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

Abstract

The present application relates to a kind of data set construction method and network reliability evaluation method for network reliability evaluation, belong to computer technology field.The present application is first by initializing basic parameter and the parameter that need batch random generation, randomly generates network maximum capacity vector and demand, randomly generates network probability matrix, according to the calculation network reliability of polymorphic minimal path algorithm, obtains the set of several networks and its reliability, then establishes multilayer residual perception machine network, and the network is trained by preprocessed data set, finally, network reliability evaluation is carried out by the multilayer residual perception machine network of training completion.The present application combines polymorphic minimal path and multilayer perception machine, to meet the characteristics of network variability and complexity, to help decision-makers when facing network reliability factor changes, or when there are multiple options, timely make optimal selection.

Description

Technical Field

[0001] The present 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 typically perform their assigned tasks under specific conditions. Their reliability refers to the probability that the network can successfully transport d units of network flow (representing goods, information, or energy, for example) from a source point s to a sink point t. However, due to internal and external uncertainties, network systems can exhibit varying levels of performance during operation. These types of network systems are referred to as polymorphic networks. They fulfill the functions of resource delivery and allocation, enabling the effective organization and coordination of logistics activities and information transfer. The capacity state of each edge in the network is a random variable, but this random variable follows a certain capacity distribution probability. Polymorphic network system models can more accurately depict and characterize the complex behavior of network systems and are gaining increasing attention and application in system performance evaluation research. Network reliability is a key factor in ensuring the stable operation of supply networks.

[0003] Network reliability is influenced by a variety of factors, including the number of edges and nodes in the network, edge capacity, and network transport requirements. Consequently, network reliability varies. Traditional reliability analysis methods are time-consuming, making it difficult for decision-makers to quickly calculate the reliability of different network structures and thus identify the most reliable one. Summary of the Invention

[0004] In light of this, the present invention aims to provide a dataset construction method and network reliability assessment method for network reliability assessment. This invention focuses on rapid network reliability assessment methods and proposes a novel approach that combines polymorphic minimal paths with a multilayer perceptron (MLP) to address the characteristics of network variability and complexity. This approach helps decision-makers make the optimal choice promptly when faced with changing network reliability factors or when multiple options are available.

[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, and obtain a set of networks and their reliabilities.

[0011] Further, in step a, the initialized basic parameters include: the number of network nodes m, the number of edges n, and the adjacency matrix G of the network topology, wherein the adjacency matrix G is represented as:

[0012]

[0013] wherein G ij ∈{0,1}, G ij =0 represents that the nodes are not connected, and G ij =1 represents that the nodes are connected;

[0014] Initialize the number of network capacity distributions x that need to be randomly generated.

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

[0016] All w i compose the maximum capacity vector W, that is:

[0017] W=(w1,w2,…,w n )

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

[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, the probability matrix corresponding to all capacity states {0,1,...,w i} of each edge e i is randomly generated according to the adjacency matrix of the network and the maximum capacity w i of each edge e i :

[0022]

[0023] In the formula, θ=max(W)+1, which represents the number of columns of 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: P i1 < P i2 < … < P iθ .

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

[0027] First, all feasible flow vectors F = (f1, f2,..., f λ ) satisfying conditions C1 and C2 are solved:

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

[0031]

[0032] Wherein, λ is the number of all minimal paths of the network from the source point to the sink point, f j is the flow through the minimal path p j , and min{W i |e i ∈p j} represents the maximum capacity of the minimal path p j .

[0033] When all polymorphic minimal paths of the network are obtained, the network reliability is calculated by using the disjoint sum algorithm.

[0034] The application further provides a network reliability evaluation method for pre-training a data set established by using the data set construction method for network reliability evaluation, and the method comprises the following steps:

[0035] S1, a corresponding data set is generated by using the data set construction method;

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

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

[0041] S21. Obtain feature data: Set 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 adjacency matrix G is input 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 with 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 11P 12 P 1j P 21 P 22 P 2j P n1 P n2 P nj

[0047] wherein j represents the column number of the probability matrix;

[0048] S24, splicing the normalized vectors and the flattened vectors to obtain a final data form:

[0049] (w'1, w'2, …, w'j) 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] The data preprocessing process is completed.

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

[0053] The main part of the multi-layer residual perception machine network adopts a multi-layer perception machine stack, the first layer takes a fully connected layer and a LeakyReLU activation function as an input layer, and maps the input feature data to a high-dimensional space to expand the feature quantity thereof;

[0054] Then, it enters a feature extraction module composed of a series of stacked residual structures to perform feature extraction, each layer of the network is composed of a fully connected layer and a LeakyReLU activation function, wherein the odd layers are residual structures, the remaining layers are used to adjust the feature dimension for the residual structures, and a shortcut branch is connected between the input layer and the feature extraction module to strengthen the representation of the features, and the shortcut branch has a fully connected layer;

[0055] ​​​Finally, after the final output layer network, the output layer network is composed of a full connection layer and a Sigmoid activation function, and the final reliability prediction result is obtained according to the extracted several groups of features, and 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 perception machine network based on the preprocessed data set, wherein the loss function is mean square error loss, the optimizer is Adam optimizer, the initial learning rate is set, the dynamic learning rate decay strategy is cosine annealing decay strategy, and the decay period in the decay process is calculated with the number of iterations as the reference, and the training is completed when the learning rate reaches the expected target.

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

[0058] The present application focuses on the rapid evaluation method of network reliability, combines the polymorphic minimal path and the multi-layer perception (MLP) to meet the characteristics of network variability and complexity. Thus, it helps decision makers to make optimal choices in the face of changes in network reliability factors or when there are multiple options.

[0059] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, and will be learned from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to make the purposes, technical solutions and advantages of the present application clearer, the preferred detailed description of the present application will be combined with the drawings to describe the present application, wherein:

[0061] Figure 1 It is a network topology diagram;

[0062] Figure 2 It is a flowchart of the network reliability evaluation method of the present application;

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

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

[0065] The present application is described and explained more fully with reference to the following detailed description. Other advantages of the present application will be realized and appreciated by those skilled in the art, and it will be understood to those skilled in the art that the application can take many different embodiments and each of the examples described herein has many different embodiments not specifically described. Thus, the following examples are non-limiting and exemplary.

[0066] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0067] The same or similar components in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only for illustrative purposes, and cannot be understood as a limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0068] Please refer to Figures 1-4 A data set construction method for network reliability evaluation and a network reliability evaluation method.

[0069] Embodiment 1

[0070] The embodiment first provides a data set construction method for network reliability evaluation, which specifically includes the following steps:

[0071] Step a: initialize the basic parameters and the parameters that need to be randomly generated in batches, wherein 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 the number of network capacity distribution x that needs to be randomly generated is initialized.

[0072] Specifically, the number of network nodes m and the number of edges n and the adjacency matrix G describing the network topology structure are initialized:

[0073]

[0074] Among them, G ij ∈{0,1},G ij =0 means 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 formed, that is, 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 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 corresponding probability matrix:

[0078]

[0079] Where 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: The network reliability is calculated using the existing polymorphic minimal path (d-minimal path for short) algorithm. A minimal path is a set of edges in the network, and 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 the present embodiment, as shown in the following table, the above scheme is described in more detail taking a network with a topology of 8 nodes and 12 edges as an example, which includes: Figure 1

[0082] In step a of the present embodiment, the parameters are inputted in batches and the parameters are randomly generated in batches: the number of nodes m = 8, the number of edges n = 12, and the number of randomly generated networks x = 24000.

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

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

[0085] In step d of the present embodiment, the reliability calculation based on the minimal path and the disjoint sum algorithm: the reliability of the 24000 generated networks is calculated by the existing multi-state minimal path (also known as d-minimal path) algorithm, the first stage of the calculation process solves all the d-minimal paths in the network, and the second stage solves the network reliability by using the disjoint sum algorithm, thereby obtaining the label data.

[0086] Among them, the process of solving all the d-minimal 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 the transport demand d = 2, and the multi-state minimal path can be obtained as follows: [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 1 1 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 0 0 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 2 0] [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] [0 2 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 2 0 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 0 2 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 2 0 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 1 0 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 0 1 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 0 1 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 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] According to the obtained polymorphic minimal path, the network reliability can be solved by using the common disjunction algorithm, and the reliability is 0.9091.

[0124] Thus, a data set containing 24000 networks and their reliabilities is obtained, and is used for training the network reliability evaluation model. The trained model is used for network reliability evaluation.

[0125] Embodiment 2

[0126] Based on the data set given in Embodiment 1, a network reliability evaluation method based on a trained multi-layer residual perception machine is proposed, specifically as shown in Figure 2 The method comprises the following steps:

[0127] S1, generating a corresponding data set by using the data set construction method described above;

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

[0129] S3, establishing a multi-layer residual perception machine network for evaluating network reliability;

[0130] S4, training the multi-layer residual perception machine network based on the preprocessed data set, to obtain a pre-trained multi-layer residual perception machine 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: Set 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 adjacency matrix G is input 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 with 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 format:

[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, Figure 4 The multilayer perceptron network with a residual and shortcut branch structure shown in the figure uses a stacked multilayer perceptron as its backbone. The first layer uses a fully connected layer and a LeakyReLU activation function as the input layer, mapping the input feature data to a high-dimensional space and expanding its feature count. The network then enters a feature extraction module consisting 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. Odd-numbered layers are residual structures, and the remaining layers are used to adjust the feature dimensions for the residual structure. Furthermore, a shortcut branch connects the input layer and the feature extraction module to strengthen feature representation and prevent the network from overfitting after feature extraction. The shortcut branch is topped with a fully connected layer. Finally, the network passes 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 from 16 features. The Sigmoid activation function is used to activate the output layer network and limit the output probability to between 0 and 1.

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

[0146] In this embodiment, the training process of the multi-layer residual perception machine network is described in more detail, and the network reliability evaluation effect of the pre-trained multi-layer residual perception machine network is verified.

[0147] In this embodiment, the maximum capacity vector W=(w1, w2, w3, …, w 12 12) and the demand d are spliced, and the spliced vector is (W, d)=(w1, w2, w3, …, w 12 , d), in order to avoid the difference between the feature values, some features dominate the model, and the accuracy is improved, so that different dimensional features can be compared fairly. Therefore, the (W, d) maximum minimum normalization processing is performed, and (W', d')=(w'1, w'2, w'3, …, w' 12 , d') is obtained. The adjacency matrix G mm (m=8) with a feature number of 64 and the probability matrix P nθ (n=12, θ=8) with a feature number of 96 are flattened, and the representation form of the flattened adjacency matrix is obtained:

[0148] G*=(G 11 , G 12 , …, G 18 , G 21 , G 22 , …, G 28 , …, G 81 , G 82 , …, G 88 )

[0149] The representation form 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] The normalized vector and the flattened vector are spliced to obtain 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] The independent data preprocessing process is completed.

[0154] The backbone of the multi-layer residual perception machine network of the embodiment adopts a 9-layer perception machine stack, the first layer takes a full connection layer and a LeakyReLU activation function as an input layer, maps 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 network are stacked, each layer of network is composed of a full connection layer and a LeakyReLU activation function, of which 3 layers (odd layers) are residual structures, and the remaining two layers are used to adjust the feature dimension for the residual structure. Specifically, 256 features first pass through the first residual layer of the feature extraction module, then pass through a layer of network to map the number of features to 128, then pass through the third layer (i.e., the second residual structure) of the feature extraction module, then pass through the fourth layer of network of the feature extraction module to map the number of features to 64, and finally pass through the last layer (i.e., the third residual structure) of the feature extraction module to complete the feature extraction process. And, the output layer is connected to the back of the feature extraction module by a shortcut branch, which is used to strengthen the representation of the features and avoid the network from falling into overfitting after feature extraction. There is a full connection layer on the shortcut branch, which is used to map the 256 features after the output layer to 64 after the feature extraction layer. Then, the 64 features after the feature extraction pass through two layers of network (i.e., the 7th layer and the 8th layer of the entire network), which respectively map the number of features to 32 and 16. Finally, the last layer of the output layer network outputs the final reliability prediction result from 16 features, and the Sigmoid activation function is used to activate the output layer network and limit the output probability to between 0-1. This design method can alleviate the gradient disappearance problem, improve the model generalization ability, and improve the accuracy.

[0155] In the training, the initial learning rate is set to 10-3, the dynamic learning rate decay strategy is cosine annealing decay strategy, the decay period in the decay process is calculated with reference to the number of iterations, and the final learning rate is 10-5.

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

[0157] For these three indicators, lower values indicate that the model has good prediction performance, because they reflect the degree of fitting between the model prediction value and the true value. Specifically, MSE and RMSE are more sensitive to outliers due to the square term, so lower values mean that the model effectively reduces large prediction errors; and 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 has high robustness and effectiveness in capturing data patterns and achieving generalization.

[0158] The final training result MSE value is 0.0027, MAE value is 0.004335, and RMSE value is 0.0045. Lower values indicate that the model has good prediction performance, which means that the method has high robustness and effectiveness in capturing data patterns and achieving generalization.

[0159] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should be covered by the claims of the present application.

Claims

1. A method for dataset construction for network reliability evaluation, 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 network maximum capacity vector and demand; Step c: randomly generate network probability matrix; Step d: calculate network reliability according to the multi-state minimal path algorithm to obtain a set of networks and their reliabilities; The set is preprocessed to train a multi-layer residual perception machine network, and a pre-trained multi-layer residual perception machine network is obtained; the real-time network is evaluated for reliability through the pre-trained multi-layer residual perception machine network; In step a, the initialized basic parameters include: the number of network nodes , the number of edges , and the adjacency matrix of the network topology , wherein the adjacency matrix is represented as: wherein, represents no connection between nodes, represents a connection between nodes; Initializing the number of network capacity distributions that need to be randomly generated ; In step b, each edge is randomly generated according to the adjacency matrix of the network the maximum capacity of wherein , and , is a positive integer; all vector of maximum capacities i.e.: Randomly generating network requirements , and is an integer, wherein: representing the maximum capacity vector of the current network the maximum value in the vector In step c, a probability matrix is generated according to the adjacency matrix of the network and the maximum capacity of each edge Randomly generate edges All capacity states Corresponding probability matrix:​ P= wherein represents the number of columns of the probability matrix; each element represents the probability that the first edge is in the state and the second edge is in the state and the third edge is in the state ; For each If Then wherein ; Probability matrix Each row element of the probability matrix satisfies the following constraint: ​​ 2. The data set construction method for network reliability evaluation according to claim 1, characterized in that: In step d, all multi-state minimal paths in the network are solved, and the process is as follows: First, solve for all feasible flow vectors that satisfy conditions C1 and C2 : Then all feasible flow vectors are converted to polymorphic minimal paths according to condition C3 :​ wherein, When all multi-state minimal paths in the network are obtained, the network reliability is calculated by using the disjoint sum algorithm. is the number of all minimal paths from the source to the sink of the network, is the flow through the minimal path , and represents the maximum capacity of the minimal path . The method comprises the following steps:

3. A method of network reliability evaluation, characterized by: S1, generating a corresponding data set by using the data set construction method for network reliability evaluation according to any one of claims 1-2; S2, preprocessing the reliability evaluation data in the data set; S3, establishing a multi-layer residual perception machine network for evaluating network reliability; S4, training the multi-layer residual perception machine network based on the preprocessed data set to obtain a pre-trained multi-layer residual perception machine network; S5, evaluating the real-time network for reliability through the pre-trained multi-layer residual perception machine network. In step S2, the generated data set is preprocessed, which comprises:

4. The network reliability evaluation method of claim 3, wherein: S23, performing flattening operation on the adjacency matrix G and the probability matrix P, and the representation of the flattened adjacency matrix is: S21. Get feature data: convert the maximum capacity vector Each of Considered as a feature data, there are feature data, and then input each element in the adjacency matrix G Considered as a feature data, there are feature data; furthermore, each element in the generated probability matrix P Considered as a feature data, there are Characteristic data; S22, the maximum capacity vector Splicing with demand d to get a new vector , the spliced data vector Maximum minimum normalization processing, the normalized vector is denoted as ; The representation of the flattened probability matrix is: S24, splicing the normalized vectors and the flattened vectors to obtain the final data form: In the formula, denotes the number of columns of the probability matrix; The data preprocessing process is completed. In step S3, the multi-layer residual perception machine network comprises a plurality of residual and shortcut branch structures, wherein, 5. The network reliability evaluation method of claim 4, wherein: The backbone of the multi-layer residual perception machine network adopts a multi-layer perception machine stack, the first layer takes a fully connected layer and a LeakyReLU activation function as an input layer, and maps the input feature data to a high-dimensional space to expand the feature quantity; Then, it enters a feature extraction module composed of a series of stacked residual structures, and performs feature extraction, each layer of the network is composed of a fully connected layer and a LeakyReLU activation function, wherein the odd layers are residual structures, and the remaining layers are used to adjust the feature dimension for the residual structures, and a shortcut branch is connected between the input layer and the feature extraction module to strengthen the representation of the features, and the shortcut branch has a fully connected layer; Finally, the final output layer network is passed through, and the output layer network is composed of a fully connected layer and a Sigmoid activation function, and the final reliability prediction result is obtained according to the extracted several groups of features, and the Sigmoid activation function is used to activate the output layer network and limit the output probability to between 0 and 1. ​ 6. The network reliability evaluation method of claim 5, wherein: The multilayer residual perception machine network is trained based on the pretreated data set by using a small batch gradient descent training method, wherein a mean square error loss is used as the loss function, an Adam optimizer is used as the optimizer, an initial learning rate is set, a cosine annealing decay strategy is used as the dynamic learning rate decay strategy, and the decay period in the decay process is calculated by taking the iteration number as the reference; and the training is completed when the learning rate reaches the expected target.

Images