention, the addition of ESN can enable the model to better handle nonlinear time series data, and the model also retains the interpretability of FCM. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 It is the overall flow chart of the prediction method based on the MR-FCM model;

[0028] Figure 2 It is an echo state network structure. DETAILED DESCRIPTION

[0029] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0031] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0032] Embodiment 1

[0033] Terminology explanation:

[0034] KPI: (Key Performance Indicator), the base station network cluster generates a large amount of data every moment, such as user experience rate, connection density, end-to-end latency, mobility, traffic density, etc. These data are called the KPI of the base station network.

[0035] FCM: Fuzzy cognitive map is a weighted directed graph with n concept nodes, which is composed of concept nodes, state values ​​and relationships. It combines fuzzy logic and neural networks. It is a powerful model for system state prediction and interpretable knowledge representation. In a fuzzy cognitive map with n nodes, each node represents a concept in the system. This concept can be an event, goal or trend of the system. Each concept characterizes its attributes through state values. The causal influence relationship between concepts is represented by directed arcs, and the degree of causal influence is reflected by weights.

[0036] ESN: Echo State Network, a type of library computing system, is easy to train and has good predictive ability for nonlinear or non-Gaussian dynamic systems. ESNs form a network hidden layer by randomly arranging large-scale sparsely connected neurons, generally called a "reserve pool".

[0037] The reserve pool has the following characteristics: 1) Compared with the classical neural network, it contains a relatively large number of neurons; 2) The connection relationship between neurons is randomly generated; 3) The connection between neurons is sparse.

[0038] like Figure 1 As shown, this embodiment provides a 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map, comprising the following steps:

[0039] S1: Obtain the KPI original sequence data of all base station networks in the 5G base station cluster;

[0040] S2: Preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network

[0041] S3: Based on the fuzzy cognitive map model, the dynamic time series is mapped to the concept nodes in the fuzzy cognitive map;

[0042] For a certain KPI, after the base station network is modeled as an FCM, the KPI of each base station in the base station network corresponds to a conceptual node.

[0043] S4: Based on the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive graph is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and multi-reservoir learning is applied to obtain the state characteristics of the KPI of each base station network;

[0044] S5: Based on the state characteristics of KPI, the KPI of the next moment is obtained by adopting iterative reasoning of fuzzy causal dynamics.

[0045] Among them, in S2, the process of obtaining the KPI dynamic time series includes:

[0046] The maximum and minimum normalization method is used to normalize the KPI indicators of each base station network obtained from S1. Taking the user experience rate as an example, the original user experience rate obtained is:

[0047]

[0048] After normalization:

[0049] Among them, x k (T) is the user experience rate of the kth base station in the base station cluster under the time length T, k = 1, 2, ..., K; the maximum and minimum normalization formula is x min , x max They are x k The minimum and maximum values ​​in (T).

[0050] Among them, the 5G base station cluster in a certain area can be regarded as a graph structure, which is similar to the structure of the fuzzy cognitive graph model. Therefore, the 5G base station cluster is analogized to the fuzzy cognitive graph model.

[0051] In S4, each concept node in the fuzzy cognitive graph is replaced by an echo state network to obtain a multi-reservoir fuzzy cognitive graph model including:

[0052] The construction formula of the fuzzy cognitive graph model is as follows:

[0053]

[0054] Among them, f(·) is the transformation function, which is used to control the state value of the node in [0,1].

[0055] Where i and j represent the i-th base station network and the j-th base station network, and the conceptual node of the i-th base station network is represented by c i Indicates that the weight of the edge between the i-th base station network node and the j-th base station network node is w ij Indicates that, w ij ∈[-1,1].

[0056] Conceptual node c of the i-th base station network i The conceptual node c of the j-th base station network j There are three kinds of relationships:

[0057] When ij >0, c i With c j Positively correlated; when w ij <0, c i With c j Negatively correlated;

[0058] When ij =0, c i With c j No causal relationship; each concept node c i At different times, there is a state value a i ∈[0,1].

[0059] Conceptual node c of the i-th base station network i The state value is a dynamic time series in The conceptual node c represents the i-th base station network i The state value at time t.

[0060] Conceptual node c of the i-th base station network i The state value at time t+1 depends on node c i The state value at time t plus the state value of node c i The state value of the conceptual node of other connected base station networks at time t is multiplied by the sum of the corresponding weights.

[0061] The multi-reservoir fuzzy cognitive graph model is:

[0062] Taking the user experience rate as an example, after substitution, formula (1) becomes the following formula (2):

[0063]

[0064] in, is the membership function, which controls the output value between [0,1]. j (·) is the calculation process of the jth ESN module, w jk is the weight of the edge connected to the kth ESN module. is the input of the jth ESN module at time t,

[0065] For the jth ESN module, the state update formula of the reserve pool is:

[0066] in, are the input weight matrix and reserve pool weight matrix of the jth ESN module, is the state of the reservoir at time t, f is the activation function of the neurons in the reservoir; the output of the jth ESN module at time (t+1) is: in, is the output weight matrix of the jth ESN module, is the output value at time t, and fout is the activation function of the output neuron. In this way, the state characteristics of the user experience rate of each base station are learned through each ESN module.

[0067] Furthermore, in the present invention, the number of input neurons M of each ESN network can be set to 1, the number of neurons in the reserve pool N can be set to 200, and the number of output neurons P can be set to 1.

[0068] like Figure 2 As shown, the state network includes an input layer, a reserve pool and an output layer;

[0069] The input signal u(t) is mapped from the low-dimensional input space to the high-dimensional state space, and the output weight W of the network is calculated using the linear regression method in the high-dimensional state space. out Training is performed, and the input weight W in , reserve pool weight W res , feedback connection weight W back It remains unchanged during network training.

[0070] The echo state network contains M input units, and the input at time t is

[0071]

[0072] N internal units, the state at time t is

[0073] There are P output units, and the output at time t is in[·] T Represents the transpose operation of a matrix.

[0074] The update of the internal state vector of the reserve pool is shown in formula (3):

[0075] x(t+1)=f[W in u(t+1)+W res x(t)] (3)

[0076] in, represents the value of the internal state vector of ESNs at time t, They represent the input connection weight matrix, the reserve pool connection weight matrix, and the output connection weight matrix respectively. f=[f1,f2,…,f N ] represents the activation function of the neurons inside the reserve pool, usually fi (i=1,2,…,N) Take the hyperbolic tangent function, as shown in formula (4):

[0077]

[0078] The output of the entire network is shown in formula (5):

[0079] y(t+1)=f out [W out (x(t+1),y(t))], (5)

[0080] Among them, f out =(f1 out ,…,f P out ) is a linear function, and (x(t+1), y(t)) is the concatenation of the input, the reservoir, and the previous output weight matrix.

[0081] In this way, each ESN module learns the state characteristics of the user experience rate of each base station. The above process learns new features through the output layer of the ESN module. The update of the reserve pool state is only the impact of the input value on the reserve pool. Its state is updated, but we want the value of the output layer as the intermediate state, that is, to learn the KPI state characteristics. The purpose of replacing the original node with the ESN module is to make up for the defect that FCM cannot model long-term dependencies.

[0082] Among them, in S5, based on the state characteristics of KPI, the KPI of the next moment is obtained by iterative reasoning using fuzzy causal dynamics, including:

[0083] Among them, ESN j The calculation process of (·) is as follows:

[0084] 1) The input of the jth ESN at time t is shown in formula (6):

[0085]

[0086] 2) The state update of the reserve pool of the jth ESN at time (t+1) is shown in formula (7):

[0087]

[0088] in, are the input weight matrix and reserve pool weight matrix of the jth ESN module, is the state of the reservoir at time t, and f is the activation function of the neurons in the reservoir;

[0089] 3) The output of the jth ESN at time (t+1) is shown in formula (8):

[0090]

[0091] in, is the output weight matrix of the jth ESN module, is the output value at time t, f out is the activation function of the output neuron.

[0092] Formula (2) can be expressed in the following vector form:

[0093]

[0094] in, is the state value of all node modules in the multi-reservoir fuzzy cognitive graph at time t, is the state value of the kth node of the multi-reservoir fuzzy cognitive graph at time (t+1), W k ={w 1k ,…,w kk ,…,w Kk} is the kth column of the weight matrix W. After the inverse transformation of formula (9), it can be expressed as:

[0095]

[0096] The learning problem of the weights of the multi-reservoir fuzzy cognitive graph model can be transformed into the solution of the least squares method. k , construct the following objective function:

[0097]

[0098] The Adam algorithm is used to optimize the objective function. After the model is trained, the weight matrix W is obtained. Through the known W and the update rules of MR-FCM, the user experience rate of the base station of the 5G base station network at the next moment can be solved. By predicting the user experience rate of the base station network in a certain area, the changing trend of the user experience rate in the area can be known in advance to a certain extent, which can assist operators in establishing reasonable emergency strategies.

[0099] Among them, the optimization process of the objective function adopts the Adam algorithm, and the specific optimization process includes:

[0100]

[0101]

[0102] In this embodiment, the prediction of the user experience rate indicator is taken as an example, and the overall idea is:

[0103] First, the user experience rate of each base station in the base station cluster is Preprocessing to obtain Secondly, the preprocessed data is passed through the calculation process ESN(·) of the ESN module to obtain the state characteristics of the user experience rate. Finally, through the iterative formula of fuzzy causal dynamics Get the user experience rate of each base station at the next moment.

[0104] Considering the difference between traffic usage during holidays and non-holidays, two metrics, growth entropy and standard deviation, were used to measure this difference. Grami angular field sequence imaging technology was then used to increase the dimension of the one-dimensional network traffic sequence. The processed images were used as input to the ConvLSTM-F layer, which can then extract the temporal and spatial features of the original sequence and predict the intelligent traffic of the bearer network. In the prediction model, the use of the ConvLSTM-F module can not only extract the temporal information of the sequence but also further extract the spatial information, thereby enriching the input information and improving the prediction accuracy of the model.

[0105] Embodiment 2

[0106] This embodiment provides a 5G base station cluster KPI prediction system based on a multi-reservoir fuzzy cognitive map, including:

[0107] The data acquisition module is configured to: acquire KPI original sequence data of all base stations in the 5G base station cluster;

[0108] The data preprocessing module is configured to: preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network;

[0109] The state feature extraction module is configured to: correspond the dynamic time series to the concept nodes in the fuzzy cognitive map based on the fuzzy cognitive map model;

[0110] On the basis of the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive graph is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and multi-repository learning is applied to obtain the state characteristics of the KPI of each base station;

[0111] The KPI prediction module of the base station is configured to obtain the KPI of the base station at the next moment by adopting iterative reasoning of fuzzy causal dynamics based on the state characteristics of the KPI.

[0112] Embodiment 3

[0113] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the bearer network traffic prediction method based on the Grami angle field as described above are implemented.

[0114] Embodiment 4

[0115] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the bearer network traffic prediction method based on the Grami angle field as described above are implemented.

[0116] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0117] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0120] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0121] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A 5G base station cluster KPI prediction method based on multi-reservoir fuzzy cognitive graph, characterized in that: The steps include: Obtain the original KPI sequence data of all base station networks in the 5G base station cluster; Preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network; Based on the fuzzy cognitive graph model, the dynamic time series is mapped to the concept nodes in the fuzzy cognitive graph; On the basis of the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive map is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive map model, and multi-reservoir learning is applied to obtain the state characteristics of the KPI of each base station; the multi-reservoir fuzzy cognitive map model is: In the formula, is the membership function, ESN j (·) is the calculation process of the jth echo state network module, w jk is the weight of the edge connected to the kth echo state network module, is the input of the j-th echo state network module at time t, where t is the current time; The learning problem of the multi-reservoir fuzzy cognitive graph model weights is expressed as: Among them, D (t) is the state value of all node modules in the multi-reservoir fuzzy cognitive graph at time t, is the state value of the kth node of the multi-reservoir fuzzy cognitive graph at time t+1, W k is the kth column of the weight matrix W, ESN is the echo state network; Based on the state characteristics of KPI, the KPI of the next moment is obtained by adopting iterative reasoning of fuzzy causal dynamics; The KPI of the next moment obtained by iterative reasoning of fuzzy causal dynamics is that the learning problem of the weight of the multi-reservoir fuzzy cognitive graph model can be converted into a least squares method for solution; The calculation process of the echo state network module is as follows: in, is the output weight matrix of the j-th echo state network module, is the output value at time t, f out is the activation function of the output neuron; in, In the formula, are the input weight matrix and reserve pool weight matrix of the jth echo state network module, is the state of the reservoir at time t, f is the activation function of the neurons in the reservoir, It is the input of the echo state network model at time t+1.

2. The 5G base station cluster KPI prediction method based on multi-reservoir fuzzy cognitive graph according to claim 1 is characterized in that: The process of obtaining the KPI dynamic time series includes: normalizing the KPI index of each base station obtained by using the maximum and minimum normalization method.

3. The 5G base station cluster KPI prediction method based on multi-reservoir fuzzy cognitive graph according to claim 1 is characterized in that: The echo state network includes an input layer, a reserve pool and an output layer; the echo state network training process includes: mapping the input signal from a low-dimensional input space to a high-dimensional state space, and using a linear regression method in the high-dimensional state space to train the output weights of the network, and the input weights, reserve pool weights and feedback connection weights remain unchanged during the network training process.

4. A 5G base station cluster KPI prediction system based on a multi-reservoir fuzzy cognitive map, using the 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map as claimed in claim 1, characterized in that: include: The data acquisition module is configured to: acquire KPI original sequence data of all base station networks in the 5G base station cluster; The data preprocessing module is configured to: preprocess the KPI original sequence data to obtain the KPI dynamic time series of the base station network; The state feature extraction module is configured to: correspond the dynamic time series to the concept nodes in the fuzzy cognitive map based on the fuzzy cognitive map model; On the basis of the original reasoning structure with fuzzy feedback, fuzzy feedback is added, and each concept node in the fuzzy cognitive graph is replaced with an echo state network to obtain a multi-reservoir fuzzy cognitive graph model, and multi-repository learning is applied to obtain the state characteristics of the KPI of each base station; The KPI prediction module of the base station is configured to obtain the KPI of the base station at the next moment by adopting iterative reasoning of fuzzy causal dynamics based on the state characteristics of the KPI.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map as described in any one of claims 1-3 are implemented.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the 5G base station cluster KPI prediction method based on a multi-reservoir fuzzy cognitive map are implemented as described in any one of claims 1-3.

Images