A method for intelligent response scheduling of a fibre channel network

Abstract

The present application relates to a kind of fiber channel network intelligent response scheduling method, belong to the fault response and adaptive control technical field in vehicle integrated electronic system.Method includes: center management node carries out network fault risk prediction based on the state report frame received in current processing period and historical state information;Determine the multiple candidate optimization strategies of each network node based on the predicted fault risk;The strategy score corresponding to each candidate optimization strategy is calculated, and the global optimal strategy combination is determined based on each candidate optimization strategy, each strategy score and strategy conflict avoidance rule;Strategy instruction chain is determined based on global optimal strategy combination;Strategy instruction chain is issued to carry out global optimization to network.The present application method solves the problem such as single fault response strategy, scheduling mode rigidity, execution mechanism solidification in large-scale fiber channel network, significantly improves the intelligent level of fault control, response accuracy and execution flexibility of vehicle fiber channel network under abnormal situation.

Description

Technical Field

[0001] This invention relates to the field of fault response and adaptive control technology in vehicle-mounted integrated electronic systems, and particularly to an intelligent response scheduling method for fiber optic channel networks. Background Technology

[0002] As in-vehicle electronic systems evolve towards intelligence and integration, network communication systems face higher performance requirements. This is particularly true in applications such as Advanced Driver Assistance Systems (ADAS), in-vehicle distributed control platforms, and mission-critical fusion scheduling. These applications demand not only high-bandwidth, low-latency physical transmission capabilities but also advanced control capabilities such as real-time performance assurance, fault self-diagnosis, and dynamic adaptation. Fibre Channel (FC) networks, due to their high reliability and high throughput, are widely used to construct the core communication framework for such in-vehicle control systems. To improve fault management capabilities, existing technologies typically employ fault diagnosis frames and status reporting frames to provide fault risk prediction, with optimization commands sent by a central management node. However, in practical applications, the in-vehicle network environment is highly dynamic, with link status and node health fluctuating dramatically. Single command triggering and static policy mapping are insufficient to handle the multi-policy coordination requirements in complex scenarios. In particular, there are significant shortcomings in the following aspects: The strategy selection lacks intelligence; most current systems adopt a static mapping method driven by fixed tables, directly matching strategy actions based on health indicator values, making it difficult to reflect the interdependence, priority adjustment, and conflict avoidance capabilities between strategies; The response combination lacks flexibility: the multi-strategy parallel scheduling capability is weak, the system struggles to select the optimal strategy set based on real-time network conditions, and lacks an interpretable combination scoring mechanism; Conflict and redundancy control is insufficient; when multiple sub-instructions are executed collaboratively, there is a lack of a unified conflict avoidance modeling system, which easily leads to strategy overlap, semantic mutual exclusion, or unclear execution logic. Summary of the Invention

[0003] Based on the above analysis, this invention aims to disclose an intelligent response scheduling method for fiber channel networks. By introducing dynamic policy scoring, conflict avoidance graph modeling, and policy instruction chain scheduling, it achieves an intelligent upgrade of fault response control, meeting the higher requirements for response efficiency, accuracy, and system stability in complex network environments.

[0004] The method of the present invention specifically includes the following steps:

[0005] Network failure risk prediction is performed based on the status report frames received in the current processing cycle and historical status information.

[0006] Multiple candidate optimization strategies for each network node are determined based on the predicted failure risk.

[0007] Calculate the policy score corresponding to each of the candidate optimization policies, and determine the global optimal policy combination based on each of the candidate optimization policies, each of the policy scores and the policy conflict avoidance rules;

[0008] The strategy instruction chain is determined based on the globally optimal strategy combination;

[0009] A chain of policy instructions is issued to perform global network optimization.

[0010] Furthermore, the method calculates the policy score corresponding to each candidate optimization policy based on the scoring function of each policy; the method also includes:

[0011] The success rate of each strategy is calculated based on the number of successful executions of each strategy and the total number of executions.

[0012] The average return rate of each strategy is calculated based on the improvement ratio of the corresponding indicators before and after each execution of each strategy.

[0013] The scoring weights and biases of each strategy are updated based on the strategy success rate and the average strategy return rate.

[0014] The scoring function of each strategy is updated based on the scoring weights and biases.

[0015] Furthermore, determining the global network optimization strategy based on the candidate optimization strategy, the strategy score, and the strategy conflict avoidance rule includes:

[0016] Based on the candidate optimization strategies, the strategy scores, and the strategy conflict avoidance rules, multiple candidate strategy combinations are determined for each network node;

[0017] The globally optimal strategy combination is determined based on multiple candidate strategy combinations and strategy conflict avoidance rules.

[0018] Furthermore, the step of determining multiple candidate strategy combinations for each network node based on the candidate optimization strategy, the strategy score, and the strategy conflict avoidance rule includes:

[0019] s1. Based on the candidate optimization strategies and strategy conflict avoidance rules, construct a conflict graph for each network node, represented as follows: Where i represents a network node. Let ε represent the set of candidate optimization strategies. (i) Let be a conflict edge set, representing conflicts between strategies;

[0020] s2. Initialize the candidate subgraph to be empty. Starting with the strategy with the highest score, check whether the strategy conflicts with the strategy in the current candidate subgraph based on the conflict graph. If there is no conflict, add the strategy to the candidate subgraph.

[0021] s3. Traverse all the candidate optimization strategies and repeat s2 to obtain multiple candidate subgraphs;

[0022] s4. Calculate the comprehensive score of each candidate subgraph based on the penalty term;

[0023] s5. Select multiple candidate subgraphs whose comprehensive scores exceed the set threshold to obtain multiple candidate strategy combinations.

[0024] Furthermore, determining the globally optimal strategy combination based on multiple candidate strategy combinations and strategy conflict avoidance rules includes:

[0025] A global conflict graph is constructed based on the candidate strategy combinations and strategy conflict avoidance rules, represented as follows: in ε represents the set of candidate policy combinations for all network nodes, and ε is the set of conflict edges that connect any two candidate policy combinations that cannot be executed in parallel logically or in terms of resources.

[0026] Define the optimization objective and constraints for the 0-1 integer linear programming problem:

[0027] The optimization goal is in, The candidate subgraph represents whether to select the j-th candidate strategy combination for network node i. Let K represent the j-th candidate strategy for network node i, where k is the current processing cycle number. i N represents the number of candidate strategy combinations for network node i, where N is the number of network nodes.

[0028] The constraints are: each network node can select at most one candidate strategy combination; if there is a conflicting edge between two candidate strategy combinations, they cannot be selected at the same time.

[0029] Solve the 0-1 integer linear programming problem based on the optimization objective and constraints to obtain the globally optimal strategy combination.

[0030] Furthermore, the network fault risk prediction based on the status report frames received in the current processing cycle and historical status information includes:

[0031] The status information of the links connected to each network node in the current processing cycle is obtained based on the status report frames received in the current processing cycle and the historical status information; wherein, the historical status information is obtained based on the historical status report frames.

[0032] Network fault risk prediction is performed based on the status information of the current processing cycle.

[0033] Furthermore, the status information includes the link frame loss rate, frame loss rate differential trend value, link round-trip time, weighted moving average of high-priority link round-trip time, trend differential value of high-priority link round-trip time delay, and bit error rate of the connected link.

[0034] The step of obtaining the status information of each network node's connected link corresponding to the current processing cycle based on the status report frame received in the current processing cycle and historical status information includes:

[0035] If a status report frame from a network node is received in the current processing cycle, the status information of the link connected to the corresponding node is obtained from the status report frame.

[0036] If no status report frame from the network node is received in the current processing cycle, then determine whether a status report frame from the same network node was received in the previous processing cycle.

[0037] If so, the estimated value of the state information of the link connected to the network node is calculated based on the state information of the previous two processing cycles of the network node, and used as the state information of the current processing cycle;

[0038] If not, then calculate the estimated value of the state information of the link connected to the node based on the node's historical state information as the state information of the current processing cycle; or, calculate the estimated value of the state information of the link connected to the node based on the node's state information and penalty coefficient in the previous processing cycle as the state information of the current processing cycle; or, set the state information of the node in the current processing cycle to null.

[0039] Furthermore, the central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle. If so, it immediately performs network failure risk prediction and ends the current processing cycle.

[0040] If not, then:

[0041] The activity level of the status report frame is calculated based on the number of network nodes corresponding to the status report frame received in the current processing cycle and the total number of network nodes.

[0042] A new processing cycle is calculated based on the activity of status report frames and the basic refresh cycle, and network fault risk prediction is performed within the current processing cycle.

[0043] Furthermore, the central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle, including:

[0044] The number of high-risk nodes in the network is determined based on the status information of the current processing cycle.

[0045] If the number of high-risk nodes exceeds a set threshold, the network is judged to be in a high-risk state.

[0046] A network node is considered a high-risk node if it meets one of the following conditions:

[0047] The link frame loss rate is greater than or equal to the frame loss rate threshold.

[0048] The cumulative value of the frame drop rate differential trend value is greater than the cumulative threshold of the frame drop rate differential trend within a first set number of consecutive processing cycles;

[0049] The link round-trip delay exceeds the link round-trip differential threshold within a second consecutive set number of processing cycles;

[0050] The round-trip weighted moving average of high-priority links is greater than or equal to the weighted moving average threshold;

[0051] The cumulative value of the high-priority link round-trip delay trend difference is greater than the high-priority trend cumulative threshold within the third consecutive set number of processing cycles;

[0052] The Bayesian probability outlier is greater than or equal to the outlier probability threshold.

[0053] The link health status value is greater than the health status threshold for two consecutive processing cycles.

[0054] Furthermore, the calculation of the new processing cycle based on the activity level of the status report frame and the basic refresh cycle includes:

[0055] T TSI (k')=T min +ρ·(1-α(k'))·T base

[0056] Among them, T min The minimum refresh interval is ρ; ρ is an adjustment factor used to control the magnitude of delay adjustment; T base The base refresh cycle; α(k') represents the activity level of the status report frame. Λ(k') represents the number of network nodes corresponding to the status report frame received in the current processing cycle, and N represents the total number of network nodes.

[0057] The present invention can achieve at least one of the following beneficial effects:

[0058] By asynchronously reporting status report frames from each network node, the central management node obtains status information based on the status report frames received in the current processing cycle, and estimates the missing status information based on the sliding window mechanism, dynamically adjusts the processing cycle, and performs network fault risk prediction. This enables more accurate identification of early signs of faults. Compared with traditional fault detection methods, it can detect potential problems earlier, reduce the impact of network communication failures on vehicle operation, and realize the linkage response of status reporting, risk prediction, and optimized control, thereby improving the timeliness and stability of the vehicle-mounted fiber optic channel network under asynchronous status upload conditions.

[0059] By introducing a dynamic strategy scoring mechanism, conflict avoidance graph modeling, and chain-execution instruction scheduling structure, it is possible to avoid situations such as strategy overlap, semantic mutual exclusion, or unclear execution logic. It can quickly select the optimal strategy combination for each network node, realize multi-strategy scheduling and coordination of the entire network, achieve intelligent upgrade of fault response control, and meet the higher requirements for response efficiency, accuracy, and system stability in complex network environments.

[0060] By employing an adaptive adjustment mechanism where each network node locally senses and drives the reporting cycle based on the status information of the connected links, the reporting cycle can be dynamically adjusted. This allows for the rapid acquisition of more information when the link status deteriorates, and reduces the monitoring load when the link is stable. It guides the rhythm of status report frames to exhibit a non-uniform distribution strongly correlated with link risk, which is beneficial for system resource scheduling. This solves the problem in existing technologies where the frequency cannot be dynamically adjusted based on link health status, resulting in no difference in reporting frequency between "healthy nodes" and "abnormal nodes," wasting network resources and failing to provide more frequent attention to high-risk nodes.

[0061] By employing a "sliding window estimation mechanism," the problem of information loss caused by asynchronous status report frame uploads is resolved, enabling the complete construction of the status matrix even in scenarios where the rhythms of multiple nodes are not aligned. This improves the continuity and timeliness of the overall network status assessment, preventing the system from being "dragged down by the slowest node." The risk prediction process is triggered as soon as a high-risk node uploads a status report frame, calculating the overall network health index value. This maintains the consistency of status information across processing cycles and provides a basis for dynamically adjusting processing cycles.

[0062] By dynamically adjusting the processing cycle based on the activity of status report frames in the current processing cycle, calculating the overall network health index, Bayesian probability outliers and Bayesian probability outlier trend values ​​for each network node, and performing fault risk prediction, this technology breaks away from the passive strategy of "fixed-cycle refresh" in existing technologies. It achieves rhythm-driven adaptive adjustment, enabling high-risk nodes that upload frequently to dominate the system's judgment rhythm, significantly improving system response efficiency. The fault risk prediction cycle is decoupled from but linked to the reporting cycle rhythm of each node, improving the system's adaptability. Furthermore, the dynamic adjustment of the processing cycle is coordinated with the sliding "sliding window estimation mechanism," allowing for accurate fault risk prediction even when some nodes fail to report status report frames, thus improving system stability and security.

[0063] By constructing dedicated fault diagnosis frames, status report frames, and network optimization command frames, the structure exhibits good compatibility with existing FC network standard frame structures, is easy to implement, and is suitable for widespread industrial applications. Through proactive preventative measures, network optimization command frames carrying network optimization instructions are issued based on predicted fault risks to optimize the vehicular network. Based on fault prediction, proactive preventative measures can rationally allocate network resources according to priority, prioritizing the transmission of critical data. This improves the overall operational efficiency and reliability of the vehicular network, avoids vehicle malfunctions caused by network communication failures, reduces vehicle operation and maintenance costs, and more accurately identifies early signs of faults. Compared to traditional fault detection methods, it can detect potential problems earlier and reduce the impact of network communication failures on vehicle operation.

[0064] Other features and advantages of the invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained from what is particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0065] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0066] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0067] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0068] One embodiment of the present invention discloses an intelligent response scheduling method for a Fibre Channel network, specifically including steps S1 to S5. The Fibre Channel network includes a central management node and various network nodes.

[0069] S1. The central management node performs network fault risk prediction based on the status report frames received in the current processing cycle and historical status information. This includes S11 to S13.

[0070] Step S11: The central management node receives status report frames from each network node; the status report frames received by the central management node in the current processing cycle are reported by each network node based on its reporting cycle; the reporting cycle is dynamically determined by each network node based on the status information of the connected links. Step S11 includes steps S11-1 to S11-3.

[0071] S11-1. The network node calculates the status information of the connected link based on the multiple fault diagnosis frames (FDF) sent and the corresponding response frames received in each reporting cycle.

[0072] The status information includes the link frame loss rate, link round-trip time, bit error rate, frame loss rate differential trend value, weighted moving average of high-priority link round-trip time, and high-priority link round-trip delay trend differential value of the connected links.

[0073] Specifically, network nodes calculate the status information of connected links based on one or more fault diagnosis frames (FDFs) sent to other network nodes and the corresponding response frames received during each reporting cycle. The fault diagnosis frame is defined based on the standard FC frame format, including: defining the TYPE field in the frame header as a dedicated type identifier and marking it as a fault diagnosis frame type; defining a priority identifier using a portion of the Parameter field from the standard FC frame; and adding a high-precision timestamp field placed before the payload. The fault diagnosis frame format is shown in Table 1.

[0074] Table 1. Fault Diagnosis Frame Format

[0075]

[0076]

[0077] Furthermore, FDF frames employ an active interaction mode. Network nodes actively send FDF frames to neighboring nodes. Within a reporting cycle, a node can send one or more FDF frames to other network nodes; that is, a reporting cycle includes one or more dynamic FDF sending cycles. A node receiving an FDF frame must immediately return an FDF response frame (ACK frame). The ACK frame has the same frame structure as the original FDF frame, but the ACK frame is marked as a response frame type (0xEF) in the TYPE field and carries the corresponding receive and send timestamps (merged and compressed in the Payload prep field). The active interaction cycle is differentially set according to different service priorities; higher priority links use shorter sending cycles to improve the real-time diagnostics of critical service links.

[0078] Furthermore, the network node calculates the link frame loss rate of the connected link based on the total number of Fault Diagnosis Frames (FDF) sent in the current reporting period and the total number of corresponding ACK frames received; the formula for calculating the link frame loss rate is:

[0079] FLR(k)=1-[N recv (k) / N send (k)];

[0080] Where FLR(k) represents the link frame loss rate in the k-th reporting period; N send (k) represents the total number of FDF frames sent in this period; N recv (k) The total number of ACK frames received in this period;

[0081] The round-trip time of the connected link is calculated based on the time of sending the fault diagnosis frame and the time of receiving the corresponding response frame in the current reporting period. If multiple fault diagnosis frames are sent in the current reporting period, the round-trip time is the average of the round-trip times of all fault diagnosis frames and the corresponding response frames.

[0082] The bit error rate (BER) of the connected link is determined based on the cumulative bit error rate per unit time or unit bit transmission volume within the current reporting period; the bit error rate BER (i) (k) Statistically obtained based on the built-in error counters in the physical layer or MAC layer of the link.

[0083] Furthermore, the frame loss rate differential trend value of the connected links is calculated based on the link frame loss rate of the current reporting period and the previous reporting period.

[0084] The weighted moving average of the high-priority links in the current reporting period is calculated based on the link round-trip delay in the current period and the weighted moving average of the high-priority links in the previous reporting period.

[0085] The round-trip delay trend difference value of the high-priority link is calculated based on the round-trip weighted moving average of the high-priority link in the current reporting period and the previous reporting period.

[0086] Furthermore, the formula for calculating the frame drop rate differential trend value ΔFLR(k) is as follows:

[0087] ΔFLR(k)=FLR(k)-FLR(k-1).

[0088] Furthermore, the formula for calculating the round-trip weighted moving average of high-priority links is as follows:

[0089] RTT EWMA (k)=β·RTT(k)+(1-β)·RTT EWMA (k-1);

[0090] Among them, RTT EWMA (k) is the round-trip weighted moving average of the high-priority link in the k-th reporting period; β is the weighting factor, with a preferred value of 0.6.

[0091] Furthermore, the high-priority link round-trip delay trend difference value ΔRTT EWMA The formula for calculating (k) is:

[0092] ΔRTT EWMA (k)=RTT EWMA (k)-RTT EWMA (k-1).

[0093] S11-2. Each network node calculates the current link health status value based on the current link frame loss rate, link round-trip time, and bit error rate of the connected link.

[0094] Specifically, the calculation method is as follows:

[0095] H (i) (k)=w1·FLR (i) (k)+w2·RTT (i) (k)+w3·BER (i) (k);

[0096] Where, the superscript i represents the network node number; k represents the reporting period number of network node i; H (i) (k) represents the link health status value of network node i during reporting period k; FLR (i) (k), RTT (i) (k), BER (i) (k) represents the link frame loss rate, link round-trip time, and bit error rate of network node i in reporting period k; w1, w2, and w3 are the corresponding weight factors, and the sum of the three is 1.

[0097] Furthermore, the values ​​of weighting factors w1, w2, and w3 can be set according to the link type and criticality. For example, high-priority links involve critical real-time control, and their requirements for link frame loss rate and link round-trip latency are relatively high. The preferred values ​​for w1, w2, and w3 are (0.48, 0.42, 0.1). For high-bandwidth links of high-definition video or sensor data, the bit error rate has a significant impact on service perception. The preferred values ​​for w1, w2, and w3 are (0.3, 0.33, 0.37). For redundant backup links and low-priority auxiliary links, the weights of link round-trip latency and bit error rate can be reduced, and the focus should be on sensitive detection of link frame loss rate. The preferred values ​​for w1, w2, and w3 are (0.64, 0.18, 0.18). For the core switching network, i.e., the distributed mesh topology with symmetrical link structure, the values ​​for w1, w2, and w3 are uniformly (1 / 3, 1 / 3, 1 / 3). The above preferred values ​​can be preset during the system design phase or dynamically adjusted by the central management node according to network traffic characteristics.

[0098] S11-3. Each network node updates its reporting cycle based on the current link health status value and the link status risk threshold.

[0099] Specifically, updating the reporting cycle based on the current link health status value and link status risk threshold includes:

[0100]

[0101] in: T represents the (k+1)th and kth reporting periods, respectively; min Minimum reporting period (e.g., 10ms); T default The default reporting period is 100 sm; ΔT is the shrinking step size during updates; δT is the increasing step size; H th The threshold value for link state risk is 0.18.

[0102] It should be noted that the link state risk threshold H th The design principle is to be sensitive and lightweight, serving as a warning system. Its goal is not to directly trigger network-level optimization, but rather to guide the node on whether to increase monitoring frequency in advance, thereby obtaining more link performance samples before potential failures occur. In actual configurations, the link status risk threshold serves to: amplify the sensitivity to small fluctuations or minor anomalies; rapidly increase the frequency of status report frame transmission, enhancing coverage density; collect multi-cycle statistical data of potentially faulty links in advance; and provide richer input for fault diagnosis at the central management node.

[0103] Furthermore, each network node reports a status report frame based on a reporting period. The frame format of the status report frame in this invention is shown in Table 2:

[0104] Table 2. Status Report Frame Format

[0105]

[0106] This embodiment implements an adaptive adjustment mechanism for each network node to drive the reporting cycle based on the status information of the connected links. This mechanism enables dynamic adjustment of the reporting cycle, allowing for the rapid capture of more information when the link status deteriorates, and reducing the monitoring load when the link is stable. It also guides the rhythm of status report frames to present a non-uniform distribution that is strongly correlated with link risk, which is beneficial for system resource scheduling.

[0107] S12. The central management node obtains the status information of the links connected to each network node in the current processing cycle based on the status report frame and historical status information received in the current processing cycle; wherein, the historical status information is obtained based on the historical status report frame.

[0108] Specifically, it includes:

[0109] If a status report frame from a network node is received in the current processing cycle, the status information of the link connected to the corresponding node is obtained from the status report frame.

[0110] If no status report frame from the network node is received in the current processing cycle, then determine whether a status report frame from the same network node was received in the previous processing cycle.

[0111] If so, the estimated value of the state information of the link connected to the network node is calculated based on the state information of the previous two processing cycles of the network node, and used as the state information of the current processing cycle;

[0112] If not, then calculate the estimated value of the state information of the link connected to the node based on the node's historical state information as the state information of the current processing cycle; or, calculate the estimated value of the state information of the link connected to the node based on the node's state information and penalty coefficient in the previous processing cycle as the state information of the current processing cycle; or, set the state information of the node in the current processing cycle to null.

[0113] Furthermore, if a status report frame from a network node is received in the current processing cycle, the status information of the link connected to the corresponding node is obtained from the status report frame. This status information includes the link frame loss rate, frame loss rate differential trend value, link round-trip time, weighted moving average of high-priority link round-trip time, high-priority link round-trip time delay trend differential value, and bit error rate. It should be noted that if multiple status report frames from the same network node are received in the current processing cycle, the status information is the average of the results reported by each status report frame.

[0114] Furthermore, if no status report frame from the network node is received in the current processing cycle, it is determined whether a status report frame from that network node was received in the previous processing cycle.

[0115] If so, the formula for calculating the estimated values ​​of each state information is:

[0116]

[0117] in, Let γ1 and γ2 be the m-th state information of network node i in the current processing period k', and let γ1 and γ2 be the weight information corresponding to the processing periods k'-1 and k'-2, respectively. γ1 + γ2 = 1, and γ1 > γ2.

[0118] If not, then for ordinary network nodes in the vehicular fiber channel network, the estimated value of the state information of the links connected to the node is calculated based on the node's historical state information. The calculation method is as follows:

[0119] Where, x m,neutral It can be the historical average of the m-th state information, or it can be the system's set value;

[0120] For critical nodes or high-risk nodes in the vehicular fiber channel network (critical nodes can be initially set by the system, while high-risk nodes are determined by the central management node through calculation, the calculation method of which will be disclosed later in this document), the estimated value of the state information of the links connected to the node is calculated based on the state information and penalty coefficient of the node in the previous processing cycle. The calculation formula is as follows:

[0121]

[0122] Where, α penalty It is a penalty factor, and α penalty A value greater than 1 indicates that "continuous missing reports may mean a hidden fault," and therefore the risk indicated by the status information of the corresponding node is amplified.

[0123] For ordinary network nodes in a vehicle-mounted fiber channel network, if the system evaluation mechanism can tolerate the condition of a small number of missing node data, the status information of the node in the current processing cycle can be set to null, thereby ensuring that the status information in the current processing cycle is completely based on the actual reported status report frame and improving the purity of the global evaluation.

[0124] In this embodiment, when no status report frame is received in the current processing cycle, a method is adopted to calculate the estimated value of the status information in the current processing cycle using historical status information. This application refers to this technical means as the "sliding window estimation mechanism" to solve the problem of information loss caused by asynchronous status report frame upload. It can still completely construct the status matrix in the scenario of non-aligned rhythm of multiple nodes; improve the continuity and timeliness of the overall network status assessment, and the system is no longer "dragged down by the slowest node"; and maintain the consistency of status information in each processing cycle.

[0125] S13. Based on the status information of the current processing cycle, perform network fault risk prediction.

[0126] include:

[0127] S13-1. The central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle. If yes, then proceed to step S13-2; otherwise, proceed to steps S13-3 to S13-4.

[0128] S13-2. Immediately perform network failure risk prediction and end the current processing cycle;

[0129] S13-3. The activity level of the status report frame is calculated based on the number of network nodes corresponding to the status report frame received in the current processing cycle and the total number of network nodes.

[0130] S13-4. Calculate a new processing cycle based on the activity level of the status report frame and the basic refresh cycle, and perform network fault risk prediction within the current processing cycle.

[0131] Furthermore, the central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle, including:

[0132] The number of high-risk nodes in the network is determined based on the status information of the current processing cycle.

[0133] If the number of high-risk nodes exceeds a set threshold, the network is judged to be in a high-risk state.

[0134] A network node is considered a high-risk node if it meets one of the following conditions:

[0135] The link frame loss rate FLR(k') is greater than or equal to the frame loss rate threshold FLR. threshold This is expressed as FLR(k')≥FLR threshold ;

[0136] Frame Drop Rate Differential Trend Value ΔFLR threshold If the cumulative value exceeds the cumulative threshold of the frame drop rate differential trend within a consecutive first set number of processing cycles (typically 3-5), it is expressed as: M is the first set number;

[0137] The link round-trip time (RTT)(k') exceeds the link round-trip differential threshold (RTT) within a second consecutive set number of processing cycles (preferably 3). threshold (k'), where RTT threshold The calculation method for (k') is: RTT threshold (k')=μ RTT (k')+α·σ RTT (k'), μ RTT (k') is the historical statistical mean. σ RTT (k') represents the historical standard deviation. W is the historical window length, which refers to the number of cycles before the k'th processing cycle, typically 20 processing cycles. α represents the differential sensitivity factor, with a preferred value of 3. The round-trip weighted moving average of high-priority links is greater than or equal to the weighted moving average threshold.

[0138] High-priority link round-trip weighted moving average RTT EWMA (k') is greater than or equal to the weighted moving average threshold RTT EWMA,threshold , represented as RTT EWMA (k)≥RTT EWMA,threshold ;

[0139] High-priority link round-trip delay trend difference ΔRTT EWMA(k') The cumulative value within the third consecutive set number (preferred value is 3) of the processing cycle is greater than the high priority trend cumulative threshold ΔRTT. EWMA,threshold , represented as Q is the third consecutive set number;

[0140] Bayesian probability outliers Greater than or equal to the abnormal probability set threshold P th , Let be the Bayesian probability outlier of network node i;

[0141] The link health status value is greater than the health status threshold H for two consecutive processing cycles. th .

[0142] Furthermore, the methods for calculating the Bayesian probability outliers of each network node include:

[0143]

[0144] Where X is a feature vector consisting of the calculated node’s link frame loss rate, link round-trip time, weighted moving average of high-priority link round-trip time, overall network health index, bit error rate, node CPU load, and node temperature.

[0145] P(fault) is the prior probability, which represents the estimated probability that the central management node is in a faulty state for the network node or link before observing the current data X. It is assigned a value based on historical statistics, empirical data or failure rate estimates in system design.

[0146] It should be noted that when calculating the probability of observed current data X, it refers to jointly calculating the probability distribution model of each indicator value in X, that is, calculating the joint probability of the observed feature combination under fault conditions. Even if the conditional probability of each feature may be small, their joint probability can effectively capture the overall fault risk through Gaussian distribution. For example, suppose a link connected to a network node has a minor performance problem, but can still send and receive data. In this case, the link round-trip latency and link frame loss rate may still be within the normal range. However, if combined with other features (such as the overall network health index value), the combination of the entire feature vector X may still effectively capture the trend of performance degradation, thus giving a higher "fault probability" assessment in Bayesian inference.

[0147] P(X|fault) represents the probability of observing the current data X when a network node or link is actually in a fault state; the value is determined based on online learning or adaptive updates of the central management node.

[0148] In the numerator of the above equation, P(fault) and P(X|fault) are multiplied together to obtain the weighted probability of observing X under the fault state. That is, assuming that the current node is really in a fault state, and given the degree of prior belief in the fault state itself, the overall probability of observing X is as follows.

[0149] P(normal) represents the estimated probability that the network node or link will remain normal when the current data X is not observed;

[0150] P(X|normal) represents the probability of observing X when a network node or link is normal. For example, if the value of P(X|normal) is large, it means that "even if it looks like a sign of failure, it may often occur under normal conditions", so the final probability of failure will be reduced. The value is determined based on the online learning or adaptive update of the central management node.

[0151] The denominator of the above formula means that all possible sources of X have been observed, which guarantees that the final result is a normalized probability value in the interval [0,1].

[0152] Furthermore, the calculation method for the overall network health index is as follows:

[0153]

[0154] Where TSI(k') represents the overall network health index value in the k'th processing cycle; N represents the number of network nodes; i and j represent the i-th and j-th network nodes, respectively; l ij (k') represents the link state of link (i,j); p ij This represents the priority weight of link (i,j).

[0155] Furthermore, the central management node determines the link status l based on the status information of each stage of the current processing cycle. ij (k'), including:

[0156] A link is considered abnormal if it meets at least one of the following conditions; otherwise, the link is considered normal. These conditions include:

[0157] The link frame loss rate is greater than or equal to the frame loss rate threshold, such as 5% for example;

[0158] The link round-trip delay exceeds the delay threshold for N processing cycles; N is a positive integer, preferably 3, and the delay threshold is preferably three times the standard deviation of the historical delay mean; it should be noted that within a reporting cycle of a network node, regardless of whether the node sends one or more FDF frames, as long as the link round-trip delay exceeds the delay threshold once, it is recorded as the link round-trip delay exceeding the delay threshold within that reporting cycle.

[0159] If a fault diagnosis frame sent through this link fails to receive a response frame twice within a specified time, specifically, if a fault diagnosis frame sent through this link fails to receive a response frame within a specified time, it is characterized by ACK_Missed_Counts and stored in the link status field of the status reporting frame.

[0160] Within the period of sending two fault diagnosis frames and receiving the corresponding response frames through the link, the bit error rate exceeds a preset multiple of the current link benchmark. For example, the preset multiple can be 100%.

[0161] It should be noted that when the central management node determines that the network is in a high-risk state based on the status information of the current processing cycle, it needs to recalculate the Bayesian probability outlier value and the Bayesian probability outlier trend value of each network node when immediately performing network fault risk prediction.

[0162] Specifically, in step S13-1, network failure risk prediction includes:

[0163] The central management node calculates the corresponding overall network health index value, Bayesian probability outlier value, and Bayesian probability outlier trend value of each network node based on the status information of the current processing cycle.

[0164] Network failure risk is predicted based on various state information, the overall network health index, the Bayesian probability outliers of each network node, and the Bayesian probability outlier trend values.

[0165] Furthermore, the Bayesian probability anomaly trend value of the network node in the k-th period is calculated using the following formula:

[0166] ΔP(fault|X) k =P(fault|X) k -P(fault|X) k-1 ;

[0167] It should be noted that ΔP(fault|X) k The trend monitoring indicates that if the posterior probability continues to increase and approaches the threshold, it means that the risk of node failure is very likely to erupt in the short term.

[0168] It should be noted that since this invention is applicable to network scenarios with a large number of nodes, there are many predicted network failure risks and many corresponding triggering conditions. They will not be listed here, but will be presented in tabular form in step S2.

[0169] Specifically, in step S13-2, the formula for calculating the activity level of the status report frame is as follows: Where Λ(k') is the number of network nodes corresponding to the status report frame received in the current processing cycle, and N is the total number of network nodes.

[0170] Specifically, in step S13-3, the new processing cycle is calculated based on the activity level of the status report frame and the basic refresh cycle, including:

[0171] T TSI (k')=T min +ρ·(1-α(k'))·T base ;

[0172] Among them, T min ρ is the minimum refresh interval (e.g., 10ms); ρ is the adjustment factor (e.g., 0.8), used to control the delay adjustment magnitude; T base Based on the refresh cycle (e.g., 100ms);

[0173] The calculation formula satisfies:

[0174] If α(k')≈1 (most nodes have reported status report frames), then T TSI (k)→T min Refresh quickly;

[0175] If α(k') is low, it indicates insufficient state information. Therefore, the refresh cycle should be appropriately delayed to avoid invalid judgments.

[0176] In this embodiment, the central management node dynamically adjusts the processing cycle based on the activity level of the status report frames in the current processing cycle, calculates the overall network health index, the Bayesian probability outlier value and the Bayesian probability outlier trend value of each network node, and performs fault risk prediction. This breaks away from the passive strategy of "fixed-cycle refresh" in the prior art, realizing rhythm-driven adaptive adjustment. This allows high-risk nodes that upload frequently to dominate the system's judgment rhythm, significantly improving system response efficiency. The fault risk prediction cycle is decoupled from the reporting cycle rhythm of each node but is adjusted in conjunction with it, improving the system's adaptability. Furthermore, the dynamic adjustment of the processing cycle is coordinated with the sliding "sliding window estimation mechanism," enabling accurate fault risk prediction even when some nodes do not report status report frames, thus improving the system's stability and security.

[0177] S2. Based on the predicted failure risk, determine multiple candidate optimization strategies for each network node.

[0178] For example, as shown in Table 3, the correspondence between predicted failure risks and candidate optimization strategies is as follows:

[0179] Table 3 shows an example of the correspondence between predicted failure risks and candidate optimization strategies.

[0180]

[0181]

[0182] It should be noted that, due to the complexity of large-scale networks in practical applications, the failure risks and corresponding candidate optimization strategies will not be listed here.

[0183] Furthermore, the candidate optimization strategies of each network node will be triggered by the task scheduling chain. For example, after the link test task is completed, the link hard recovery strategy will be triggered; if the task scheduling process requires redundant bandwidth, the backup link strategy will be triggered.

[0184] Furthermore, there are collaborative triggering relationships among the candidate optimization strategies. For example, if the implementation of the topology reconstruction strategy fails, the topology rollback strategy will be automatically triggered. Before enabling the backup node strategy, the node state sealing strategy needs to be implemented first.

[0185] As can be seen from the above description, multiple candidate optimization strategies for each network node are determined based on the predicted failure risk. However, the candidate optimization strategies for each network node may also include candidate optimization strategies triggered by task scheduling chains or cooperative triggering relationships.

[0186] Table 4 illustrates multiple candidate optimization strategies corresponding to the predicted fault risks, task scheduling chains, and collaborative triggering relationships of strategies for each network node. It should be noted that in practical applications, networks with large node scales are highly complex, and the strategies corresponding to each risk can be dynamically formulated and adjusted based on the actual network conditions.

[0187] Table 4. Example of the correspondence between fault risk triggering conditions and candidate optimization strategies.

[0188]

[0189]

[0190]

[0191] In this embodiment, by taking proactive preventative measures and determining candidate optimization strategies based on predicted fault risks, early signs of faults can be identified more accurately. Compared with traditional fault detection methods, potential problems can be discovered earlier, reducing the impact of network communication failures on vehicle operation.

[0192] S3. Calculate the policy score corresponding to each candidate optimization policy, and determine the globally optimal policy combination based on each candidate optimization policy, each policy score, and the policy conflict avoidance rule. Specifically, this includes S31 to S33.

[0193] S31. Calculate the strategy score corresponding to each candidate optimization strategy based on the scoring function of each strategy.

[0194] Specifically, the scoring function is expressed as:

[0195]

[0196] in, For the j-th candidate strategy s of network node i j Scoring in cycle k'; For candidate strategies s j The importance weights for feature m; d represents the number of features; For vector Y (i) The m-th feature of (k'); Y (i) (k') is a feature vector consisting of the link frame loss rate, link round-trip time, weighted moving average of high-priority links, bit error rate, node CPU load, node temperature, overall network health index, and Bayesian probability outliers for the processing period k'; f m (·) is the computation function corresponding to feature m; σ(·) is the normalization function; For candidate strategies s j The bias term.

[0197] Furthermore, f m The specific forms of (·) are shown in Table 5:

[0198] Table 5. Calculation function f corresponding to feature m m (·)

[0199]

[0200] Furthermore, the normalization function σ(·) is the sigmoid function, i.e. The sigmoid function implementation prevents the output from becoming too extreme due to an extremely large or small input.

[0201] Furthermore, The initial value is determined based on historical experience, and the value can also be optimized through feedback and learning during strategy execution (see step S6 of this application).

[0202] For example, preferred values ​​for some measurements are shown in Table 6:

[0203] Table 6. Examples of Optimal Strategy Values

[0204]

[0205]

[0206] S32. Based on the candidate optimization strategies, the strategy scores, and the strategy conflict avoidance rules, determine multiple candidate strategy combinations for each network node. Specifically, this includes s1 to s5.

[0207] s1. Based on the candidate optimization strategies and strategy conflict avoidance rules, construct a conflict graph for each network node, denoted as G. (i) =(s (i) ,ε (i) ), where i represents a network node, s (i) Let ε represent the set of candidate optimization strategies. ( i ) Let be a conflict edge set, representing conflicts between strategies;

[0208] s2. Initialize the candidate subgraph to be empty. Starting with the strategy with the highest score, check whether the strategy conflicts with the strategy in the current candidate subgraph based on the conflict graph. If there is no conflict, add the strategy to the candidate subgraph.

[0209] s3. Traverse all the candidate optimization strategies and repeat s2 to obtain multiple candidate subgraphs;

[0210] s4. Calculate the comprehensive score of each candidate subgraph based on the penalty term. The calculation method is as follows:

[0211] in, Candidate subgraph for network node i In processing the overall score of period k', This is the strategy in the candidate subgraph. The score for the processing cycle k' of network node i. Let γ be the number of strategies in the candidate subgraph, and let γ be the penalty intensity for controlling the number of strategies, where γ > 0;

[0212] s5. Since each candidate subgraph corresponds to a candidate strategy combination, multiple candidate subgraphs with a combined score exceeding a set threshold are selected to obtain multiple candidate strategy combinations, denoted as follows: K i This represents the number of candidate strategy combinations for network node i.

[0213] For example, Table 7 shows some typical strategy conflict avoidance rules:

[0214] Table 7 Examples of Strategy Conflict Avoidance Rules

[0215]

[0216]

[0217] Furthermore, in practical applications, it can be automatically expanded and generated based on the system's strategy set and task graph.

[0218] Furthermore, the multiple candidate policy combinations (multiple candidate subgraphs) obtained for each network node in step S32 satisfy the constraints of no conflict edges between nodes, including at least one policy, and being as concise as possible (the size of the candidate policy combination is controlled by a penalty term).

[0219] In this embodiment, by introducing a strategy scoring mechanism and conflict avoidance graph modeling, the optimal strategy combination for each network node can be quickly selected, laying the foundation for determining the globally optimal strategy combination.

[0220] S33. Determine the globally optimal strategy combination based on multiple candidate strategy combinations and strategy conflict avoidance rules.

[0221] A global conflict graph is constructed based on the candidate strategy combinations and strategy conflict avoidance rules, represented as follows: in ε represents the set of candidate policy combinations for all network nodes, and ε is the set of conflict edges that connect any two candidate policy combinations that cannot be executed in parallel logically or in terms of resources.

[0222] Define the optimization objective and constraints for the 0-1 integer linear programming problem:

[0223] The optimization goal is in, The candidate subgraph represents whether to select the j-th candidate strategy combination for network node i. Let K represent the j-th candidate strategy for network node i, where k is the current processing cycle number. i N represents the number of candidate strategy combinations for network node i, where N is the number of network nodes.

[0224] The constraint is that each network node can select at most one candidate strategy combination (candidate subgraph), denoted as: If there are conflicting edges between two candidate strategy combinations (candidate subgraphs), they cannot be selected simultaneously, as shown below.

[0225] Solving the 0-1 integer linear programming problem based on the optimization objective and constraints yields the globally optimal strategy combination, ensuring the lowest execution cost and the least conflict risk.

[0226] This embodiment, by using a policy scoring mechanism and conflict avoidance graph modeling, can avoid situations such as policy overlap, semantic mutual exclusion, or unclear execution logic, and realize multi-policy scheduling and coordination across the entire network, achieving intelligent upgrades in fault response control. For network scenarios with a large number of nodes, it can especially meet the higher requirements for response efficiency, accuracy, and system stability in complex network environments. For example, the large-scale network generally includes a network scale of no less than 30 nodes and 6 switches.

[0227] S4. Determine the policy instruction chain based on the globally optimal policy combination.

[0228] The globally optimal strategy combination can be represented as:

[0229] in The optimal candidate strategy combination selected for network node i.

[0230] Furthermore, the globally optimal strategy combination is orchestrated into a strategy instruction chain, represented as:

[0231]

[0232] in, For the strategy instruction chain, the subscript L indicates the number of instructions, s indicates the instruction corresponding to the strategy, and θ indicates the corresponding execution control parameter group.

[0233] Furthermore, the execution control parameters include five categories: triggering conditions, execution delay, rollback conditions, execution window, and priority flags, as shown in the following examples:

[0234] (1) Triggering conditions: such as "RTT exceeding the threshold ratio reaches 50% or more";

[0235] (2) Execution delay: such as "delay for 1 cycle before restarting";

[0236] (3) Rollback conditions: such as "rollback if bit error rate does not improve";

[0237] (4) Execution window: such as "Executable only during the XX-XX time period";

[0238] (5) Priority flag: such as "execute with high priority" or "can be interrupted by high priority tasks".

[0239] S5. Issue a chain of policy instructions to perform global network optimization.

[0240] Specifically, the central management node issues a chain of policy instructions to optimize the network globally.

[0241] Furthermore, in implementation, to achieve chained strategy scheduling control, a scheduling state machine (FSM) is introduced on the network's end nodes, with the following state set:

[0242] PENDING: Waiting for the trigger condition to be met;

[0243] ACTIVE: The current policy is being executed;

[0244] SUCCESS: The strategy is completed and the goal is achieved;

[0245] FAIL: Policy execution failed, triggering a rollback;

[0246] ROLLBACK: Cancels the preceding strategy according to the set conditions;

[0247] SKIPPED: The strategy was skipped due to failure of the preceding steps.

[0248] Through this state mechanism, the policy instruction chain can dynamically jump to different execution flows, supporting "condition triggering", "stage control", "failure handling" and "policy rollback".

[0249] Furthermore, during implementation, policy instructions are issued through network optimization instruction frames, as shown in Table 8. A policy instruction chain consists of multiple consecutively sent network optimization instruction frames.

[0250] Table 8 Network Optimization Command Frame Format

[0251]

[0252] The instruction parameters are extended fields, including:

[0253] field name illustrate Chain_ID The strategy chain number uniquely identifies a chain scheduling task. Step_Index The current step number is used for phased execution control. Strategy_ID Atomic strategy numbering Trigger_Cond Triggering conditions (monitoring indicators such as link round-trip latency) Rollback_Cond Rollback criteria Delay_Cycle Delay period Priority_Flag Execution priority or interruption permission Exec_Status The node's current execution status is reported (returned by SRF).

[0254] The following is an example of the scheduling and execution process:

[0255] Suppose the system selects the following strategy combination for network node i:

[0256] (i) ={s1: Activate backup link, s2: Softly isolate primary link, s3: Link load transfer};

[0257] The scheduling structure is as follows:

[0258] step Strategy Triggering conditions Delay Rollback conditions 1 Enable backup link The current main link round-trip latency exceeds the limit by 50%. 0 processing cycle Backup link disconnected 2 Soft-isolated main link Backup link is stable online for 1 processing cycle 1 processing cycle The main link was not restored. 3 Link load balancing Backup link bit error rate is below threshold 1 processing cycle System load imbalance

[0259] The policy chain is encapsulated in a chained network optimization instruction frame, which is sequentially scheduled and issued by the central management node controller, and then parsed and executed by the network nodes through their local state machines, providing state feedback.

[0260] In this embodiment, by adopting a chain-execution instruction scheduling structure, the efficiency and execution flexibility of the central management node in intelligent network control are improved.

[0261] Furthermore, in one specific embodiment of the present invention, step S6 is also included.

[0262] S6. Strategy execution feedback and learning, specifically including:

[0263] The success rate of each strategy is calculated based on the number of successful executions of each strategy and the total number of executions.

[0264] The average return rate of each strategy is calculated based on the improvement ratio of the corresponding indicators before and after each execution of each strategy.

[0265] The scoring weights and biases of each strategy are updated based on the strategy success rate and the average strategy return rate.

[0266] The scoring function of each strategy is updated based on the scoring weights and biases.

[0267] Furthermore, the success rate of the strategy is calculated as follows:

[0268]

[0269] in, Let be the success rate of strategy s for network node i within the specified time period. For the number of successes, This represents the total number of times the strategy was executed.

[0270] Furthermore, the average strategy return is calculated by averaging the returns of multiple individual strategies. For example, the calculation method for the individual strategy return of the overall network health index is as follows:

[0271]

[0272] in, To examine the single-strategy return of strategy s of network node i with respect to the overall network health index during the time period under consideration, TSI before and TSI after These are the overall network health index values ​​before and after execution, respectively.

[0273] Furthermore, the scoring weights and biases for each strategy are updated based on strategy success rate and average strategy return, including:

[0274] Scoring weight:

[0275] Bias term:

[0276] in, and These represent the strategies s for the (T+1)th and Tth time periods, respectively. j The importance weights for feature m, and η is the learning rate (usually between 0.001 and 0.01), used to control the update magnitude. and Let s represent the strategies for the Tth time period. j The strategy success rate and average strategy return rate, The calculation function representing the T-th observation time period corresponding to feature m is calculated as follows: based on the calculation function described in step S31. Calculate the average of the results for each treatment cycle during the T-th observation period;

[0277] and These represent the strategies s for the (T+1)th and Tth time periods, respectively. j The bias term, η b The learning rate is the bias term; it can be negative for strategies that fail frequently.

[0278] Furthermore, a periodic policy feedback and learning process is set up, and within each period, the scoring function of each policy is updated based on the scoring weights and bias terms. For example, a scoring model for policy execution feedback and learning can be constructed in a simulation environment or an offline replica, using the minimum mean squared error as the loss function for gradient updates.

[0279] This embodiment, through a dynamic scoring mechanism, meets the needs of automatically evaluating policy adaptability, avoiding policy conflicts, and constructing optimal policy combinations in complex network environments, and has strong adaptability and robustness.

[0280] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent response scheduling in fiber optic channel networks, characterized in that, Includes the following steps: Network failure risk prediction is performed based on the status report frames received in the current processing cycle and historical status information. Multiple candidate optimization strategies for each network node are determined based on the predicted failure risk. Calculate the policy score corresponding to each of the candidate optimization policies, and determine the global optimal policy combination based on each of the candidate optimization policies, each of the policy scores and the policy conflict avoidance rules; The step of determining the globally optimal strategy combination based on each candidate optimization strategy, each strategy score, and strategy conflict avoidance rule includes: determining multiple candidate strategy combinations for each network node based on each candidate optimization strategy, each strategy score, and strategy conflict avoidance rule; and determining the globally optimal strategy combination based on the multiple candidate strategy combinations and strategy conflict avoidance rule. Specifically, determining multiple candidate strategy combinations for each network node based on each candidate optimization strategy, each strategy score, and strategy conflict avoidance rule includes: s1, constructing a conflict graph for each network node based on each candidate optimization strategy and strategy conflict avoidance rule, represented as... ,in Represents a network node. Represents the set of candidate optimization strategies. s2, Initialize the candidate subgraph to empty, starting with the strategy with the highest score, check whether this strategy conflicts with the strategy in the current candidate subgraph based on the conflict graph. If there is no conflict, add the strategy to the candidate subgraph; s3, Traverse all the candidate optimization strategies, repeat s2 to obtain multiple candidate subgraphs; s4, Calculate the comprehensive score of each candidate subgraph based on the penalty term; s5, Select multiple candidate subgraphs with comprehensive scores exceeding a set threshold to obtain multiple candidate strategy combinations; The step of determining the globally optimal strategy combination based on multiple candidate strategy combinations and strategy conflict avoidance rules includes: constructing a global conflict graph based on the candidate strategy combinations and strategy conflict avoidance rules, represented as... ,in This represents the set of candidate policy combinations for all network nodes. Let the set of conflicting edges connect any two pairs of candidate strategies that are logically or resource-incompatible for parallel execution; define the optimization objective and constraints of a 0-1 integer linear programming problem: the optimization objective is... ,in, Indicates whether to select a network node. The The combination of candidate strategies constitutes a candidate subgraph. Represents network nodes The One candidate strategy, This is the current processing cycle number. Represents network nodes The number of candidate strategy combinations Let be the number of network nodes; the constraints are: each network node can select at most one candidate strategy combination; if there is a conflicting edge between two candidate strategy combinations, they cannot be selected at the same time; based on the optimization objective and constraints, solve the 0-1 integer linear programming problem to obtain the globally optimal strategy combination; Determine the policy instruction chain based on the globally optimal policy combination; A chain of policy instructions is issued to perform global network optimization.

2. The intelligent response scheduling method according to claim 1, characterized in that, The method further includes calculating the strategy score corresponding to each candidate optimization strategy based on the scoring function of each strategy; the method also includes: The success rate of each strategy is calculated based on the number of successful executions of each strategy and the total number of executions. The average return rate of each strategy is calculated based on the improvement ratio of the corresponding indicators before and after each execution of each strategy. The scoring weights and biases of each strategy are updated based on the strategy success rate and the average return of the strategy. The scoring function of each strategy is updated based on the scoring weights and biases.

3. The intelligent response scheduling method for fiber optic channel networks according to any one of claims 1-2, characterized in that, The network fault risk prediction based on the status report frames received in the current processing cycle and historical status information includes: The status information of the links connected to each network node in the current processing cycle is obtained based on the status report frames received in the current processing cycle and the historical status information; wherein, the historical status information is obtained based on the historical status report frames. Network fault risk prediction is performed based on the status information of the current processing cycle.

4. The intelligent response scheduling method for fiber optic channel networks according to claim 3, characterized in that, The status information includes the link frame loss rate, frame loss rate differential trend value, link round-trip time, weighted moving average of high-priority link round-trip time, trend differential value of high-priority link round-trip time, and bit error rate of the connected link. The step of obtaining the status information of each network node's connected link corresponding to the current processing cycle based on the status report frame received in the current processing cycle and historical status information includes: If a status report frame from a network node is received in the current processing cycle, the status information of the link connected to the corresponding node is obtained from the status report frame. If no status report frame from the network node is received in the current processing cycle, then determine whether a status report frame from the same network node was received in the previous processing cycle. If so, the estimated value of the state information of the link connected to the network node is calculated based on the state information of the previous two processing cycles of the network node, and used as the state information of the current processing cycle; If not, then calculate the estimated value of the state information of the link connected to the node based on the node's historical state information as the state information of the current processing cycle; or, calculate the estimated value of the state information of the link connected to the node based on the node's state information and penalty coefficient in the previous processing cycle as the state information of the current processing cycle; or, set the state information of the node in the current processing cycle to null.

5. The intelligent response scheduling method for fiber optic channel networks according to claim 4, characterized in that, The central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle. If so, it immediately performs network failure risk prediction and ends the current processing cycle. If not, then: The activity level of the status report frame is calculated based on the number of network nodes corresponding to the status report frame received in the current processing cycle and the total number of network nodes. A new processing cycle is calculated based on the activity of status report frames and the basic refresh cycle, and network fault risk prediction is performed within the current processing cycle.

6. The intelligent response scheduling method for fiber optic channel networks according to claim 5, characterized in that, The central management node determines whether the network is in a high-risk state based on the status information of the current processing cycle, including: The number of high-risk nodes in the network is determined based on the status information of the current processing cycle. If the number of high-risk nodes exceeds a set threshold, the network is judged to be in a high-risk state. A network node is considered a high-risk node if it meets one of the following conditions: The link frame loss rate is greater than or equal to the frame loss rate threshold; The cumulative value of the frame drop rate differential trend value is greater than the cumulative threshold of the frame drop rate differential trend within a first set number of consecutive processing cycles; The link round-trip delay exceeds the link round-trip differential threshold within a second consecutive set number of processing cycles; The round-trip weighted moving average of high-priority links is greater than or equal to the weighted moving average threshold; The cumulative value of the high-priority link round-trip delay trend difference is greater than the high-priority trend cumulative threshold within the third consecutive set number of processing cycles; The Bayesian probability outlier is greater than or equal to the outlier probability threshold. The link health status value is greater than the health status threshold for two consecutive processing cycles.

7. The intelligent response scheduling method for fiber optic channel networks according to claim 6, characterized in that, The new processing cycle calculated based on the activity level of the status report frame and the basic refresh cycle includes: ； in, Minimum refresh interval; This is an adjustment factor used to control the magnitude of the delay adjustment; Based on the refresh cycle; For status report frame activity, , This represents the number of network nodes corresponding to the status report frames received in the current processing cycle. This represents the total number of network nodes.

Images