Optical transmission cross-domain service real-time configuration method and device and computer program product

Abstract

The invention discloses an optical transmission cross-domain service real-time configuration method, an optical transmission cross-domain service real-time configuration device and a computer program product, and aims to realize unified planning and dynamic output of cross-domain service end-to-end service quality targets by constructing a multi-domain quality strategy fusion model and integrating service quality strategies and resource states of different network domains. By collecting performance data of each domain in real time and combining with a time sequence analysis algorithm, the network performance degradation trend can be accurately identified; based on quality propagation path analysis of a cross-domain service topological relation graph, a quality degradation root can be quickly positioned, the influence range of the quality degradation root can be quantified, and the service quality stability of a key service in a resource limited scene is ensured; by establishing a cross-domain service quality mapping system, heterogeneous network performance parameters are mapped to end-to-end service quality indexes, and the cross-domain collaborative operation and maintenance efficiency is greatly improved. A continuous iterative optimization mechanism driven by reinforcement learning is adopted, adaptive adjustment of network resource allocation and a routing strategy is realized, and a cross-domain service quality dynamic optimization closed loop is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and particularly relates to a real-time configuration method, device, and computer program product for optical transmission cross-domain services. Background Art

[0002] Ensuring the quality of cross-domain services faces multiple technical challenges, and its core problem stems from the heterogeneity of network domains and the lack of coordination mechanisms. In cross-domain transmission scenarios, there are significant differences in the quality of service strategies and network performance parameters of different network domains. For example, the setting thresholds and implementation methods of key indicators such as delay, packet loss rate, and bandwidth are different. This difference causes service data to encounter quality degradation phenomena such as delay jitter, sudden packet loss, or bandwidth bottlenecks when crossing multiple network domains, especially at the network domain boundary, where performance mutations caused by parameter mismatches are likely to occur.

[0003] The diversity of service data further exacerbates the complexity of quality assurance. Interactive services such as real-time audio and video have strict requirements for delay and jitter, while file transfer services are more concerned with packet loss rate and throughput stability. This differential demand makes the quality degradation problems in the same network environment present diverse manifestations in different service types, requiring targeted evaluation indicators and optimization strategies. In addition, there is an accumulative effect on the processing delay of cross-domain service requests. When the request path involves multiple network domains, the superposition of the processing delay and transmission delay within each domain will significantly affect the end-to-end service quality, and the quality degradation characteristics will show non-linear propagation characteristics with the dynamic change of the network path.

[0004] The heterogeneity of monitoring data has become the main obstacle to accurate evaluation. The monitoring data formats adopted by each network domain lack a unified standard, and there are differences in timestamp accuracy at the millisecond or even second level, resulting in difficulties in cross-domain data correlation analysis. The existing monitoring system is difficult to effectively integrate multi-domain data and cannot construct a complete end-to-end service quality portrait. At the same time, the actual transmission path of service requests is affected by network topology and load balancing strategies, and may pass through different network devices and links at different times, making traditional static quality mapping rules unable to adapt to the dynamically changing network environment. Summary of the Invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a real-time configuration method, device, and computer program product for optical transmission cross-domain services to achieve accurate evaluation and dynamic optimization of end-to-end service quality and improve the level of cross-domain service quality assurance.

[0006] To solve the above technical problem, the present invention provides a real-time configuration method for optical transmission cross-domain services, including:

[0007] Step S1: Obtain quality of service (QoS) policies, network performance parameters, and end-to-end paths of cross-domain services from each network domain, construct a multi-domain QoS policy fusion model for analyzing quality degradation propagation and optimizing resource allocation, and output end-to-end service quality objectives.

[0008] Step S2: Determine the collection metrics based on the sensitivity of the network performance to the service type, and collect the performance and quality data of each domain on the cross-domain service path in real time.

[0009] Step S3: Use time series analysis algorithms to identify the changing trends of network performance and service quality in each domain, and detect potential quality degradation risks.

[0010] Step S4: When the performance data for multiple consecutive cycles deteriorates continuously, determine that quality degradation has occurred in the target domain. Combine the network topology and service association data to construct a cross-domain service topology relationship graph, and analyze the propagation path and scope of influence of the quality degradation.

[0011] Step S5: Dynamically adjust the resource allocation and traffic scheduling of relevant domains according to the propagation path, and formulate service quality guarantee priorities for different domains and services.

[0012] Step S6: Establish a cross-domain service quality mapping system, map the network performance parameters of each domain to end-to-end service quality indicators, and form a unified view of the multi-domain quality status.

[0013] Step S7: Use reinforcement learning algorithms to optimize resource allocation and routing strategies, and continuously iterate to narrow the gap between the actual service quality and the target.

[0014] Preferably, the specific steps of Step S1 include:

[0015] Step S11: Obtain historical network bandwidth usage data from the network path selection scheme through a deep neural network, set inter-domain resource usage thresholds for the resource occupancy metric set, and obtain an inter-domain resource constraint table.

[0016] Step S12: Extract inter-domain connectivity association degree data from the service process definition set according to the inter-domain resource constraint table, and analyze and classify the inter-domain connectivity association degree data through a graph convolutional algorithm to obtain a service transmission priority level sequence.

[0017] Step S13: Extract the inter-domain link state information set from each network domain for the service transmission priority level sequence, and perform cumulative calculation on the inter-domain service level indicators to obtain an initial service quality matrix.

[0018] Step S14: Establish a service quality objective function based on the initial service quality matrix, and use a neural network adaptive regulator to optimize the parameters of the service quality objective function, and output the end-to-end service quality benchmark data for cross-domain services.

[0019] Preferably, step S2 specifically includes:

[0020] Step S21: Determine the delay / jitter index of real-time audio and video data and the packet loss rate index of file transmission according to the sensitivity difference of network performance to service types.

[0021] Step S22: Construct a service type recognition parameter table according to the network performance sensitive type, and classify and label the service data stream features through a deep learning classifier to obtain the service sensitivity classification result;

[0022] Step S23: Extract the index monitoring endpoint group from the service sensitivity classification result, and use the support vector machine algorithm to obtain the basic network quality data within the acquisition time period;

[0023] Step S24: Collect network packet loss data from the basic network quality data to obtain an end-to-end index collection set;

[0024] Step S25: Construct a performance benchmark curve through the end-to-end index collection set, and obtain the delay jitter reference threshold from the data sampling granularity configuration table;

[0025] Step S26: Establish a performance monitoring rule sequence according to the delay jitter reference threshold, and use an adaptive evaluator to perform hierarchical statistics on the network quality data, and output the end-to-end path performance evaluation result;

[0026] Step S27: Construct a real-time monitoring function for the end-to-end path performance evaluation result, extract the quality degradation warning parameters from the performance monitoring rule sequence, and generate real-time network performance monitoring data.

[0027] Preferably, step S3 specifically includes:

[0028] Step S31: Obtain the original performance data according to the data acquisition timing sequence, obtain the performance index reference value through data standardization processing, and extract the timing characteristics of the performance index reference value through a long short-term memory network to obtain a performance characteristic sequence;

[0029] Step S32: Construct a time window sliding set for the performance characteristic sequence, and compare the time window sliding set with the trend recognition reference threshold after classification by the random forest algorithm to obtain a trend feature vector;

[0030] Step S33: Statistically calculate the performance change parameters within the domain according to the trend feature vector, and obtain a multi-dimensional trend curve after calculating the index correlation weight and collecting multi-domain network quality data;

[0031] Step S34: Extract quality degradation features from the multi-dimensional trend curve. Match the quality degradation features with the quality decay mode sequences in the degradation feature library. If the match is successful, update the quality degradation warning rule and output the network quality risk identification result.

[0032] Preferably, the specific steps of step S4 are as follows:

[0033] Step S41: Perform node encoding on the inter-domain link state data through a graph neural network to obtain an inter-domain connectivity state parameter set, which includes network inter-domain link feature information.

[0034] Step S42: Use the gradient boosting tree to perform path analysis on the service traffic data according to the inter-domain connectivity state parameter set to generate an inter-domain service association graph.

[0035] Step S43: Construct a propagation path depth matrix for the inter-domain service association graph and calculate the quality decay propagation rate from the propagation path depth matrix.

[0036] Step S44: Establish a quality impact assessment function according to the quality decay propagation rate, and use an adaptive predictor to dynamically adjust the quality impact assessment function and output the service quality impact prediction result.

[0037] Preferably, the specific steps of step S5 are as follows:

[0038] Step S51: Analyze the intra-domain traffic distribution characteristics through a deep neural network to obtain network load benchmark data, which includes traffic fluctuation period values.

[0039] Step S52: Classify different service types according to the network load benchmark data using a support vector machine to generate a service quality grade table, which contains service priority weight values.

[0040] Step S53: Calculate the resource allocation benchmark value for the service quality grade table, collect the inter-domain traffic data to obtain an initial resource allocation plan, which contains resource guarantee parameters.

[0041] Step S54: Establish a resource dynamic allocation function according to the initial resource allocation plan, and use an adaptive scheduler to obtain a scheduling instruction from the resource scheduling sequence and then output the resource allocation result, which contains resource utilization data.

[0042] Preferably, the specific steps of step S6 are as follows:

[0043] Step S61: Classify and label the intra-domain performance parameters through a deep neural network to obtain a performance index classification sequence.

[0044] Step S62: Extract quality features according to the performance index classification sequence, and use a support vector machine to perform mapping transformation on the quality features to generate an index mapping relation table;

[0045] Step S63: Calculate the in-domain performance benchmark value according to the index mapping relation table, set quality mapping thresholds for different service types, collect cross-domain performance data, and obtain a multi-domain quality status table;

[0046] Step S64: Construct a quality evaluation function through the multi-domain quality status table, extract evaluation parameters from the performance benchmark value, perform cumulative statistics on the quality mapping thresholds to generate a quality difference sequence, and construct a unified status view according to the quality difference sequence.

[0047] Preferably, the specific steps of step S7 include:

[0048] Step S71: Construct a reinforcement learning state space according to the resource optimization reward value, and perform state evaluation on the network topology through a reinforcement learning agent to obtain a routing optimization parameter set;

[0049] Step S72: Extract the resource allocation benchmark value from the routing optimization parameter set, and perform multi-dimensional evaluation on the network resources through a dynamic planner to generate a resource allocation scheme;

[0050] Step S73: Construct an optimization execution sequence for the resource allocation scheme, collect resource competition data, and obtain an initial optimization instruction;

[0051] Step S74: Construct an optimization iteration function through the initial optimization instruction, extract performance indicators from the resource competition data to generate an optimization feedback sequence, and output a resource optimization instruction.

[0052] The present invention also provides an optical transmission cross-domain service real-time configuration device, including:

[0053] One or more processors;

[0054] A memory;

[0055] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the optical transmission cross-domain service real-time configuration method.

[0056] The present invention also provides a computer program product, including computer instructions, and the computer instructions direct a computer device to perform the operations corresponding to the method.

[0057] Implementing the present invention has the following beneficial effects: By constructing a multi-domain quality policy fusion model and integrating service quality policies and resource status of different network domains, the present invention realizes the unified planning and dynamic output of end-to-end service quality objectives for cross-domain services, effectively solving the problem of low collaborative efficiency caused by service quality policy differences in a multi-domain network environment. By collecting performance data of each domain in real time and combining with time series analysis algorithms, the trend of network performance degradation can be accurately identified, the risk of quality degradation can be warned in advance, and the fault prevention ability can be significantly improved. Based on the quality propagation path analysis of the cross-domain service topology graph, the root cause of quality degradation can be quickly located and its influence range can be quantified. Combining with the dynamic resource scheduling and priority guarantee mechanism, the service quality stability of critical services in resource-constrained scenarios can be ensured. By establishing a cross-domain service quality mapping system, heterogeneous network performance parameters are mapped to end-to-end service quality indicators, forming a unified view of multi-domain quality status, and greatly improving the cross-domain collaborative operation and maintenance efficiency. By adopting a reinforcement learning-driven continuous iterative optimization mechanism, the adaptive adjustment of network resource allocation and routing strategies is realized, forming a dynamic optimization closed-loop for cross-domain service quality. The present invention breaks through the technical limitations of traditional single-domain service quality guarantee, realizes the global perception, intelligent decision-making and real-time control of service transmission quality in a multi-domain network environment, and significantly improves the network reliability, resource utilization efficiency and service quality guarantee ability in complex cross-domain service scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0059] Figure 1 FIG. is a schematic flowchart of a real-time configuration method for an optical transmission cross-domain service in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The following descriptions of the embodiments are with reference to the drawings to exemplify specific embodiments in which the present invention can be implemented.

[0061] Please refer to Figure 1 as shown, Embodiment 1 of the present invention provides a real-time configuration method for an optical transmission cross-domain service, including:

[0062] Step S1: Obtain service quality policies, network performance parameters, and end-to-end paths of cross-domain services from each network domain, construct a multi-domain quality policy fusion model for analyzing quality degradation propagation and optimizing resource allocation, and output end-to-end service quality objectives;

[0063] Step S2: Determine the collection metrics based on the sensitivity of network performance to business types, and collect the performance and quality data of each domain on the cross-domain service path in real time;

[0064] Step S3: Use time series analysis algorithms to identify the changing trends of network performance and service quality in each domain, and detect potential quality degradation risks;

[0065] Step S4: When the performance data for multiple consecutive cycles deteriorates continuously, determine that the quality degradation has occurred in the target domain, construct a cross-domain service topology relationship graph by combining network topology and service association data, and analyze the propagation path and influence scope of the quality degradation;

[0066] Step S5: Dynamically adjust the resource allocation and traffic scheduling of relevant domains according to the propagation path, and formulate service quality guarantee priorities for different domains and services;

[0067] Step S6: Establish a cross-domain service quality mapping system, map the network performance parameters of each domain to the end-to-end service quality indicators, and form a unified view of the multi-domain quality status;

[0068] Step S7: Use reinforcement learning algorithms to optimize resource allocation and routing strategies, and continuously iterate to narrow the gap between the actual service quality and the target.

[0069] Specifically, in the embodiment of the present invention, Step S1 specifically includes:

[0070] Step S11: Obtain historical network bandwidth usage data from the network path selection scheme through a deep neural network, set the inter-domain resource usage threshold for the resource occupancy metric set, and obtain the inter-domain resource constraint table;

[0071] Step S12: Extract the inter-domain connectivity association degree data from the service process definition set according to the inter-domain resource constraint table, and analyze and classify the inter-domain connectivity association degree data through a graph convolutional algorithm to obtain the service transmission priority level sequence;

[0072] Step S13: Extract the inter-domain link state information set from each network domain for the service transmission priority level sequence, and perform cumulative calculation on the inter-domain service level indicators to obtain the initial service quality matrix;

[0073] Step S14: Establish a service quality objective function according to the initial service quality matrix, and use a neural network adaptive regulator to optimize the parameters of the service quality objective function, and output the end-to-end service quality benchmark data for cross-domain services.

[0074] Specifically, an inter-domain quality of service (QoS) level indicator mapping table is established according to the network topology description language. Historical network bandwidth usage data is obtained from the network path selection scheme through a deep neural network. An inter-domain resource usage threshold is set for the set of resource occupancy metric indicators, and an inter-domain resource constraint table is obtained. When the quality degradation evaluation indicator is at the inter-domain resource usage threshold, the inter-domain connectivity correlation data is extracted from the set of business process definitions. The different inter-domain service level indicators are analyzed and classified through a graph convolutional algorithm to distinguish the business transmission priority level sequence. If the resource allocation restriction rule satisfies the business transmission priority level sequence, the inter-domain link status information set is extracted from each network domain, and the inter-domain service level indicators are accumulated and calculated to obtain an initial QoS matrix. An inter-domain resource interruption probability table is constructed based on the initial QoS matrix, and real-time resource allocation data is obtained from the inter-domain resource constraint table, and an inter-domain resource allocation reference value is calculated. A QoS objective function is established through the inter-domain resource allocation reference value, and the inter-domain link carrying capacity is dynamically calculated to obtain an end-to-end QoS parameter set. A quality policy fusion function is constructed for the end-to-end QoS parameter set, and a neural network adaptive regulator is used for parameter optimization to output cross-domain service QoS benchmark data.

[0075] The QoS between network domains has an important impact on business transmission. When cross-domain services are transmitted in a financial transaction scenario, the QoS level indicator mapping table includes three dimensions: response latency, packet loss rate, and throughput.

[0076] Examples are as follows:

[0077] In the power system, the dispatching center needs to monitor and control the operating status of each substation and power plant in real time. By establishing an inter-domain QoS level indicator mapping table according to the network topology description language, the communication requirements and QoS requirements between different domains can be clarified;

[0078] The communication link between the dispatching center and the substation needs to have high reliability and low latency to ensure the timely transmission of control instructions;

[0079] By obtaining historical network bandwidth usage data through a deep neural network, an inter-domain resource usage threshold can be set, and an inter-domain resource constraint table can be obtained;

[0080] When the quality degradation evaluation indicator is close to the inter-domain resource usage threshold, the system can analyze different inter-domain service level indicators through a graph convolutional algorithm to distinguish the business transmission priority level sequence;

[0081] The transmission priority of dispatching instructions is higher than that of data acquisition and monitoring information. If the resource allocation restriction rule satisfies the business transmission priority level sequence, the inter-domain link status information set is extracted from each network domain, and the inter-domain service level indicators are accumulated and calculated to obtain an initial QoS matrix;

[0082] Construct an inter-domain resource interruption probability table based on the initial quality of service matrix, and calculate the reference value of inter-domain resource allocation by combining the real-time data in the inter-domain resource constraint table;

[0083] Establish a quality of service objective function through the reference value of inter-domain resource allocation, dynamically calculate the inter-domain link capacity, and obtain the end-to-end quality of service parameter set, so as to ensure the efficient operation of power dispatching and control services.

[0084] Further, in the embodiment of the present invention, step S2 specifically includes:

[0085] Step S21, determine the delay / jitter index of real-time audio and video data and the packet loss rate index of file transfer according to the sensitivity difference of network performance to service types.

[0086] Step S22, construct a service type recognition parameter table according to the network performance sensitive type, classify and mark the service data stream features through a deep learning classifier, and obtain the service sensitivity classification result. At the same time, set the data acquisition frequency parameter for the network domain identification sequence.

[0087] Step S23, extract the index monitoring endpoint group from the service sensitivity classification result, and use the support vector machine algorithm to obtain the basic network quality data within the acquisition time period. At the same time, generate an initial sequence of performance indicators through data sampling granularity configuration.

[0088] Step S24, collect network packet loss data from the basic network quality data to obtain an end-to-end index collection set. Specifically, monitor the real-time audio and video data path according to the initial sequence of performance indicators, collect the delay reference value and jitter reference value for different network domains, and collect network packet loss data for file transfer services to obtain the end-to-end index collection set.

[0089] Step S25, construct a performance benchmark curve through the end-to-end index collection set, obtain the real-time sampling sequence from the data sampling granularity configuration table, and obtain the delay jitter reference threshold;

[0090] Step S26, establish a performance monitoring rule sequence according to the delay jitter reference threshold, and use an adaptive evaluator to perform hierarchical statistics on the network quality data, and output the end-to-end path performance evaluation result.

[0091] Step S27, construct a real-time monitoring function for the end-to-end path performance evaluation result, extract the quality degradation warning parameters from the performance monitoring rule sequence, and generate real-time network performance monitoring data.

[0092] Examples are as follows:

[0093] In the power dispatching and control scenario, the sensitivities of different types of services to network performance vary significantly. The transmission of power dispatching instructions has extremely high requirements for latency and reliability, while data acquisition and monitoring information (SCADA) has relatively high requirements for data integrity and real-time performance;

[0094] By constructing a service type identification parameter table, the sensitivities of various services to network performance can be clarified. Using a deep learning classifier to classify and label the characteristics of service data streams to obtain the service sensitivity classification results;

[0095] For the network domain identification sequence, set the data acquisition frequency parameter to ensure the real-time performance and accuracy of key service data. The acquisition frequency of power dispatching instructions may be set to multiple times per second, while the acquisition frequency of data acquisition and monitoring information may be set to once per minute;

[0096] Extract the index monitoring endpoint groups from the service sensitivity classification results. These endpoint groups include key nodes such as power dispatching centers, substations, and power plants;

[0097] Use the support vector machine algorithm to obtain the basic network quality data within the acquisition time period, such as latency, jitter, packet loss rate, etc.;

[0098] Generate the initial sequence of performance indicators through data sampling granularity configuration to ensure the accuracy and real-time performance of the data. On the communication link between the power dispatching center and the substation, collect latency and jitter data to evaluate the real-time performance of the link;

[0099] Monitor the real-time audio and video data path according to the initial sequence of performance indicators, and collect the latency reference value and jitter reference value for different network domains;

[0100] In the video surveillance data transmission between the power dispatching center and the substation, collect latency and jitter data to ensure the real-time performance and smoothness of video surveillance. For file transfer services, collect network packet loss data to evaluate the reliability of file transfer. Through these data, obtain the end-to-end index acquisition set, providing a basis for subsequent performance evaluation;

[0101] Construct a performance benchmark curve through the end-to-end index acquisition set. This curve reflects the performance level of the power communication network under normal operating conditions. Obtain the real-time sampling sequence from the data sampling granularity configuration table and calculate the latency jitter reference threshold;

[0102] Set the latency jitter reference threshold for power dispatching instruction transmission to 10 milliseconds to ensure the timely transmission of dispatching instructions;

[0103] Establish a sequence of performance monitoring rules based on the reference threshold of delay jitter. These rule sequences are used to monitor the performance status of the power communication network in real time. An adaptive evaluator is used to classify and statistically analyze the network quality data, and the end-to-end path performance evaluation result is output;

[0104] When it is monitored that the delay jitter of a certain link exceeds the reference threshold, the system will issue a warning to prompt the network administrator to take measures for optimization. In this way, performance problems in the power communication network can be discovered and solved in a timely manner to ensure the efficient operation of power dispatching and control services.

[0105] Furthermore, in the embodiment of the present invention, step S3 specifically includes:

[0106] Step S31, obtain the original performance data according to the data acquisition timing sequence, obtain the performance index reference value through data standardization processing, and perform time series feature extraction on the performance index reference value through a long short-term memory network to obtain a performance feature sequence;

[0107] Step S32, construct a time window sliding set for the performance feature sequence. The time window sliding set is classified by a random forest algorithm and compared with the trend recognition reference threshold to obtain a trend feature vector;

[0108] Step S33, statistically analyze the performance change parameters within the domain according to the trend feature vector. After the performance change parameters are calculated by the index correlation weight and the multi-domain network quality data is collected, a multi-dimensional trend curve is obtained;

[0109] Step S34, extract the quality degradation features from the multi-dimensional trend curve. The quality degradation features are matched with the quality attenuation mode sequence in the degradation feature library. If the match is successful, the quality degradation warning rule is updated, and the network quality risk identification result is output.

[0110] Specifically, extract the quality degradation features through the multi-dimensional trend curve, obtain the quality attenuation mode sequence from the degradation feature library, perform cumulative calculation on the index correlation weight to obtain the risk degree data; establish a quality degradation warning rule according to the risk degree data, use an adaptive warning device to monitor the quality attenuation trend in real time, and match the degradation risk features from the quality attenuation mode sequence; construct a risk warning function for the degradation risk features, dynamically update the quality degradation warning rule, and output the network quality risk identification result.

[0111] The example is as follows:

[0112] In the power dispatching and control scenario, in the communication network between the power dispatching center and each substation and power plant, obtain the original performance data from the network nodes in each domain according to the data acquisition timing sequence, such as delay, jitter, packet loss rate, etc.;

[0113] By performing data standardization on this data, the benchmark values of performance metrics are obtained. The long short-term memory network is used to extract temporal features from the benchmark values of performance metrics, generating performance feature sequences;

[0114] On the power dispatching instruction transmission link, by collecting delay data at multiple time points, after standardization processing, the long short-term memory network is used to extract temporal features, generating performance feature sequences reflecting the trend of delay changes;

[0115] A time window sliding set is constructed through the performance feature sequences. The time window is set to 5 minutes and the sliding step is 1 minute. The random forest algorithm is used to classify the network performance data within each time window, and by comparing with the pre-set trend recognition benchmark threshold, a trend feature vector is obtained;

[0116] When the delay within a certain time window continuously increases and exceeds the trend recognition benchmark threshold, the random forest algorithm marks this window as an upward trend in delay, generating the corresponding trend feature vector;

[0117] According to the trend feature vector, performance change parameters within the statistical domain are calculated, such as the frequency and amplitude of the increase in delay, etc. The index correlation weights are calculated for different performance data intervals;

[0118] Within the delay data interval [10ms, 20ms], the impact of delay changes on the power dispatching instruction transmission service is relatively small, and the correlation weight is set to 0.6; while within the delay data interval [20ms, 30ms], the impact of delay changes on the service is relatively large, and the correlation weight is set to 0.9;

[0119] Multi-domain network quality data is collected to generate multi-dimensional trend curves. Delay, jitter and other data are collected from multiple domains such as the power dispatching center, substation A, and substation B to generate multi-dimensional trend curves, reflecting the changing trends of network performance in each domain;

[0120] Quality degradation feature extraction is performed through the multi-dimensional trend curves. The quality attenuation mode sequence is obtained from the degradation feature library. Features such as continuous increase in delay and increase in jitter are extracted from the multi-dimensional trend curves and matched with the quality attenuation mode sequence (such as link congestion mode) in the degradation feature library;

[0121] The index correlation weights are cumulatively calculated to obtain the risk degree data. According to the delay correlation weight of 0.9 and the jitter correlation weight of 0.7, the risk degree data is obtained as 0.8 through cumulative calculation (such as weighted summation);

[0122] Based on the risk degree data, a quality degradation early warning rule is established. When the risk degree data reaches or exceeds 0.8, the quality degradation early warning rule is triggered;

[0123] Use an adaptive early warning device to monitor the quality decay trend in real time, and match the degradation risk features from the quality decay mode sequence, such as link congestion mode;

[0124] Construct a risk early warning function for the degradation risk features, such as a linear regression function or a logistic regression function;

[0125] Dynamically update the quality degradation early warning rules, such as adjusting the early warning threshold according to the latest network performance data and service requirements, and output the network quality risk identification results, including risk level, risk description, etc.;

[0126] The risk level is "medium-high", and the risk description is "The quality of Link A has decreased, which may cause delays in the transmission of scheduling instructions."

[0127] Furthermore, in the embodiment of the present invention, step S4 specifically includes:

[0128] Step S41, perform node encoding on the inter-domain link state data through a graph neural network to obtain an inter-domain connectivity state parameter set, and the inter-domain connectivity state parameter set includes network inter-domain link feature information;

[0129] Step S42, perform path analysis on the service traffic data by using a gradient boosting tree according to the inter-domain connectivity state parameter set to generate an inter-domain service association graph;

[0130] Step S43, construct a propagation path depth matrix for the inter-domain service association graph, and calculate the quality decay propagation rate from the propagation path depth matrix;

[0131] Step S44, establish a quality impact evaluation function according to the quality decay propagation rate, and dynamically adjust the quality impact evaluation function by using an adaptive predictor to output the service quality impact prediction result.

[0132] Specifically, quality monitoring is performed on the target network domain according to the number of performance degradation cycles and the quality degradation determination threshold. Node encoding is performed on the inter-domain link status data through a graph neural network to obtain an inter-domain connectivity status parameter set. Business traffic distribution features are extracted from the inter-domain connectivity status parameter set, and gradient boosting trees are used to perform path analysis on the business traffic data to generate an inter-domain business association graph. A propagation path depth matrix is constructed based on the inter-domain business association graph, the quality decay propagation rate is calculated for different propagation depths, the quality data of each network domain is collected to obtain an initial propagation direction sequence. A propagation influence area graph is constructed through the initial propagation direction sequence, propagation trend features are extracted from the quality decay propagation rate, and the propagation path depth matrix is updated to generate a propagation range prediction sequence. A quality impact assessment function is established based on the propagation range prediction sequence, an adaptive predictor is used to analyze the quality decay trend, and influence degree features are extracted from the propagation influence area graph. An end-to-end service impact model is constructed for the influence degree features, and the quality impact assessment function is dynamically adjusted to output the predicted result of service quality impact.

[0133] The example is as follows:

[0134] Quality monitoring is performed on the target network domain according to the number of performance degradation cycles and the quality degradation determination threshold. The number of performance degradation cycles is set to 3 cycles, and the quality degradation determination threshold is an increase in delay of 50% or a packet loss rate exceeding 1%.

[0135] Node encoding is performed on the inter-domain link status data through a graph neural network to obtain an inter-domain connectivity status parameter set. In the communication network between the power dispatching center and Substation A and Substation B, the graph neural network encodes the link status data of each node to generate a parameter set reflecting the connectivity status between nodes, including information such as link delay, packet loss rate, and bandwidth utilization.

[0136] Business traffic distribution features are extracted from the inter-domain connectivity status parameter set, and the traffic distribution of services such as power dispatching instructions and Supervisory Control and Data Acquisition (SCADA) information is extracted.

[0137] Gradient boosting trees are used to perform path analysis on the business traffic data to generate an inter-domain business association graph. The gradient boosting trees analyze the transmission path of power dispatching instructions from the dispatching center to the substations to generate a business association graph, showing the business traffic distribution and link status of different paths.

[0138] A propagation path depth matrix is constructed based on the inter-domain business association graph. The propagation path depth is set to 3 layers. The first layer is the power dispatching center, the second layer is Substation A and Substation B, and the third layer is the power plant and lower-level substations.

[0139] Calculate the quality decay propagation rate for different propagation depths. During the propagation from the first layer to the second layer, the rate of increase in delay is 10 ms / cycle, and the rate of increase in packet loss rate is 0.5% / cycle; during the propagation from the second layer to the third layer, the rate of increase in delay is 15 ms / cycle, and the rate of increase in packet loss rate is 1% / cycle. Collect the quality data of each network domain to obtain the initial propagation direction sequence.

[0140] The initial propagation direction sequence is Power Dispatching Center → Substation A → Power Plant. Construct a propagation influence area graph through the initial propagation direction sequence. The propagation influence area graph shows the influence range of the performance change of the Power Dispatching Center on Substation A and the Power Plant.

[0141] Extract the propagation trend characteristics from the quality decay propagation rate, such as the slope of the increase in delay and the slope of the increase in packet loss rate.

[0142] Update the propagation path depth matrix to generate a propagation range prediction sequence. According to the propagation trend characteristics, predict that within the next 3 cycles, the performance degradation will spread from Substation A to the lower-level substations, and generate a propagation range prediction sequence.

[0143] Establish a quality impact assessment function based on the propagation range prediction sequence. Set the quality impact assessment function as the weighted sum of delay and packet loss rate, and the weights are determined according to the service type.

[0144] Use an adaptive predictor to analyze the quality decay trend. The adaptive predictor dynamically adjusts the prediction parameters according to historical data and real-time data to improve the prediction accuracy.

[0145] Extract the impact degree characteristics from the propagation influence area graph, such as the number of nodes in the affected area and the proportion of traffic volume.

[0146] Construct an end-to-end service impact model for the impact degree characteristics, evaluate the impact of the delay and packet loss rate of power dispatching instructions on the operation of the power system.

[0147] Dynamically adjust the quality impact assessment function and output the predicted results of service quality impact. According to the evaluation results of the service impact model, dynamically adjust the parameters of the quality impact assessment function and output the predicted results of service quality impact of power dispatching instruction transmission, including the amplitude of increase in delay and the amplitude of increase in packet loss rate.

[0148] Furthermore, when the quality degrades in the target network domain, analyze the network topology structure and service association relationship of the target network domain, judge whether the quality degradation spreads to the next domain along the cross-domain service path. If it causes the network performance and service quality of the next domain not to meet the requirements, send a warning to the next domain and dynamically adjust and optimize its network resources.

[0149] Establish quality parameter monitoring points according to the degree of network topology structure, and obtain quality change parameters by collecting inter-domain propagation probability data through the quality parameter monitoring points;

[0150] Obtain quality degradation status data through the quality change parameters, and use a deep learning model to analyze the quality degradation status data to obtain inter-domain link status marking data;

[0151] Construct an inter-domain propagation prediction table for the inter-domain link status marking data, calculate the performance impact value according to the inter-domain propagation prediction table, and collect inter-domain link load data to generate a resource allocation benchmark table;

[0152] Construct a resource adjustment interval sequence through the resource allocation benchmark table, use an adaptive decision maker to process the performance impact value to obtain a performance impact level, and obtain a resource adjustment plan from the resource adjustment interval sequence;

[0153] Specifically, construct an inter-domain service association graph according to the degree of network topology structure and quality parameter monitoring points, extract quality change parameters from the inter-domain propagation probability data through a support vector machine, and judge the quality degradation status based on the quality parameter threshold;

[0154] Construct a quality attenuation trend graph through the quality degradation status data, use deep learning to classify and mark the inter-domain link status data, and combine with the network performance benchmark value to generate an inter-domain propagation prediction table;

[0155] Calculate the next-domain performance impact value according to the inter-domain propagation prediction table, set performance benchmark thresholds for different service types, collect inter-domain link load data, and obtain a resource allocation benchmark table;

[0156] Construct a resource adjustment interval sequence through the resource allocation benchmark table, extract adjustment parameters from the performance impact value, and analyze the inter-domain link load data to generate an initial warning rule;

[0157] Establish a warning level sequence according to the initial warning rule, use an adaptive decision maker to classify the performance impact degree, and obtain an adjustment plan from the resource adjustment interval sequence;

[0158] Construct a resource optimization function for the adjustment plan, dynamically update the warning level sequence, and output resource adjustment instructions and warning information.

[0159] The example is as follows:

[0160] When constructing an inter-domain service association graph according to the degree of network topology structure and quality parameter monitoring points, it can be achieved by analyzing the connection relationship and quality data between each node in the network;

[0161] Exemplarily, assume a network consisting of a core layer, an aggregation layer, and an access layer. The core layer has 3 nodes, the aggregation layer has 5 nodes, and the access layer has 10 nodes;

[0162] For the data collected by the monitoring points, such as bandwidth utilization rate and latency, constructing a correlation graph can clearly show the dependency relationships of inter-domain traffic flows. For example, the bandwidth utilization rate of the link from core node A to aggregation node B is 70%, and the latency is 5 milliseconds. The correlation graph can mark the quality status of this link;

[0163] In a possible implementation, when using a support vector machine to extract quality change parameters from inter-domain propagation probability data, the model can be trained based on historical data. Assume that the data in the past 7 days shows that the bandwidth utilization rate of a certain link has increased from 60% to 90%, and the latency has increased from 4 milliseconds to 15 milliseconds. The support vector machine will identify the trend of quality degradation and extract change parameters such as the rising rate;

[0164] Combined with the quality parameter threshold, for example, when the latency exceeds 10 milliseconds, it is regarded as degradation, and it can be determined that this link has entered the quality degradation state. The advantage of this method is that it can quickly locate the problem link;

[0165] It should be noted that when constructing a quality decay trend graph, deep learning can be used to classify and mark the inter-domain link status data. For example, for a certain link, the latencies of three consecutive samples are 8, 12, and 15 milliseconds respectively, and the packet loss rate has increased from 0.005% to 0.02%. The model can mark its status as "decaying";

[0166] Combined with the network performance benchmark value, for example, the normal latency is less than 10 milliseconds, generating an inter-domain propagation prediction table can estimate the performance of this link within the next 1 hour, which helps to perceive potential risks in advance;

[0167] Specifically, when calculating the performance impact value of the next domain, the business impact can be analyzed according to the prediction table. Assume that a certain real-time service requires a latency lower than 4 milliseconds, and the prediction table shows that the latency in the next domain may reach 6 milliseconds, then the impact value is negative;

[0168] Collecting inter-domain link load data, such as the current load is 80%, a resource allocation benchmark table can be generated, and an adjustment interval such as 75% - 85% can be set;

[0169] In an embodiment, when extracting adjustment parameters from the performance impact value to generate an initial warning rule, a rule such as "when the latency exceeds 5 milliseconds and lasts for 10 minutes, trigger a warning" can be set;

[0170] According to the load data analysis, the warning level sequence can be divided into three levels: low, medium, and high, corresponding to adjustment schemes such as increasing the bandwidth by 10% or switching to a backup link respectively. The adaptive decision-maker dynamically grades according to the impact degree to ensure the accuracy of the warning;

[0171] When constructing the resource optimization function, the resources can be adjusted by integrating the load and delay data. For example, when the load of a certain link reaches 90%, the optimization function will first reduce the load to 80% and update the warning level sequence;

[0172] The output resource adjustment instruction may be "allocate 50 Mbps of standby bandwidth", and the warning message will prompt "the link is about to be congested". For example, through multiple aspects of verification, if the delay and load of a certain link exceed the standard at the same time, the optimization function will generate a consistent solution by combining the historical trend and the current state. This multi-dimensional analysis ensures the reliability of the solution and improves the continuity of business transmission at the same time.

[0173] Furthermore, in the embodiment of the present invention, step S5 specifically includes:

[0174] Step S51, analyze the traffic distribution characteristics within the domain through a deep neural network to obtain network load reference data, and the network load reference data includes traffic fluctuation period values;

[0175] Step S52, classify different service types using a support vector machine according to the network load reference data to generate a service quality grade table, and the service quality grade table contains service priority weight values;

[0176] Step S53, calculate the resource allocation reference value for the service quality grade table, and obtain the initial resource allocation plan after collecting the inter-domain traffic data. The initial resource allocation plan contains resource guarantee parameters;

[0177] Step S54, establish a resource dynamic allocation function according to the initial resource allocation plan, and use an adaptive scheduler to obtain a scheduling instruction from the resource scheduling sequence and then output the resource allocation result. The resource allocation result contains resource utilization data.

[0178] Specifically, construct a resource evaluation table based on the propagation path depth value and the propagation influence area, analyze the characteristics of the in-domain traffic distribution through a deep neural network to obtain the network load benchmark data; construct a service priority sequence from the network load benchmark data, use a support vector machine to classify different service types, and generate a service quality level table; calculate the resource allocation benchmark value according to the service quality level table, set resource guarantee parameters for different priority levels, collect inter-domain traffic data, and obtain the initial resource allocation plan; construct a resource scheduling sequence through the initial resource allocation plan, extract the priority weights from the service quality level table, adjust the resource guarantee parameters, and generate a scheduling rule set; establish a resource dynamic allocation function according to the scheduling rule set, use an adaptive scheduler to classify the network resources by priority, and obtain the scheduling instructions from the resource scheduling sequence; perform resource reallocation on the scheduling instructions, update the scheduling rule set in real time, and output the resource allocation result; collect performance feedback data according to the resource allocation result, verify the resource scheduling effect through a quality monitor, and update the resource evaluation table.

[0179] The example is as follows:

[0180] Construct a resource evaluation table based on the propagation path depth value and the propagation influence area. Exemplarily, it can be understood as evaluating resource requirements by counting the propagation levels and coverage ranges of data streams in the power network;

[0181] The propagation path depth value reflects the number of hops of service data from the starting point to the end point. For example, scheduling and command services may involve 3 hops, while management information services may be 5 hops;

[0182] The propagation influence area refers to the number of nodes affected by the data stream. For example, scheduling and command services cover 10 key nodes. Use these data to generate a resource evaluation table, which records the depth value and the size of the influence area of each service type;

[0183] Analyze the characteristics of the in-domain traffic distribution through a deep neural network. Specifically, the traffic data of the power network can be input into the neural network to identify the load patterns during peak hours. For example, the traffic of scheduling and command services surges at 9 am, and the bandwidth requirement reaches 10 Mbps, while the management information services are relatively stable in the afternoon and only require 1 Mbps;

[0184] After analysis, obtain the network load benchmark data. For example, the total load benchmark of a certain area is 20 Mbps, and the scheduling and command services account for 50%. Construct a service priority sequence from the network load benchmark data. In a possible implementation, use a support vector machine to classify the service types;

[0185] Scheduling and command services are classified as the highest level due to high real-time requirements, with a priority value of 1; protection and control services are the second, with a value of 2; management information services are 3;

[0186] Generate a service quality level table, which records the priorities and quality requirements of each type of service. For example, the dispatching and command service requires low latency and high reliability;

[0187] Calculate the baseline value of resource allocation according to the service quality level table. Preferably, allocate 10 Mbps bandwidth for the dispatching and command service, 2 Mbps for the protection and control service, and 1 Mbps for the management information service;

[0188] After collecting the inter-domain traffic data, the initial resource allocation plan may show that the total bandwidth of a certain link is 15 Mbps, which needs to be further optimized;

[0189] Construct a resource scheduling sequence through the initial resource allocation plan. For example, extract the priority weights from the service quality level table. The weight of the dispatching and command service is 0.5, the protection and control service is 0.3, and the management information service is 0.2. After adjusting the resource guarantee parameters, generate a scheduling rule set. For example, the dispatching and command service preferentially occupies 80% of the link resources;

[0190] Establish a resource dynamic allocation function according to the scheduling rule set. It can be understood that the adaptive scheduler adjusts resources according to the real-time load. For example, when the load of a certain node exceeds 15 Mbps, the dispatching and command service is preferentially guaranteed;

[0191] The scheduling instruction may indicate to temporarily transfer the excess bandwidth from the management information service to the protection and control service. After reallocating resources for the scheduling instruction, update the scheduling rule set in real time;

[0192] In one embodiment, if the latency of a certain link rises to 50 ms due to high load, the rule set will dynamically reduce the bandwidth of the management information service to 0.5 Mbps to ensure the stability of the dispatching and command service;

[0193] Collect performance feedback data according to the resource allocation results, and verify the effect through a quality monitor. For example, the latency of the dispatching and command service drops to 10 ms, and the packet loss rate is lower than 0.1%, indicating that the resource scheduling is effective. The updated resource evaluation table shows the optimized depth value and influence area, and the loop supports the next scheduling.

[0194] Furthermore, in the embodiment of the present invention, step S6 specifically includes:

[0195] Step S61, classify and label the intra-domain performance parameters through a deep neural network to obtain a performance index classification sequence;

[0196] Step S62, extract quality features according to the performance index classification sequence, and use a support vector machine to perform mapping transformation on the quality features to generate an index mapping relationship table;

[0197] Step S63: Calculate the in-domain performance benchmark value according to the index mapping relation table, set the quality mapping threshold for different business types, collect cross-domain performance data, and obtain the multi-domain quality status table;

[0198] Step S64: Construct a quality evaluation function through the multi-domain quality status table, extract evaluation parameters from the performance benchmark value, perform cumulative statistics on the quality mapping threshold, generate a quality difference sequence, and construct a unified status view according to the quality difference sequence.

[0199] Specifically, construct an initial mapping table according to the end-to-end quality target data and the index collection period, classify and label the in-domain performance parameters through a deep neural network to obtain a performance index classification sequence; extract business quality features from the performance index classification sequence, and use a support vector machine to perform mapping transformation on different quality features to generate an index mapping relation table; calculate the in-domain performance benchmark value according to the index mapping relation table, set the quality mapping threshold for different business types, collect cross-domain performance data, and obtain the multi-domain quality status table; construct a quality evaluation function through the multi-domain quality status table, extract evaluation parameters from the performance benchmark value, perform cumulative statistics on the quality mapping threshold, and generate a quality difference sequence; establish a unified status view according to the quality difference sequence, use an adaptive evaluator to classify the quality differences, and obtain real-time data from the multi-domain quality status table; calculate the end-to-end quality difference for the real-time data, dynamically update the quality evaluation function, and output the quality evaluation result; construct a quality status display graph according to the quality evaluation result, and perform visual processing on the quality difference data through a multi-dimensional renderer to form a multi-domain quality unified view.

[0200] The example is as follows:

[0201] In the power dispatching and control scenario, construct a resource evaluation table according to the propagation path depth value and the propagation influence area;

[0202] In the communication network between the power dispatching center and substations and power plants, the propagation path depth value reflects the hierarchical relationship from the dispatching center to each substation and power plant, and the propagation influence area indicates the influence range of the performance change of a certain node on other nodes;

[0203] Analyze the in-domain traffic distribution characteristics through a deep neural network to obtain network load benchmark data. The deep neural network analyzes the traffic distribution of services such as power dispatching instructions and supervisory control and data acquisition (SCADA) information, and obtains network load benchmark data such as the average traffic and peak traffic of each node;

[0204] Construct a business priority sequence from the network load benchmark data. The transmission priority of power dispatching instructions is the highest, followed by the transmission of supervisory control and data acquisition information, and finally the transmission of other auxiliary services;

[0205] Use a support vector machine to classify different service types and generate a service quality grade table. The support vector machine classifies power dispatching instructions into the highest priority level (priority 1), data acquisition and monitoring information into the second highest priority level (priority 2), and other auxiliary services into a lower priority level (priority 3) according to characteristics such as the delay sensitivity and packet loss sensitivity of the services;

[0206] Calculate the resource allocation benchmark value according to the service quality grade table. The benchmark bandwidth for services with priority 1 (power dispatching instructions) is 100 Mbps, the benchmark bandwidth for services with priority 2 (data acquisition and monitoring information) is 50 Mbps, and the benchmark bandwidth for services with priority 3 (other auxiliary services) is 20 Mbps;

[0207] Set resource guarantee parameters for different priority levels. For example, the resource guarantee parameter for services with priority 1 is a bandwidth guarantee ratio of 80%, the resource guarantee parameter for services with priority 2 is a bandwidth guarantee ratio of 60%, and the resource guarantee parameter for services with priority 3 is a bandwidth guarantee ratio of 40%. Collect inter-domain traffic data to obtain an initial resource allocation plan;

[0208] According to the traffic requirements and resource guarantee parameters of each node, the initial resource allocation plan is as follows: The power dispatching center is allocated 200 Mbps of bandwidth, Substation A is allocated 100 Mbps of bandwidth, Substation B is allocated 80 Mbps of bandwidth, and the power plant is allocated 50 Mbps of bandwidth;

[0209] Construct a resource scheduling sequence through the initial resource allocation plan. The resource scheduling sequence is arranged in priority order. Services with priority 1 are allocated resources first, followed by services with priority 2, and finally services with priority 3;

[0210] Extract the priority weights from the service quality grade table. For example, the weight of priority 1 is 0.6, the weight of priority 2 is 0.3, and the weight of priority 3 is 0.1;

[0211] Adjust the resource guarantee parameters to generate a scheduling rule set. When the network load is high, the scheduling rule set stipulates that the bandwidth of services with priority 1 should be guaranteed first, and the bandwidth of services with priority 3 can be appropriately compressed;

[0212] Establish a resource dynamic allocation function according to the scheduling rule set. The resource dynamic allocation function dynamically adjusts the bandwidth allocation of each node according to the real-time network load and service priority;

[0213] Use an adaptive scheduler to divide the network resources into priorities and obtain scheduling instructions from the resource scheduling sequence;

[0214] The adaptive scheduler monitors the network resource usage in real time according to the resource scheduling sequence. When it is found that the bandwidth of high-priority services at a certain node is insufficient, it issues a scheduling instruction to allocate bandwidth resources from low-priority services to ensure the transmission quality of high-priority services.

[0215] Furthermore, in the embodiment of the present invention, step S7 specifically includes:

[0216] Step S71, construct a reinforcement learning state space according to the resource optimization reward value, and perform a state evaluation on the network topology through a reinforcement learning agent to obtain a set of routing optimization parameters;

[0217] Step S72, extract the resource allocation reference value from the set of routing optimization parameters, and perform a multi-dimensional evaluation on the network resources through a dynamic planner to generate a resource allocation plan;

[0218] Step S73, construct an optimization execution sequence for the resource allocation plan, collect resource competition data, and obtain an initial optimization instruction;

[0219] Step S74, construct an optimization iteration function through the initial optimization instruction, extract performance indicators from the resource competition data, generate an optimization feedback sequence, and output a resource optimization instruction.

[0220] Specifically, construct a reinforcement learning state space according to the resource optimization reward value and the quality gap index, perform a state evaluation on the network topology through a reinforcement learning agent to obtain a set of routing optimization parameters; extract the resource allocation reference value from the set of routing optimization parameters, use a dynamic planner to perform a multi-dimensional evaluation on the network resources to generate a resource allocation plan; construct an optimization execution sequence according to the resource allocation plan, set an optimization adjustment threshold for different service types, collect resource competition data, and obtain an initial optimization instruction; construct an optimization iteration function through the initial optimization instruction, extract performance indicators from the resource competition data, dynamically update the optimization adjustment threshold, and generate an optimization feedback sequence; establish a convergence determination rule according to the optimization feedback sequence, use an adaptive evaluator to quantify the optimization effect, and obtain the execution result from the optimization execution sequence; construct an iterative optimization function for the execution result, update the convergence determination rule in real time, and output a resource optimization instruction; verify the plan according to the resource optimization instruction, verify the optimization effect through a quality evaluator, and update the reinforcement learning reward value.

[0221] Examples are as follows:

[0222] In the power dispatching and control scenario, construct a reinforcement learning state space according to the resource optimization reward value and the quality gap index;

[0223] The resource optimization reward value can be set as the reduction in the transmission delay of power dispatching instructions, the reduction in the packet loss rate, etc. The quality gap index represents the gap between the current network performance and the ideal state, such as the delay gap, the packet loss rate gap, etc.;

[0224] The reinforcement learning agent conducts a state evaluation of the network topology to obtain a routing optimization parameter set. The reinforcement learning agent evaluates the communication link status between the power dispatching center, substations, and power plants, and generates a routing optimization parameter set, including parameters such as link delay, packet loss rate, and bandwidth utilization;

[0225] Extract the resource allocation reference value from the routing optimization parameter set. The reference bandwidth for power dispatching instruction transmission is extracted as 100 Mbps, and the reference bandwidth for data acquisition and monitoring information transmission is 50 Mbps;

[0226] The dynamic planner conducts a multi-dimensional evaluation of network resources to generate a resource allocation plan. The dynamic planner comprehensively considers factors such as network load, service priority, and link status, and generates a resource allocation plan, allocating 200 Mbps of bandwidth to the power dispatching center, 100 Mbps of bandwidth to Substation A, 80 Mbps of bandwidth to Substation B, and 50 Mbps of bandwidth to the power plant;

[0227] Construct an optimization execution sequence according to the resource allocation plan. The optimization execution sequence is executed in the order of service priority, first ensuring the transmission of power dispatching instructions, and then the transmission of data acquisition and monitoring information;

[0228] Set optimization adjustment thresholds for different service types. The delay adjustment threshold for power dispatching instruction transmission is 10 ms, and the packet loss rate adjustment threshold is 0.1%;

[0229] The delay adjustment threshold for data acquisition and monitoring information transmission is 20 ms, and the packet loss rate adjustment threshold is 0.5%;

[0230] Collect resource competition data to obtain an initial optimization instruction. When resource competition is detected on the link between the power dispatching center and Substation A, generate an initial optimization instruction, suggesting adjusting the bandwidth of Substation A from 100 Mbps to 120 Mbps;

[0231] Construct an optimization iteration function through the initial optimization instruction. The optimization iteration function sets the iteration step size and the number of iterations according to the initial optimization instruction, and gradually adjusts the resource allocation;

[0232] Extract performance indicators from the resource competition data, such as delay, packet loss rate, etc., and dynamically update the optimization adjustment threshold;

[0233] When the time delay of power dispatching instruction transmission continuously exceeds 10 ms, the time delay adjustment threshold is dynamically updated to 15 ms to adapt to the change of network performance, generate an optimized feedback sequence, and record the results of each optimization iteration, such as resource allocation and network performance metrics;

[0234] Establish a convergence determination rule according to the optimized feedback sequence. When the results of three consecutive optimization iterations meet the conditions that the time delay is less than 10 ms and the packet loss rate is less than 0.1%, it is determined that the optimization process converges;

[0235] Quantify the optimization effect using an adaptive evaluator. The adaptive evaluator calculates the quantified value of the optimization effect based on the network performance metrics before and after optimization, such as the reduction in time delay and the reduction in packet loss rate;

[0236] Obtain the execution results from the optimization execution sequence, and record the specific operations and results of each optimization execution, such as bandwidth adjustment and link switching;

[0237] Construct an iterative optimization function for the execution results. The iterative optimization function adjusts the optimization strategy and parameters according to the execution results to further improve the optimization effect;

[0238] Update the convergence determination rule in real time. When the network load changes, update the time delay and packet loss rate thresholds in the convergence determination rule in real time to adapt to the new network environment;

[0239] Output resource optimization instructions. According to the results of the iterative optimization function, generate the final resource optimization instructions, such as bandwidth allocation schemes and link switching instructions;

[0240] Verify the solution according to the resource optimization instructions. Apply the optimized resource allocation scheme to the power dispatching and control network, and monitor network performance metrics, such as time delay and packet loss rate;

[0241] Verify the optimization effect through a quality evaluator. The quality evaluator compares the network performance metrics before and after optimization to evaluate whether the optimization effect meets the expected goal;

[0242] Update the reinforcement learning reward value. According to the evaluation results of the optimization effect, adjust the reinforcement learning reward value to guide the reinforcement learning agent to better select optimization strategies in subsequent optimization processes.

[0243] Corresponding to the optical transmission cross-domain service real-time configuration method described in the foregoing Embodiment 1 of the present invention, Embodiment 2 of the present invention further provides an optical transmission cross-domain service real-time configuration device, including:

[0244] One or more processors;

[0245] A memory;

[0246] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to execute the optical transmission cross-domain service real-time configuration method described in the foregoing Embodiment 1 of the present invention.

[0247] Corresponding to the optical transmission cross-domain service real-time configuration method described in the foregoing Embodiment 1 of the present invention, Embodiment 3 of the present invention further provides a computer program product, including computer instructions, and the computer instructions instruct a computer device to execute operations corresponding to the optical transmission cross-domain service real-time configuration method described in the foregoing Embodiment 1 of the present invention.

[0248] Preferably, the processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor. The processor is the control center of the device and connects various parts of the device through various interfaces and lines.

[0249] The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, applications required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory may be a high-speed random access memory, or may also be a non-volatile memory, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory may also be other volatile solid-state storage devices.

[0250] It should be noted that the above device may include but is not limited to a processor and a memory, which can be understood by those skilled in the art.

[0251] As can be seen from the above description, compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a multi-domain quality strategy fusion model and integrating the service quality strategies and resource status of different network domains, the present invention realizes the unified planning and dynamic output of end-to-end service quality objectives for cross-domain services, effectively solving the problem of low collaboration efficiency caused by service quality strategy differences in a multi-domain network environment. By collecting performance data of each domain in real time and combining with time series analysis algorithms, the deterioration trend of network performance can be accurately identified, and the risk of quality degradation can be predicted in advance, significantly improving the fault prevention ability. Based on the quality propagation path analysis of the cross-domain service topology graph, the root cause of quality degradation can be quickly located and its influence scope can be quantified. Combining with the dynamic resource scheduling and priority guarantee mechanism, the service quality stability of critical services in resource-constrained scenarios can be ensured. By establishing a cross-domain service quality mapping system, heterogeneous network performance parameters are mapped to end-to-end service quality indicators, forming a unified view of multi-domain quality status, and greatly improving the cross-domain collaborative operation and maintenance efficiency. By adopting a reinforcement learning-driven continuous iterative optimization mechanism, the adaptive adjustment of network resource allocation and routing strategies is realized, forming a dynamic optimization closed-loop for cross-domain service quality. The present invention breaks through the technical limitations of traditional single-domain service quality assurance, realizes the global perception, intelligent decision-making and real-time control of service transmission quality in a multi-domain network environment, and significantly improves the network reliability, resource utilization efficiency and service quality assurance ability in complex cross-domain service scenarios.

[0252] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A real-time configuration method for optical transmission cross-domain services, characterized in that, Including: Step S1: Obtain service quality policies, network performance parameters, and end-to-end paths of cross-domain services from each network domain, construct a multi-domain quality policy fusion model for analyzing quality degradation propagation and optimizing resource allocation, and output end-to-end service quality objectives; Step S2: Determine the collection metrics based on the sensitivity of the service type to network performance, and collect the performance and quality data of each domain on the cross-domain service path in real time; Step S3: Use time series analysis algorithms to identify the changing trends of network performance and service quality in each domain, and detect potential quality degradation risks; Step S4: When the performance data for multiple consecutive periods deteriorates continuously, determine that quality degradation has occurred in the target domain, construct a cross-domain service topology relationship diagram in combination with network topology and service association data, and analyze the propagation path and influence scope of quality degradation; Step S5: Dynamically adjust the resource allocation and traffic scheduling of relevant domains according to the propagation path, and formulate service quality guarantee priorities for each domain and service; Step S6: Establish a cross-domain service quality mapping system, map the network performance parameters of each domain to end-to-end service quality indicators, and form a unified view of multi-domain quality status; Step S7: Use reinforcement learning algorithms to optimize resource allocation and routing strategies, and continuously iterate to narrow the gap between the actual service quality and the target; 2. The method according to claim 1, characterized in that, The specific steps of Step S1 include: Step S11: Obtain historical network bandwidth usage data from the network path selection scheme through a deep neural network, set inter-domain resource usage thresholds for the resource occupancy metric set, and obtain an inter-domain resource constraint table; Step S12: Extract inter-domain connectivity association degree data from the business process definition set according to the inter-domain resource constraint table, and analyze and classify the inter-domain connectivity association degree data through a graph convolutional algorithm to obtain a service transmission priority level sequence; Step S13: Extract an inter-domain link status information set from each network domain for the service transmission priority level sequence, and accumulate and calculate the inter-domain service level indicators to obtain an initial service quality matrix; Step S14: Establish a service quality objective function based on the initial service quality matrix, use a neural network adaptive regulator to optimize the parameters of the service quality objective function, and output the end-to-end service quality benchmark data for cross-domain services; 3. The method according to claim 1, wherein The specific steps of Step S2 include: Step S21: Determine the delay / jitter indicators for real-time audio and video data and the packet loss rate indicators for file transfer according to the sensitivity differences of the service type to network performance; Step S22: Construct a service type recognition parameter table according to the network performance sensitive type, classify and label the service data stream features through a deep learning classifier, and obtain the service sensitivity classification result; Step S23: Extract an index monitoring endpoint group from the service sensitivity classification result, and use a support vector machine algorithm to obtain basic network quality data during the collection time period; Step S24: Collect network packet loss data for the basic network quality data to obtain an end-to-end index collection set; Step S25: Construct a performance benchmark curve through the end-to-end index collection set, and obtain a real-time sampling sequence from the data sampling granularity configuration table to obtain a delay jitter reference threshold; Step S26: Establish a sequence of performance monitoring rules based on the delay jitter reference threshold, perform hierarchical statistics on the network quality data using an adaptive evaluator, and output the end-to-end path performance evaluation result; Step S27: Construct a real-time monitoring function for the end-to-end path performance evaluation result, extract the quality degradation warning parameters from the sequence of performance monitoring rules, and generate real-time network performance monitoring data.

4. The method according to claim 1, wherein The specific steps of step S3 include: Step S31: Obtain the original performance data according to the data acquisition time sequence, obtain the performance index reference value through data standardization processing, and extract the time sequence features of the performance index reference value through a long short-term memory network to obtain the performance feature sequence; Step S32: Construct a time window sliding set for the performance feature sequence, classify the time window sliding set through a random forest algorithm and compare it with the trend recognition reference threshold to obtain the trend feature vector; Step S33: Statistically calculate the performance change parameters within the domain according to the trend feature vector, calculate the index correlation weight of the performance change parameters and collect the multi-domain network quality data to obtain a multi-dimensional trend curve; Step S34: Extract the quality degradation features from the multi-dimensional trend curve, match the quality degradation features with the quality attenuation mode sequence in the degradation feature library, and if the match is successful, update the quality degradation warning rule and output the network quality risk identification result.

5. The method according to claim 1, characterized in that The specific steps of step S4 include: Step S41: Perform node encoding on the inter-domain link state data through a graph neural network to obtain an inter-domain connectivity state parameter set, and the inter-domain connectivity state parameter set includes network inter-domain link feature information; Step S42: Perform path analysis on the service traffic data using a gradient boosting tree according to the inter-domain connectivity state parameter set to generate an inter-domain service association graph; Step S43: Construct a propagation path depth matrix for the inter-domain service association graph, and calculate the quality attenuation propagation rate from the propagation path depth matrix; Step S44: Establish a quality impact evaluation function according to the quality attenuation propagation rate, dynamically adjust the quality impact evaluation function using an adaptive predictor, and output the service quality impact prediction result.

6. The method according to claim 1, characterized in that, The specific steps of step S5 include: Step S51: Analyze the intra-domain traffic distribution characteristics through a deep neural network to obtain the network load reference data, and the network load reference data includes the traffic fluctuation period value; Step S52: Classify different service types using a support vector machine according to the network load reference data to generate a service quality grade table, and the service quality grade table contains service priority weight values; Step S53: Calculate the resource allocation reference value for the service quality grade table, collect the inter-domain traffic data to obtain the initial resource allocation plan, and the initial resource allocation plan includes resource guarantee parameters; Step S54: Establish a resource dynamic allocation function according to the initial resource allocation plan, and use an adaptive scheduler to obtain the scheduling instruction from the resource scheduling sequence and then output the resource allocation result, and the resource allocation result includes resource utilization data.

7. The method according to claim 1, characterized in that, The specific steps of step S6 include: Step S61: Classify and label the intra-domain performance parameters through a deep neural network to obtain the performance index classification sequence; Step S62: Extract quality features according to the performance metric classification sequence, and use a support vector machine to perform mapping transformation on the quality features to generate an index mapping relationship table; Step S63: Calculate the in-domain performance benchmark value according to the index mapping relationship table, set quality mapping thresholds for different service types, collect cross-domain performance data, and obtain a multi-domain quality status table; Step S64: Construct a quality evaluation function through the multi-domain quality status table, extract evaluation parameters from the performance benchmark value, perform cumulative statistics on the quality mapping thresholds to generate a quality difference sequence, and construct a unified status view according to the quality difference sequence.

8. The method according to claim 1, characterized in that The specific steps of step S7 include: Step S71: Construct a reinforcement learning state space according to the resource optimization reward value, and perform state evaluation on the network topology through a reinforcement learning agent to obtain a routing optimization parameter set; Step S72: Extract the resource allocation benchmark value from the routing optimization parameter set, and perform multi-dimensional evaluation on the network resources through a dynamic planner to generate a resource allocation plan; Step S73: Construct an optimization execution sequence for the resource allocation plan, collect resource competition data, and obtain an initial optimization instruction; Step S74: Construct an optimization iteration function through the initial optimization instruction, extract performance metrics from the resource competition data, generate an optimization feedback sequence, and output a resource optimization instruction.

9. A real-time configuration device for optical transmission cross-domain services, characterized in that, Comprising: One or more processors; A memory; One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the real-time configuration method for optical transmission cross-domain services according to any one of claims 1 to 8.

10. A computer program product, characterized in that, Comprising computer instructions, the computer instructions instructing the computer device to perform the operations corresponding to the method according to any one of claims 1 to 8.

Images