Data set construction method for network reliability evaluation and network reliability evaluation method

Abstract

The invention relates to a data set construction method for network reliability evaluation and a network reliability evaluation method, and belongs to the technical field of computers. The method comprises the following steps: firstly, initializing basic parameters and parameters needing to be randomly generated in batches, randomly generating a network maximum capacity vector and demand, randomly generating a network probability matrix, calculating the network reliability according to a polymorphic minimum path algorithm to obtain a plurality of networks and a set of the reliability thereof, and then establishing a multi-layer residual perceptron network. And training the network through the preprocessed data set, and finally performing network reliability evaluation through the trained multi-layer residual perceptron network. According to the method, the polymorphic minimum path and the multi-layer perceptron are combined, so that the characteristics of network variability and complexity are met, and a decision maker is helped to make an optimal choice in time when network reliability factors are changed or multiple schemes are selected.

Description

Technical Field

[0001] The invention belongs to the field of computer technology and relates to a data set construction method for network reliability evaluation and a network reliability evaluation method. Background Art

[0002] Network systems usually perform given tasks under specific conditions. Their reliability refers to the probability that the network can smoothly transport d units of network flow (representing goods, information or energy, etc.) from the source point s to the sink point t. However, due to the influence of internal or external uncertainties, network systems will show a variety of different performance levels during operation. Such network systems are called polymorphic network systems. They undertake the functions of resource transmission and allocation, and are used to achieve effective organization and coordination of logistics activities or information transmission. The capacity state of each edge in the network is a random variable, but this random variable obeys a certain capacity distribution probability. The polymorphic network system model can more accurately describe and characterize the complex behavior of the network system, and has received increasing attention and application in the field of system performance evaluation research. Network reliability is a key factor in ensuring the stable operation of the supply network.

[0003] There are many factors that affect network reliability, such as the number of edges and nodes in the network itself, the capacity status of the edges, network transportation requirements, etc. Therefore, the reliability of the network varies. Traditional reliability analysis methods are time-consuming, and decision makers often find it difficult to quickly calculate the reliability of different network structures, and thus cannot quickly determine the network structure with the highest reliability. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a data set construction method for network reliability evaluation and a network reliability evaluation method. The present invention focuses on a rapid evaluation method for network reliability and proposes a novel method that combines a polymorphic minimal path and a multilayer perceptron (MLP) to meet the characteristics of network variability and complexity. This helps decision makers make the best choice in a timely manner when faced with changes in network reliability factors or when there are multiple options.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for constructing a data set for network reliability assessment, the method comprising the following steps:

[0007] Step a: Initialize basic parameters and parameters that need to be randomly generated in batches;

[0008] Step b: Randomly generate the network maximum capacity vector and demand;

[0009] Step c: Randomly generate a network probability matrix;

[0010] Step d: Calculate the network reliability according to the polymorphic minimal path algorithm to obtain a set of several networks and their reliabilities.

[0011] Furthermore, in step a, the basic parameters initialized include: the number of network nodes m, the number of edges n, and the adjacency matrix G of the network topology structure, where the adjacency matrix G is expressed as:

[0012]

[0013] Among them, G ij ∈{0,1},G ij =0 means that there is no connection between nodes, G ij =1 indicates that the nodes are connected;

[0014] Initialization requires randomly generated network capacity distribution quantity x.

[0015] Furthermore, in step b, each edge e is randomly generated according to the adjacency matrix of the network. i The maximum capacity w i , where 1≤i≤n, and w i ∈[1,m-1],w i is a positive integer;

[0016] All i To form the maximum capacity vector W, that is:

[0017] W=(w1,w2,…,w n )

[0018] Randomly generate network demand d, 1≤d≤max(W) and d is an integer, where:

[0019] max(W)=max{w i |1≤i≤n}

[0020] max(W) represents the maximum value in the current network maximum capacity vector W.

[0021] Further, in step c, according to the adjacency matrix of the network and each edge e i The maximum capacity w i Randomly generate edge e i All capacity states {0,1,...,w i The probability matrix corresponding to} is:

[0022]

[0023] Where θ = max(W) + 1, which represents the number of columns in the probability matrix; each element P ij represents the i-th edge e iThe probability of being in the jth capacity state, and

[0024] For each e i , if w i <max(W),1≤i≤n, then P ij =0, where j>w i +1;

[0025] Each row element of the probability matrix P satisfies the following constraints: i1 <P i2 <… <P iθ .

[0026] Furthermore, in step d, all polymorphic minimal paths in the network are solved, and the process is as follows:

[0027] First, solve all feasible flow vectors F=(f1,f2,...,f λ ):

[0028] C1:f1+f2+...+f λ =d

[0029] C2:0≤f j ≤min{W i |e i ∈p j},j=1,2,...,λ

[0030] Then, according to condition C3, all feasible flow vectors F are transformed into polymorphic minimal paths X = (x1, x2, ..., x m ):

[0031]

[0032] Among them, λ is the number of all minimal paths from the source to the sink in the network, and f j It is through a very small road j The flow rate, min{W i |e i ∈p j} represents the minimum path p j Maximum capacity;

[0033] After all the multi-state minimal paths of the network are obtained, the network reliability is calculated using the disjoint sum algorithm.

[0034] The present invention also proposes a network reliability assessment method using a data set established by the aforementioned data set construction method for network reliability assessment for pre-training, the method comprising the following steps:

[0035] S1, using the dataset construction method to generate the corresponding dataset;

[0036] S2, preprocessing the reliability assessment data in the data set;

[0037] S3. Establish a multi-layer residual perceptron network for evaluating network reliability;

[0038] S4, training the multi-layer residual perceptron network based on the preprocessed data set to obtain a pre-trained multi-layer residual perceptron network;

[0039] S5. Reliability evaluation of the real-time network is performed through a pre-trained multi-layer residual perceptron network.

[0040] Further, in step S2, the generated data set is preprocessed, which includes:

[0041] S21, obtain feature data: the maximum capacity vector W = (w1, w2, ..., w n ) for each w i As a feature data, there are n feature data in total. Then, each element G in the input adjacency matrix G ij As a feature data, there are m×m feature data in total; furthermore, each element P in the generated probability matrix P ij Considered as one feature data, there are n×(max(W)+1) feature data in total;

[0042] S22, concatenate the maximum capacity vector W and the demand d to obtain a new vector (W, d) = (w1, w2, ..., w n ,d), perform maximum and minimum normalization on the concatenated data vector (W,d), and the normalized vector is recorded as (W′,d′)=(w′1,w′2,…,w′ n ,d′);

[0043] S23, flatten the adjacency matrix G and the probability matrix P. The adjacency matrix after flattening is expressed as:

[0044] G * =(G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm )

[0045] The representation of the flattened probability matrix is:

[0046] P * =(P 11,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj )

[0047] In the formula, j represents the number of columns of the probability matrix;

[0048] S24. Concatenate the normalized vectors with the flattened vectors to obtain the final data form:

[0049] (w′1,w′2,…,w′ n ,d′,G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm ,

[0050] P 11 ,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj )

[0051] Complete the data preprocessing process.

[0052] Further, in step S3, the established multi-layer residual perceptron network includes a plurality of residual and shortcut branch structures, wherein:

[0053] The backbone of the multi-layer residual perceptron network uses a multi-layer perceptron stack. The first layer uses a fully connected layer and a LeakyReLU activation function as the input layer to map the input feature data to a high-dimensional space and expand the number of features.

[0054] Then it enters the feature extraction module composed of a series of stacked residual structures for feature extraction. Each layer of the network consists of a fully connected layer and a LeakyReLU activation function. The odd-numbered layers are residual structures, and the remaining layers are used to adjust the feature dimensions for the residual structure. In addition, the input layer and the feature extraction module are connected by a shortcut branch to strengthen the representation of the features. There is a fully connected layer on the shortcut branch.

[0055] Finally, after passing through the final output layer network, which consists of a fully connected layer and a Sigmoid activation function, the final reliability prediction result is obtained based on several sets of extracted features. The Sigmoid activation function is used to activate the output layer network and limit the output probability between 0 and 1.

[0056] Further, in step S4, a small batch gradient descent training method is used to train the established multi-layer residual perceptron network based on the preprocessed data set, wherein the loss function is the mean square error loss, the optimizer is the Adam optimizer, the initial learning rate is set, the dynamic learning rate decay strategy is the cosine annealing decay strategy, and the decay cycle calculation in the decay process is based on the number of iterations. When the learning rate reaches the expected target, the training is completed.

[0057] The beneficial effects of the present invention are:

[0058] The present invention focuses on a rapid evaluation method for network reliability, combining polymorphic minimal paths with multi-layer perceptrons (MLPs) to meet the characteristics of network variability and complexity, thereby helping decision makers make the best choice in a timely manner when faced with changes in network reliability factors or when there are multiple options to choose from.

[0059] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0061] Figure 1 A schematic diagram of a network topology structure;

[0062] Figure 2 A schematic diagram of a flow chart of a network reliability evaluation method of the present invention;

[0063] Figure 3 It is a schematic diagram of the data preprocessing process of the reliability evaluation method of the present invention;

[0064] Figure 4 It is a schematic diagram of the structure of the multi-layer residual perceptron of the present invention. DETAILED DESCRIPTION

[0065] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0066] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0067] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0068] See also Figure 1 to Figure 4 , which is a data set construction method for network reliability evaluation and a network reliability evaluation method.

[0069] Example 1

[0070] This embodiment first provides a method for constructing a data set for network reliability assessment, which specifically includes the following steps:

[0071] Step a: Initialize basic parameters and parameters that need to be randomly generated in batches, where the basic parameters include the number of network nodes M, the number of edges n, and the adjacency matrix G of the network topology structure, and then initialize the number of network capacity distributions x that need to be randomly generated.

[0072] Specifically, initialize the number of network nodes m, the number of edges n, and the adjacency matrix G that describes the network topology:

[0073]

[0074] Among them, G ij ∈{0,1},G ij =0 means that there is no connection between nodes, G ij =1 indicates that the nodes are connected.

[0075] Step b: Randomly generate the network maximum capacity vector and demand, where

[0076] According to the adjacency matrix of the network, each edge e is randomly generated i The maximum capacity w i , where 1≤i≤n, and w i The value of is a positive integer between [1, m-1], all w i The maximum capacity vector W is composed of W = (w1,w2,…,w n ); Randomly generate network demand d, the value of d is a positive integer between 1 and max(W), that is, 1≤d≤max(W), where max(W)=max{w i |1≤i≤n} represents the maximum value in the current network maximum capacity vector W;

[0077] Step c: Randomly generate a network probability matrix: According to the network adjacency matrix and each edge e i The maximum capacity w i Randomly generate edge e i All capacity states {0,1,...,w i The probability matrix corresponding to} is:

[0078]

[0079] In the formula, θ=max(W)+1, each element p ij represents the i-th edge e i The probability of being in the jth capacity state, and For e i For example, if w i <max(W),1≤i≤n, then P ij =0, where j>w i +1. In addition, each row element of the probability matrix P satisfies the following constraints: P i1 <P i2 <… <P iθ .

[0080] Step d: Calculate reliability: Calculate network reliability through the existing polymorphic minimal path (d-minimal path for short) algorithm. The minimal path is a set of edges in the network. The edges in the set form a path from the source to the sink. The d-minimal path is a special capacity state vector. In the first stage, all d-minimal paths in the network are solved. In the second stage, the network reliability is calculated using the disjoint sum algorithm.

[0081] More specifically, in this embodiment, if Figure 1 As shown, taking a network with a topology of 8 nodes and 12 edges as an example, the above scheme is described in more detail, including:

[0082] In step a of this embodiment, batch input parameters and batch random generation parameters are as follows: the number of input nodes m=8, the number of edges n=12, and the number of randomly generated networks x=24000.

[0083] In step b of this embodiment, the network maximum capacity vector W and the demand d are randomly generated: the transportation demand d of the network is randomly generated, and the range of d is [1,4]. The randomly generated maximum capacity vector W = (w1, w2, w3, w4, w5, w6, w7, w8, w9, w 10 , w 11 , w 12 ).

[0084] In step c of this embodiment, the probability matrix P is randomly generated: according to the number of nodes, the number of edges and each edge e i The maximum capacity w i (1≤i≤n), the number of rows of the probability matrix P is n=12 and the number of columns is θ=8.

[0085] In step d of this embodiment, the reliability calculation is based on the minimal path and disjoint sum algorithm: the reliability of the generated 24,000 networks is calculated using the existing polymorphic minimal path (also known as d-minimal path) algorithm. In the first stage of the calculation process, all d-minimal paths in the network are solved, and in the second stage, the disjoint sum algorithm is used to solve the network reliability, thereby obtaining label data.

[0086] The process of solving all d-minimum paths in the network includes: taking one of the randomly generated networks as an example: the randomly generated parameters are the maximum capacity vector W = (2,2,2,3,3,4,5,4,5,3,5,4), and when the transportation demand d = 2, its polymorphic minimum path can be obtained as: [0 2 0 0 2 0 0 0 0 2 0 2] [0 2 0 0 2 0 0 1 0 1 1 1] [0 2 0 0 2 0 0 1 1 1 0 2] [0 2 0 0 2 0 0 2 0 0 2 0] [0 2 0 0 2 0 0 2 1 0 1 1] [0 2 0 0 2 0 0 2 2 0 0 2] [0 2 0 1 1 0 1 0 0 2 0 2][0 2 0 1 1 0 1 1 0 1 1 1][0 2 0 1 1 0 1 1 11 0 2][0 2 0 1 1 0 1 2 0 0 2 0][0 2 0 1 1 0 1 2 1 0 1 1][0 2 0 1 1 0 1 2 2 00 2][0 2 0 1 1 1 0 0 0 1 1 1][0 2 0 1 1 1 0 0 1 1 0 2][0 2 0 1 1 1 0 1 0 0 20][0 2 0 1 1 1 0 1 1 0 1 1][0 2 0 1 1 1 0 1 2 0 0 2][0 2 0 2 0 0 2 0 0 2 0 2][0 2 0 2 0 0 2 1 0 1 1 1][0 2 0 2 0 0 2 1 1 1 0 2][0 2 0 2 0 0 2 2 0 0 2 0][02 0 2 0 0 2 2 1 0 1 1][0 2 0 2 0 0 2 2 2 0 0 2][0 2 0 2 0 1 1 0 0 1 1 1][0 20 2 0 1 1 0 1 1 0 2][0 2 0 2 0 1 1 1 0 0 2 0][0 2 0 2 0 1 1 1 1 0 1 1][0 2 02 0 1 1 1 2 0 0 2][0 2 0 2 0 2 0 0 0 0 2 0][0 2 0 2 0 2 0 0 1 0 1 1][0 2 0 20 2 0 0 2 0 0 2][1 1 1 0 1 0 1 0 0 2 0 2][1 1 1 0 1 0 1 1 0 1 1 1][1 1 1 0 10 1 1 1 1 0 2][1 1 1 0 1 0 1 2 0 0 2 0][1 1 1 0 1 0 1 2 1 0 1 1][1 1 1 0 1 01 2 2 0 0 2][1 1 1 0 1 1 0 0 0 1 1 1][1 1 1 0 1 1 0 0 1 1 0 2][1 1 1 0 1 1 01 0 0 2 0][1 1 1 0 1 1 0 1 1 0 1 1] [1 1 1 0 1 1 0 1 2 0 0 2] [1 1 1 1 0 0 2 0 0 2 0 2] [1 1 1 1 0 0 2 1 0 1 1 1] [1 1 1 1 0 0 2 1 1 1 0 2] [1 1 1 1 0 0 2 2 0 0 2 0] [1 1 1 1 0 0 2 2 1 0 1 1] [1 1 1 1 0 0 2 2 2 0 0 2] [1 1 1 1 0 1 1 0 0 1 1 1] [1 1 1 1 0 1 1 0 1 1 0 2] [1 1 1 1 0 1 1 1 0 0 2 0] [1 1 1 1 0 1 1 1 1 0 1 1] [1 1 1 1 0 1 1 1 2 0 0 2] [1 1 1 1 0 2 0 0 0 0 2 0] [1 1 1 1 0 2 0 0 1 0 1 1] [1 1 1 1 0 2 0 0 2 0 0 2] [2 0 2 0 0 0 2 0 0 2 0 2] [2 0 2 0 0 0 2 1 0 1 1 1] [2 0 2 0 0 0 2 1 1 1 0 2] [2 0 2 0 0 0 2 2 0 0 2 0] [2 0 2 0 0 0 2 2 1 0 1 1] [2 0 2 0 0 0 2 2 2 0 0 2] [2 0 2 0 0 1 1 0 0 1 1 1] [2 0 2 0 0 1 1 0 1 1 0 2] [2 0 2 0 0 1 1 1 0 0 2 0] [2 0 2 0 0 1 1 1 1 0 1 1] [2 0 2 0 0 1 1 1 2 0 0 2] [2 0 2 0 0 2 0 0 0 0 2 0] [2 0 2 0 0 2 0 0 1 0 1 1] [2 0 2 0 0 2 0 0 2 0 0 2]

[0123] Based on the obtained multi-state minimal path, the common disjoint sum algorithm can be used to solve the network reliability: the reliability is: 0.9091.

[0124] In this way, a data set containing 24,000 networks and their reliability is obtained, which is used to train a network reliability assessment model. The trained model is used for network reliability assessment.

[0125] Example 2

[0126] This embodiment is based on the data set given in Example 1 and proposes a network reliability evaluation method based on a trained multi-layer residual perceptron. Specifically, Figure 2 As shown, it includes the following steps:

[0127] S1. Generate a corresponding data set using the aforementioned data set construction method;

[0128] S2, preprocessing the reliability assessment data in the data set;

[0129] S3. Establish a multi-layer residual perceptron network for evaluating network reliability;

[0130] S4, training the multi-layer residual perceptron network based on the preprocessed data set to obtain a pre-trained multi-layer residual perceptron network;

[0131] S5. Reliability evaluation of the real-time network is performed through a pre-trained multi-layer residual perceptron network.

[0132] Specifically, in step S2, the preprocessing process is as follows: Figure 3 As shown, it includes the following operations:

[0133] S21, obtain feature data: the maximum capacity vector W = (w1, w2, ..., w n ) for each w i As a feature data, there are n feature data in total. Then, each element G in the input adjacency matrix G ij As a feature data, there are m×m feature data in total; furthermore, each element P in the generated probability matrix P ij If it is regarded as one feature data, there are n×(max(W)+1) feature data in total.

[0134] S22, concatenate the maximum capacity vector W and the demand d to obtain a new vector (W, d) = (w1, w2, ..., w n ,d), in order to avoid some characteristic data dominating the experimental results, the concatenated data vector (W, d) is normalized to the maximum and minimum values, and the normalized vector is recorded as (W′, d′) = (w′1, w′2, …, w′ n ,d′).

[0135] S23, flatten the adjacency matrix G and the probability matrix P. The adjacency matrix after flattening is expressed as:

[0136] G * =(G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm )

[0137] The representation of the flattened probability matrix is:

[0138] P * =(P 11 ,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj )

[0139] Where j represents the number of columns in the probability matrix.

[0140] S24. Concatenate the normalized vectors with the flattened vectors to obtain the final data form:

[0141] (w′1,w′2,…,w′ n ,d′,G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm ,

[0142] P 11 ,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj )

[0143] Complete the independent data preprocessing process.

[0144] In step S3, if Figure 4 The multi-layer perceptron network with residual and shortcut branch structure shown in the figure adopts multi-layer perceptron stacking as its backbone. The first layer uses a fully connected layer and a LeakyReLU activation function as the input layer to map the input feature data to a high-dimensional space and expand its feature quantity. Then, it enters a feature extraction module composed of a series of stacked residual structures for feature extraction. Each layer of the network is composed of a fully connected layer and a LeakyReLU activation function. The odd-numbered layers are residual structures, and the remaining layers are used to adjust the feature dimensions for the residual structure. In addition, the input layer and the feature extraction module are connected by a shortcut branch to strengthen the representation of the features and avoid the network from falling into overfitting after feature extraction. There is a fully connected layer on the shortcut branch. Finally, after passing through the final output layer network, the output layer network is composed of a fully connected layer and a Sigmoid activation function, and the final reliability prediction result is obtained from 16 features. The Sigmoid activation function is used to activate the output layer network and limit the output probability between 0 and 1.

[0145] In step S4, a small batch gradient descent training method is used to train the established multi-layer residual perceptron network based on the preprocessed data set, wherein the loss function is the mean square error loss, the optimizer is the Adam optimizer, the initial learning rate is set, the dynamic learning rate decay strategy is the cosine annealing decay strategy, and the decay cycle calculation in the decay process is based on the number of iterations.

[0146] In this embodiment, a network with a topological structure of 8 nodes and 12 edges is also taken as an example to explain the training process of the multi-layer residual perceptron network in more detail, and to verify the network reliability evaluation effect of the multi-layer residual perceptron network after pre-training.

[0147] In this embodiment, the maximum capacity vector W with a feature number of 12 is W=(w1, w2, w3, ..., w 12 ) and the demand d, the concatenated vector is (W, d) = (w1, w2, w3, …, w 12 , d), in order to avoid some features dominating the model due to the difference in eigenvalues, improve accuracy, and make features of different dimensions fairly compared. Therefore, the maximum and minimum normalization processing of (W, d) is performed to obtain (W', d') = (w'1, w'2, w'3, ..., w' 12 , d'). The adjacency matrix G with 64 features mm (m=8) and the probability matrix P with 96 features nθ (n=12, θ=8) flattened to obtain the representation of the flattened adjacency matrix:

[0148] G*=(G 11 , G 12 , …, G18 , G 21 , G 22 , …, G 28 , …, G 81 , G 82 , …, G 88 )

[0149] The representation of the flattened probability matrix is:

[0150] P*=(P 11 , P 12 ,…,P 18 , P 21 , P 22 ,…,P28,…,P 12 1, P 12 2, ..., P 12 8)

[0151] Concatenate the normalized vector with the flattened vector to get the final feature data form:

[0152] (w'1, w'2, w'3, ..., w' 12 , d', G 11 , G 12 , …, G 18 , G 21 , G 22 , …, G 28 , …, G 81 , G 82 , …, G 88 , P 11 , P 12 ,…,P 1j , P 21 , P 22 ,…,P 2j ,…,P 12 1, P 12 2, ..., P 12 8)

[0153] Complete the independent data preprocessing process.

[0154] The backbone of the multi-layer residual perceptron network of this embodiment adopts a 9-layer perceptron stack. The first layer uses a fully connected layer and a LeakyReLU activation function as the input layer to map the input features to a high-dimensional space. The number of features is expanded to 256, and then enters a feature extraction module composed of a series of stacked residual structures for feature extraction. A total of 5 layers of networks are stacked, each of which is composed of a fully connected layer and a LeakyReLU activation function, of which 3 layers (odd-numbered layers) are residual structures, and the remaining two layers are used to adjust the feature dimensions for the residual structure. Specifically, the 256 features are firstly processed by the first layer of the feature extraction module. The residual layer, then passes through a layer of network to map the number of features to 128, and then passes through the third layer of the feature extraction module (i.e., the second residual structure), and then passes through the fourth layer of the feature extraction module to map the number of features to 64, and finally passes through the last layer of the feature extraction module (i.e., the third residual structure) to complete the feature extraction process. In addition, the output layer and the feature extraction module are connected by a shortcut branch to strengthen the representation of the features and avoid the network from falling into overfitting after feature extraction. There is a fully connected layer on the shortcut branch to map the 256 features after the output layer to the 64 after the feature extraction layer. Then, the 64 features after feature extraction pass through two layers of network (i.e., the 7th and 8th layers of the entire network) to map the number of features to 32 and 16 respectively. Finally, after the last layer of the output layer network, the output layer network consists of a fully connected layer and a Sigmoid activation function, and the final reliability prediction result is obtained from 16 features. The Sigmoid activation function is used to activate the output layer network and limit the output probability between 0 and 1. This design method can alleviate the gradient vanishing problem, improve the generalization ability of the model, and improve the accuracy.

[0155] During training, the initial learning rate is set to 10-3, the dynamic learning rate decay strategy is the cosine annealing decay strategy, the decay cycle calculation in the decay process is based on the number of iterations, and the final learning rate is 10-5.

[0156] For the trained multi-layer residual perceptron network, three indicators, namely mean square error, mean absolute error and root mean square error, are used to test and evaluate it on the test set.

[0157] For these three indicators, lower values ​​indicate that the model has good prediction performance because they reflect the fit between the model prediction and the true value. Specifically, MSE and RMSE are more sensitive to outliers due to the influence of the square term, so lower values ​​mean that the model effectively reduces large prediction errors; while lower MAE values ​​indicate that the model is consistent in average prediction accuracy. In the evaluation, if the model obtains lower MSE, MAE, and RMSE values ​​on the test set, it means that the method is highly robust and effective in capturing data patterns and achieving generalization.

[0158] The final training results have an MSE value of 0.0027, a MAE value of 0.004335, and an RMSE value of 0.0045. The lower values ​​indicate that the model has good predictive performance, indicating that the method is highly robust and effective in capturing data patterns and achieving generalization.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a data set for network reliability assessment, characterized in that: The method comprises the following steps: Step a: Initialize basic parameters and parameters that need to be randomly generated in batches; Step b: Randomly generate the network maximum capacity vector and demand; Step c: Randomly generate a network probability matrix; Step d: Calculate the network reliability according to the polymorphic minimal path algorithm to obtain a set of several networks and their reliabilities.

2. The method for constructing a data set for network reliability assessment according to claim 1, characterized in that: In step a, the basic parameters initialized include: the number of network nodes m, the number of edges n, and the adjacency matrix G of the network topology structure, where the adjacency matrix G is expressed as: Among them, G ij ∈{0,1},G ij =0 means that there is no connection between nodes, G ij =1 indicates that the nodes are connected; Initialization requires randomly generated network capacity distribution quantity x.

3. The method for constructing a data set for network reliability assessment according to claim 2, characterized in that: In step b, each edge e is randomly generated according to the adjacency matrix of the network i The maximum capacity w i , where 1≤i≤n, and w i ∈[1,m-1],w i is a positive integer; All i To form the maximum capacity vector W, that is: In=(in1,in2,…,in n ) Randomly generate network demand d, 1≤d≤max(W) and d is an integer, where: max(W)=max{w i |1≤i≤n} max(W) represents the maximum value in the current network maximum capacity vector W.

4. The method for constructing a data set for network reliability assessment according to claim 3, characterized in that: In step c, according to the adjacency matrix of the network and each edge e i The maximum capacity w i Randomly generate edge e i All capacity states {0,1,...,w i The probability matrix corresponding to} is: Where θ = max(W) + 1, which represents the number of columns in the probability matrix; each element P ij represents the i-th edge e i The probability of being in the jth capacity state, and For each e i , if w i <max(W),1≤i≤n, then P ij =0, where j>w i +1; Each row element of the probability matrix P satisfies the following constraints: i1 <P i2 <… <P iθ .

5. The method for constructing a data set for network reliability assessment according to claim 4, characterized in that: In step d, all polymorphic minimal paths in the network are solved, and the process is: First, solve all feasible flow vectors F=(f1,f2,...,f λ ): C1:f1+f2+...+f λ =d C2:0≤f j ≤min{in i |e i ∈p j },j=1,2,...,λ Then, according to condition C3, all feasible flow vectors F are transformed into polymorphic minimal paths X = (x1, x2, ..., x m ): Among them, λ is the number of all minimal paths from the source to the sink in the network, and f j It is through a very small road j The flow rate, min{W i |e i ∈p j } represents the minimum path p j Maximum capacity; After all the multi-state minimal paths of the network are obtained, the network reliability is calculated using the disjoint sum algorithm.

6. A network reliability assessment method, characterized in that: The method comprises the following steps: S1. Generate a corresponding data set using the data set construction method for network reliability assessment according to any one of claims 1 to 5; S2, preprocessing the reliability assessment data in the data set; S3. Establish a multi-layer residual perceptron network for evaluating network reliability; S4, training the multi-layer residual perceptron network based on the preprocessed data set to obtain a pre-trained multi-layer residual perceptron network; S5. Reliability evaluation of the real-time network is performed through a pre-trained multi-layer residual perceptron network.

7. The network reliability evaluation method according to claim 6, characterized in that: In step S2, the generated data set is preprocessed, which includes: S21, obtain feature data: the maximum capacity vector W = (w1, w2, ..., w n ) for each w i As a feature data, there are n feature data in total. Then, each element G in the input adjacency matrix G ij As a feature data, there are m×m feature data in total; furthermore, each element P in the generated probability matrix P ij Considered as one feature data, there are n×(max(W)+1) feature data in total; S22, concatenate the maximum capacity vector W and the demand d to obtain a new vector (W, d) = (w1, w2, ..., w n ,d), perform maximum and minimum normalization on the concatenated data vector (W,d), and the normalized vector is recorded as (W′,d′)=(w′1,w′2,…,w′ n ,d′); S23, flatten the adjacency matrix G and the probability matrix P. The adjacency matrix after flattening is expressed as: G * =(G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm ) The representation of the flattened probability matrix is: P * =(P 11 ,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj ) In the formula, j represents the number of columns of the probability matrix; S24. Concatenate the normalized vectors with the flattened vectors to obtain the final data form: (w′1,w′2,…,w′ n ,d′,G 11 ,G 12 ,…,G 1m ,G 21 ,G 22 ,…,G 2m ,…,G m1 ,G m2 ,…,G mm , P 11 ,P 12 ,…,P 1j ,P 21 ,P 22 ,…,P 2j ,…,P n1 ,P n2 ,…,P nj ) Complete the data preprocessing process.

8. The network reliability evaluation method according to claim 7, characterized in that: In step S3, the established multi-layer residual perceptron network includes several residual and shortcut branch structures, where: The backbone of the multi-layer residual perceptron network uses a multi-layer perceptron stack. The first layer uses a fully connected layer and a LeakyReLU activation function as the input layer to map the input feature data to a high-dimensional space and expand the number of features. Then it enters the feature extraction module composed of a series of stacked residual structures for feature extraction. Each layer of the network consists of a fully connected layer and a LeakyReLU activation function. The odd-numbered layers are residual structures, and the remaining layers are used to adjust the feature dimensions for the residual structure. In addition, the input layer and the feature extraction module are connected by a shortcut branch to strengthen the representation of the features. There is a fully connected layer on the shortcut branch. Finally, after passing through the final output layer network, which consists of a fully connected layer and a Sigmoid activation function, the final reliability prediction result is obtained based on several sets of extracted features. The Sigmoid activation function is used to activate the output layer network and limit the output probability between 0 and 1.

9. The network reliability evaluation method according to claim 8, characterized in that: The small batch gradient descent training method is adopted to train the established multi-layer residual perceptron network based on the preprocessed data set. The loss function is the mean square error loss, the optimizer is the Adam optimizer, the initial learning rate is set, and the dynamic learning rate decay strategy is the cosine annealing decay strategy. The decay cycle calculation in the decay process is based on the number of iterations. The training is completed when the learning rate reaches the expected target.

Images