Communication performance evaluation method for intelligent Internet of Things networking and terminal equipment

Abstract

The invention provides a communication performance evaluation method for intelligent Internet of Things networking and terminal equipment, and relates to the technical field of communication. The method comprises the following steps: acquiring communication performance index data of a to-be-evaluated intelligent Internet of Things networking; inputting the communication performance index data into a pre-trained neural network model to obtain a communication performance evaluation result of the intelligent Internet of Things networking to be evaluated; wherein the pre-trained evaluation model is obtained by training according to output data of the pre-trained distributed cloud model; the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the to-be-evaluated intelligent Internet of Things networking. According to the method, the accuracy and the efficiency of communication performance evaluation of the intelligent Internet of Things networking can be improved on the whole level, and the problems of low accuracy and low efficiency when a traditional cloud model is used for evaluating the communication performance of the intelligent Internet of Things networking are solved.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular, to a method for evaluating the communication performance of an intelligent Internet of Things (IoT) network and a terminal device. Background Art

[0002] With the increasing application requirements of intelligent IoT networks in fields such as unmanned clusters and smart homes, higher requirements are put forward for the communication performance of intelligent IoT networks. In these applications, the evaluation of the communication performance of intelligent IoT networks is the key to ensuring the quality of wireless communication. By evaluating key indicators such as signal quality, access capacity, and network performance, potential problems can be detected and solved early, thereby improving the reliability and stability of the intelligent interconnection system.

[0003] However, with the continuous growth of the intelligent IoT system, the amount of communication interaction data has increased sharply, which poses new challenges to the evaluation of the communication performance of intelligent IoT networks.

[0004] In the broad field of data science and machine learning, cloud model evaluation has achieved remarkable results in intelligent control, frequency-hopping radios, and large system effectiveness evaluation. However, when dealing with massive data, traditional cloud model evaluation methods may encounter performance bottlenecks such as memory limitations and calculation delays, which limit the wide application of cloud models in the big data field. In addition, the accuracy and reliability of cloud model evaluation may also be affected. Since the evaluation process involves sampling and calculation of multiple random variables, the evaluation results may have certain fluctuations and instabilities, and such fluctuations may weaken the accuracy and reliability of the evaluation results, thereby affecting the accuracy of subsequent decisions. Summary of the Invention

[0005] Embodiments of the present invention provide a method for evaluating the communication performance of an intelligent IoT network and a terminal device, so as to solve the problems of low accuracy and low efficiency in evaluating the communication performance of an intelligent IoT network by using a traditional cloud model.

[0006] In a first aspect, embodiments of the present invention provide a method for evaluating the communication performance of an intelligent IoT network, including:

[0007] Obtaining communication performance index data of the intelligent IoT network to be evaluated;

[0008] Inputting the communication performance index data into a pre-trained neural network model to obtain an evaluation result of the communication performance of the intelligent IoT network to be evaluated;

[0009] Among them, the pre-trained evaluation model is trained according to the output data of a pre-trained distributed cloud model; the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the intelligent IoT network to be evaluated.

[0010] In a possible implementation, the output data of the pre-trained distributed cloud model is obtained through the following steps:

[0011] Input the historical communication performance metric data and its corresponding comprehensive weight into the pre-trained distributed cloud model;

[0012] In the pre-trained distributed cloud model, calculate the comprehensive eigenvalue of each cloud node according to the historical communication performance metric data and its corresponding comprehensive weight; obtain the evaluation result of each historical communication performance metric data according to the comprehensive eigenvalue of each cloud node;

[0013] Take each historical communication performance metric data and its corresponding evaluation result as the output data of the pre-trained distributed cloud model.

[0014] In a possible implementation, in the pre-trained distributed cloud model, calculating the comprehensive eigenvalue of each cloud node according to the historical communication performance metric data and its corresponding comprehensive weight includes:

[0015] In the pre-trained distributed cloud model, evaluate each historical communication performance metric data using a preset evaluation method to obtain the first eigenvalue of each cloud node; where each historical communication performance metric data corresponds to one cloud node;

[0016] Calculate the second eigenvalue according to the first eigenvalue of each cloud node and the comprehensive weight corresponding to each historical communication performance metric data;

[0017] Calculate the comprehensive eigenvalue of each cloud node according to the second eigenvalue.

[0018] In a possible implementation, obtaining the evaluation result of each historical communication performance metric data according to the comprehensive eigenvalue of each cloud node includes:

[0019] For any category of historical communication performance metric data, calculate the cloud model membership degree of this category of historical communication performance metric data according to the comprehensive eigenvalue of each cloud node corresponding to this category of historical communication performance metric data;

[0020] Take the comprehensive eigenvalue of the cloud node whose membership degree meets the preset condition as the evaluation eigenvalue of this category of historical communication performance metric data;

[0021] Obtain the evaluation result of this category of historical communication performance metric data according to this evaluation eigenvalue;

[0022] Obtain the evaluation result of each historical communication performance metric data according to the evaluation result of each category of historical communication performance metric data.

[0023] In a possible implementation, the pre-trained neural network model is trained in the following way:

[0024] Using the historical communication performance index data as the input layer of the neural network model, and using the evaluation results corresponding to the historical communication performance index data as the output layer of the neural network model, model training is performed to obtain a pre-trained evaluation model.

[0025] In a possible implementation, the comprehensive weights corresponding to each communication performance index data are determined by the following method:

[0026] Using the subjective weighting method, calculate the subjective weights corresponding to each communication performance index data;

[0027] Using the objective weighting method, calculate the objective weights corresponding to each communication performance index data;

[0028] Combine the subjective weights and objective weights corresponding to each communication performance index data to obtain the comprehensive weights corresponding to each communication performance index data.

[0029] In a possible implementation, using the subjective weighting method to calculate the subjective weights corresponding to each communication performance index data includes:

[0030] Using the analytic hierarchy process to evaluate each communication performance index data to obtain a discriminant matrix;

[0031] According to the discriminant matrix, obtain the subjective weights corresponding to each communication performance index data.

[0032] In a possible implementation, using the objective weighting method to calculate the objective weights corresponding to each communication performance index data includes:

[0033] Calculate the comparison intensity corresponding to each communication performance index data, and the conflict quantification index between each communication performance index data and other communication performance index data;

[0034] According to the comparison intensity and conflict quantification index corresponding to each communication performance index data, obtain the objective weights corresponding to each communication performance index data.

[0035] In a possible implementation, combining the subjective weights and objective weights corresponding to each communication performance index data to obtain the comprehensive weights corresponding to each communication performance index data includes:

[0036] For any type of communication performance index data, perform the following steps:

[0037] Using the particle swarm optimization algorithm, combine the subjective weights and objective weights corresponding to this type of communication performance index data to obtain multiple weight combinations;

[0038] Evaluate each weight combination to obtain the optimal weight combination;

[0039] Take the optimal weight combination as the comprehensive weight corresponding to the communication performance index data of this category.

[0040] In a second aspect, an embodiment of the present invention provides a terminal device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method in the first aspect above or any possible implementation manner of the first aspect.

[0041] In the embodiment of the present invention, considering that the distributed cloud model can not only greatly improve the evaluation rate, but also analyze and process different cloud nodes, the embodiment of the present invention constructs a distributed cloud model according to the characteristics and requirements of the intelligent IoT networking to be evaluated; enabling it to accurately locate the communication performance of nodes, enhancing the accuracy of evaluation, and making the accuracy of the finally obtained output data higher. Correspondingly, the accuracy of the neural network model obtained by using the output data of the pre-obtained distributed cloud model as the training set is also correspondingly improved. In addition, combining the distributed cloud model with the neural network model can reduce problems such as memory limitations and calculation delays caused by large amounts of data in the traditional cloud model, thereby improving the accuracy and efficiency of the communication performance evaluation of the intelligent IoT networking from the overall level, and solving the problems of low accuracy and low efficiency in evaluating the communication performance of the intelligent IoT networking by the traditional cloud model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is the implementation flowchart of the communication performance evaluation method for the intelligent IoT networking provided by the embodiment of the present invention;

[0043] Figure 2 is the overall architecture diagram of the communication performance evaluation method for the intelligent IoT networking provided by the embodiment of the present invention;

[0044] Figure 3 is the implementation flowchart of the communication performance evaluation method for the intelligent IoT networking provided by another embodiment of the present invention;

[0045] Figure 4 is the structural schematic diagram of the communication performance evaluation device for the intelligent IoT networking provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The following will describe the embodiments of the present invention in detail with reference to the accompanying drawings.

[0047] Figure 1 is the implementation flowchart of the communication performance evaluation method for the intelligent IoT networking provided by the embodiment of the present invention. As Figure 1 shown, the method may include:

[0048] Step 110: Obtain the communication performance index data of the intelligent IoT networking to be evaluated.

[0049] In this embodiment, the communication performance index data of the intelligent IoT network to be evaluated can be divided into primary index data and secondary index data; among them, the primary index data is determined based on the secondary index data.

[0050] Specifically, the secondary index data may include data such as packet delivery ratio, packet byte efficiency, average number of routing hops, end-to-end average delay, delay jitter, end-to-end average throughput, peak throughput, frequency band interference test, signal stability, and energy per unit packet.

[0051] The primary index data may include the reliability, delay, throughput, anti-interference ability, and energy of the intelligent IoT network to be evaluated. Among them, the reliability is determined by the packet delivery ratio, packet byte efficiency, and average number of routing hops; the delay is determined by the end-to-end average delay and delay jitter; the throughput is determined by the end-to-end average throughput and peak throughput; the anti-interference ability is determined by the frequency band interference test and signal stability; the energy is determined by the energy per unit packet.

[0052] Step 120: Input the communication performance index data into a pre-trained neural network model to obtain the communication performance evaluation result of the intelligent IoT network to be evaluated.

[0053] Among them, the pre-trained evaluation model is trained according to the output data of the pre-trained distributed cloud model; the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the intelligent IoT network to be evaluated.

[0054] In this embodiment, each corresponding secondary index data in the communication performance index data is input into the pre-trained neural network model, and through the calculation of the neural network model, the communication performance evaluation result of the intelligent IoT network to be evaluated can be obtained.

[0055] In this embodiment, the pre-trained neural network model can be a lightweight Convolutional Neural Networks (CNN)-Transformer hybrid neural network model; among them, the CNN part is used to extract the local features of communication data, and the Transformer part is used to capture the global dependencies of the data. This hybrid architecture can handle local details and global structures simultaneously, improving the evaluation accuracy of the model. Constructing a lightweight CNN-Transformer hybrid neural network model combines the advantages of convolutional neural networks (CNN) and Transformer architectures for in-depth learning analysis of the communication performance of intelligent IoT networking. In addition, to improve the real-time performance and adaptability of the model, the present invention adopts a lightweight design. By reducing the number of model parameters and computational complexity, the computational resource consumption of the model is reduced to solve the problem that traditional cloud model evaluation methods may encounter performance bottlenecks when processing massive data.

[0056] It can be seen that the accuracy and reliability of traditional cloud model evaluation may be affected due to the sampling and calculation of multiple random variables during the evaluation process, while the neural network model can solve this problem.

[0057] Furthermore, when the neural network model is trained, a large amount of training data is required, and the data needs corresponding labels. If the labels are set manually, it will increase the time cost. Therefore, in this embodiment, the historical communication performance index data is not directly used as the training set to train the pre-trained neural network model.

[0058] In this embodiment, a pre-trained distributed cloud model is used to preliminarily evaluate the communication performance index data, calculate the corresponding evaluation results for each communication performance index data, that is, the corresponding labels, and use these data as the training set, which can solve the problem of the large time cost required for traditional neural network training.

[0059] Furthermore, the CNN-Transformer hybrid neural network model used in this application can overcome the problem that the evaluation results of traditional cloud models may have certain fluctuations and instabilities due to the sampling and calculation of multiple random variables during the evaluation process, and thus can improve the accuracy and reliability of the evaluation results.

[0060] In this embodiment, to implement the method provided in this embodiment, it is necessary to pre-determine the comprehensive weights of each communication performance index data, pre-train the distributed cloud model, and train the neural network model. Figure 2 It is the overall architecture diagram of the communication performance evaluation method for intelligent IoT networking provided by the embodiments of the present invention. The following combines Figure 2The process of determining the comprehensive weights of each communication performance index data, training the distributed cloud model, and training the neural network model in this embodiment will be described.

[0061] First, the process of determining the comprehensive weights of each communication performance index data will be described:

[0062] In an alternative embodiment, the comprehensive weights corresponding to each communication performance index data are determined by the following method:

[0063] Step 210: Use the subjective weighting method to calculate the subjective weights corresponding to each communication performance index data.

[0064] Step 220: Use the objective weighting method to calculate the objective weights corresponding to each communication performance index data.

[0065] Step 230: Combine the subjective weights and objective weights corresponding to each communication performance index data to obtain the comprehensive weights corresponding to each communication performance index data.

[0066] In this embodiment, the subjective and objective weighting methods can be used respectively to calculate the subjective weights and objective weights of each communication performance index data. Then, by integrating the advantages of the subjective and objective weights, the deviation caused by a single method can be reduced, and the comprehensive weights corresponding to each communication performance index data can be obtained.

[0067] In an alternative embodiment, in step 210, using the subjective weighting method to calculate the subjective weights corresponding to each communication performance index data may include:

[0068] Use the analytic hierarchy process to evaluate each communication performance index data to obtain a judgment matrix.

[0069] According to the judgment matrix, obtain the subjective weights corresponding to each communication performance index data.

[0070] In this embodiment, using the analytic hierarchy process, each communication performance index data is evaluated according to expert opinions to construct a judgment matrix. Among them, the scale and meaning of the analytic hierarchy process are shown in Table 1:

[0071] Table 1 Scale and meaning of the analytic hierarchy process

[0072]

[0073]

[0074] As shown in Table 1, the values of the elements in the judgment matrix reflect people's understanding of the relative importance of each element. Generally, a scale method of 1-9 and their reciprocals is adopted. However, when the importance of the compared factors can be explained by a ratio with practical significance, the value of the corresponding element in the judgment matrix takes this ratio. That is, the judgment matrix A=(a ij ) n×n , i, j = 1, 2,..., n; where, a ij represents the elements in the matrix.

[0075] Perform a consistency test on the discrimination matrix to improve scientificity and consistency. Specifically, it can include: calculating the maximum eigenvalue λ max of the judgment matrix S, and its corresponding eigenvector A; where, the eigenvector A is the importance ranking of each communication performance index data, and it is also the basis for the distribution of weight coefficients.

[0076] Calculate the consistency index of the discrimination matrix:

[0077]

[0078] Judge whether the consistency index meets the following conditions:

[0079]

[0080] Among them, RI is the average random consistency index. It is constructed by randomly generating 500 sample matrices. The construction method is to randomly fill the upper triangular terms of the matrix with the scale and their reciprocals. The values of the main diagonal terms are always 1, and the terms in the corresponding transposed positions adopt the reciprocals of the random numbers in the above corresponding positions. Then calculate the consistency index for each random sample matrix, and average these CI values to obtain the average random consistency index RI value;

[0081] If it is satisfied, it is considered that the result of the hierarchical analysis ranking has satisfactory consistency, that is, the distribution of weight coefficients is reasonable; otherwise, adjust the values of the elements in the judgment matrix, re-allocate the values of the weight coefficients, and then obtain the subjective weights corresponding to each communication performance index data.

[0082] In an alternative embodiment, in step 220, an objective weighting method is adopted to calculate the objective weights corresponding to each communication performance index data, which may include:

[0083] Calculate the comparison intensity corresponding to each communication performance index data, and the conflict quantification index between each communication performance index data and other communication performance index data.

[0084] According to the comparison intensity and conflict quantification index corresponding to each communication performance index data, obtain the objective weights corresponding to each communication performance index data.

[0085] In this embodiment, the Criteria Importance Through Intercriteria Correlation (CRITIC) method can be used to calculate the objective weights of the communication performance index data.

[0086] Specifically, through the following formula, the contrast intensity corresponding to each communication performance index data is calculated:

[0087]

[0088] where n represents the total number of categories of communication performance index data; x ij represents the comparison value of the jth communication performance index data relative to other communication performance index data; represents the average comparison value of the jth communication performance index data.

[0089] In the CRITIC method, the conflict of indicators is expressed in the form of a correlation coefficient. The larger the correlation coefficient between indicators, the stronger the correlation with other indicators, the more the same information is reflected, and the smaller the weight. The correlation coefficient r ij The specific calculation formula is:

[0090]

[0091] Correspondingly, the conflict quantification index corresponding to each communication performance index data is:

[0092]

[0093] According to the contrast intensity and conflict quantification index corresponding to each communication performance index data, the amount of information is calculated:

[0094]

[0095] where the larger the amount of information, the greater the weight of the communication performance index data.

[0096] According to the amount of information, the objective weights corresponding to each communication performance index data are obtained:

[0097]

[0098] In an alternative embodiment, combining the subjective weights and objective weights corresponding to each communication performance index data in step 230 to obtain the comprehensive weights corresponding to each communication performance index data may include:

[0099] For any category of communication performance index data, the following steps are performed:

[0100] Using the particle swarm optimization algorithm, the subjective weights and objective weights corresponding to the communication performance index data of this type are combined to obtain multiple weight combinations.

[0101] Evaluate each weight combination to obtain the optimal weight combination.

[0102] Take the optimal weight combination as the comprehensive weight corresponding to the communication performance index data of this type.

[0103] In this embodiment, the particle swarm optimization algorithm can be selected to combine the subjective weights and objective weights corresponding to each communication performance index data, so as to obtain the comprehensive weights corresponding to each communication performance index data.

[0104] Specifically, select the particle swarm optimization algorithm to optimize the combination of the weights of different evaluation methods to obtain the weights of the combined evaluation method.

[0105] Exemplarily, when using m evaluation methods to compare k schemes, an evaluation method weight vector matrix H will be obtained. m×k .

[0106] Among them, H = (H ij ), where i = 1, 2,..., m; j = 1, 2,..., k; m×k , i = 1, 2,…, m; j = 1, 2,…, k;

[0107] In the formula,

[0108] The weight vector of the combined evaluation is h = [h 1 , h 2 , …, h k , where the smaller the deviation between h and different evaluation methods, the better. Using the principle of the minimum sum of squared deviations, the objective function for optimizing the evaluation method is constructed as shown in the following formula:

[0109]

[0110] In the formula

[0111] Based on the obtained weight combinations, perform evaluation optimization to obtain the optimal weight combination:

[0112] (1) Parameter initialization: Initialize the particle swarm size Ns to 30, the maximum number of iterations T to 100, and the particle dimension d to the dimension of the weight vector.

[0113] (2) Initialize the velocity and position of the particles: Improve the particle swarm algorithm in terms of initialization and inertia weight according to the characteristics of the objective function. The objective function adopts the principle of the minimum sum of squared deviations. According to the weight distribution of each scheme, in accordance with the constraint condition h j ∈(min(H ij), max(H ij ), i = 1, 2, …, m, randomly generate the initial positions of each particle. According to the characteristics of the particle dimensions, according to Equation Initialize the particle velocity to ensure that the individual moves within the weight dimension.

[0114] (3) Calculate the fitness value of each particle according to the above objective function formula.

[0115] (4) Update the positions and velocities of the particles: To ensure that the algorithm has good global and local optimization capabilities, the sigmoid function principle is used for the inertia weight and updated in a decreasing manner according to Equation Perform asynchronous improvement operations on the learning factors according to the following formula. For the individual learning factor c1 with a larger initial value, it continuously decreases as the number of iterations increases, and for the group learning factor c2 with a smaller initial value, it continuously increases as the number of iterations increases, improving the optimization ability and speed of the algorithm. Update the positions and velocities of each particle using the improved inertia weight and learning factors.

[0116]

[0117] In the formula: ω is the inertia weight, t is the current number of iterations, T is the maximum number of iterations, c 1s is c 1 The initial value of is taken as 1.5, c 1e is the final value of c1 taken as 0.5, c 2s is c 2 The initial value of is taken as 0.5, c 2e is c 2 The final value of is taken as 1.5.

[0118] (5) Obtain the historical optimal solutions and global optimal solutions of the particles.

[0119] (6) Determine the iteration termination condition. If the termination condition is not satisfied, return to step (3). If it is satisfied, end and obtain the optimal weight combination corresponding to each communication performance index data.

[0120] Take the optimal weight combination as the comprehensive weight corresponding to each communication performance index data

[0121] The following will illustrate the training process of the distributed cloud model through the following related embodiments:

[0122] In an alternative embodiment, the output data of the pre-trained distributed cloud model is obtained through the following steps:

[0123] Step 310: Input the historical communication performance index data and its corresponding comprehensive weight into the pre-trained distributed cloud model;

[0124] Step 320: In the pre-trained distributed cloud model, calculate the comprehensive eigenvalue of each cloud node according to the historical communication performance index data and its corresponding comprehensive weight; obtain the evaluation results of each historical communication performance index data according to the comprehensive eigenvalues of each cloud node;

[0125] Step 330: Use each historical communication performance index data and its corresponding evaluation result as the output data of the pre-trained distributed cloud model.

[0126] In this embodiment, the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the intelligent IoT networking to be evaluated. The model consists of multiple cloud nodes. Each cloud node is responsible for processing a part of the communication data and collaborates through inter-cloud communication. Utilizing the advantages of the distributed cloud model, the evaluation task is assigned to each cloud node for parallel processing. Each cloud node performs calculations and analyses based on its own data and evaluation metrics to obtain local evaluation results. The local evaluation results of each cloud node are integrated and calculated to obtain the overall evaluation conclusion.

[0127] In this embodiment, the use of the distributed cloud model has the following advantages:

[0128] Through distributed processing, the evaluation task is assigned to multiple cloud nodes for parallel processing, greatly improving the evaluation rate. Using the distributed cloud model for evaluation can process and analyze different nodes, and can more accurately locate the node communication performance, thereby enhancing the accuracy of the evaluation. Efficiently process complex data: The distributed cloud model can effectively handle the ambiguity and randomness in the evaluation of wireless networking communication performance and is applicable to large-scale complex network environments. By using the output data of the distributed cloud model as the training data of the neural network model, the automation of the evaluation task can be achieved, reducing the cost and complexity of manual evaluation.

[0129] In an optional embodiment, in step 320, in the pre-trained distributed cloud model, calculating the comprehensive eigenvalue of each cloud node according to the historical communication performance index data and its corresponding comprehensive weight may include:

[0130] In the pre-trained distributed cloud model, use a preset evaluation method to evaluate each historical communication performance index data to obtain the first eigenvalue of each cloud node; among them, each historical communication performance index data corresponds to a cloud node.

[0131] Calculate the second eigenvalue according to the first eigenvalue of each cloud node and the comprehensive weight corresponding to each historical communication performance index data.

[0132] Calculate the comprehensive eigenvalue of each cloud node according to the second eigenvalue.

[0133] In this embodiment, first, the cloud nodes can be evaluated according to their digital characteristics to obtain the standard cloud. The digital characteristics can be the expected value Ex, entropy En, and hyper-entropy He. The evaluation levels can be determined by dividing the corresponding scoring intervals as needed. For example, excellent (90 - 100], good (75 - 90], medium (50 - 75], acceptable (25 - 50], poor [0.0 - 25.0]. The calculation formula for the standard cloud can be:

[0134]

[0135] where Ex v is the expected value of the standard cloud model; En v is the entropy of the standard cloud model; He v is the hyper-entropy of the standard cloud model; is the upper limit of the scoring interval corresponding to the standard cloud model; is the lower limit of the scoring interval corresponding to the standard cloud model; k is a constant, and the empirical value can be taken as 0.1.

[0136] The preset evaluation method can be the expert scoring method. By using the expert scoring method to evaluate each index, the index evaluation matrix can be obtained.

[0137] Exemplarily, there are m experts and n communication performance index data, where z ij represents the evaluation result of the i-th expert on the j-th index. i = 1, 2,..., m; j = 1, 2,..., n. By processing the index evaluation matrix and calculating according to the following formula, the cloud node corresponding to the j-th communication performance index data is obtained as C j (Ex j , En j , He j ), j = 1, 2,..., n. Each cloud node is responsible for calculating the first eigenvalue corresponding to a secondary index data. The calculation formula for the first eigenvalue is:

[0138]

[0139] where is the data variance corresponding to the j-th communication performance index data.

[0140] Since each cloud node corresponds to a communication performance index data, after calculating the corresponding eigenvalues of each cloud node, according to the first eigenvalue of each cloud node and the comprehensive weight corresponding to the communication performance index data represented by this cloud node, the second eigenvalue of the first-level index corresponding to this cloud node is calculated;

[0141] where the calculation formula is:

[0142]

[0143] Wherein, n is the number of secondary indicators; W j is the comprehensive weight of the j-th secondary indicator.

[0144] According to the obtained second eigenvalue of each cloud node, that is, the eigenvalue corresponding to the primary indicator of each cloud node, the comprehensive eigenvalue is calculated through the following formula:

[0145]

[0146] Wherein, Ex i , En i , He i are the comprehensive eigenvalues of the i-th cloud node; q is the number of primary indicators; W b is the comprehensive weight of the b-th primary indicator.

[0147] In an optional embodiment, obtaining the evaluation results of each historical communication performance index data according to the comprehensive eigenvalues of each cloud node in step 320 may include:

[0148] For any type of historical communication performance index data, according to the comprehensive eigenvalues of the cloud nodes corresponding to this type of historical communication performance index data, calculate the cloud model membership degree of this type of historical communication performance index data.

[0149] Use the comprehensive eigenvalue of the cloud node whose membership degree meets the preset condition as the evaluation eigenvalue of this type of historical communication performance index data.

[0150] According to this evaluation eigenvalue, obtain the evaluation result of this type of historical communication performance index data.

[0151] According to the evaluation results of each type of historical communication performance index data, obtain the evaluation results of each historical communication performance index data.

[0152] In this embodiment, each type of historical communication performance index data has multiple cloud nodes, that is, multiple cloud models. For any type of historical communication performance index data, calculate the membership degree between any two cloud models through the following formula:

[0153]

[0154] Wherein, V 1 , V 2 respectively represent two cloud models; V(·) represents the similarity calculation between two cloud models; p i represents the membership degree.

[0155] According to the above method, select the comprehensive eigenvalue corresponding to the cloud node with the largest membership degree as the evaluation eigenvalue of each type of historical communication performance index data.

[0156] The training process of the neural network model is described below through the following related embodiments:

[0157] In an alternative embodiment, the pre-trained neural network model is trained in the following manner:

[0158] Use each historical communication performance metric data as the input layer of the neural network model, and use the evaluation results corresponding to each historical communication performance metric data as the output layer of the neural network model to perform model training to obtain a pre-trained evaluation model.

[0159] In this embodiment, the neural network model can be obtained through the following steps:

[0160] 1. Structure design:

[0161] (1) Use each communication performance metric data obtained by the distributed cloud model and the evaluation results corresponding to each communication performance metric data as the training set for training the neural network model.

[0162] First, perform normalization processing on the output data of the distributed cloud model. Based on its comprehensive evaluation result, the output data of the cloud model is defined as five types of labels.

[0163] Then, use each communication performance metric data in the output data, that is, each secondary index data, as the input layer of the neural network, and use the comprehensive evaluation results corresponding to each secondary data as the output layer of the neural network. Optionally, the number of data points in the input layer can be 10, and the number of data points in the output layer can be 5.

[0164] (2) Perform lightweight design on the neural network model:

[0165] The lightweight design includes using Depthwise Separable Convolution (DSC) and Attention Mechanism (AM) technologies to ensure that the neural network model can still operate efficiently in a resource-constrained environment.

[0166] In this embodiment, the CNN layer is composed of multiple depthwise separable convolution layers. Each depthwise separable convolution layer is followed by a ReLU activation function to extract local features of the input data. The output of the convolution layer is downsampled through a max pooling layer to reduce the dimension of the features while retaining important feature information.

[0167] Among them, the convolution layer can be expressed as:

[0168] O (l) =σ(W (l) *I(l) +b (l) )

[0169] In the formula, O (l) is the output of the l-th layer, W (l) is the convolution kernel, I (l) is the input feature map, b (l) is the bias term, σ is the ReLU activation function, and * represents the convolution operation.

[0170] The ReLU activation function can be expressed as:

[0171] σ(x) = max(0, x)

[0172] In the formula, x is the input value. If the input value is positive, the value is output; if the input value is negative, the output is 0.

[0173] The max pooling layer can be expressed as:

[0174] P (l) = max(O (l) )

[0175] (3) Training the Transformer layer:

[0176] In this embodiment, training the Transformer layer based on the self-attention mechanism can capture the global dependency relationships between input features. Each Transformer layer consists of a multi-head self-attention module (Multi-Head Attention, MHA) and a feed-forward neural network (Feed-Forward Network, FFN).

[0177] Among them, the multi-head self-attention module can be expressed as:

[0178] MultiHead(Q, K, V) = Concat(head 1 , …, head h )W O

[0179] In the formula, W i Q 、 are learnable weight matrices, and W O is the output weight matrix.

[0180] The self-attention function can be expressed as:

[0181]

[0182] In the formula, Q, K, and V are the query, key, and value matrices respectively, dk is the dimension of the key vector.

[0183] The feedforward neural network can be expressed as:

[0184] FFN(x) = max(0, xW 1 + b 1 )W 2 + b 2

[0185] In the formula, W 1 , W 2 are weight matrices, and b 1 , b 2 are biases.

[0186] (4) Generate the feature fusion layer:

[0187] The feature fusion layer fuses the output features of the CNN layer and the Transformer layer to generate a comprehensive feature vector.

[0188] (5) Output layer generation:

[0189] The output layer maps the comprehensive feature vector to the final evaluation metric through a fully connected layer (Dense Layer). The output of the fully connected layer is normalized by the softmax function to obtain the probability distribution of each evaluation metric.

[0190] Among them, the fully connected layer can be expressed as:

[0191] y = W (out) h + b (out)

[0192] Among them, y is the output, W (out) is the weight matrix, h is the output of the feature fusion layer, and b (out) is the bias term.

[0193] The Softmax function can be expressed as:

[0194]

[0195] In the formula, n is the number of output nodes 5.

[0196] 2. Divide the dataset

[0197] Take 80% of the output data of the distributed cloud model as the training set and 20% as the validation set.

[0198] 3. Model training

[0199] The model is trained using a training dataset. During the training process, the cross-entropy loss function is used to calculate the loss of the model, and the Adam optimizer is used for gradient descent to update the parameters of the model. During the training process, the validation set is also used for validation. When the performance on the validation set no longer improves or begins to decline, the training is stopped to prevent overfitting and avoid over-training the network on the training set.

[0200] The cross-entropy loss function can be expressed as:

[0201]

[0202] In the formula, y i is the true value, is the predicted value, and n is the number of samples.

[0203] The Adam optimizer can be expressed as:

[0204]

[0205] In the formula, θ t is the parameter at the t-th iteration, α is the learning rate, m t is the exponentially weighted average of the gradients at the t-th iteration, v t is the exponentially weighted average of the squares of the gradients at the t-th iteration, and ∈ is a constant to avoid division by zero.

[0206] 4. Real-time performance evaluation

[0207] In the actual operation of the intelligent IoT networking, the parameter values of the secondary indicators are collected and input into the trained lightweight CNN-Transformer hybrid neural network model, and then the real-time performance evaluation of the networking can be carried out.

[0208] Finally, based on the evaluation conclusion of the neural network model, the evaluation conclusion is analyzed in depth to find out the communication performance bottleneck and optimization direction, generate a performance evaluation report, and put forward optimization suggestions

[0209] Figure 3 is the implementation flowchart of the communication performance evaluation method for the intelligent IoT networking provided by another embodiment of the present invention. Figure 3 The method provided is a detailed description of the process from data preparation to model training and model use. As Figure 3 shown, the method may include:

[0210] First, collect the communication performance index data of the intelligent IoT networking to be evaluated. Subjective weighting and objective weighting are respectively carried out on the communication performance index data of the intelligent IoT networking to be evaluated; among them, the analytic hierarchy process can be used for subjective weighting, and the CRITIC method can be used for objective weighting.

[0211] Subjective weighting is based on processes such as constructing a judgment matrix, detecting consistency, and determining index weights to obtain the subjective weights of each communication performance index data.

[0212] Objective weighting is based on processes such as calculating the comparison intensity, conflict calculation, and determining index weights of the index data to obtain the objective weights of each communication performance index data.

[0213] The subjective and objective weights of each communication performance index data obtained are used to calculate the comprehensive weight by the PSO combination method to obtain the comprehensive weight of each communication performance index data.

[0214] Collect historical communication performance index data for distributed cloud model training; the training process of the distributed cloud model can be expressed as:

[0215] First, determine the division results of the evaluation levels and the standard cloud characteristic values in advance. Calculate the characteristic values of each secondary index (i.e., the first characteristic values of each historical communication performance index data) in a distributed manner. According to the first characteristic values of each historical communication performance index data and the comprehensive weight, calculate the characteristic values of the primary index, i.e., the second characteristic values of each cloud node. According to the second characteristic values of each cloud node, determine the characteristic values of the overall evaluation cloud model, i.e., the comprehensive characteristic values. Based on the comprehensive characteristic values, calculate the membership degree, and take the comprehensive characteristic value of the cloud node with the largest membership degree as the evaluation characteristic value of this type of historical communication performance index data to obtain the evaluation result.

[0216] The communication performance index data corresponding to each cloud node and the evaluation result are used as the training data of the neural network model. Among them, the neural network model can be a lightweight CNN-Transformer hybrid neural network model, and its construction process is as follows:

[0217] First, perform structure setting, and then divide the output data of the cloud model into datasets for model training. During the training process, perform real-time performance evaluation on the obtained neural network model to increase the reliability of the neural network model.

[0218] Finally, input the collected communication performance index data as secondary indexes into the trained neural network model for performance evaluation, and optimize the intelligent IoT networking and the model in a timely manner according to the evaluation conclusion.

[0219] In summary, the embodiments of the present invention consider that the distributed cloud model can not only greatly improve the evaluation rate, but also analyze and process different cloud nodes. The embodiments of the present invention construct a distributed cloud model according to the characteristics and requirements of the intelligent IoT networking to be evaluated, enabling it to accurately locate the communication performance of nodes, enhancing the accuracy of evaluation, and making the accuracy of the finally obtained output data higher. Correspondingly, the accuracy of the neural network model obtained by using the output data of the pre-obtained distributed cloud model as the training set is also correspondingly improved. In addition, by combining the distributed cloud model with the neural network model, problems such as memory limitations and calculation delays caused by large amounts of data in the traditional cloud model can be reduced, thereby improving the accuracy and efficiency of the communication performance evaluation of the intelligent IoT networking as a whole and solving the problems of low accuracy and low efficiency in the evaluation of the communication performance of the intelligent IoT networking by the traditional cloud model.

[0220] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0221] The following are the device embodiments of the present invention. For the details not described in detail therein, reference may be made to the corresponding method embodiments above.

[0222] Figure 4 The structure diagram of the communication performance evaluation device for the intelligent IoT networking provided by the embodiments of the present invention is shown. For the sake of convenience of description, only the parts related to the embodiments of the present invention are shown and are described in detail as follows:

[0223] As Figure 4 shown, the communication performance evaluation device 4 for the intelligent IoT networking includes:

[0224] An acquisition module 41, configured to acquire the communication performance index data of the intelligent IoT networking to be evaluated;

[0225] An evaluation module 42, configured to input the communication performance index data into a pre-trained neural network model to obtain the communication performance evaluation result of the intelligent IoT networking to be evaluated;

[0226] Among them, the pre-trained evaluation model is trained according to the output data of the pre-trained distributed cloud model; the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the intelligent IoT networking to be evaluated.

[0227] In a possible implementation manner, the communication performance evaluation device 4 for the intelligent IoT networking further includes a training module 43; the training module 43 is specifically configured to:

[0228] Input the historical communication performance index data and its corresponding comprehensive weight into the pre-trained distributed cloud model;

[0229] In a pre-trained distributed cloud model, according to the historical communication performance metric data and their corresponding comprehensive weights, calculate the comprehensive eigenvalue of each cloud node; according to the comprehensive eigenvalue of each cloud node, obtain the evaluation result of each historical communication performance metric data.

[0230] Take each historical communication performance metric data and its corresponding evaluation result as the output data of the pre-trained distributed cloud model.

[0231] In a possible implementation, the training module 43 is specifically used for:

[0232] In a pre-trained distributed cloud model, use a preset evaluation method to evaluate each historical communication performance metric data to obtain the first eigenvalue of each cloud node; where each historical communication performance metric data corresponds to a cloud node.

[0233] Calculate the second eigenvalue according to the first eigenvalue of each cloud node and the comprehensive weight corresponding to each historical communication performance metric data.

[0234] Calculate the comprehensive eigenvalue of each cloud node according to the second eigenvalue.

[0235] In a possible implementation, the training module 43 is specifically used for:

[0236] For any category of historical communication performance metric data, calculate the cloud model membership degree of this category of historical communication performance metric data according to the comprehensive eigenvalue of each cloud node corresponding to this category of historical communication performance metric data.

[0237] Take the comprehensive eigenvalue of the cloud node whose membership degree meets the preset condition as the evaluation eigenvalue of this category of historical communication performance metric data.

[0238] Obtain the evaluation result of this category of historical communication performance metric data according to this evaluation eigenvalue.

[0239] Obtain the evaluation results of each historical communication performance metric data according to the evaluation results of each category of historical communication performance metric data.

[0240] In a possible implementation, the training module 43 is specifically used for:

[0241] Take each historical communication performance metric data as the input layer of the neural network model, take the evaluation result corresponding to each historical communication performance metric data as the output layer of the neural network model, and perform model training to obtain a pre-trained evaluation model.

[0242] In a possible implementation, the training module 43 is specifically used for:

[0243] Use the subjective weighting method to calculate the subjective weights corresponding to the data of each communication performance index;

[0244] Use the objective weighting method to calculate the objective weights corresponding to the data of each communication performance index;

[0245] Combine the subjective weights and objective weights corresponding to the data of each communication performance index to obtain the comprehensive weights corresponding to the data of each communication performance index.

[0246] In a possible implementation manner, the training module 43 is specifically configured to:

[0247] Adopt the analytic hierarchy process to evaluate the data of each communication performance index to obtain a judgment matrix;

[0248] According to the judgment matrix, obtain the subjective weights corresponding to the data of each communication performance index.

[0249] In a possible implementation manner, the training module 43 is specifically configured to:

[0250] Calculate the comparison intensity corresponding to the data of each communication performance index, and the conflict quantification index between the data of each communication performance index and the data of other communication performance indexes;

[0251] According to the comparison intensity and conflict quantification index corresponding to the data of each communication performance index, obtain the objective weights corresponding to the data of each communication performance index.

[0252] In a possible implementation manner, the training module 43 is specifically configured to:

[0253] For any type of communication performance index data, perform the following steps:

[0254] Adopt the particle swarm optimization algorithm to combine the subjective weights and objective weights corresponding to the data of this type of communication performance index to obtain multiple weight combinations;

[0255] Evaluate each weight combination to obtain the optimal weight combination;

[0256] Take the optimal weight combination as the comprehensive weight corresponding to the data of this type of communication performance index.

[0257] An embodiment of the present invention further provides a terminal device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the methods in the above method embodiments are implemented. Exemplarily, the terminal device may be a desktop computer, a notebook, a palm computer, a cloud server, etc., which is not limited herein.

[0258] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not elaborated or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. Without special instructions and logical conflicts, the terms and / or descriptions between different embodiments are consistent and can be mutually referred to. The technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0259] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A communication performance evaluation method for intelligent Internet of Things networking, characterized in that: include: Obtain the communication performance indicator data of the smart IoT network to be evaluated; Inputting the communication performance indicator data into a pre-trained neural network model to obtain a communication performance evaluation result of the smart IoT network to be evaluated; Among them, the pre-trained evaluation model is trained according to the output data of the pre-trained distributed cloud model; the pre-trained distributed cloud model is constructed according to the characteristics and requirements of the intelligent Internet of Things network to be evaluated.

2. The communication performance evaluation method of the intelligent Internet of Things network according to claim 1 is characterized in that: The output data of the pre-trained distributed cloud model is obtained by the following steps: Input historical communication performance indicator data and their corresponding comprehensive weights into the pre-trained distributed cloud model; In the pre-trained distributed cloud model, the comprehensive feature value of each cloud node is calculated according to the historical communication performance indicator data and its corresponding comprehensive weight; According to the comprehensive characteristic values ​​of each cloud node, the evaluation results of each historical communication performance indicator data are obtained; Each historical communication performance indicator data and its corresponding evaluation results are used as the output data of the pre-trained distributed cloud model.

3. The communication performance evaluation method of the intelligent Internet of Things network according to claim 2 is characterized in that: In the pre-trained distributed cloud model, the comprehensive characteristic value of each cloud node is calculated according to the historical communication performance indicator data and its corresponding comprehensive weight, including: In the pre-trained distributed cloud model, each historical communication performance indicator data is evaluated by a preset evaluation method to obtain a first characteristic value of each cloud node; wherein each historical communication performance indicator data corresponds to a cloud node; Calculate the second eigenvalue according to the first eigenvalue of each cloud node and the comprehensive weight corresponding to each historical communication performance indicator data; Based on the second eigenvalue, the comprehensive eigenvalue of each cloud node is calculated.

4. The communication performance evaluation method of the intelligent Internet of Things network according to claim 2 is characterized in that: The evaluation results of each historical communication performance indicator data are obtained according to the comprehensive characteristic values ​​of each cloud node, including: For any type of historical communication performance indicator data, the cloud model membership of the historical communication performance indicator data is calculated according to the comprehensive characteristic values ​​of each cloud node corresponding to the historical communication performance indicator data; The comprehensive characteristic value of the cloud node whose membership degree meets the preset conditions is used as the evaluation characteristic value of the historical communication performance index data of this type; According to the evaluation characteristic value, an evaluation result of the historical communication performance indicator data of this type is obtained; According to the evaluation results of each type of historical communication performance indicator data, the evaluation results of each historical communication performance indicator data are obtained.

5. The communication performance evaluation method of the intelligent Internet of Things network according to claim 2 is characterized in that: The pre-trained neural network model is trained in the following way: Each historical communication performance indicator data is used as the input layer of the neural network model, and the evaluation result corresponding to each historical communication performance indicator data is used as the output layer of the neural network model, and model training is performed to obtain the pre-trained evaluation model.

6. The communication performance evaluation method of the intelligent Internet of Things network according to claim 2 is characterized in that: The comprehensive weights corresponding to the communication performance indicator data are determined in the following manner: The subjective weighting method is used to calculate the subjective weight corresponding to each communication performance indicator data; The objective weighting method is used to calculate the objective weight corresponding to each communication performance indicator data; The subjective weight and the objective weight corresponding to each communication performance indicator data are combined to obtain the comprehensive weight corresponding to each communication performance indicator data.

7. The communication performance evaluation method of the intelligent Internet of Things network according to claim 6 is characterized in that: The subjective weighting method is used to calculate the subjective weight corresponding to each communication performance indicator data, including: The analytic hierarchy process is used to evaluate the data of each communication performance indicator and obtain the discriminant matrix; According to the discriminant matrix, the subjective weight corresponding to each communication performance indicator data is obtained.

8. The communication performance evaluation method of the intelligent Internet of Things network according to claim 6 is characterized in that: The objective weighting method is used to calculate the objective weight corresponding to each communication performance indicator data, including: Calculate the contrast intensity corresponding to each communication performance indicator data, and the conflict quantification index between each communication performance indicator data and other communication performance indicator data; According to the comparison strength and conflict quantification index corresponding to each communication performance indicator data, the objective weight corresponding to each communication performance indicator data is obtained.

9. The communication performance evaluation method of the intelligent Internet of Things network according to claim 6 is characterized in that: The subjective weight and the objective weight corresponding to each communication performance indicator data are combined to obtain the comprehensive weight corresponding to each communication performance indicator data, including: For any type of communication performance indicator data, perform the following steps: The particle swarm algorithm is used to combine the subjective weights and objective weights corresponding to the communication performance indicator data to obtain multiple weight combinations; Evaluate each weight combination to obtain the optimal weight combination; The optimal weight combination is used as the comprehensive weight corresponding to this type of communication performance indicator data.

10. A terminal device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 9 when executing the computer program.

Images