A network slice performance prediction method and device based on digital twinning

Abstract

The application provides a network slice performance prediction method and device based on digital twinning, and belongs to the technical field of network slices. The network slice performance prediction method based on digital twinning is applied to a digital twinning network and comprises the following steps: collecting real-time network state parameters and multiple sets of network slice candidate resource configuration parameters; inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training sample network state parameters, sample slice resource configuration parameters and sample slice performance results of a preset sample network slice; and performing uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network slicing, in particular to a network slicing performance prediction method and device based on digital twinning. BACKGROUND

[0002] Network slicing customizes a virtual transmission channel on the same physical network through soft-defined network and network function virtualization technology. Generally, network slicing technology is used to provide differentiated network transmission capacity on demand to meet the diversified transmission needs of vertical industries. In order to dynamically meet the service level agreement (SLA) requirements, it is necessary to effectively predict the performance of network slicing to avoid performance degradation of network slicing. Network slicing performance prediction methods mainly include network slicing performance prediction methods based on network models, network slicing performance prediction methods based on time series and network slicing performance prediction methods based on digital twinning.

[0003] Among them, the network slicing performance prediction method based on digital twinning can monitor the network state changes in real time through frequent interaction between the physical network, the digital twinning model, the database and the network control system, and has strong adaptive ability to different environmental traffic models and transmission paths. However, the existing network slicing performance prediction method based on digital twinning is difficult to analyze and predict uncertainty, and cannot provide high credible network slicing performance prediction results. SUMMARY

[0004] The present application provides a network slicing performance prediction method and device based on digital twinning, which solves the defect that it is difficult to analyze and predict uncertainty in the prior art, and improves the credibility of network slicing performance prediction results.

[0005] The present application provides a network slicing performance prediction method based on digital twinning, applied to a digital twinning network, comprising: collecting real-time network state parameters and a plurality of sets of network slicing candidate resource configuration parameters; inputting the real-time network state parameters and the plurality of sets of network slicing candidate resource configuration parameters into a network slicing performance prediction model to obtain a network slicing preliminary performance prediction result output by the network slicing performance prediction model; wherein the network slicing performance prediction model is obtained by training a preset sample network slicing through sample network state parameters, sample slicing resource configuration parameters and sample slicing performance results; and performing uncertainty analysis on the network slicing preliminary performance prediction result to obtain a network slicing final performance prediction result.

[0006] The method for predicting network slice performance based on digital twinning provided by the application comprises the following steps: performing uncertainty analysis on a preliminary performance prediction result of the network slice to obtain a final performance prediction result of the network slice, and calculating a multi-dimensional uncertainty parameter of the preliminary performance prediction result of the network slice; wherein the multi-dimensional uncertainty parameter comprises at least two of mean value, variance, confidence interval and confidence probability; and performing weighted calculation and processing on the multi-dimensional uncertainty parameter to obtain a one-dimensional final performance prediction result of the network slice.

[0007] The method for predicting network slice performance based on digital twinning provided by the application further comprises a physical network, wherein the physical network is electrically connected with the digital twinning network.

[0008] Before the real-time network state parameter and the multiple sets of network slice candidate resource configuration parameters are input into the network slice performance prediction model to obtain a preliminary performance prediction result of the network slice output by the network slice performance prediction model, the method further comprises the following steps: collecting real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters from the physical network; processing the real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters to obtain slice configuration experience parameters; the slice configuration experience parameters comprise sample network state parameters, sample slice resource configuration parameters and sample slice performance results in the same slice time window; and updating a slice configuration experience dataset with a preset number of sets of slice configuration experience parameters based on the slice configuration experience parameters.

[0009] The method for predicting network slice performance based on digital twinning provided by the application further comprises a physical network, wherein the physical network is electrically connected with the digital twinning network.

[0010] According to the present invention, a network slice performance prediction method based on digital twins includes: inputting each set of empirical parameters of the slice configuration experience dataset, the sample network state parameters, and the sample slice configuration parameters into an input layer, and processing them sequentially through a preset number of hidden layers and an output layer to output sample slice prediction performance results; performing supervised learning training on the network slice performance prediction model based on the sample network state parameters, sample slice resource configuration parameters, and sample slice performance results of the preset sample network slices; comparing the sample slice prediction performance results with the corresponding sample slice performance results, and adjusting the parameters of the input layer, the preset number of hidden layers, and the output layer according to the comparison results; and comparing the sample slice prediction performance results with the corresponding sample slice performance results, and adjusting the parameters of the input layer, the preset number of hidden layers, and the output layer according to the comparison results using a backpropagation algorithm.

[0011] According to the present invention, a network slice performance prediction method based on digital twins includes the following steps: after inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain the preliminary network slice performance prediction results output by the network slice performance prediction model, the method further includes: collecting real-time network slice performance parameters from the physical network, statistically analyzing the real-time network slice performance parameters to obtain the corresponding actual network slice performance results; obtaining a slice performance prediction error based on the actual network slice performance results and the predicted network slice performance results; if the slice performance prediction error is greater than a first threshold and less than a second threshold, obtaining the latest first set of slice configuration experience parameters, and fine-tuning the network slice performance prediction model based on the latest first set of slice configuration experience parameters.

[0012] According to the present invention, a network slice performance prediction method based on digital twins is provided. After obtaining the slice performance prediction error based on the actual performance result of the network slice and the predicted performance result of the network slice, the method further includes: if the slice performance prediction error is greater than a second threshold, obtaining the latest slice configuration experience dataset, and retraining the network slice performance prediction model based on the latest slice configuration experience dataset.

[0013] According to the present invention, a network slice performance prediction method based on digital twins, after collecting real-time network slice performance parameters from the physical network and statistically analyzing the real-time network slice performance parameters to obtain the corresponding actual network slice performance results, the method further includes: collecting slice performance requirement parameters from the physical network, obtaining network slice requirement performance results based on the slice performance requirement parameters; comparing the network slice requirement performance results with the network slice actual performance results to determine whether the network slice meets the service requirements.

[0014] The network slice performance prediction method based on digital twinning provided by the application comprises the following steps: inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model, and obtaining a network slice preliminary performance prediction result output by the network slice performance prediction model.

[0015] The application further provides a network slice performance prediction device based on digital twinning, which comprises: a network slice data acquisition module, which is used for acquiring real-time network state parameters and multiple sets of network slice candidate resource configuration parameters; a network slice performance prediction module, which is used for inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model, and obtaining a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training sample network state parameters, sample slice resource configuration parameters and sample slice performance results of a preset sample network slice; and an uncertainty analysis module, which is used for performing uncertainty analysis on the network slice preliminary performance prediction result, and obtaining a network slice final performance prediction result.

[0016] The network slice performance prediction method and device based on digital twinning provided by the application can consider the influence of statistical characteristics of different network slice preliminary performance prediction results by performing uncertainty analysis on the network slice preliminary performance prediction result, obtain a network slice final performance prediction result, and improve the credibility of the network slice performance prediction result. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0018] Figure 1 is a flowchart of the network slice performance prediction method based on digital twinning provided by the embodiments of the application.

[0019] Figure 2 is one of the framework diagrams of the network slice performance prediction method based on digital twinning provided by the embodiments of the application.

[0020] Figure 3Fig. 2 is a schematic diagram of a framework of a network slice performance prediction method based on digital twinning according to an embodiment of the present application.

[0021] Figure 4 Fig. 3 is a schematic diagram of a network slice performance prediction device module based on digital twinning according to an embodiment of the present application.

[0022] Figure 5 Fig. 4 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0024] The present application will be described below with reference to the drawings. Figures 1-5 The network slice performance prediction method and device based on digital twinning of the present application will be described.

[0025] A network slice can customize a virtual transmission channel on the same physical network to provide differentiated network transmission capabilities on demand. In order to achieve more efficient resource allocation and network management, the performance of the network slice needs to be predicted when the network slice is applied. The resource demand and performance of the network slice can be predicted by a network model-based prediction method and a time series-based prediction method.

[0026] The network model-based prediction method generally predicts the performance of the network slice by using mathematical models such as queuing theory and network calculus under certain restrictive assumptions, such as assuming that the traffic flow is a simple traffic flow model with simple statistical characteristics, and the network transmission path is a static transmission path that is pre-set and unchanged. The network model-based prediction method has high theoretical value, but it is not suitable for the case where the traffic flow model is complex and unknown, and the network transmission path is dynamically reconstructed.

[0027] The network slice performance prediction method based on time series generally predicts the performance of the network slice by analyzing the historical monitored network slice performance data and extracting the time correlation of the network slice performance data at different times. The network slice performance prediction method based on time series is simple to implement, but the prediction accuracy is low, and it is difficult to dynamically perceive the influence of different network environments and slice configurations on the change of the slice performance.

[0028] In the case of complex and unknown business traffic model, dynamic reconstruction of network transmission path, and high requirement for prediction accuracy, the performance of network slice can be predicted by a network slice performance prediction method based on digital twinning. The network slice performance prediction method based on digital twinning can dynamically establish a high-fidelity network prediction model by real-time monitoring of network state changes through frequent interaction between the physical network, the digital twinning network, the database and the network control system, so as to realize closed-loop optimization. The network slice performance prediction method based on digital twinning can adaptively construct a network slice performance prediction model, and pre-evaluate the influence of different network states and slice resource configurations on slice performance in the digital space. However, the current slice performance prediction method based on digital twinning is difficult to analyze and predict uncertainty, and cannot provide high-reliable network slice performance prediction results. In view of this problem, the embodiment of the present application provides a network slice performance prediction method based on digital twinning. Figure 1 is a flowchart of the network slice performance prediction method based on digital twinning provided by the present application, applied to a digital twinning network, as shown in Figure 1 The method comprises the following steps:

[0029] Step 110, collecting real-time network state parameters and multiple sets of network slice candidate resource configuration parameters.

[0030] The network state parameters are a series of parameters representing the network state. The real-time network state parameters are a series of parameters representing the network state directly collected from the physical network in real time, which reflect the running status of the network at the current time. For example, the real-time network state parameters can include uplink signal to interference plus noise ratio (SINR), channel quality indication (CQI), modulation and coding scheme (MCS), user buffer size, uplink traffic, resource block (RB) resource utilization, central processing unit (CPU) resource utilization, memory resource utilization, etc.

[0031] The network slice resource configuration parameters are a series of parameters defining and managing the performance and characteristics of the network slice. For example, the network slice resource configuration parameters can include the number of physical resource blocks, the number of computing processing units, the number of memories, etc. These network slice resource configuration parameters can be adjusted according to business requirements, network conditions and performance targets to achieve the optimal performance of the network slice.

[0032] The network slice candidate resource configuration parameter is a series of possible network slice resource configuration parameters proposed to meet specific business requirements or optimize network performance. The network slice resource configuration parameter can be selected from the network slice candidate resource configuration parameter.

[0033] In step 120, the real-time network state parameter and the multiple sets of network slice candidate resource configuration parameters are input into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model. The network slice performance prediction model is trained by sample network state parameters, sample slice resource configuration parameters, and sample slice performance results of a preset sample network slice.

[0034] The network slice preliminary performance prediction result is a series of parameters obtained by predicting the performance and characteristics of a network slice by a network slice performance prediction model under real-time network state parameters and network slice candidate resource configuration parameters. For example, the network slice preliminary performance prediction result can include an SLA (Service-Level Agreement) delay satisfaction rate, a slice SLA rate satisfaction rate, a slice SLA packet loss rate satisfaction rate, and a slice SLA satisfaction rate.

[0035] In step 130, an uncertainty analysis is performed on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

[0036] The uncertainty analysis is a process of evaluating the reliability and accuracy of the network slice preliminary performance prediction result. For example, the uncertainty analysis can be performed on the network slice preliminary performance prediction result when network environment changes rapidly due to traffic bursts, or the uncertainty analysis can be performed on the network slice preliminary performance prediction result when network environment changes rapidly due to channel quality degradation.

[0037] The network slice performance prediction method based on digital twinning provided by the embodiment can consider the influence of different network slice preliminary performance prediction result statistical characteristics by performing an uncertainty analysis on the network slice preliminary performance prediction result, obtain a network slice final performance prediction result, and improve the reliability of the network slice performance prediction result.

[0038] Based on the above embodiment, the uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result includes: calculating a multi-dimensional uncertainty parameter of the network slice preliminary performance prediction result; the multi-dimensional uncertainty parameter includes at least two of mean, variance, confidence interval, and confidence probability; and performing weighted calculation processing on the multi-dimensional uncertainty parameter to obtain a one-dimensional network slice final performance prediction result.

[0039] Exemplarily, the multi-dimensional uncertainty parameters of the network slice preliminary performance prediction result can be obtained by statistical analysis and calculation on the network slice preliminary performance prediction result. A multi-factor comprehensive evaluation method can be used to set different weights for the mean, variance, upper confidence interval, lower confidence interval and confidence probability, etc. to perform weighted calculation and obtain a one-dimensional network slice final performance prediction result.

[0040] The network slice performance prediction method based on digital twinning provided by the embodiment provides quantitative evaluation of the network slice preliminary performance prediction result by obtaining multi-dimensional uncertainty parameters, helps to identify the performance of the network slice performance prediction model under real-time network state parameters and different network slice candidate resource configuration parameters, and provides a highly reliable final performance prediction result for obtaining network slice resource configuration parameters through comprehensive analysis of the multi-dimensional uncertainty parameters, guides the optimized configuration of the network slice, avoids high-risk network slice resource configuration actions, and reduces the risk of network slice performance degradation.

[0041] Based on the above embodiments, the network slice preliminary performance prediction result can also be analyzed for uncertainty to obtain a network slice final performance prediction result, which can include: using the Bootstrap method in ensemble learning to randomly down-sample multiple sets of slice configuration experience data sets with replacement, and training multiple slice performance prediction models based on deep neural networks in parallel, to evaluate the sample network state parameters, sample slice resource configuration parameters and sample slice performance results collected from the physical network for volatility and uncertainty.

[0042] Based on the above embodiments, the network slice preliminary performance prediction result can also be analyzed for uncertainty to obtain a network slice final performance prediction result, which can include: using the Dropout method in Bayesian theory to test the output of multiple DNN models, and calculating the mean, variance, confidence interval and confidence probability of the network slice preliminary performance prediction result of the trained DNN output on each network slice, to facilitate quantitative evaluation of the reliability of the model prediction result.

[0043] Based on the above embodiments, the network slice performance prediction method based on digital twinning further includes a physical network, which is electrically connected to the digital twin network.

[0044] Specifically, the physical network refers to the actual hardware and devices that provide infrastructure and transmission media. For example, the physical network can include network devices such as routers, switches, and transmission media such as optical fiber cables, twisted pairs, etc. The digital twin network refers to the frequent interaction between the physical network and the digital twin network by establishing a digital mapping model of the physical network, thereby dynamically monitoring the network state changes, and based on this, real-time construction of a high-fidelity network slice performance prediction model is realized to achieve closed-loop optimization.

[0045] Before the real-time network state parameters and the plurality of sets of network slice candidate resource configuration parameters are input into the network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model, the method further comprises: collecting real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters from the physical network; processing the real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters to obtain slice configuration experience parameters; the slice configuration experience parameters comprise sample network state parameters, sample slice resource configuration parameters and sample slice performance results in the same slice time window; and updating a slice configuration experience dataset with a preset number of sets of slice configuration experience parameters based on the slice configuration experience parameters.

[0046] Specifically, the real-time network slice performance parameters refer to a series of parameters directly collected from the physical network in real time for describing and evaluating the performance of the network slice. For example, the real-time network slice performance parameters can include rate, latency, bit error rate, packet loss rate, etc. The sample slice performance results refer to a series of parameters obtained after network slice performance statistics are performed on the real-time network slice performance parameters. The sample slice performance results can include SLA latency satisfaction rate, slice SLA rate satisfaction rate, slice SLA packet loss rate satisfaction rate, slice SLA satisfaction rate, etc.

[0047] Exemplarily, the real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters can be processed through slice time window alignment, mean calculation, data normalization and standardization, etc. to obtain corresponding slice configuration experience parameters. By making the sample network state parameters, sample slice resource configuration parameters and sample slice performance results in the slice configuration experience parameters in the same time window, errors caused by different synchronization of data collection times can be effectively avoided, and the accuracy of the network slice preliminary performance prediction result output by the network slice performance prediction model can be improved.

[0048] The slice time window is a set of network state parameters, network slice configuration parameters and network slice performance parameters of the network slice in a specific period of time. The duration of each slice window can be set according to specific business requirements, for example, 1 second.

[0049] The specific number of sets of slice configuration experience parameters in the slice configuration experience dataset needs to be able to reflect the current network state and performance, so that the network slice performance prediction model can better adapt to the rapid changes of the network environment during the training process, thereby improving the accuracy and reliability of the network slice performance prediction model. Exemplarily, the specific number of sets of slice configuration experience parameters in the slice configuration experience dataset can be 3000 sets.

[0050] The network slice performance prediction method based on digital twinning provided by the embodiment can update the slice configuration experience dataset by using the slice configuration experience parameters obtained based on the real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters, so that the configuration experience dataset can provide the latest network state and performance. In this way, the network slice performance prediction model trained based on the slice configuration experience dataset can better adapt to the rapid changes of the network environment, thereby improving the accuracy and reliability of the network slice performance prediction model.

[0051] The network slice performance prediction model can use a machine learning algorithm, such as a deep neural network (DNN).

[0052] Based on the above embodiments, the network slice performance prediction model includes an input layer, a preset number of hidden layers and an output layer.

[0053] Specifically, the input layer is the part of the network slice performance prediction model that receives raw data, the hidden layer is the part of the network slice performance prediction model that processes data and extracts features, and the output layer is the part of the network slice performance prediction model that generates network slice prediction performance results.

[0054] The specific number of hidden layers can be adjusted according to actual needs and the complexity of the network slice performance prediction model. For example, the hidden layer can be set to two to adapt to simpler prediction scenarios, and the hidden layer can be set to four to adapt to more complex prediction requirements and process higher-dimensional and more diversified sample network state parameters and sample slice configuration parameters.

[0055] The training method of the network slice performance prediction model includes: obtaining the slice configuration experience dataset; inputting each set of slice configuration experience parameters, sample network state parameters and sample slice configuration parameters of the slice configuration experience dataset into the input layer, and sequentially processing them through the preset number of hidden layers and the output layer to output sample slice prediction performance results; comparing the sample slice prediction performance results with the corresponding sample slice performance results, and adjusting the parameters of the input layer, the preset number of hidden layers and the output layer according to the comparison results until the difference between the sample slice prediction performance results and the corresponding sample slice performance results is less than a first threshold.

[0056] The difference between the sample slice prediction performance result and the corresponding sample slice performance result can be a mean relative error (MRE), which represents the proportion of the error between the sample slice prediction performance result and the corresponding sample slice performance result relative to a true value. The true value here can be the sample slice performance result. The first threshold value can be determined based on the accuracy requirement of the network slice performance prediction model and factors such as the complexity of the network slice performance prediction model. An exemplary first threshold value can be set to 1%.

[0057] The network slice performance prediction method based on digital twinning provided in this embodiment can predict the performance under the current state, quantify the prediction deviation of the model under the current state, and control the prediction deviation of the network slice performance prediction model by making the difference between the sample slice prediction performance result and the corresponding sample slice performance result less than the threshold value, thereby improving the accuracy of the network slice prediction performance result output by the network slice performance prediction model.

[0058] Based on the above embodiments, each set of slice configuration experience parameters, sample network state parameters and sample slice configuration parameters of the slice configuration experience data set is input to the input layer and sequentially processed via the preset number of hidden layers and the output layer, and a sample slice prediction performance result is output, including: based on the sample network state parameters, sample slice resource configuration parameters and sample slice performance result of the preset sample network slice, the network slice performance prediction model is supervised learning training; the sample slice prediction performance result is compared with the corresponding sample slice performance result, and the parameters of the input layer, the preset number of hidden layers and the output layer are adjusted according to the comparison result, including: comparing the sample slice prediction performance result with the corresponding sample slice performance result, and adjusting the parameters of the input layer, the preset number of hidden layers and the output layer according to the comparison result through the back propagation algorithm.

[0059] Specifically, supervised learning is a machine learning method. In supervised learning, the training data is composed of input data and expected output. In this embodiment, the input data is the sample network state parameter and the sample slice configuration parameter, and the expected output is the sample slice performance result. In supervised learning, the model learns the pattern in the training data.

[0060] In this embodiment, the network slice performance prediction model can learn the internal relationship between different sample network state parameters, sample slice configuration parameters and sample slice performance results, adjust the parameters of the input layer, the preset number of hidden layers and the output layer, and gradually improve the prediction accuracy of the network slice performance prediction model, so as to be able to predict the network slice performance corresponding to the real-time network state parameter and the network slice candidate resource configuration parameter.

[0061] Further, the network weights and bias parameters of the network slice performance prediction model trained through supervised learning are initialized randomly.

[0062] Adjusting the parameters of the input layer, the preset number of hidden layers and the output layer through the back propagation algorithm can gradually reduce the prediction error of the network slice preliminary performance prediction result, help the model better capture complex nonlinear relationships and improve the prediction ability of the network slice performance prediction model.

[0063] Based on the above embodiments, after inputting the real-time network state parameters and the plurality of sets of network slice candidate resource configuration parameters into the network slice performance prediction model and obtaining the network slice preliminary performance prediction result output by the network slice performance prediction model, the method further comprises: collecting real-time network slice performance parameters from the physical network, and statistically obtaining corresponding network slice actual performance results based on the real-time network slice performance parameters; obtaining a slice performance prediction error based on the network slice actual performance results and the network slice predicted performance results; if the slice performance prediction error is greater than a first threshold and less than a second threshold, obtaining the latest first number of sets of slice configuration experience parameters, and fine-tuning the network slice performance prediction model based on the latest first number of sets of slice configuration experience parameters.

[0064] For example, the ratio of the number of data packets meeting the delay requirement in the data packets generated by the service in the current slice window to the total number of all data packets generated in the current slice window can be calculated to obtain the slice SLA delay satisfaction rate. The ratio of the number of data packets meeting the user rate requirement in the data packets generated by the service in the current slice window to the total number of all data packets generated in the current slice window can be calculated to obtain the slice SLA rate satisfaction rate. Whether the packet loss rate in the current slice window is not higher than the packet loss rate threshold can be calculated, and if yes, the value is 1, otherwise, the value is 0, to obtain the slice SLA packet loss rate satisfaction rate. The ratio of the number of data packets meeting the delay and user rate requirements in the data packets generated by the service in the current slice window to the total number of all data packets generated in the current slice window can be calculated, and multiplied by the slice SLA packet loss rate satisfaction rate in the slice window to obtain the slice SLA satisfaction rate. The value ranges of the slice SLA delay satisfaction rate, the slice SLA rate satisfaction rate, the slice SLA packet loss rate satisfaction rate and the slice SLA satisfaction rate are all [0, 1].

[0065] The second threshold is greater than the first threshold. The second threshold is a threshold capable of effectively balancing the accuracy of the network slice preliminary performance prediction result output by the network slice performance prediction model and the update frequency of the network slice performance prediction model, and can be determined according to the prediction accuracy requirement of the network slice preliminary performance prediction result output by the network slice performance prediction model, the complexity of the network slice performance prediction model, and the like. For example, the size of the slice performance prediction error can be measured by the absolute error percentage, so that the second threshold can be 5%.

[0066] The number of the latest slice configuration experience parameters used for fine-tuning training of the network slice performance prediction model can be determined according to the prediction accuracy requirement of the network slice preliminary performance prediction result output by the network slice performance prediction model, the complexity of the network slice performance prediction model, and the like. For example, the first number group can be 30 groups.

[0067] The slice performance prediction error greater than the first threshold indicates that the difference between the slice prediction performance result and the corresponding slice performance result does not meet the accuracy requirement of the slice prediction performance result, and the parameters of the network slice performance prediction model need to be adjusted.

[0068] The slice performance prediction error less than the second threshold indicates that the deviation from the accuracy requirement of the slice prediction performance result is small, and the parameters of the network slice performance prediction model can be adjusted by fine-tuning training, so that the difference between the slice prediction performance result and the corresponding slice performance result meets the accuracy requirement of the slice prediction performance result.

[0069] The latest first number group of slice configuration experience parameters can reflect the changes in the current network environment. Therefore, the network slice performance prediction method based on digital twinning provided in this embodiment can fine-tune the network slice performance prediction model based on the latest first number group of slice configuration experience parameters, so that the network slice performance prediction model can better adapt to the changing network environment and the network slice resource configuration parameters corresponding to the service requirement.

[0070] In addition, fine-tuning the network slice performance prediction model can enhance the self-evolution ability of the digital twinning model. Through frequent small-scale updates, the network slice performance prediction model can be continuously optimized without frequent comprehensive retraining, thereby reducing the large-scale training overhead of the network slice performance prediction model.

[0071] Based on the above embodiments, after obtaining the slice performance prediction error based on the network slice actual performance result and the network slice prediction performance result, the method further includes: if the slice performance prediction error is greater than a second threshold, obtaining the latest slice configuration experience dataset, and retraining the network slice performance prediction model based on the latest slice configuration experience dataset.

[0072] The slice performance prediction error greater than the second threshold value indicates that there is a large deviation from the slice prediction performance result precision requirement, and the parameters of the network slice performance prediction model need to be adjusted by retraining, so that the difference between the slice prediction performance result and the corresponding slice performance result meets the slice prediction performance result precision requirement.

[0073] The latest slice configuration experience data set can more accurately and comprehensively reflect the changes of the current network environment. The network slice performance prediction method based on digital twinning provided in the embodiment can retrain the prediction model based on the latest slice configuration experience data set, which can better adapt to the changing network environment and the network slice resource configuration parameters corresponding to the business demand, especially the scenarios of sudden burst of business traffic, time-varying channel quality, etc. leading to rapid changes in the network environment.

[0074] For example, the network slice performance prediction model can include a training module and an inference module. The training module is used for training or fine-tuning of the network slice performance prediction model, and the inference module is used for receiving real-time network state parameters and network slice candidate resource configuration parameters and outputting network slice preliminary performance prediction results. After the training or fine-tuning of the network slice performance prediction model is completed in the training module, the corresponding network slice performance prediction model file can be obtained and sent to the inference module, so that the inference module obtains the corresponding network slice performance prediction model, and uses the corresponding network slice performance prediction model to infer and predict the received real-time network state parameters and network slice candidate resource configuration parameters. The network slice performance prediction model file can include the network structure, the excitation function, the connection weight and the bias parameter of the prediction model, etc.

[0075] Based on the above embodiments, after collecting real-time network slice performance parameters from the physical network, and obtaining the corresponding network slice actual performance result by counting the real-time network slice performance parameters, the method further comprises: collecting slice performance demand parameters from the physical network, obtaining network slice demand performance results based on the slice performance demand parameters; comparing the network slice demand performance results with the network slice actual performance results, and determining whether the network slice meets the business demand.

[0076] Specifically, the slice performance demand parameters refer to specific indicators and requirements that define and describe the performance of the network slice, for example, the slice performance demand parameters can include rate, delay, bit error rate, packet loss rate, etc.

[0077] The network slice demand performance result can be obtained by statistically processing the slice performance demand parameters. Whether the network slice meets the service demand can be determined by comparing whether the maximum delay, the minimum user rate, the highest packet loss rate, etc. of the actual performance result of the network slice falls within the range of the maximum delay, the minimum user rate, the highest packet loss rate, etc. of the corresponding network slice demand performance result.

[0078] The network slice performance prediction method based on digital twinning provided in the embodiment can discover the performance risk of the network slice in time by judging whether the network slice meets the service demand, so as to guide the slice resource dynamic configuration module to make slice resource configuration decisions and dynamically adjust resource configuration, and ensure the performance of the network slice.

[0079] Based on the above embodiments, the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters are input into the network slice performance prediction model to obtain the network slice preliminary performance prediction result output by the network slice performance prediction model, including: inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into an input layer and sequentially processing via a preset number of hidden layers and an output layer to output the network slice preliminary performance prediction result.

[0080] The network slice performance prediction method based on digital twinning provided in the embodiment processes the real-time network state parameters and the network slice candidate resource configuration parameters via the multi-layer network structure of the preset number of hidden layers and the output layer, which can better capture complex nonlinear relationships and features and increase the accuracy of the output network slice preliminary performance prediction result.

[0081] In order to specifically describe the network slice performance prediction method based on digital twinning provided in the embodiment, a specific example is provided below.

[0082] As shown in Figure 2 The network slice performance prediction method based on digital twinning can include a physical network, a digital twin network, and a slice resource dynamic configuration module, and the physical network, the digital twin network, and the slice resource dynamic configuration module are electrically connected to each other in pairs.

[0083] The digital twin network can include a network data acquisition and processing module, a network slice performance prediction module, a network slice performance prediction model evolution module, and a network slice performance prediction result uncertainty analysis module.

[0084] The network data collection and processing module can collect network state parameters, network slice resource configuration parameters, network slice performance parameters and slice performance demand parameters from the physical network, and can process the network state parameters, network slice resource configuration parameters and network slice performance parameters, obtain slice configuration experience parameters, and send the slice configuration experience parameters to the network slice performance prediction model evolution module. The network data collection and processing module can send the collected real-time network state parameters to the slice resource dynamic configuration module.

[0085] The network slice performance prediction model evolution module can obtain a slice configuration experience dataset based on the received slice configuration experience parameters, and update the slice configuration experience dataset based on the latest received slice configuration experience parameters. The network slice performance prediction model evolution module can send the slice configuration experience dataset to the network slice performance prediction module.

[0086] The network slice performance prediction module can include a training module and an inference module. The training module can train a network slice performance prediction model based on the received slice configuration experience dataset, send a file of the network slice performance prediction model to the inference module, and the inference module can obtain a corresponding network slice performance prediction model based on the file of the network slice performance prediction model. In this way, after receiving the real-time network state parameters and the plurality of sets of network slice candidate resource configuration parameters sent by the slice resource dynamic configuration module, the inference module can output a corresponding network slice preliminary performance prediction result, and transmit the network slice preliminary performance prediction result to the network slice performance prediction result uncertainty analysis module.

[0087] The uncertainty analysis module can obtain multi-dimensional uncertainty parameters of the network slice preliminary performance prediction result, further obtain one-dimensional network slice final performance prediction result based on the multi-dimensional uncertainty parameters, and transmit the final performance prediction result to the slice resource dynamic configuration module, so that the slice resource dynamic configuration module selects network slice resource configuration parameters from the plurality of sets of network slice candidate resource configuration parameters, and transmits the network slice resource configuration parameters to the physical network.

[0088] After the physical network receives the network slice resource configuration parameters, the network data collection and processing module can also transmit the slice performance demand parameters collected from the physical network to the performance judgment module in the network slice performance prediction module. Since the performance judgment module can obtain the corresponding network slice prediction performance result from the inference module, the performance judgment module can determine whether the network slice meets the business demand.

[0089] The network data acquisition and processing module can also transmit network slice performance parameters acquired from the physical network to the performance statistics module in the network slice performance prediction module to obtain the actual performance results of the network slices. Since the performance statistics module can also obtain the corresponding network slice prediction performance results from the inference module, it can obtain the slice performance prediction error and transmit the slice performance prediction error to the network slice performance prediction model evolution module.

[0090] The network slice performance prediction model evolution module can send the slice configuration experience dataset and the fine-tuning training instructions or retraining instructions from the network slice performance prediction module to the training module of the network slice performance prediction module based on the received slice performance prediction error, so that the training module can fine-tune or retrain the network slice performance prediction model.

[0091] Network slicing can include different types of service slices, such as enhanced mobile broadband (eMBB) service slices, massive machine-type communications (mMTC) service slices, and ultra-reliable low latency communications (URLLC) service slices.

[0092] Network status parameters may include traffic values ​​and uplink SINR within the sliding slice window. The sliding slice window contains several consecutive slice windows preceding the current slice window, such as 10, and the duration of the corresponding sliding slice window is 10 seconds.

[0093] Network slice candidate resource configuration parameters can be combinations of network resources allocated to different slices within the current slice window, including the number of physical resource blocks, the number of computing processing units, and the number of memory resources. Slice performance can include slice SLA latency satisfaction rate, slice SLA rate satisfaction rate, slice SLA packet loss rate satisfaction rate, and slice SLA satisfaction rate.

[0094] For example, such as Figure 3 As shown, when predicting the performance of mobile enhanced broadband service slices, the current sliding slice window service traffic, the current sliding slice window uplink SINR, and the current sliding slice window slice resource configuration parameters can be obtained and input into the network slice performance prediction model. The predicted values ​​of the slice SLA satisfaction rate, slice SLA delay satisfaction rate, slice SLA rate satisfaction rate, and slice SLA packet loss rate satisfaction rate of the current slice window are then obtained from the network slice performance prediction model.

[0095] The following describes the network slice performance prediction device based on digital twins provided by the present invention. The network slice performance prediction device based on digital twins described below can be referred to in correspondence with the network slice performance prediction method based on digital twins described above.

[0096] Figure 4 is a module schematic diagram of a network slice performance prediction device based on digital twinning provided by the present application, as shown in the figure, the device comprises a network slice data acquisition module 210, a network slice performance prediction module 220 and an uncertainty analysis module 230. Figure 4

[0097] The network slice data acquisition module 210 is used to acquire real-time network state parameters and multiple sets of network slice candidate resource configuration parameters.

[0098] The network slice performance prediction module 220 is used to input the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training a sample network state parameter, a sample slice resource configuration parameter and a sample slice performance result of a preset sample network slice.

[0099] The uncertainty analysis module 230 is used to perform uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

[0100] Based on any of the above embodiments, the uncertainty analysis module 230 is specifically used to calculate multiple-dimensional uncertainty parameters of the network slice preliminary performance prediction result; wherein the multiple-dimensional uncertainty parameters include at least two of mean, variance, confidence interval and confidence probability; and the multiple-dimensional uncertainty parameters are processed by weighted calculation to obtain a one-dimensional network slice final performance prediction result.

[0101] Based on any of the above embodiments, the network slice performance prediction device based on digital twinning further comprises a parameter acquisition unit, which is used to acquire real-time network state parameters, real-time network slice configuration parameters and real-time network slice performance parameters from the physical network; a slice configuration experience parameter acquisition unit, which is used to process the real-time network state parameters, the real-time network slice configuration parameters and the real-time network slice performance parameters to obtain slice configuration experience parameters; the slice configuration experience parameters include sample network state parameters, sample slice resource configuration parameters and sample slice performance results in the same slice time window; and a slice configuration experience dataset acquisition unit, which is used to acquire the slice configuration experience parameters to update a slice configuration experience dataset with a preset number of sets of slice configuration experience parameters.

[0102] ​Based on any of the above embodiments, the network slice performance prediction model comprises an input layer, a preset number of hidden layers, and an output layer; the network slice performance prediction device based on digital twinning further comprises a model training unit configured to obtain the slice configuration experience dataset; each set of slice configuration experience parameters, sample network state parameters, and sample slice configuration parameters of the slice configuration experience dataset are input into the input layer and sequentially processed through the preset number of hidden layers and the output layer, and a sample slice prediction performance result is output; the sample slice prediction performance result is compared with the corresponding sample slice performance result, and the parameters of the input layer, the preset number of hidden layers, and the output layer are adjusted according to the comparison result until the difference between the sample slice prediction performance result and the corresponding sample slice performance result is less than a first threshold.

[0103] Based on any of the above embodiments, the model training unit is specifically configured to perform supervised learning training on the network slice performance prediction model based on the sample network state parameters, sample slice resource configuration parameters, and sample slice performance result of the preset sample network slice; the sample slice prediction performance result is compared with the corresponding sample slice performance result, and the parameters of the input layer, the preset number of hidden layers, and the output layer are adjusted according to the comparison result through a back propagation algorithm.

[0104] Based on any of the above embodiments, the network slice performance prediction device based on digital twinning further comprises a slice performance statistical unit configured to collect real-time network slice performance parameters from the physical network, and statistically obtain a corresponding network slice actual performance result based on the real-time network slice performance parameters; a slice performance prediction error is obtained based on the network slice actual performance result and the network slice prediction performance result; a fine-tuning and retraining setting unit is configured to, if the slice performance prediction error is greater than a first threshold and less than a second threshold, obtain a latest first number of sets of slice configuration experience parameters, and perform fine-tuning training on the network slice performance prediction model based on the latest first number of sets of slice configuration experience parameters.

[0105] Based on any of the above embodiments, the fine-tuning and retraining setting unit is further configured to, if the slice performance prediction error is greater than the second threshold, obtain a latest slice configuration experience dataset, and perform retraining on the network slice performance prediction model based on the latest slice configuration experience dataset.

[0106] Based on any of the above embodiments, the network slice performance prediction device based on digital twinning further comprises a slice performance judgment unit configured to collect slice performance demand parameters from the physical network, and obtain a network slice demand performance result based on the slice performance demand parameters; the network slice demand performance result is compared with the network slice actual performance result, and it is judged whether the network slice meets the business demand.

[0107] Based on any of the above embodiments, the network slice performance prediction module 220 is specifically configured to input the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into an input layer, and sequentially process via a preset number of hidden layers and an output layer to output a network slice preliminary performance prediction result.

[0108] Figure 5 An example of a schematic diagram of the physical structure of an electronic device is shown in Figure 5 The electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can invoke the logical instructions in the memory 330 to execute a network slice performance prediction method based on digital twinning, which includes: collecting real-time network state parameters and multiple sets of network slice candidate resource configuration parameters; inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training a preset sample network slice with sample network state parameters, sample slice resource configuration parameters, and sample slice performance results; performing uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

[0109] In addition, the logical instructions in the memory 330 described above can be implemented in the form of a software functional unit and sold or used as an independent product, which can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium, includes several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0110] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the network slice performance prediction method based on digital twinning provided by the above method, which comprises: collecting real-time network state parameters and multiple sets of network slice candidate resource configuration parameters; inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training preset sample network slice sample network state parameters, sample slice resource configuration parameters and sample slice performance results; performing uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

[0111] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, which can be executed by a processor to implement the network slice performance prediction method based on digital twinning provided by the above method, which comprises: collecting real-time network state parameters and multiple sets of network slice candidate resource configuration parameters; inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into a network slice performance prediction model to obtain a network slice preliminary performance prediction result output by the network slice performance prediction model; wherein the network slice performance prediction model is obtained by training preset sample network slice sample network state parameters, sample slice resource configuration parameters and sample slice performance results; performing uncertainty analysis on the network slice preliminary performance prediction result to obtain a network slice final performance prediction result.

[0112] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment scheme. Those skilled in the art can understand and implement it without creative labor.

[0113] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0114] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A network slicing performance prediction method based on digital twins, characterized in that, Applications in digital twin networks include: Collect real-time network status parameters and configuration parameters for multiple sets of candidate network slice resources; The real-time network state parameters and the multiple sets of candidate network slice resource configuration parameters are input into the network slice performance prediction model to obtain the preliminary network slice performance prediction results output by the network slice performance prediction model; wherein, the network slice performance prediction model is trained by preset sample network state parameters, sample slice resource configuration parameters and sample slice performance results of sample network slices; Uncertainty analysis is performed on the preliminary performance prediction results of the network slice to obtain the final performance prediction results of the network slice; Uncertainty analysis is performed on the preliminary performance prediction results of the network slices to obtain the final performance prediction results of the network slices, including: Calculate the multidimensional uncertainty parameters for obtaining the preliminary performance prediction results of the network slice; wherein the multidimensional uncertainty parameters include at least two of the following: mean, variance, confidence interval, and confidence probability; The multidimensional uncertainty parameters are weighted and calculated to obtain the final performance prediction result of the one-dimensional network slice.

2. The network slicing performance prediction method based on digital twins according to claim 1, characterized in that, It also includes a physical network, which is electrically connected to the digital twin network; Before inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into the network slice performance prediction model to obtain the preliminary network slice performance prediction results output by the network slice performance prediction model, the method further includes: Real-time network status parameters, real-time network slice configuration parameters, and real-time network slice performance parameters are collected from the physical network. The real-time network status parameters, real-time network slice configuration parameters, and real-time network slice performance parameters are processed to obtain slice configuration experience parameters; the slice configuration experience parameters include sample network status parameters, sample slice resource configuration parameters, and sample slice performance results within the same slice time window. The slice configuration experience dataset with a preset number of slice configuration experience parameters is updated based on the slice configuration experience parameters.

3. The network slicing performance prediction method based on digital twins according to claim 2, characterized in that, The network slicing performance prediction model includes an input layer, a preset number of hidden layers, and an output layer; The training method for the network slice performance prediction model includes: Obtain the slice configuration experience dataset; Each set of the slice configuration experience parameters, sample network state parameters, and sample slice configuration parameters of the slice configuration experience dataset are input to the input layer, and then processed sequentially through a preset number of hidden layers and output layers to output the sample slice prediction performance results. The predicted performance results of the sample slices are compared with the corresponding performance results of the sample slices, and the parameters of the input layer, a preset number of hidden layers and the output layer are adjusted according to the comparison results until the difference between the predicted performance results of the sample slices and the corresponding performance results of the sample slices is less than a first threshold.

4. The network slicing performance prediction method based on digital twins according to claim 3, characterized in that, Each set of the slice configuration experience parameters, sample network state parameters, and sample slice configuration parameters of the slice configuration experience dataset are input to the input layer and processed sequentially through a preset number of hidden layers and output layers to output the sample slice prediction performance results, including: supervised learning training of the network slice performance prediction model based on the sample network state parameters, sample slice resource configuration parameters, and sample slice performance results of the preset sample network slices; The process of comparing the predicted performance results of the sample slices with the corresponding performance results of the sample slices, and adjusting the parameters of the input layer, a preset number of hidden layers, and the output layer based on the comparison results, includes: comparing the predicted performance results of the sample slices with the corresponding performance results of the sample slices, and adjusting the parameters of the input layer, a preset number of hidden layers, and the output layer based on the comparison results using a backpropagation algorithm.

5. The network slicing performance prediction method based on digital twins according to claim 2, characterized in that, After inputting the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into the network slice performance prediction model, and obtaining the preliminary network slice performance prediction results output by the network slice performance prediction model, the method further includes: Real-time network slice performance parameters are collected from the physical network, and the actual performance results of the network slices are obtained by statistically analyzing the real-time network slice performance parameters. Based on the actual performance results of the network slice and the predicted performance results of the network slice, the slice performance prediction error is obtained; If the slice performance prediction error is greater than a first threshold and less than a second threshold, then the latest first group of slice configuration experience parameters are obtained, and the network slice performance prediction model is fine-tuned and trained based on the latest first group of slice configuration experience parameters.

6. The network slicing performance prediction method based on digital twins according to claim 5, characterized in that, After obtaining the slice performance prediction error based on the actual performance results and the predicted performance results of the network slice, the method further includes: If the slice performance prediction error is greater than the second threshold, then the latest slice configuration experience dataset is obtained, and the network slice performance prediction model is retrained based on the latest slice configuration experience dataset.

7. The network slicing performance prediction method based on digital twins according to claim 5, characterized in that, After collecting real-time network slice performance parameters from the physical network and statistically analyzing these parameters to obtain the corresponding actual network slice performance results, the method further includes: Slice performance requirement parameters are collected from the physical network, and network slice performance requirement results are obtained based on the slice performance requirement parameters; By comparing the required performance results of the network slice with the actual performance results of the network slice, it can be determined whether the network slice meets the business requirements.

8. The network slicing performance prediction method based on digital twins according to claim 3, characterized in that, The real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters are input into the network slice performance prediction model to obtain the preliminary network slice performance prediction results output by the network slice performance prediction model, including: The real-time network state parameters and the configuration parameters of the multiple sets of network slice candidate resources are input to the input layer, and then processed through a preset number of hidden layers and output layers to output the preliminary performance prediction results of the network slice.

9. A network slicing performance prediction device based on digital twins, characterized in that, include: The network slice data acquisition module is used to collect real-time network status parameters and configuration parameters of multiple sets of network slice candidate resources; The network slice performance prediction module is used to input the real-time network state parameters and the multiple sets of network slice candidate resource configuration parameters into the network slice performance prediction model to obtain the preliminary performance prediction results of the network slice output by the network slice performance prediction model; wherein, the network slice performance prediction model is trained by preset sample network state parameters, sample slice resource configuration parameters and sample slice performance results of sample network slices; The uncertainty analysis module is used to perform uncertainty analysis on the preliminary performance prediction results of the network slice and obtain the final performance prediction results of the network slice. The uncertainty analysis module is specifically used to calculate the multidimensional uncertainty parameters for obtaining the preliminary performance prediction results of the network slice; wherein, the multidimensional uncertainty parameters include at least two of the following: mean, variance, confidence interval, and confidence probability; and the multidimensional uncertainty parameters are weighted and calculated to obtain the final performance prediction results of the network slice in one dimension.

Images