Network congestion prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a network congestion prediction method and device, electronic equipment and a storage medium, and relates to the technical field of quantum communication. The method comprises the following steps: acquiring multi-dimensional real-time network state data of a quantum cryptography network; predicting the real-time network state data of each dimension by adopting the hybrid prediction model of each dimension to obtain first network state data and second network state data of each dimension at a prediction moment, the first network state data and the second network state data are prediction results of a first prediction model and a second prediction model in the hybrid prediction model respectively; calculating target network state data of each dimension at the prediction moment according to the first network state data and the second network state data of each dimension at the prediction moment; and determining the congestion probability of the quantum cryptography network at the prediction moment according to the target network state data of the multiple dimensions. According to the invention, the network congestion of the quantum cryptography network can be predicted, and the performance and reliability of the quantum cryptography network are improved.

Description

Technical Field

[0001] The present application relates to the field of quantum communication technology, and in particular to a network congestion prediction method, device, electronic device and storage medium. Background Art

[0002] Quantum Key Distribution (QKD) technology is a secure key distribution technology based on the basic principles of quantum mechanics. It is used to ensure the high security of communications and has great application prospects in the field of information security.

[0003] However, with the promotion of QKD technology in practical applications, congestion problems in QKD networks have gradually emerged. Traditional network congestion problems also exist in QKD networks. Due to bandwidth limitations or insufficient device performance, data transmission may encounter delays or packet loss, which not only reduces the efficiency of key distribution, but also may bring security risks. More seriously, the congestion problem in quantum networks directly affects the ability to generate and distribute keys, thereby threatening the stability of the entire communication service.

[0004] The routing selection of most existing QKD networks is based on static network conditions and is difficult to use in dynamic network environments, thus failing to effectively solve the congestion problem in QKD networks. Summary of the invention

[0005] The purpose of this application is to provide a network congestion prediction method, device, electronic device and storage medium to address the deficiencies in the above-mentioned prior art, so as to predict the network congestion of the quantum cryptography network and improve the performance and reliability of the quantum cryptography network.

[0006] To achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:

[0007] In a first aspect, an embodiment of the present application provides a network congestion prediction method, the method comprising:

[0008] Acquire real-time network status data of multiple dimensions of the quantum cryptography network, wherein the real-time network status data of multiple dimensions at least includes: part or all of network delay, bandwidth utilization rate, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate;

[0009] Use a pre-trained hybrid prediction model for each dimension to predict the real-time network status data of each dimension, and obtain first network status data and second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model;

[0010] Calculating target network status data for each dimension at the prediction time according to the first network status data and the second network status data for each dimension at the prediction time;

[0011] The congestion probability of the quantum cryptography network at the prediction moment is determined according to the target network state data of the multiple dimensions at the prediction moment.

[0012] Optionally, the modeling steps of the first prediction model in the hybrid prediction model of each dimension are as follows:

[0013] Obtaining historical network status data at multiple historical moments in each dimension;

[0014] Performing differential processing on the historical network status data of multiple historical moments in each dimension to obtain a stable network status data sequence in each dimension, and determining the differential order of each dimension;

[0015] Calculating the autocorrelation function and the partial autocorrelation function of each dimension according to the stationary network state data sequence of each dimension;

[0016] Determine the autoregressive order and the sliding average order of each dimension according to the autocorrelation function and the partial autocorrelation function of each dimension;

[0017] A first prediction model for each dimension is established according to the difference order, autoregressive order and sliding average order of each dimension.

[0018] Optionally, the training steps of the second prediction model in the hybrid prediction model of each dimension are as follows:

[0019] Extracting features from the historical network status data at multiple historical moments in each dimension to obtain historical network status features in each dimension;

[0020] Using the first prediction model of each dimension to predict the historical network status characteristics of each dimension, and determining the first prediction data at the historical prediction time;

[0021] Calculating a first prediction data residual according to the first prediction data and the historical network status data at the historical prediction time;

[0022] Using a preset long short-term memory network model to predict the historical network state characteristics of each dimension and the first prediction data residual, and determine the second prediction data at the historical prediction moment;

[0023] Calculating a second prediction data residual according to the second prediction data and the historical network status data at the historical prediction time;

[0024] The preset long short-term memory network model is optimized according to the second prediction data residual to obtain the second prediction model.

[0025] Optionally, calculating the autocorrelation function of each dimension according to the stationary network state data sequence of each dimension includes:

[0026] The autocorrelation function of each dimension is calculated according to the steady network state data sequence of each dimension and the preset time lag parameter.

[0027] Optionally, the extracting features of the historical network status data at multiple historical moments in each dimension to obtain the historical network status features of each dimension includes:

[0028] Extracting periodic features of the historical network status data at multiple historical moments in each dimension to determine the periodic features of the historical network in each dimension;

[0029] Statistical features are extracted from the historical network status data at multiple historical moments in each dimension to determine the historical network statistical features of each dimension.

[0030] Optionally, determining the network congestion probability at the prediction time according to the target network status data of the multiple dimensions at the prediction time includes:

[0031] Calculating the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction moment and the mapping function of each dimension;

[0032] The network congestion probability at the prediction moment is determined according to the congestion impact factors of the multiple dimensions.

[0033] Optionally, calculating the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction moment and the mapping function of each dimension includes:

[0034] Calculating a congestion impact factor of the network delay according to the predicted network delay at the predicted time and a positive correlation mapping function of the network delay;

[0035] Calculating a congestion impact factor of the bandwidth usage rate according to a positive correlation mapping function of the predicted bandwidth usage rate at the predicted time and the bandwidth usage rate;

[0036] Calculating a congestion impact factor of the packet loss rate according to the predicted packet loss rate at the predicted time and a positive correlation mapping function of the packet loss rate;

[0037] Calculating a congestion impact factor of the remaining amount of the key pool according to a negative correlation mapping function between the predicted remaining amount of the key pool at the prediction moment and the remaining amount of the key pool;

[0038] Calculating a congestion impact factor of the key generation rate according to a negative correlation mapping function between the predicted key generation rate at the predicted moment and the key generation rate;

[0039] The congestion impact factor of the key consumption rate is calculated according to the predicted key consumption rate at the predicted moment and a positive correlation mapping function of the key consumption rate.

[0040] In a second aspect, an embodiment of the present application further provides a network congestion prediction device, the device comprising:

[0041] A data acquisition module is used to acquire real-time network status data of multiple dimensions of the quantum cryptography network, wherein the real-time network status data of multiple dimensions at least includes: part or all of network delay, bandwidth utilization rate, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate;

[0042] A model prediction module, used to use a pre-trained hybrid prediction model for each dimension to predict the real-time network status data of each dimension, and obtain first network status data and second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model;

[0043] A data calculation module, configured to calculate target network status data of each dimension at the prediction moment according to the first network status data and the second network status data of each dimension at the prediction moment;

[0044] A probability calculation module is used to determine the congestion probability of the quantum cryptography network at the prediction moment according to the target network state data of the multiple dimensions at the prediction moment.

[0045] Optionally, the first prediction model modeling module is used to obtain historical network status data at multiple historical moments of each dimension; perform differential processing on the historical network status data at multiple historical moments of each dimension to obtain a steady network status data sequence of each dimension, and determine the differential order of each dimension; calculate the autocorrelation function and partial autocorrelation function of each dimension according to the steady network status data sequence of each dimension; determine the autoregressive order and sliding average order of each dimension according to the autocorrelation function and partial autocorrelation function of each dimension; and establish the first prediction model of each dimension according to the differential order, autoregressive order and sliding average order of each dimension.

[0046] Optionally, a second prediction model training module is used to extract features from the historical network status data of multiple historical moments in each dimension to obtain the historical network status features of each dimension; use the first prediction model of each dimension to predict the historical network status features of each dimension to determine the first prediction data at the historical prediction moment; calculate the first prediction data residual based on the first prediction data and the historical network status data at the historical prediction moment; use a preset long short-term memory network model to predict the historical network status features of each dimension and the first prediction data residual to determine the second prediction data at the historical prediction moment; calculate the second prediction data residual based on the second prediction data and the historical network status data at the historical prediction moment; optimize the preset long short-term memory network model based on the second prediction data residual to obtain the second prediction model.

[0047] Optionally, the first prediction model building module is further used to calculate the autocorrelation function of each dimension based on the stable network status data sequence of each dimension and a preset time lag parameter.

[0048] Optionally, the second prediction model training module is also used to extract periodic features of the historical network status data at multiple historical moments in each dimension to determine the periodic features of the historical network in each dimension; and to extract statistical features of the historical network status data at multiple historical moments in each dimension to determine the statistical features of the historical network in each dimension.

[0049] Optionally, the probability calculation module is specifically used to calculate the congestion impact factor of each dimension based on the target network status data of each dimension at the prediction moment and the mapping function of each dimension; and determine the network congestion probability at the prediction moment based on the congestion impact factors of the multiple dimensions.

[0050] Optionally, the probability calculation module is further used to calculate the congestion impact factor of the network delay based on the predicted network delay at the prediction moment and the positive correlation mapping function of the network delay; calculate the congestion impact factor of the bandwidth utilization rate based on the predicted bandwidth utilization rate at the prediction moment and the positive correlation mapping function of the bandwidth utilization rate; calculate the congestion impact factor of the packet loss rate based on the predicted packet loss rate at the prediction moment and the positive correlation mapping function of the packet loss rate; calculate the congestion impact factor of the key pool remaining amount based on the predicted key pool remaining amount at the prediction moment and the negative correlation mapping function of the key pool remaining amount; calculate the congestion impact factor of the key generation rate based on the predicted key generation rate at the prediction moment and the negative correlation mapping function of the key generation rate; calculate the congestion impact factor of the key consumption rate based on the predicted key consumption rate at the prediction moment and the positive correlation mapping function of the key consumption rate.

[0051] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium through the bus, and the processor executes the program instructions to perform the steps of the network congestion prediction method as described in any one of the first aspects.

[0052] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the network congestion prediction method as described in any one of the first aspects are executed.

[0053] The beneficial effects of this application are:

[0054] The network congestion prediction method, device, electronic device and storage medium provided in the present application predict the real-time network status data of the quantum cryptographic network in multiple dimensions based on a pre-trained hybrid prediction model, and obtain the first network status data and the second network status data in multiple dimensions. The first network status data and the second network status data respectively predict the linear change and the nonlinear change of the real-time network status data. The target network status data calculated according to the first network status data and the second network status data can more accurately predict the network status change of the quantum cryptographic network in different dimensions, so that the congestion probability of the quantum cryptographic network at the prediction time can be accurately predicted according to the target network status data in multiple dimensions, and the prediction accuracy is guaranteed, so as to dynamically adjust the route according to the congestion probability, improve the ability of the quantum cryptographic network to cope with the load environment, and reduce the key transmission problem caused by congestion; and by identifying and avoiding congestion problems in advance, the delay and failure rate can be reduced to ensure a smooth key distribution process. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 1 ;

[0057] Figure 2 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 2 ;

[0058] Figure 3 An autocorrelation diagram of the network delay dimension provided in an embodiment of the present application;

[0059] Figure 4 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 3 ;

[0060] Figure 5 A schematic diagram of a hybrid prediction model provided in an embodiment of the present application;

[0061] Figure 6 A comparison chart of network delay predictions provided in an embodiment of the present application;

[0062] Figure 7 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 4 ;

[0063] Figure 8 A prediction graph of network congestion probability provided by an embodiment of the present application;

[0064] Fig. 9 A schematic diagram of the structure of a network congestion prediction device provided in an embodiment of the present application;

[0065] Fig.10 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0067] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0068] In addition, the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0069] It should be noted that, in the absence of conflict, the features in the embodiments of the present application may be combined with each other.

[0070] A quantum cryptography network is a secure communication network that uses quantum cryptography technology. It is constructed jointly by a classical communication network and a quantum key distribution network. The classical communication network uses quantum keys to encrypt and decrypt data and transmit encrypted data. The quantum key distribution network consists of QKD terminals and quantum links and is used for key distribution.

[0071] The network nodes of the quantum cryptography network include user nodes (User Node, UN) and relay nodes (RelayNode, RN). The user terminal is mounted on the user node. Two communicating user nodes are connected through the relay node. The relay node is used to extend the key transmission distance to transmit the key between two remote user terminals. The link formed between the relay nodes is the relay link (Relay Link, RL), which realizes the communication function between the two relay nodes. The user node is essentially a relay node, but it is called a user node because it needs to mount the user terminal.

[0072] The network nodes of the quantum cryptography network are composed of classical communication devices and QKD devices. The classical communication devices encrypt and decrypt data and transmit encrypted data through the classical communication network, and the QKD devices are used for key distribution. Generally, QKD devices are used to distribute keys between relay nodes and between relay nodes and user nodes.

[0073] Figure 1Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 1 ,like Figure 1 As shown, the method may include:

[0074] S101. Acquire real-time network status data of multiple dimensions of the quantum cryptography network, where the real-time network status data of multiple dimensions at least include: part or all of: network delay, bandwidth utilization, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate.

[0075] In this embodiment, the real-time network status data of multiple dimensions of the quantum cryptography network are real-time network status data of multiple dimensions of the relay nodes of the quantum cryptography network, including real-time network status data of the classical communication network and real-time network status data of the quantum key distribution network.

[0076] Network Latency is the time delay required for data to be transmitted between the relay links of relay nodes, that is, the time delay for data transmission between relay nodes through the classical communication network in the quantum cryptography network; bandwidth utilization rate is the bandwidth utilization rate of the classical communication network for data transmission between relay nodes; packet loss rate is the loss rate of data packets for data transmission between relay nodes through the classical communication network.

[0077] The remaining amount in the key pool (Key Remaining) is the number of quantum keys that have not been used by the relay node or can be further distributed through the quantum key distribution network; the key generation rate (Key Generation Rate) is the rate at which new keys are generated through negotiation between relay nodes, and the key consumption rate (Key Consumption Rate) is the rate at which relay nodes use or distribute keys.

[0078] It should be noted that the real-time network status data used for network congestion prediction may include part or all of the data in the above-mentioned multiple dimensions, and may also include data in other dimensions besides the data in the above-mentioned multiple dimensions, which is not limited in this embodiment.

[0079] S102. Use a pre-trained hybrid prediction model for each dimension to predict the real-time network status data of each dimension to obtain the first network status data and the second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model.

[0080] In this embodiment, a hybrid prediction model is pre-trained for the network status data of each dimension, and the hybrid prediction model consists of a first prediction model and a second prediction model. The first prediction model and the second prediction model predict the network status respectively.

[0081] Among them, the first prediction model is used to capture the linear change trend of the network status data of each dimension, and the second prediction model is used to process the nonlinear changes in the network status data of each dimension. The real-time network status data of each dimension is input into the hybrid prediction model of each dimension. The first prediction model of each dimension predicts the linear change trend of the real-time network status data of each dimension to obtain the first network status data under the linear change of the real-time network status data. The second prediction model of each dimension predicts the nonlinear change in the real-time network status data of each dimension to obtain the second network status data under the nonlinear change of the real-time network status data.

[0082] S103: Calculate target network status data of each dimension at the prediction time according to the first network status data and the second network status data of each dimension at the prediction time.

[0083] In this embodiment, the linear first network state data and the nonlinear second network state data predicted by the hybrid prediction model are combined to obtain the target network state data of each dimension at the prediction time.

[0084] S104. Determine the congestion probability of the quantum cryptography network at the prediction time based on the target network state data in multiple dimensions at the prediction time.

[0085] In this embodiment, the congestion probability of the quantum cryptography network at the prediction time is calculated based on the predicted network delay, predicted bandwidth usage, predicted packet loss rate, predicted key pool remaining amount, predicted key generation rate and / or predicted key consumption rate at the prediction time.

[0086] Furthermore, based on the congestion probability at the predicted time, it can be determined whether to select a new network route for the quantum cryptography network. If the congestion probability at the predicted time indicates that the quantum cryptography network has a greater probability of being congested at the predicted time, a new network route can be selected for the quantum cryptography network in advance to avoid network congestion in the quantum cryptography network.

[0087] In one possible implementation, Figure 2 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 2 ,like Figure 2 As shown, the modeling steps of the first prediction model in the hybrid prediction model of each dimension are as follows:

[0088] S201. Obtain historical network status data at multiple historical moments in each dimension.

[0089] In this embodiment, the network state detection module in the quantum cryptography network collects historical network state data of multiple historical moments in each dimension related to the network node state, and the historical network state data carries timestamp information. The historical network state data of multiple historical moments in each dimension constitute a single variable target sequence.

[0090] In some embodiments, after collecting historical network status data at multiple historical moments in each dimension, it is necessary to clean abnormal data and noise.

[0091] Specifically, the standard score (Z-Score) method is used to process the historical network status data at each historical moment to determine whether the historical network status data at each historical moment is an outlier. If it is an outlier, it is removed. For missing data, linear interpolation and polynomial interpolation methods can be used to supplement it.

[0092] For example, the formula of the standard score (Z-Score) method is as follows:

[0093]

[0094] Among them, X is the historical network status data at each historical moment, μ X is the mean of the historical state data at multiple historical moments, σ X is the standard deviation of multiple historical moments. When |Z| is greater than a preset value, such as 3, the corresponding historical network status data is determined to be abnormal data.

[0095] S202: Perform differential processing on the historical network status data of multiple historical moments in each dimension to obtain a stable network status data sequence in each dimension, and determine the differential order of each dimension.

[0096] In this embodiment, the auto-regressive integrated moving average model (ARIMA) is used to construct the first prediction model.

[0097] The modeling process of the ARIMA model is the process of determining the difference order d, autoregression order p and moving average order q of the ARIMA model.

[0098] In the process of determining the differential order d, it is determined whether the historical network status data of multiple historical moments in each dimension are stable. If the historical network status data of multiple historical moments in each dimension are stable, the differential order of each dimension is determined to be 0. If the historical network status data of multiple historical moments in each dimension are not stable, the historical network status data of multiple historical moments in each dimension are differentially processed to determine the differential order d that changes the historical network status data of multiple historical moments in each dimension from being not stable to being stable.

[0099] Each time a first-order difference process is performed, a stationary test is performed on the data after the difference. If it is stable, the difference order is determined. If it is not stable, the difference order is increased until the data after the difference is stable and the difference order is obtained.

[0100] For example, the first-order difference processing is to use the historical network status data at the second historical moment to subtract the historical network status data at the first historical moment, as the first-order difference data at the first historical moment. The second-order difference processing is to use the difference between the historical network status data at the third historical moment and the historical network status data at the second historical moment, minus the difference between the historical network status data at the second historical moment and the historical network status data at the first historical moment, as the second-order difference data at the first historical moment.

[0101] S203. Calculate the autocorrelation function and partial autocorrelation function of each dimension according to the stationary network state data sequence of each dimension.

[0102] In this embodiment, the steady network state sequence of each dimension includes steady network state data of multiple historical moments, and the correlation between the steady network state data of different historical moments in the steady network state sequence of each dimension is analyzed to determine the autocorrelation function (ACF) and partial autocorrelation function (PACF) of each dimension.

[0103] For example, for the first historical moment and the fourth historical moment, the correlation between the historical network status data of the first historical moment and the historical network status data of the fourth historical moment analyzed by the autocorrelation function includes the influence of the historical network status data of the second historical moment and the historical network status data of the third historical moment, and the correlation between the historical network status data of the first historical moment and the historical network status data of the fourth historical moment analyzed by the partial autocorrelation function eliminates the influence of the historical network status data of the second historical moment and the historical network status data of the third historical moment.

[0104] S204: Determine the autoregressive order and the sliding average order of each dimension according to the autocorrelation function and the partial autocorrelation function of each dimension.

[0105] In this embodiment, an autocorrelation graph is generated according to the autocorrelation function of each dimension, and the autoregressive order p is determined from the autocorrelation graph. A partial autocorrelation graph is generated according to the partial autocorrelation function of each dimension, and the sliding average order q is determined from the partial autocorrelation graph.

[0106] S205. Establish a first prediction model for each dimension according to the difference order, autoregressive order and sliding average order of each dimension.

[0107] In this embodiment, after determining the difference order d, the autoregressive order p, and the sliding average order q of the ARIMA model, the modeling of the ARIMA model is completed to obtain the first prediction model of each dimension.

[0108] In some embodiments, the ARIMA model is used to predict the network status data at the historical prediction moment, and the autoregressive order p and the sliding average order q are modified according to the difference between the predicted data and the actual detected historical network status to improve the accuracy of the first prediction model.

[0109] In a possible implementation, the process of calculating the autocorrelation function of each dimension in S203 according to the stationary network state data sequence of each dimension may include:

[0110] According to the stationary network state data sequence of each dimension and the preset time lag parameter, the autocorrelation function of each dimension is calculated.

[0111] In this embodiment, a time lag parameter of a steady network status data sequence is determined, and a correlation analysis is performed on the steady network status data at the current moment and the steady network status data at a moment after the time lag parameter at the current moment according to the time lag parameter, to determine the autocorrelation function of each dimension.

[0112] In some embodiments, if the historical network status data of each dimension changes periodically and the preset time lag parameter is a fixed value, the change period is determined to be the preset time lag parameter. For example, when analyzing the historical network status data of each day or each week, the preset time lag parameter is one day or one week.

[0113] If the historical network status data of each dimension changes dynamically, its preset time lag parameter is a dynamic value. By continuously adjusting the time lag parameter, the correlation between the historical network status data is analyzed from the autocorrelation diagram of the autocorrelation function corresponding to different time lag parameters, so as to determine the autoregressive order p.

[0114] For example, the calculation formula of the autocorrelation function is as follows:

[0115]

[0116] Among them, τ is the time lag parameter, and T is the total time length of multiple historical moments.

[0117] For example, Figure 3 The autocorrelation diagram of the network delay dimension provided in the embodiment of the present application is as follows: Figure 3 As shown, after the autocorrelation function is plotted, the autocorrelation coefficient in the autocorrelation graph oscillates and decays at an exponential rate, and the autocorrelation function is tailed, and the autoregressive order is determined to be 0. If the autocorrelation function is truncated, the autoregressive order is determined according to the truncation order.

[0118] In one possible implementation, Figure 4 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 3 ,like Figure 4 As shown, the training steps of the second prediction model in the hybrid prediction model of each dimension are as follows:

[0119] S301 , extracting features from historical network status data at multiple historical moments in each dimension to obtain historical network status features in each dimension.

[0120] In this embodiment, feature extraction is performed on the change pattern, distribution and other features of the historical network status data at multiple historical moments in each dimension to obtain the historical network status features of each dimension.

[0121] In some embodiments, the above S301 extracts features from the historical network status data of multiple historical moments in each dimension to obtain the historical network status features of each dimension, including:

[0122] Perform periodic feature extraction on the historical network status data at multiple historical moments in each dimension to determine the periodic features of the historical network in each dimension; perform statistical feature extraction on the historical network status data at multiple historical moments in each dimension to determine the statistical features of the historical network in each dimension.

[0123] In this embodiment, the network status data of multiple historical moments in each dimension are analyzed in the time dimension to determine the time information corresponding to the recurring states in the real-time network status data of each dimension, thereby determining the historical network periodicity characteristics of each dimension, such as daily traffic peaks, weekly usage habits, etc.

[0124] In some embodiments, the following sine and cosine functions may be used to convert time information of multiple historical moments t into periodic information, determine the period length T, analyze the historical network status data within the period length T, and determine the periodic characteristics of the historical network.

[0125]

[0126] For example, the length of the cycle may be a day, a week, a month, or the like.

[0127] By combining historical network status data at multiple historical moments with historical network periodic characteristics, the model can capture time-related change trends in quantum cryptography networks and enhance the ability to predict periodic congestion risks.

[0128] Statistics are collected on the network status data at multiple historical moments in each dimension to determine historical network statistical features to evaluate the overall distribution of the network status data at multiple historical moments. The historical network statistical features may be, for example, one or more of the mean, variance, maximum value, and minimum value.

[0129] S302: Use the first prediction model of each dimension to predict the historical network status characteristics of each dimension, and determine the first prediction data at the historical prediction moment.

[0130] In this embodiment, the historical network status features of each dimension are input into the first prediction model of each dimension, and the first prediction model of each dimension performs linear trend modeling on the historical network status features of each dimension to output the first prediction data at the historical prediction moment.

[0131] In some embodiments, the historical network periodic characteristics and historical network statistical characteristics of each dimension are input into the first prediction model of each dimension, and the first prediction model of each dimension performs linear trend modeling on the historical network periodic characteristics and historical network statistical characteristics of each dimension to output the first prediction data at the historical prediction moment.

[0132] S303: Calculate the first prediction data residual according to the first prediction data and the historical network status data at the historical prediction time.

[0133] In this embodiment, historical network status data at the historical prediction time is obtained, the difference between the historical network status data at the historical prediction time and the first prediction data is calculated, and the first prediction data residual is extracted.

[0134] For example, the residual calculation formula is:

[0135]

[0136] Among them, X(t) is the historical network status data at the historical prediction time t, is the first prediction data at the historical prediction time t. The first preset data residual reflects the nonlinear components in the historical network status data at multiple historical moments.

[0137] In some embodiments, historical network status features corresponding to historical network status data at multiple historical moments in different historical time periods are respectively input into a first prediction model, and first prediction data at multiple historical prediction moments are output. Based on the difference between the historical network status data at multiple historical prediction moments and the first prediction data at the corresponding historical prediction moments, a first prediction data residual sequence is constructed.

[0138] S304: Use a preset long short-term memory network model to predict the historical network state characteristics of each dimension and the residual of the first prediction data to determine the second prediction data at the historical prediction moment.

[0139] In this embodiment, the Long Short-Term Memory (LSTM) network model is a time recursive neural network, which inputs the historical network state characteristics of each dimension and the first prediction data residual at the historical prediction moment into the LSTM model, learns the nonlinear time dependency therein, predicts the nonlinear trend of the network state data of each dimension, and outputs the second prediction data at the historical prediction moment.

[0140] S305: Calculate the second prediction data residual according to the second prediction data and the historical network status data at the historical prediction time.

[0141] In this embodiment, historical network status data at the historical prediction time is obtained, the difference between the historical network status data at the historical prediction time and the second prediction data is calculated, and the second prediction data residual is determined.

[0142] S306: Optimize the preset long short-term memory network model according to the second prediction data residual to obtain a second prediction model.

[0143] In this embodiment, the parameters of the LSTM model are optimized according to the second prediction data residual, and the LSTM model is iterated multiple times. After the number of iterations reaches a preset number or the second prediction data residual converges, the training of the LSTM model is completed to obtain the second prediction model.

[0144] For example, Figure 5 A schematic diagram of a hybrid prediction model provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the single variable target sequence, i.e., the historical network status data of multiple historical moments, is preprocessed and then input into the ARIMA model. The ARIMA model is modeled as the first prediction model, and the historical network periodic characteristics and historical network statistical characteristics are predicted by the first prediction model to output linear first network status data.

[0145] A residual sequence is constructed according to the first network status data output by the ARIMA model, and the residual sequence and the historical network status data are divided into data sets to determine the training set and the test set. The parameters of the LSTM model are optimized to obtain the second prediction model. The historical network periodic characteristics and historical network statistical characteristics are predicted by the second prediction model, and nonlinear second network status data is output.

[0146] The prediction results of the first prediction model and the second prediction model, namely the first network status data and the second network status data, are combined to obtain the target network status data.

[0147] For example, the calculation formula of the target network status data can be expressed as:

[0148]

[0149] Furthermore, in addition to inputting the real-time network status data into the hybrid prediction model to predict the network status data for the next time period, in order to ensure that the hybrid prediction model always reflects the latest network status, the model parameters need to be updated regularly.

[0150] Specifically, at every preset time period, the latest real-time network status data is added to the training set, and the earliest historical network status data is removed to keep the size of the training set constant. Then, the model parameters are retrained using the updated training set to ensure that the model can adapt to the new data distribution and trends. The trained model is evaluated using the validation set to verify whether its prediction performance meets expectations, thereby ensuring the effectiveness and accuracy of the model update.

[0151] For example, Figure 6 A comparison diagram of the prediction of network delay provided in the embodiment of the present application, such as Figure 6 As shown in the figure, the solid line represents the actual monitored network delay, and the dotted line represents the predicted network delay. It can be seen that the network delay predicted by the hybrid prediction model based on ARIMA and LSTM has a higher accuracy.

[0152] In one possible implementation, Figure 7 Schematic diagram of the network congestion prediction method provided in the embodiment of the present application Figure 4 ,like Figure 7 As shown, the process of determining the network congestion probability at the prediction time according to the target network status data of multiple dimensions at the prediction time in S104 may include:

[0153] S401. Calculate the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction time and the mapping function of each dimension.

[0154] S402: Determine the network congestion probability at the prediction time according to congestion impact factors in multiple dimensions.

[0155] In this embodiment, in order to reflect the actual network status of the quantum cryptography network, it is necessary to map the network status data of each dimension to calculate the network congestion probability.

[0156] Each dimension of network status data has a corresponding mapping function f i (X(t)) is used to map the target network status data of each dimension into the congestion impact factor of the network congestion probability. According to the mapping function of each dimension, the target network status data of each dimension is calculated to determine the congestion impact factor of each dimension. Then, the congestion impact factors of multiple dimensions are weighted to obtain the network congestion probability at the prediction time.

[0157] In some embodiments, the process of calculating the congestion impact factor of each dimension in S401 according to the target network status data of each dimension at the prediction time and the mapping function of each dimension may include:

[0158] According to the predicted network delay at the prediction time and the positive correlation mapping function of the network delay, the congestion impact factor of the network delay is calculated.

[0159] The congestion impact factor of the bandwidth utilization rate is calculated according to the predicted bandwidth utilization rate at the prediction time and the positive correlation mapping function of the bandwidth utilization rate.

[0160] According to the predicted packet loss rate at the prediction time and the positive correlation mapping function of the packet loss rate, the congestion impact factor of the packet loss rate is calculated.

[0161] The congestion impact factor of the remaining amount of the key pool is calculated according to the predicted remaining amount of the key pool at the prediction moment and the negative correlation mapping function of the remaining amount of the key pool.

[0162] The congestion impact factor of the key generation rate is calculated according to the predicted key generation rate at the prediction moment and the negative correlation mapping function of the key generation rate.

[0163] According to the predicted key consumption rate at the prediction moment and the positive correlation mapping function of the key consumption rate, the congestion impact factor of the key consumption rate is calculated.

[0164] In this embodiment, the greater the network delay L(t), the greater the congestion probability, that is, the mapping function of the impact of network delay on network congestion probability is a positive correlation mapping function. The positive correlation mapping function of network delay L(t) can be set to the following linear relationship:

[0165]

[0166] The higher the bandwidth utilization rate B(t), the greater the congestion probability, that is, the mapping function of the bandwidth utilization rate on the network congestion probability is a positive correlation mapping function. The positive correlation mapping function of the bandwidth utilization rate B(t) can be set as the following linear relationship:

[0167]

[0168] The higher the packet loss rate P(t), the greater the congestion probability, that is, the mapping function of the packet loss rate on the network congestion probability is a positive correlation mapping function. The positive correlation mapping function of the packet loss rate P(t) can be set as the following linear relationship:

[0169]

[0170] The remaining amount of key pool K r The smaller (t), the greater the congestion probability, that is, the mapping function of the remaining amount of the key pool to the network congestion probability is a negative correlation mapping function. r The negative correlation mapping function of (t) can be set as the following linear relationship:

[0171]

[0172] Key generation rate K g The slower (t), the greater the congestion probability, that is, the mapping function of the key generation rate to the network congestion probability is a negative correlation mapping function. g The negative correlation mapping function of (t) can be set as the following linear relationship:

[0173]

[0174] Key consumption rate K c The faster (t), the greater the congestion probability, that is, the mapping function of the key consumption rate on the network congestion probability is a positive correlation mapping function. The positive correlation mapping function of the key consumption rate Kc(t) can be set as the following linear relationship:

[0175]

[0176] The final calculation formula for the congestion probability can be expressed as:

[0177]

[0178] Among them, wi is the congestion weight of each dimension, and the cumulative value of all congestion weights is 1. i (X(t)) is the mapping function of each dimension, and N is the number of dimensions, which is 6 in this embodiment.

[0179] For example, Figure 8 The prediction diagram of network congestion probability provided by the embodiment of the present application is as follows: Figure 8 As shown, by using the network congestion prediction method provided in this embodiment, the network congestion probability (congestion probability) of the quantum cryptography network at each moment can be predicted.

[0180] The network congestion prediction method provided in the above embodiment predicts the real-time network status data of the quantum cryptography network in multiple dimensions based on a pre-trained hybrid prediction model, and obtains first network status data and second network status data in multiple dimensions. The first network status data and the second network status data respectively predict the linear change and nonlinear change of the real-time network status data. The target network status data calculated according to the first network status data and the second network status data can more accurately predict the network status changes of the quantum cryptography network in different dimensions, so that the congestion probability of the quantum cryptography network at the prediction time can be accurately predicted according to the target network status data in multiple dimensions, and the prediction accuracy is guaranteed, so as to dynamically adjust the route according to the congestion probability, improve the ability of the quantum cryptography network to cope with the load environment, and reduce the key transmission problem caused by congestion; and by identifying and avoiding congestion problems in advance, the delay and failure rate can be reduced to ensure a smooth key distribution process.

[0181] Based on the above method embodiments, the embodiments of the present application also provide a network congestion prediction device. Fig. 9 A schematic diagram of the structure of a network congestion prediction device provided in an embodiment of the present application, such as Fig. 9 As shown, the device may include:

[0182] The data acquisition module 501 is used to acquire real-time network status data of multiple dimensions of the quantum cryptography network, where the real-time network status data of multiple dimensions at least includes: part or all of network delay, bandwidth utilization rate, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate;

[0183] The model prediction module 502 is used to predict the real-time network status data of each dimension using the pre-trained hybrid prediction model of each dimension to obtain the first network status data and the second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model;

[0184] A data calculation module 503, configured to calculate target network status data of each dimension at the prediction moment according to the first network status data and the second network status data of each dimension at the prediction moment;

[0185] The probability calculation module 504 is used to determine the congestion probability of the quantum cryptography network at the prediction time according to the target network state data of multiple dimensions at the prediction time.

[0186] Optionally, the first prediction model modeling module is used to obtain historical network status data at multiple historical moments in each dimension; perform differential processing on the historical network status data at multiple historical moments in each dimension to obtain a steady network status data sequence for each dimension, and determine the differential order for each dimension; calculate the autocorrelation function and partial autocorrelation function of each dimension based on the steady network status data sequence for each dimension; determine the autoregressive order and sliding average order of each dimension based on the autocorrelation function and partial autocorrelation function of each dimension; and establish a first prediction model for each dimension based on the differential order, autoregressive order and sliding average order of each dimension.

[0187] Optionally, a second prediction model training module is used to extract features from historical network status data at multiple historical moments in each dimension to obtain historical network status features of each dimension; use the first prediction model of each dimension to predict the historical network status features of each dimension to determine the first prediction data at the historical prediction moment; calculate the first prediction data residual based on the first prediction data and the historical network status data at the historical prediction moment; use a preset long short-term memory network model to predict the historical network status features of each dimension and the first prediction data residual to determine the second prediction data at the historical prediction moment; calculate the second prediction data residual based on the second prediction data and the historical network status data at the historical prediction moment; optimize the preset long short-term memory network model based on the second prediction data residual to obtain a second prediction model.

[0188] Optionally, the first prediction model building module is also used to calculate the autocorrelation function of each dimension based on the stable network status data sequence of each dimension and a preset time lag parameter.

[0189] Optionally, the second prediction model training module is also used to extract periodic features of the historical network status data at multiple historical moments in each dimension to determine the periodic features of the historical network in each dimension; and to extract statistical features of the historical network status data at multiple historical moments in each dimension to determine the statistical features of the historical network in each dimension.

[0190] Optionally, the probability calculation module 504 is specifically used to calculate the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction time and the mapping function of each dimension; and determine the network congestion probability at the prediction time according to the congestion impact factors of multiple dimensions.

[0191] Optionally, the probability calculation module 504 is further used to calculate the congestion impact factor of the network delay based on the predicted network delay at the prediction moment and the positive correlation mapping function of the network delay; calculate the congestion impact factor of the bandwidth utilization rate based on the predicted bandwidth utilization rate at the prediction moment and the positive correlation mapping function of the bandwidth utilization rate; calculate the congestion impact factor of the packet loss rate based on the predicted packet loss rate at the prediction moment and the positive correlation mapping function of the packet loss rate; calculate the congestion impact factor of the key pool remaining amount based on the predicted key pool remaining amount at the prediction moment and the negative correlation mapping function of the key pool remaining amount; calculate the congestion impact factor of the key generation rate based on the predicted key generation rate at the prediction moment and the negative correlation mapping function of the key generation rate; calculate the congestion impact factor of the key consumption rate based on the predicted key consumption rate at the prediction moment and the positive correlation mapping function of the key consumption rate.

[0192] The above-mentioned device is used to execute the method provided by the aforementioned embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0193] The above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASICs), or one or more microprocessors, or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0194] Fig.10 A schematic diagram of an electronic device provided in an embodiment of the present application, such as Fig.10 As shown, the electronic device 600 may include: a processor 601, a storage medium 602 and a bus, the storage medium 602 stores program instructions executable by the processor 601, when the electronic device 600 is running, the processor 601 and the storage medium 602 communicate through the bus, and the processor 601 executes the program instructions to execute the above method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0195] Optionally, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor at runtime, it is used to execute the above method embodiment.

[0196] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0197] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0198] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0199] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (English: Read-Only Memory, abbreviated: ROM), random access memory (English: Random Access Memory, abbreviated: RAM), disk or optical disk and other media that can store program codes.

[0200] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A network congestion prediction method, characterized in that: The method comprises: Acquire real-time network status data of multiple dimensions of the quantum cryptography network, wherein the real-time network status data of multiple dimensions at least includes: part or all of network delay, bandwidth utilization rate, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate; Use a pre-trained hybrid prediction model for each dimension to predict the real-time network status data of each dimension, and obtain first network status data and second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model; Calculating target network status data for each dimension at the prediction time according to the first network status data and the second network status data for each dimension at the prediction time; The congestion probability of the quantum cryptography network at the prediction moment is determined according to the target network state data of the multiple dimensions at the prediction moment.

2. The method according to claim 1, characterized in that The modeling steps of the first prediction model in the hybrid prediction model of each dimension are as follows: Obtaining historical network status data at multiple historical moments in each dimension; Performing differential processing on the historical network status data of multiple historical moments in each dimension to obtain a stable network status data sequence in each dimension, and determining the differential order of each dimension; Calculating the autocorrelation function and the partial autocorrelation function of each dimension according to the stationary network state data sequence of each dimension; Determine the autoregressive order and the sliding average order of each dimension according to the autocorrelation function and the partial autocorrelation function of each dimension; A first prediction model for each dimension is established according to the difference order, autoregressive order and sliding average order of each dimension.

3. The method according to claim 2, characterized in that The training steps of the second prediction model in the hybrid prediction model of each dimension are as follows: Extracting features from the historical network status data at multiple historical moments in each dimension to obtain historical network status features in each dimension; Using the first prediction model of each dimension to predict the historical network status characteristics of each dimension, and determining the first prediction data at the historical prediction time; Calculating a first prediction data residual according to the first prediction data and the historical network status data at the historical prediction time; Using a preset long short-term memory network model to predict the historical network state characteristics of each dimension and the first prediction data residual, and determine the second prediction data at the historical prediction moment; Calculating a second prediction data residual according to the second prediction data and the historical network status data at the historical prediction time; The preset long short-term memory network model is optimized according to the second prediction data residual to obtain the second prediction model.

4. The method according to claim 2, characterized in that The calculating the autocorrelation function of each dimension according to the stationary network state data sequence of each dimension comprises: The autocorrelation function of each dimension is calculated according to the steady network state data sequence of each dimension and the preset time lag parameter.

5. The method according to claim 3, characterized in that The extracting features of the historical network status data at multiple historical moments in each dimension to obtain the historical network status features of each dimension includes: Extracting periodic features of the historical network status data at multiple historical moments in each dimension to determine the periodic features of the historical network in each dimension; Statistical features are extracted from the historical network status data at multiple historical moments in each dimension to determine the historical network statistical features of each dimension.

6. The method according to claim 1, characterized in that The determining, according to the target network status data of the multiple dimensions at the prediction time, the network congestion probability at the prediction time comprises: Calculating the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction moment and the mapping function of each dimension; The network congestion probability at the prediction moment is determined according to the congestion impact factors of the multiple dimensions.

7. The method according to claim 6, characterized in that The calculating the congestion impact factor of each dimension according to the target network status data of each dimension at the prediction moment and the mapping function of each dimension includes: Calculating a congestion impact factor of the network delay according to the predicted network delay at the predicted time and a positive correlation mapping function of the network delay; Calculating a congestion impact factor of the bandwidth usage rate according to a positive correlation mapping function of the predicted bandwidth usage rate at the predicted time and the bandwidth usage rate; Calculating a congestion impact factor of the packet loss rate according to the predicted packet loss rate at the predicted time and a positive correlation mapping function of the packet loss rate; Calculating a congestion impact factor of the remaining amount of the key pool according to a negative correlation mapping function between the predicted remaining amount of the key pool at the prediction moment and the remaining amount of the key pool; Calculating a congestion impact factor of the key generation rate according to a negative correlation mapping function between the predicted key generation rate at the predicted moment and the key generation rate; The congestion impact factor of the key consumption rate is calculated according to the predicted key consumption rate at the predicted moment and a positive correlation mapping function of the key consumption rate.

8. A network congestion prediction device, characterized in that: The device comprises: A data acquisition module is used to acquire real-time network status data of multiple dimensions of the quantum cryptography network, wherein the real-time network status data of multiple dimensions at least includes: part or all of network delay, bandwidth utilization rate, packet loss rate, key pool remaining amount, key generation rate, and key consumption rate; A model prediction module, used to use a pre-trained hybrid prediction model for each dimension to predict the real-time network status data of each dimension, and obtain first network status data and second network status data of each dimension at the prediction moment, wherein the first network status data and the second network status data are respectively prediction results output by the first prediction model and the second prediction model in the hybrid prediction model; A data calculation module, configured to calculate target network status data of each dimension at the prediction moment according to the first network status data and the second network status data of each dimension at the prediction moment; A probability calculation module is used to determine the congestion probability of the quantum cryptography network at the prediction moment according to the target network state data of the multiple dimensions at the prediction moment.

9. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of the network congestion prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the network congestion prediction method according to any one of claims 1 to 7 are executed.

Images