Data flow priority management method and system for optical communication device

Abstract

The present application provides a data flow priority management method and system for an optical communication device, which relates to the field of optical communication data transmission. The method includes: extracting the network load parameters of the optical communication device and the data flow attribute characteristics corresponding to each data flow according to the sampling window; calculating the corresponding membership combination of the data flow based on the fuzzy reasoning model; defuzzifying the membership combination of each data flow respectively to convert it into the corresponding priority weights; determining the priority correction coefficient corresponding to the network load parameter, and calibrating each priority weight according to the priority correction coefficient; determining the priority of each corresponding data flow according to the calibrated priority weights. In this way, the priority allocation is dynamically adjusted based on the network status and data flow demand, which improves the transmission efficiency of the data flow and the overall service quality of the network.

Description

Technical Field

[0001] The present application relates to the technical field of optical communication data transmission, and in particular to a data flow priority management method and system for an optical communication device. Background Art

[0002] With the continuous explosive growth of global data traffic, there are many types of data streams carried in today's networks, from video conferencing, real-time voice calls to file transfers, and cloud service access. Different data streams have different requirements for network resources. For example, video streaming and voice calls are sensitive to latency, while file transfers require high bandwidth. The diversity of data streams makes it difficult for the network to take into account the needs of each service when allocating resources, resulting in a waste of network resources or a degradation of the performance of key services.

[0003] The network load and status in optical communication networks often change dynamically, but the current data flow priority management methods of most optical communication networks generally adopt static priority settings. When faced with rapid changes in network status, static priority settings cannot be adjusted in time, resulting in the network being unable to respond quickly. In addition, due to insufficient precision in priority settings, low-priority data flows may be delayed for a long time or even unable to be transmitted under high load conditions; while high-priority data flows may occupy too many resources when the network load is light, resulting in bandwidth waste.

[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention

[0005] The present application provides a data flow priority management method, system, storage medium, computer program product and electronic device for an optical communication device, which are used to at least solve the problem of resource waste and key business performance degradation caused by the static priority setting method of the optical communication device in the current related technology.

[0006] In a first aspect, an embodiment of the present application provides a data flow priority management method for an optical communication device, comprising: extracting network load parameters of the optical communication device and data flow attribute characteristics corresponding to each data flow according to a sampling window, wherein the data flow attribute characteristics include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement; the network load parameters include bandwidth utilization, link delay time, packet loss rate and flow fluctuation degree; for each of the data flows, inputting the network load parameters and the data flow attribute characteristics of the data flow into a fuzzy reasoning model, so that the fuzzy reasoning model calculates the membership of the data flow to each fuzzy reasoning rule in a preset fuzzy reasoning rule library, and combining each of the memberships to generate a corresponding membership combination; Each of the fuzzy inference rules in the fuzzy inference rule base comprises a corresponding condition part and a result part, wherein the condition part is defined by the association between the attribute characteristics of the data flow and the network load parameter, and the result part is used to describe the membership value corresponding to the condition part; the membership combinations of each of the data flows are defuzzified respectively to be converted into corresponding priority weights; the priority correction coefficient corresponding to the network load parameter is determined, and each of the priority weights is calibrated according to the priority correction coefficient; there is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient; and the priority of each of the corresponding data flows is determined according to the calibrated priority weights.

[0007] In a second aspect, an embodiment of the present application provides a data flow priority management device for an optical communication device, comprising: a feature extraction unit, for extracting network load parameters of the optical communication device and data flow attribute characteristics corresponding to each data flow according to a sampling window, wherein the data flow attribute characteristics include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement; the network load parameters include bandwidth utilization, link delay time, packet loss rate and flow fluctuation degree; a fuzzy reasoning unit, for inputting the network load parameters and the data flow attribute characteristics of the data flow into a fuzzy reasoning model for each of the data flows, so that the fuzzy reasoning model calculates the membership of the data flow to each fuzzy reasoning rule in a preset fuzzy reasoning rule library, and combines each of the memberships to generate a corresponding membership combination; the fuzzy reasoning Each of the fuzzy inference rules in the rule base includes a corresponding condition part and a result part, wherein the condition part is defined by the association between the attribute characteristics of the data flow and the network load parameter, and the result part is used to describe the membership value corresponding to the condition part; a priority determination unit is used to defuzzify the membership combination of each of the data flows respectively to convert it into corresponding priority weights; a priority calibration unit is used to determine the priority correction coefficient corresponding to the network load parameter, and calibrate each of the priority weights according to the priority correction coefficient; there is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient; a priority determination unit is used to determine the priority of each of the corresponding data flows according to the calibrated priority weights.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the data flow priority management method of an optical communication device of any embodiment of the present application.

[0009] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the data flow priority management method of the optical communication device of any embodiment of the present application are implemented.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the data flow priority management method of an optical communication device of any embodiment of the present application.

[0011] The data flow priority management method and system of an optical communication device provided by the present application can produce at least the following technical effects:

[0012] (1) By combining network load parameters and data flow attribute characteristics and using fuzzy reasoning models to dynamically calculate the priority weights of data flows, the network can dynamically adjust the priority weights of each data flow according to the real-time network status and the data flow characteristics of different services. It can respond promptly to rapid changes in network status, thereby improving the adaptability and flexibility of the network. For example, under high load conditions, it can give priority to data flows that are sensitive to latency and have high bandwidth requirements, ensuring the effective use of network resources.

[0013] (2) By setting the association between the condition part and the result part in the fuzzy reasoning rule base, the priority weight of each data flow can be accurately evaluated, and the priority correction coefficient can be used to calibrate when the network load changes. This calibration mechanism can further amplify the weight advantage of high-priority data flows over low-priority data flows under high load conditions, and can allocate more resources to higher-priority data flows to ensure the transmission of critical data flows. When the network load is light, the gap between the weight of low-priority data flows and the weight of high-priority data flows is narrowed, preventing high-priority data flows from occupying too many resources and avoiding excessive delays of low-priority data flows in low-load networks. In this way, personalized priority settings based on real-time perception of network load status are realized, effectively optimizing the utilization efficiency of network resources and achieving the purpose of balancing the needs of various data flows and resource allocation.

[0014] Through this technical solution, by comprehensively utilizing the fuzzy reasoning model and the priority correction coefficient, it is possible to dynamically adjust the priority allocation in real time according to the network status and data flow requirements, thereby ensuring the service quality of key data flows to the greatest extent, and being able to accurately identify and respond to the business needs of different data flows, significantly improving the transmission efficiency of data flows and the overall service quality of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A flowchart showing an example of a data stream priority management method for an optical communication device according to an embodiment of the present application;

[0017] Figure 2 An operation flow chart of an example of converting the membership combinations of various data streams into corresponding priority weights according to the implementation of the present application is shown;

[0018] Figure 3 A schematic diagram of the structural connection of an example of an adaptive defuzzification model according to an embodiment of the present application is shown;

[0019] Figure 4 A structural block diagram showing an example of a data flow priority management system for an optical communication device according to an embodiment of the present application;

[0020] Figure 5 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0023] Figure 1 A flowchart showing an example of a data flow priority management method for an optical communication device according to an embodiment of the present application is shown.

[0024] Regarding the execution subject of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. It is based on automated fuzzy reasoning and dynamic adjustment mechanism, and does not require frequent manual adjustments by network administrators when static priority settings are required. Through dynamic priority adjustment mechanism, refined priority setting and optimization of resource utilization efficiency, the transmission efficiency and service quality of different data streams in the optical communication network are significantly improved.

[0025] In some examples, it may be integrated and configured in an electronic device, terminal or server through software, hardware or a combination of software and hardware, and the types of the terminal, electronic device or server may be diverse, such as a mobile phone, tablet computer or desktop computer, etc.

[0026] like Figure 1 As shown, in step S110, the network load parameters of the optical communication device and the data flow attribute characteristics corresponding to each data flow are extracted according to the sampling window.

[0027] In one example of the embodiment of the present application, the size of the sampling window can be a preset fixed size. In another example of the embodiment of the present application, the size of the sampling window can be flexibly and dynamically set, and an adaptive time window mechanism can also be used to adjust the data collection frequency according to the fluctuation of the network state and data flow characteristics, so as to better track the dynamic changes of the network and data flow, and more details will be described below in conjunction with other examples.

[0028] On the one hand, by calling the network monitoring component in the optical communication device to collect various network load parameters in real time, including bandwidth utilization, link delay time, packet loss rate and traffic fluctuation degree, specifically, the real-time bandwidth occupancy of the link is collected through the network equipment to determine the bandwidth utilization. The delay time of the current network link is collected through network probes or other delay monitoring tools. The current packet loss rate is determined by analyzing the statistical data of the network equipment. By analyzing the historical traffic data, the volatility of the traffic, such as the coefficient of variation of the traffic over a period of time, is calculated.

[0029] On the other hand, by extracting and analyzing the relevant working parameters of each data stream being transmitted in the optical communication device, the corresponding data stream attribute characteristics are obtained, which include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement. For the service type of the data stream, it can be real-time video streaming, voice calls, file transfers, web browsing, etc. For the delay sensitivity of the data stream, it measures the sensitivity or tolerance of the data stream to delays. For the minimum bandwidth requirement of the data stream, it refers to the minimum bandwidth required for the data stream to ensure the quality of service during the transmission process. For the transmission time efficiency requirement of the data stream, it indicates the requirement for the data stream to complete the transmission within a specific time, which is used to indicate the urgency of the data stream.

[0030] In some examples of the embodiments of the present application, the service type is determined based on the application layer protocol information and / or port number information corresponding to the data stream. Specifically, the type of the data stream is identified by sampling data packets from the data stream and analyzing the application layer protocol (such as HTTP, FTP, etc.) of the data packets. In addition, the type of the data stream can also be effectively determined by the port number. For example, ports 80 and 443 usually correspond to HTTP / HTTPS traffic, and port 554 corresponds to RTSP traffic.

[0031] On the other hand, the delay sensitivity is determined based on the service type and / or historical transmission delay of the data flow. Here, the service type can preliminarily determine the delay sensitivity of the data flow. For example, real-time video and voice calls are marked as high delay sensitivity, while file transfers, emails, etc. are marked as low delay sensitivity. In addition, by performing statistical analysis on the historical delay data of the data flow, the average value, maximum value, standard deviation and other indicators of the delay are calculated. If the historical delay data shows that the data flow is more sensitive to delay (such as large fluctuations and low delay values), its delay sensitivity can be adjusted to a higher level.

[0032] On the other hand, the minimum bandwidth requirement is determined based on the service type and / or historical bandwidth occupancy of the data stream. Specifically, a model library is maintained in the optical communication device to store bandwidth requirement models for common service types, and can be updated and optimized according to the actual network operation. For example, the minimum bandwidth requirement for high-definition video streams is set to 5Mbps and that for audio streams is 100Kbps. In addition, by performing statistical analysis on the historical bandwidth usage data of the data stream, the average value, peak value, etc. of bandwidth usage are calculated, and future bandwidth requirements are predicted based on these data. If historical data indicates that the bandwidth usage of the data stream is high, its minimum bandwidth requirement value is adjusted accordingly.

[0033] On the other hand, the transmission timeliness requirement is determined based on the service type and / or application timeliness setting of the data stream. Specifically, the transmission timeliness requirement of the data stream is preliminarily determined based on the service type of the data stream. For example, services such as video conferencing and online games are marked as high timeliness requirements, while file transfers, emails, etc. are marked as low timeliness requirements. In addition, by analyzing whether the application layer protocol of the data stream contains timeliness setting information, such as the existence of transmission timeliness requirements configured by the application or user, the system can read these configurations and adjust the timeliness label of the data stream accordingly.

[0034] Therefore, through flexible and efficient feature extraction methods, it can adapt to a variety of network application scenarios. Whether it is an enterprise-level data center or a high-traffic Internet backbone network, it can maintain efficient operation and meet the needs of various business scenarios.

[0035] In step S120, for each data flow, the network load parameters and the data flow attribute characteristics of the data flow are input into the fuzzy reasoning model, so that the fuzzy reasoning model calculates the membership of the data flow to each fuzzy reasoning rule in the preset fuzzy reasoning rule library, and combines each membership to generate a corresponding membership combination.

[0036] Here, each fuzzy inference rule in the fuzzy inference rule base includes a corresponding condition part and a result part. The condition part is defined by the association between the attribute characteristics of the data flow and the network load parameters, and the result part is used to describe the membership value corresponding to the condition part.

[0037] It should be noted that the fuzzy reasoning rule base is the core component of the fuzzy reasoning mechanism, which contains a set of fuzzy reasoning rules. These rules model the complex relationship between data flow attribute characteristics and network load parameters through fuzzy logic to describe how to calculate the membership of data flows under different conditions. Specifically, the fuzzy rule can be an "if-then" rule, whose conditional part (precondition) is used to describe the combination of network load parameters and data flow attribute characteristics, and its result part (consequent) is used to describe the membership value corresponding to the conditional part. Exemplarily, a fuzzy reasoning rule may be expressed as: if the bandwidth utilization is "high" (0.8) and the delay sensitivity is "high" (0.7), then the membership value of the data flow is "0.7". Furthermore, for each pair of network load parameters and data flow attribute characteristics, corresponding fuzzy rules are defined respectively, and the corresponding fuzzy reasoning rule base is constructed by combination.

[0038] Specifically, the fuzzy reasoning model converts the fuzzified input (network load parameters and data flow attribute characteristics) into membership output by using the rules in the fuzzy reasoning rule base, and combines the membership values ​​obtained by matching various rules to obtain the corresponding membership combination. In addition, the fuzzy reasoning engine can use Min-Max reasoning, multiplication reasoning and other methods, which are not limited here.

[0039] In step S130, the membership combinations of the various data streams are defuzzified to be converted into corresponding priority weights.

[0040] Here, various preset defuzzification algorithms, such as centroid method, maximum membership method, etc., can be used to perform defuzzification calculation on the membership combination obtained by matching the fuzzy rules of the data stream to convert it into a specific priority weight value, which is not limited here. In some preferred implementations, the corresponding defuzzification algorithms are dynamically matched according to the actual conditions of different data streams to meet the intelligent dynamic defuzzification processing, and more details will be launched in conjunction with other examples below.

[0041] In step S140, a priority correction coefficient corresponding to the network load parameter is determined, and each priority weight is calibrated according to the priority correction coefficient, wherein there is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient.

[0042] In some embodiments, the current network load level is calculated by weighted statistics of various network load parameters or other network load analysis algorithms. Then, the corresponding priority correction coefficient is obtained through the preset positive correlation conversion. Here, the type of positive correlation can be diverse, for example, the priority correction coefficient can be positively correlated with the network load level in a linear relationship or other relationships. In this way, when the network load level is high (such as when the network is congested), the gap between the weight of the high-priority data flow and the weight of the low-priority data flow is enlarged by a correspondingly larger priority correction coefficient to ensure the transmission of critical data flows. On the contrary, when the network load level is low, the gap between the weight of the low-priority data flow and the weight of the high-priority data flow is reduced by a smaller priority correction coefficient, so as to efficiently utilize resources while avoiding long-term delays in low-priority data flows.

[0043] In conjunction with some examples of the embodiments of the present application, assuming that the network load level is relatively light, the corresponding priority correction coefficient obtained by calculation is also correspondingly small, for example, 0.7. The initial priority weight of data stream A is 0.85, and after calibration by the correction coefficient 0.7, its corresponding priority weight is 0.595. The initial priority weight of data stream B is 0.6, and its weight becomes 0.42 through adjustment of the same correction coefficient. The initial priority weight of data stream C is 0.75, and the adjusted weight is 0.525. By applying the above correction coefficients, under the network load state, the priority weights of data streams A and C, which originally have higher weights, are weakened, while data stream B, which has a lower priority weight, can obtain relatively more resource allocation, effectively avoiding high-priority data streams from occupying too many resources under light load conditions.

[0044] In step S150, the priority of each corresponding data stream is determined according to each calibrated priority weight.

[0045] In some implementations, the data streams are sorted according to the calibrated priority weights to determine the final priority of each data stream. Then, according to the sorting results, actual operations such as bandwidth allocation and queue scheduling are performed on the data streams to ensure that high-priority data streams can obtain better network resource guarantees. In addition, the priority weights of each data stream are matched with each weight interval respectively, so that the priority corresponding to each data stream is defined according to the sorting results and the matching weight intervals, for example, high, medium or low priority, and through the above-mentioned calibration process based on the priority correction coefficient, the priority level of the data stream can also be dynamically changed according to the network load.

[0046] Through the embodiments of the present application, by combining network load parameters and data flow attribute characteristics, and using the fuzzy reasoning model to dynamically calculate the priority weight of the data flow, it is possible to respond to the rapid changes in the network status in a timely manner, thereby improving the adaptability and flexibility of the network. In addition, the priority calibration mechanism based on the network load realizes the personalized setting of the priority based on the real-time perception of the network load status, effectively optimizes the utilization efficiency of network resources, and achieves the purpose of balancing the requirements of various data flows and resource allocation.

[0047] Regarding the priority correction coefficient in the above step S140, in some examples of the embodiments of the present application, the priority correction coefficient is determined by an exponential decay function based on a dynamic decay factor.

[0048] Specifically, the priority correction factor is calculated by the following formula:

[0049] , formula (1);

[0050] , formula (2);

[0051] In the formula, Indicates The priority correction factor corresponding to each data stream; represents the dynamic attenuation factor, Indicates the network load level, represents the basic attenuation factor; Represents the adjustment coefficient, which is used to control the dynamic adjustment range of the attenuation factor; Indicates the current network load level. Indicates the historical average network load level.

[0052] It should be noted that It can respond to the current network load changes in real time and make dynamic adjustments, and determine the sensitivity of the current network load relative to historical load changes. Represents the default decay rate. When the network load is stable or close to the historical average level, the system can Corrected as standard.

[0053] In addition, by introducing the historical average network load corresponding to the preset time period in the past as a benchmark, overreaction caused by short-term fluctuations can be avoided, thereby achieving a smooth transition, ensuring the continuity and stability of network resource allocation, and reducing unnecessary resource waste or sudden resource shortages.

[0054] Through the embodiments of the present application, the decay rate of the correction coefficient can be adjusted in real time according to the ratio of the current network load to the historical average load through the dynamically adjusted decay factor. In this way, in the case of high load fluctuations, the correction coefficient will decay faster to ensure that resources are preferentially allocated to high-priority data streams; however, in the case of low load fluctuations, the decay rate slows down, and available resources can be used more flexibly. As a result, the optical communication device can quickly respond to changes in network load, whether it is a sudden high load or a load drop, it can automatically adapt and adjust the resource allocation strategy to improve the overall utilization efficiency of the network.

[0055] It should be noted that in optical communication networks, the dynamics of network load and data flow characteristics determine the need for a flexible sampling window to accurately capture these changes, while the traditional fixed sampling window is difficult to adapt to complex network environments and may result in resource waste or data lag. In view of this, for the sampling window in the above step S110, in some examples of the embodiments of the present application, it can be determined through an adaptive window adjustment mechanism, which adopts an adaptive sampling window adjustment mechanism that combines network load parameters and data flow attributes.

[0056] Specifically, the adaptive sampling window mechanism dynamically adjusts the sampling frequency according to changes in network load parameters and data flow attribute characteristics, allowing the system to more sensitively perceive changes in network status. When the network load is heavy, shortening the sampling window and increasing the sampling frequency can help quickly capture changes in network status and adjust the priority of data flows in a timely manner, thereby ensuring that key data flows are given priority in complex network environments. Through dynamic adjustment, the accuracy and response speed of data flow priority management are improved, and delays and packet losses caused by sudden changes in network conditions are reduced.

[0057] Exemplarily, the adaptive window adjustment mechanism can be expressed by the following formula:

[0058] , formula (3);

[0059] , formula (4);

[0060] , formula (5);

[0061] , formula (6);

[0062] In the formula, Represents the adjusted sampling window time interval, represents the time interval of adjacent historical sampling windows; represents the first sampling window adjustment value calculated based on the network load parameter, represents the second sampling window adjustment value calculated based on the data stream attribute characteristics, and Respectively represent the weight coefficients of the corresponding adjustment values; Represents the smoothing coefficient, which is used to control the speed of sampling window adjustment; Indicates The normalized value of a network load parameter, Indicates The influence weight of each network load parameter on the sampling window adjustment is is the total number of network load parameters; Indicates The normalized value of the attribute feature of the data stream, Indicates The influence weight of each data stream attribute feature on the sampling window adjustment, is the total number of data stream attribute features; is the preset basic smoothing coefficient, is the adjustment factor, Indicates The standard deviation of the network load parameter, Indicates The standard deviation of the attribute characteristics of the data stream.

[0063] Thus, when the network load is heavy (e.g. high bandwidth utilization, high packet loss rate), The value of should be smaller to shorten the sampling window and enable the system to obtain updated network status data more frequently. On the contrary, when the network load is light, The value should be larger to extend the sampling window and reduce the usage of system resources.

[0064] In addition, data flows with high latency sensitivity and time-criticality require more precise priority management. It should be small to allow for more frequent sampling and stream prioritization. Streams with low latency sensitivity or bandwidth requirements can allow for larger sampling windows, thus reducing system burden.

[0065] By weight coefficient and right and A weighted combination is performed to comprehensively consider the impact of network load and data flow attributes. The final sampling window is based on the adjacent historical sampling window The smooth combination of the recommended adjustment value ensures that the sampling window adjustment process is smooth without sudden changes.

[0066] It should be noted that due to the fixed smoothing coefficient It cannot adapt to the dynamic changes of network status and data flow attributes. Therefore, in this embodiment, a dynamic adaptive smoothing coefficient mechanism is adopted to make It can be dynamically adjusted according to the real-time network status and data flow attribute characteristics. Specifically, when it is detected that the network load parameter fluctuates greatly (for example, by calculating the standard deviation of the network load parameter), The sampling window will automatically increase to quickly adjust to the sudden network state. When the network load parameter changes more steadily, will be reduced, making the sampling window adjustment process smoother and avoiding resource waste caused by over-adjustment. On the other hand, if the delay sensitivity or transmission timeliness of a data stream changes suddenly, The sampling window is automatically adjusted to ensure that it can quickly adapt to the dynamic needs of the data stream. On the contrary, for a stable data stream, the sampling window can be lowered. , in order to reduce the frequent adjustment of the sampling window. Therefore, by dynamically adjusting the smoothing coefficient , can quickly respond to changes in network status and data flow, and maintain network stability and efficient transmission.

[0067] Through the embodiments of the present application, network load parameters and data flow attribute characteristics are introduced as key factors for adjusting the sampling window. The adaptive sampling window mechanism can flexibly respond to diverse network environments and application requirements. When the network load suddenly increases or the data flow attributes change, the sampling frequency can be adjusted in time to ensure that the system responds quickly, thereby reducing transmission delays and packet loss rates, thereby improving adaptability and reliability in different business scenarios. In addition, the dynamically adaptive smoothing coefficient can be automatically adjusted according to real-time changes in network status. When the network load or data flow status fluctuates violently, the smoothing coefficient will be rapidly increased so that the sampling window can shrink quickly and the sampling frequency can be increased. As a result, sudden changes in network status and data flow status can be captured more timely, thereby quickly adjusting the data flow priority, achieving rapid response, and maintaining network stability and efficient transmission.

[0068] Figure 2 An operational flow chart of an example of converting the membership combinations of various data streams into corresponding priority weights according to the implementation of the present application is shown.

[0069] like Figure 2 As shown, in step S210, for each data flow, the network load parameters and the data flow attribute characteristics of the data flow and the corresponding membership combination are input into the adaptive defuzzification model to determine the corresponding target defuzzification algorithm and defuzzification parameter settings.

[0070] Here, the adaptive defuzzification model adopts a neural network model, and the defuzzification parameter setting includes a threshold adjustment parameter and a weight allocation parameter. The threshold adjustment parameter defines the threshold of the influence of different membership values ​​on the priority weight, and the weight allocation parameter defines the relative proportion of each membership value in the defuzzification process. The target defuzzification algorithm includes any one of the following: centroid method, weighted average method and maximum membership method.

[0071] Through the neural network-based optimization algorithm, continuous learning and optimization of the defuzzification processing method, it can adaptively respond to changes in the network environment, quickly adjust priority weights, and ensure that key services can still maintain efficient transmission in the event of sudden traffic or network fluctuations.

[0072] Here, through the adaptively determined threshold adjustment parameters, according to the current network conditions, the neural network model can suggest adjusting the threshold in the defuzzification process, based on the threshold adjustment parameters, ensuring that only membership values ​​above a certain threshold can significantly affect the priority weight. For example, when the network load increases, the threshold is raised to enhance the influence of the membership of high-priority data flows, ensuring that high-priority data flows can obtain sufficient resource support in the network environment. Through the adaptively determined weight allocation parameters, the weight allocation of different memberships can be adjusted in real time, ensuring that the most important data flows obtain more resource priority in the defuzzification process, allowing for refined control of the criticality of different data flows.

[0073] Through this embodiment, a neural network-based optimization algorithm is introduced to dynamically select the optimal defuzzification method, and the threshold adjustment parameters and weight allocation parameters of the defuzzification can be adjusted in real time according to the current network load situation.

[0074] It should be noted that different types of defuzzification algorithms have their own technical advantages when facing different types of input data. Specifically, the centroid method is suitable for complex scenarios and high-load scenarios where multiple data streams are mixed. When the network load is high (such as bandwidth utilization exceeding 80%), the centroid method can effectively balance the priority weights of each data stream. In addition, when there are multiple data streams in the network at the same time (such as real-time video, online games, and large file downloads) and the delay sensitivity and bandwidth requirements of these data streams vary greatly, the centroid method can balance the needs of all data streams by comprehensively considering the impact of different memberships, and will not be overly biased towards a certain type of traffic.

[0075] On the other hand, the weighted average method is more suitable for low- to medium-level load scenarios or scenarios with concentrated membership distribution. When the network load is low or medium (such as bandwidth utilization of 30%-70%), the weighted average method can allow more data flows to participate in priority calculation and use network resources more effectively. In addition, when the membership distribution of data flows is relatively concentrated and the difference is not large, the weighted average method can smooth the priority weight distribution.

[0076] On the other hand, the maximum membership method is more suitable for scenarios with high decision clarity and low load and high real-time requirements. Specifically, when a certain type of data flow has a clear priority requirement in the network (such as emergency communications, critical business data transmission), the maximum membership method can directly select the fuzzy set with the largest membership and quickly determine the priority, especially ensuring the priority of critical data flows. In scenarios with low load (such as bandwidth utilization less than 30%) but high real-time requirements, such as real-time video conferencing or online games, the maximum membership method can directly ensure that these critical data flows receive the highest priority.

[0077] In step S220, based on the target defuzzification algorithm and defuzzification parameter settings corresponding to each data stream, the membership combinations of the corresponding data streams are defuzzified to obtain the corresponding priority weights.

[0078] Here, by using the defuzzification algorithm and corresponding defuzzification parameters (including threshold adjustment and weight allocation) selected by the neural network model, the priority weight of each data stream is calculated accordingly. In this way, the most appropriate defuzzification algorithm and its parameter settings can be dynamically selected according to the real-time network status and data stream characteristics, and the priority weight of each data stream can be calculated, which can greatly improve the accuracy and adaptability of priority calculation.

[0079] The following will take the centroid method as an example to describe the details of solving the priority weight of the corresponding data flow. , assuming that its membership function is The centroid method determines the corresponding priority weight by calculating the centroid of the membership function .

[0080] Adjusting parameters by thresholding To filter membership values:

[0081] , formula (7);

[0082] Determine appropriate weight distribution parameters for key data flows , weight the membership function:

[0083] , formula (8);

[0084] , formula (9);

[0085] In the formula, is a continuous possible priority value, representing the possible output value during the defuzzification process. is the membership function, indicating the The data flow has a possible priority value The degree of membership on . Represents that by all possible priority values Perform weighted integration. is a normalization factor, which represents the value of all possible priority values Upper membership Integrate the sum of .

[0086] Through the embodiment of the present application, an adaptive defuzzification model is introduced, which can analyze the network status and data flow attribute characteristics in real time, dynamically select the defuzzification algorithm that best suits the current scenario, and adjust the threshold and weight allocation according to actual needs, thereby improving the accuracy and adaptability of priority calculation. By reasonably setting the threshold and weight, more refined and differentiated weight allocation for each data flow is supported, which can effectively reduce congestion and conflict between data flows and improve the performance and stability of the overall network.

[0087] Figure 3 A schematic diagram of the structural connection of an example of an adaptive defuzzification model according to an embodiment of the present application is shown.

[0088] like Figure 3 As shown, the adaptive defuzzification model 300 adopts a hybrid attention neural network, which includes a cascaded input layer 310, a feature extraction layer 320, a hybrid attention mechanism layer 330 and an output layer 340, wherein the feature extraction layer 320 includes a convolutional layer 321 and a fully connected layer 322.

[0089] The input layer 310 is used to receive network load parameters and data flow attribute characteristics.

[0090] Exemplarily, the input layer 310 is used to receive various network load parameters, such as bandwidth utilization, link delay time, packet loss rate and traffic fluctuation degree, and receive various data flow attribute characteristics, such as service type, delay sensitivity, minimum bandwidth requirement and transmission timeliness, and then perform standardized preprocessing on these parameters to facilitate subsequent feature fusion and unified analysis.

[0091] In the feature extraction layer 320, the time series data in the network load parameters and data flow attribute features are processed by the convolution layer 321 to determine the corresponding time series convolution features The network load parameters and non-time series data in the data flow attribute characteristics are processed by the fully connected layer 322 to determine the corresponding non-time series features .

[0092] Specifically, the convolution layer 321 is used to process timing parameters in the input data, such as link delay time and traffic fluctuation degree, while the fully connected layer 322 is used to process non-timing parameters in the input data, such as service type and minimum bandwidth requirement.

[0093] The hybrid attention mechanism layer 330 combines the self-attention mechanism and the cross-modal attention mechanism. By introducing the hybrid attention mechanism, the model can dynamically focus on key network load parameters and data flow attribute characteristics, ensuring that the most important factors for priority judgment are given priority in the defuzzification process, making the calculation of priority weights more accurate, thereby achieving more refined priority management. It is specifically used to perform the following operations:

[0094] First, calculate and The corresponding self-attention weight:

[0095] , formula (10);

[0096] , formula (11);

[0097] , formula (12);

[0098] , formula (13);

[0099] , formula (14);

[0100] In the formula, The query vector representing the temporal convolutional features, Represents the query weight matrix for temporal convolution features; The key vector representing the temporal convolutional features, Represents the key weight matrix for temporal convolution features; represents the attention weight matrix for temporal convolution features, represents the softmax activation function; The query vector representing non-temporal features, represents the query weight matrix for non-temporal features; The key vector representing the non-temporal features, represents the key weight matrix for non-temporal features; Represents the attention weight matrix for non-temporal features; and Respectively The vector transpose and The vector transpose of .

[0101] Then, the features are weighted using the attention weights:

[0102] , formula (15);

[0103] , formula (16);

[0104] , formula (17);

[0105] In the formula, A vector of values ​​representing temporal convolution features, Represents the value weight matrix for the temporal convolution feature; A vector of values ​​representing non-time series features, represents the value weight matrix for non-temporal features; Represents the time series convolution feature after attention weighting; Represents the non-temporal features after attention weighting.

[0106] Afterwards, the attention weights between cross-modal features are calculated:

[0107] , formula (18);

[0108] , formula (19);

[0109] , formula (20);

[0110] In the formula, represents the query vector of temporal convolutional features after self-attention processing, represents the self-attention query weight matrix; represents the key vector of non-temporal features after self-attention processing, represents the self-attention key weight matrix; represents the cross-modal attention weight matrix.

[0111] Finally, the feature fusion operation is performed through the cross-modal attention weight matrix:

[0112] , formula (21);

[0113] In the formula, Represents cross-modal fusion features.

[0114] By adopting a hybrid attention mechanism and cross-modal feature fusion, the model can effectively process heterogeneous data from different sources and extract the most valuable information for priority judgment, achieving strong generalization capabilities. This allows the model to maintain a stable priority judgment effect when dealing with different types of data streams and network load scenarios, avoiding the deviation caused by over-reliance on a certain type of feature.

[0115] The output layer 340 is used to process the cross-modal fusion features to determine the corresponding target defuzzification algorithm and defuzzification parameter settings.

[0116] Specifically, the output layer 340 may include multiple layers of fully connected layers, and determine the corresponding target defuzzification algorithm and defuzzification parameter settings by processing cross-modal fusion features.

[0117] The feature processing operation of the first fully connected layer is used to predict the corresponding target defuzzification algorithm:

[0118] , formula (22);

[0119] The feature processing operation is performed through the second fully connected layer to predict the corresponding defuzzification parameter settings:

[0120] , formula (23);

[0121] In the formula, Representing the selection probability of each defuzzification algorithm, and determining the target defuzzification algorithm based on the defuzzification algorithm with the highest selection probability; and Represent the weight matrix and bias term of the first fully connected output layer respectively; represents the output vector of defuzzification parameter settings, and They represent the weight matrix and bias term of the second fully connected output layer respectively.

[0122] In this way, the model has strong adaptive capabilities. By dynamically selecting defuzzification algorithms and parameter settings, it can automatically adjust strategies under different network load conditions and data flow states, enabling the system to maintain efficient operation in complex and changeable network environments and data flows with diverse states.

[0123] Through the embodiments of the present application, a hybrid attention mechanism is introduced, so that the model can quickly respond to changes in network status, adjust priority judgments and resource allocation strategies in real time, and dynamically balance the resource requirements of different data streams during transmission, thereby greatly improving the user experience of network transmission.

[0124] In some examples of the embodiments of the present application, the past network load conditions, data flow attribute characteristics and corresponding priority weight results of the optical communication device can be collected to construct a data sample set, and the data sample set can be used to train and optimize the adaptive defuzzification model.

[0125] In some embodiments, in the process of training the adaptive defuzzification model, it is expected to achieve multiple optimization goals. Specifically, the loss function of the adaptive defuzzification model includes defuzzification algorithm selection loss, defuzzification parameter setting loss and priority management effect loss. The defuzzification algorithm selection loss is used to measure the accuracy of the model in selecting the optimal defuzzification algorithm, so that the model can more accurately select the appropriate defuzzification algorithm. The defuzzification parameter setting loss is used to measure the accuracy of the model in predicting defuzzification parameters, which is used to guide the model to optimize the accuracy of parameter prediction. The priority management effect loss evaluates the actual performance of the model in priority management, which is mainly measured by the adaptive weighted delay loss term and the network resource utilization loss term. Finally, the comprehensive multi-objective loss function integrates the above-mentioned losses through weighted summation to balance the optimization goals of the model between algorithm selection, parameter setting and priority management effect.

[0126] Exemplarily, the loss function of the adaptive defuzzification model is:

[0127] , formula (24);

[0128] , formula (25);

[0129] , formula (26);

[0130] , formula (27);

[0131] , formula (28);

[0132] In the formula, represents the loss function of the adaptive defuzzification model; , , and They represent the defuzzification algorithm selection loss term, the defuzzification parameter setting loss term, the adaptive weighted delay loss term, and the network bandwidth utilization loss term respectively; , and Respectively represent the weight adjustment coefficients of the corresponding loss terms; Indicates the total number of samples in the data sample set; represents the total number of categories of the defuzzification algorithm, Indicates The samples correspond to The true label of the class defuzzification algorithm, The model predicts the The samples belong to Probability of class defuzzification algorithms; represents the total number of defuzzified parameters corresponding to each sample, Indicates In the sample The actual defuzzified parameter values, The model predicts the In the sample defuzzified parameter values; Indicates the total number of data streams corresponding to each sample, represents the RELU nonlinear activation function, Indicates In the sample Adaptive weights for each data stream, Indicates In the sample The model predicts the delay of each data stream. Indicates the delay threshold; Indicates In the sample The actual network bandwidth used by each data flow is Indicates the total amount of available network bandwidth. The model is In the sample The predicted bandwidth allocation ratio for each data flow.

[0133] Here, through the adaptive weighted delay loss term, the model can focus on optimizing the transmission quality of delay-sensitive data flows and reduce the delay of key data flows. In addition, by introducing the measurement of bandwidth resource utilization in the loss function, the model can optimize bandwidth allocation so that data flows can use network resources more efficiently. In addition, by optimizing delay and resource utilization, it is ensured that high-priority data flows are given priority during transmission. Therefore, through multi-objective comprehensive optimization, the model can significantly improve the overall network performance, especially in complex and high-load network environments. It can intelligently adjust the priority management strategy, improve the stability and service quality of the entire network, and enable the network to continue to operate efficiently in complex environments.

[0134] Through the embodiments of the present application, combined with the defuzzification algorithm selection loss, the defuzzification parameter setting loss, and the priority management effect loss, the loss function can comprehensively measure and optimize each key link of the model, ensuring that the model can simultaneously improve its performance in multiple dimensions, ensuring that all aspects of priority management are fully optimized, and avoiding deviations and suboptimal results that may be caused by single-target optimization.

[0135] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0136] Figure 4 A structural block diagram of an example of a data flow priority management system for an optical communication device according to an embodiment of the present application is shown.

[0137] like Figure 4 As shown, the data stream priority management system 400 of the optical communication device includes a feature extraction unit 410 , a fuzzy reasoning unit 420 , a priority determination unit 430 , a priority calibration unit 440 and a priority determination unit 450 .

[0138] The feature extraction unit 410 is used to extract the network load parameters of the optical communication device and the data flow attribute characteristics corresponding to each data flow according to the sampling window, wherein the data flow attribute characteristics include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement; the network load parameters include bandwidth utilization, link delay time, packet loss rate and flow fluctuation degree.

[0139] The fuzzy reasoning unit 420 is used to input the network load parameters and the data flow attribute characteristics of the data flow into the fuzzy reasoning model for each of the data flows, so that the fuzzy reasoning model calculates the membership of the data flow for each fuzzy reasoning rule in a preset fuzzy reasoning rule library, and combines each of the memberships to generate a corresponding membership combination; each of the fuzzy reasoning rules in the fuzzy reasoning rule library includes a corresponding condition part and a result part, the condition part is defined by the association relationship between the attribute characteristics of the data flow and the network load parameters, and the result part is used to describe the membership value corresponding to the condition part.

[0140] The priority determination unit 430 is used to perform defuzzification processing on the membership combinations of the data streams respectively, so as to convert them into corresponding priority weights.

[0141] The priority calibration unit 440 is used to determine the priority correction coefficient corresponding to the network load parameter, and calibrate each priority weight according to the priority correction coefficient; there is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient.

[0142] The priority determination unit 450 is used to determine the priority of each of the data flows according to the calibrated priority weights.

[0143] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of the data flow priority management method of any of the above-mentioned optical communication devices of the present application.

[0144] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of any of the above-mentioned data flow priority management methods for optical communication devices.

[0145] In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the data flow priority management method of the optical communication device.

[0146] Figure 5 is a schematic diagram of the hardware structure of an electronic device for executing a data stream priority management method for an optical communication device provided by another embodiment of the present application, such as Figure 5 As shown, the device includes:

[0147] One or more processors 510 and memory 520, Figure 5 A processor 510 is taken as an example.

[0148] The apparatus for executing the data flow priority management method of the optical communication device may further include: an input device 530 and an output device 540 .

[0149] The processor 510, the memory 520, the input device 530 and the output device 540 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0150] The memory 520, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the data flow priority management method of the optical communication device in the embodiment of the present application. The processor 510 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 520, that is, implements the data flow priority management method of the optical communication device in the above method embodiment.

[0151] The memory 520 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 520 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 520 may optionally include a memory remotely arranged relative to the processor 510, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0152] The input device 530 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 540 may include a display device such as a display screen.

[0153] The one or more modules are stored in the memory 520 , and when executed by the one or more processors 510 , perform the data flow priority management method for an optical communication device in any of the above method embodiments.

[0154] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0155] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0156] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0157] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0158] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0159] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0160] The device embodiments described above are merely illustrative, wherein 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 modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0161] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A data flow priority management method for an optical communication device, comprising: According to the sampling window, extract the network load parameters of the optical communication device and the data flow attribute characteristics corresponding to each data flow, wherein the data flow attribute characteristics include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement; the network load parameters include bandwidth utilization, link delay time, packet loss rate and flow fluctuation degree; For each of the data flows, the network load parameters and the data flow attribute characteristics of the data flows are input into a fuzzy reasoning model, so that the fuzzy reasoning model calculates the membership of the data flows for each fuzzy reasoning rule in a preset fuzzy reasoning rule base, and combines each of the memberships to generate a corresponding membership combination; each of the fuzzy reasoning rules in the fuzzy reasoning rule base includes a corresponding condition part and a result part, the condition part is defined by the association relationship between the attribute characteristics of the data flow and the network load parameters, and the result part is used to describe the membership value corresponding to the condition part; Defuzzifying the membership combinations of the data streams to convert them into corresponding priority weights; Determining a priority correction coefficient corresponding to the network load parameter, and calibrating each priority weight according to the priority correction coefficient; There is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient; Determining the priority of each of the data streams according to the calibrated priority weights; The membership combinations of the data streams are defuzzified to convert them into corresponding priority weights, including: For each of the data flows, the network load parameters and the data flow attribute characteristics of the data flows and the corresponding membership combinations are input into the adaptive defuzzification model to determine the corresponding target defuzzification algorithm and defuzzification parameter settings; the adaptive defuzzification model adopts a neural network model, and the defuzzification parameter settings include a threshold adjustment parameter and a weight allocation parameter; wherein the threshold adjustment parameter defines the impact threshold of different membership values ​​on the priority weight, and the weight allocation parameter defines the relative proportion of each membership value in the defuzzification process; wherein the target defuzzification algorithm includes any one of the following: a centroid method, a weighted average method, and a maximum membership method; Based on the target defuzzification algorithm and defuzzification parameter settings corresponding to each data stream, the membership combination of each corresponding data stream is defuzzified to obtain the corresponding priority weights; The adaptive defuzzification model adopts a hybrid attention neural network, which includes an input layer, a feature extraction layer, a hybrid attention mechanism layer and an output layer; the input layer is used to receive network load parameters and data flow attribute characteristics; The feature extraction layer includes a convolution layer and a fully connected layer; the convolution layer is used to process the time series data in the network load parameters and data flow attribute features to determine the corresponding time series convolution features The fully connected layer is used to process the non-time series data in the network load parameters and data flow attribute characteristics to determine the corresponding non-time series features ; The hybrid attention mechanism layer is used to perform the following operations: Calculate separately and The corresponding self-attention weight: ， In the formula, The query vector representing the temporal convolutional features, Represents the query weight matrix for temporal convolution features; The key vector representing the temporal convolutional features, Represents the key weight matrix for temporal convolution features; represents the attention weight matrix for temporal convolution features, represents the softmax activation function; The query vector representing non-temporal features, represents the query weight matrix for non-temporal features; The key vector representing the non-temporal features, represents the key weight matrix for non-temporal features; Represents the attention weight matrix for non-temporal features; and Respectively The vector transpose and The vector transpose of ; Use attention weights to weight features: In the formula, A vector of values ​​representing the temporal convolution features, Represents the value weight matrix for the temporal convolution feature; A vector of values ​​representing non-time series features, represents the value weight matrix for non-temporal features; Represents the time series convolution feature after attention weighting; Represents the non-temporal features after attention weighting; Calculate the attention weights between cross-modal features: In the formula, represents the query vector of temporal convolutional features after self-attention processing, Shows the self-attention query weight matrix; represents the key vector of non-temporal features after self-attention processing, represents the self-attention key weight matrix; represents the cross-modal attention weight matrix; The feature fusion operation will be performed through the cross-modal attention weight matrix: In the formula, Represents cross-modal fusion features; The output layer is used to process the cross-modal fusion features to determine the corresponding target defuzzification algorithm and defuzzification parameter settings: In the formula, Representing the selection probability of each defuzzification algorithm, and determining the target defuzzification algorithm based on the defuzzification algorithm with the highest selection probability; and Represent the weight matrix and bias term of the first fully connected output layer respectively; represents the output vector of defuzzification parameter settings, and Represent the weight matrix and bias term of the second fully connected output layer respectively; The loss function of the adaptive defuzzification model is: In the formula, represents the loss function of the adaptive defuzzification model; , , and They represent the defuzzification algorithm selection loss term, the defuzzification parameter setting loss term, the adaptive weighted delay loss term, and the network bandwidth utilization loss term respectively; , and Respectively represent the weight adjustment coefficients of the corresponding loss terms; Indicates the total number of samples in the data sample set; represents the total number of categories of the defuzzification algorithm, Indicates The samples correspond to The true label of the class defuzzification algorithm, The model predicts the The samples belong to Probability of class defuzzification algorithms; represents the total number of defuzzified parameters corresponding to each sample, Indicates In the sample The actual defuzzified parameter values, The model predicts the In the sample defuzzified parameter values; Indicates the total number of data streams corresponding to each sample, represents the RELU nonlinear activation function, Indicates In the sample Adaptive weights for each data stream, Indicates In the sample The model predicts the delay of each data stream. Indicates the delay threshold; Indicates In the sample The actual network bandwidth used by each data flow is Indicates the total amount of available network bandwidth. The model shown is In the sample The bandwidth allocation ratio predicted for each data flow; The priority correction factor is calculated by the following formula: In the formula, Indicates The priority correction factor corresponding to each data stream; represents the dynamic attenuation factor, Indicates the network load level, represents the basic attenuation factor; Represents the adjustment coefficient, which is used to control the dynamic adjustment range of the attenuation factor; Indicates the current network load level. Indicates the historical average network load level.

2. The method according to claim 1, wherein: The sampling window is determined by an adaptive window adjustment mechanism: In the formula, Represents the adjusted sampling window time interval, represents the time interval of adjacent historical sampling windows; represents the first sampling window adjustment value calculated based on the network load parameter, represents the second sampling window adjustment value calculated based on the data stream attribute characteristics, and Respectively represent the weight coefficients of the corresponding adjustment values; Represents the smoothing coefficient, which is used to control the speed of sampling window adjustment; Indicates The normalized value of a network load parameter, Indicates The influence weight of each network load parameter on the sampling window adjustment is is the total number of network load parameters; Indicates The normalized value of the attribute feature of the data stream, Indicates The weight of the influence of the data stream attribute characteristics on the sampling window adjustment, is the total number of data stream attribute features; is the preset basic smoothing coefficient, is the adjustment factor, Indicates The standard deviation of the network load parameter, Indicates The standard deviation of the attribute characteristics of the data stream.

3. The method according to claim 1, wherein: The service type is determined based on the application layer protocol information and / or port number information corresponding to the data flow, the delay sensitivity is determined based on the service type of the data flow and / or the historical transmission delay, the minimum bandwidth requirement is determined based on the service type of the data flow and / or the historical bandwidth occupancy, and the transmission timeliness requirement is determined based on the service type of the data flow and / or the application timeliness setting.

4. A data flow priority management device for an optical communication device, used to execute the data flow priority management method for an optical communication device according to any one of claims 1 to 3, the device comprising: A feature extraction unit is used to extract network load parameters of the optical communication device and data flow attribute characteristics corresponding to each data flow according to the sampling window, wherein the data flow attribute characteristics include service type, delay sensitivity, minimum bandwidth requirement and transmission time efficiency requirement; the network load parameters include bandwidth utilization, link delay time, packet loss rate and flow fluctuation degree; A fuzzy reasoning unit, for each of the data flows, inputs the network load parameters and the data flow attribute characteristics of the data flow into a fuzzy reasoning model, so that the fuzzy reasoning model calculates the membership of the data flow to each fuzzy reasoning rule in a preset fuzzy reasoning rule library, and combines each of the memberships to generate a corresponding membership combination; each of the fuzzy reasoning rules in the fuzzy reasoning rule library includes a corresponding condition part and a result part, the condition part is defined by the association relationship between the attribute characteristics of the data flow and the network load parameter, and the result part is used to describe the membership value corresponding to the condition part; A priority determination unit, configured to perform defuzzification processing on the membership combinations of the data streams to convert them into corresponding priority weights; A priority calibration unit, used to determine a priority correction coefficient corresponding to the network load parameter, and calibrate each priority weight according to the priority correction coefficient; There is a positive correlation between the network load level indicated by the network load parameter and the priority correction coefficient; The priority determination unit is used to determine the priority of each of the data flows according to the calibrated priority weights.

Images