An intelligent decision method for a dual-active switching management and control platform

Abstract

The application discloses an intelligent decision method for a dual-active switching management and control platform, which comprises the following steps: acquiring dual-active switching decision original data, multi-source data preprocessing, dual-active master node health assessment, self-adaptive dynamic intervention judgment mechanism, selecting an optimal switching target and intelligent switching execution. The application relates to the technical field of data information processing, and specifically relates to an intelligent decision method for a dual-active switching management and control platform. The application innovatively proposes a self-adaptive dynamic intervention judgment mechanism and a design problem ability adaptation model, realizes rapid perception and accurate judgment of the abnormal state of a master node, designs a channel-level self-adaptive abnormal threshold mechanism and introduces an expansion convolution structure, enhances the practicability of the model in an intelligent decision scene, adopts a dynamic inertia weight updating mechanism and a position disturbance mechanism to improve a particle swarm optimization algorithm, acquires dynamic master node switching trigger judgment parameters, and realizes accurate identification of a reasonable timing of master node switching.

Description

Technical Field

[0001] The present invention relates to the field of data information processing technology, and in particular to an intelligent decision-making method for an active-active switching management and control platform. Background Art

[0002] An intelligent decision-making method for an active-active switching management and control platform refers to a means of automating and intelligently judging and making decisions on whether to perform active-active switching and how to perform active-active switching in the context of an active-active data center, using artificial intelligence technology, data analysis algorithms, and predefined rule strategies, to ensure the continuity, stability, and data consistency of the business system in abnormal scenarios.

[0003] However, the intelligent decision-making method of the traditional active-active switching management and control platform has technical problems such as reliance on fixed switching thresholds, inaccurate switching timing, and rough selection of backup node targets, which can easily lead to misjudgment and business interruption and data loss after switching; the existing node health assessment model has technical problems such as susceptibility to noise interference and insufficient sensitivity to trend changes, which leads to low accuracy of node health assessment results and unstable abnormality judgment; the existing optimization algorithm for obtaining switching judgment parameters has the problem of falling into local optimal solutions, resulting in inaccurate switching judgment parameters. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides an intelligent decision-making method for a dual-active switching management and control platform. In view of the technical problems in the intelligent decision-making method of the traditional dual-active switching management and control platform, such as reliance on fixed switching thresholds, inaccurate switching timing, and rough selection of backup node targets, which easily lead to misjudgment and service interruption and data loss after switching, this solution innovatively proposes an adaptive dynamic intervention judgment mechanism and a design problem capability adaptation model; among them, the adaptive dynamic intervention judgment mechanism has a high degree of adaptability, breaking through the single-point triggering limitation of the traditional static threshold, and performing multi-parameter joint judgment, which significantly improves the system's response sensitivity and judgment accuracy to different fault modes; The problem capability adaptation model introduces a master-slave matching strategy, which can make context-aware adaptation selections based on the anomaly type of the master node and the takeover capability characteristics of the backup node, avoiding the traditional rough judgment based on static priority, and significantly improving the accuracy of backup node selection and business continuity assurance capabilities; it realizes rapid perception, accurate judgment and high-availability switching of the abnormal state of the master node under the platform architecture, ensuring that the business system can continue to run stably and continuously when a fault occurs, avoiding data loss and service interruption; in response to the technical problems of existing node health assessment models being susceptible to noise interference and insufficient sensitivity to trend changes, which leads to low accuracy of node health assessment results and unstable anomaly judgment, this solution is innovatively designed The introduction of channel-level adaptive anomaly threshold mechanism and dilated convolution structure; among them, the channel-level adaptive anomaly threshold mechanism dynamically calculates the anomaly boundary value of each operating indicator channel, combines the indicator fluctuation intensity and channel uncertainty factors, realizes the accurate distinction between abnormal signals and background noise, and obtains key state characteristics; and the dilated convolution structure captures the trend changes of operating indicators over time by introducing multi-scale receptive fields and residual connections, which significantly improves the model's ability to identify chronic degeneration-type anomalies; this solution improves the model's dual perception ability of sudden anomalies and trend changes, enhances the stability of health scores in dynamic and complex environments, and effectively avoids the situation where abnormal jitters and trend changes are misjudged as health anomalies. The proposed scheme uses a dynamic inertia weight update mechanism and a position perturbation mechanism to improve the particle swarm optimization algorithm, which improves the global search capability and the ability to jump out of local extreme values ​​of the algorithm, effectively enhances the optimization accuracy and stability of the switching trigger judgment parameters, and ensures that the final judgment parameter group has higher environmental adaptability and strategy matching, and finally realizes the accurate identification and dynamic control of the reasonable timing of the master node switching.

[0005] The technical solutions adopted by the present application are as follows: The present application provides an intelligent decision-making method for a dual-active switching management and control platform, which comprises the following steps:

[0006] Step S1: obtaining dual-active switching decision-making original data;

[0007] Step S2: multi-source data preprocessing;

[0008] Step S3: dual-active master node health assessment;

[0009] Step S4: self-adaptive dynamic intervention judgment mechanism;

[0010] Step S5: selecting an optimal switching target;

[0011] Step S6: intelligent switching execution.

[0012] Further, in step S1, the dual-active switching decision-making original data is obtained by collecting the running state data of the master node and the plurality of backup nodes in the platform through the dual-active switching management and control platform, and the dual-active switching decision-making original data is obtained; the dual-active switching decision-making original data includes historical switching decision-making data and real-time switching decision-making data; the historical switching decision-making data and the real-time switching decision-making data both include system resource usage data, network state data and service layer running state data of each node; the historical switching decision-making data further includes historical master node health assessment results, historical backup node switching data and historical switching success data.

[0013] Further, in step S2, the multi-source data preprocessing is used for data preprocessing of the dual-active switching decision-making original data, specifically for data integrity detection, outlier detection and elimination, multi-source data time alignment and multi-source data standardization processing, to obtain dual-active switching decision-making optimization data; comprising the following steps:

[0014] Step S21: data integrity detection, used for field level and time dimension integrity detection of original data, specifically including field integrity detection and time alignment integrity detection;

[0015] Step S22: outlier detection and elimination, used for identifying and removing extreme values, error sampling or abnormal fluctuation data that may exist in the data set, specifically using Z-score anomaly detection method to calculate the standard deviation of each field, identifying the data points deviating from the mean value by more than a set multiple threshold, and performing the elimination operation;

[0016] Step S23: Multi-source data time alignment is used to perform formal alignment and format unification operations on data from different sources. Specifically, it constructs a unified reference time axis, performs downsampling window averaging processing on data with inconsistent sampling frequencies, and performs time zone unification and timestamp standardization operations on data deployed in different time zones and across nodes;

[0017] Step S24: multi-source data standardization processing, used to perform standardization processing on all structured numerical fields, specifically using the maximum and minimum normalization method to map the data value of each field to the [0,1] interval.

[0018] Furthermore, in step S3, the active-active master node health assessment is used to comprehensively assess the operating status of the current master node in the platform, identify whether there are any anomalies that affect system stability, and output a health score and abnormal problems. Specifically, the health assessment model is established, the model is trained, and the master node health is assessed in real time to obtain the real-time health assessment result of the master node. The steps include:

[0019] Step S31: Establishing a node health assessment model, which includes the following steps:

[0020] Step S311: constructing key status features, which is used to obtain key status features of the master node, including the following steps:

[0021] Step S3111: Construct a feature map input, and splice the multi-dimensional data in the switching decision data collected within a fixed time window by channel to form an input feature map , where C represents the number of node operation indicator channels, and W represents the sliding window time step;

[0022] Step S3112: Local change feature extraction is used to identify local mutation points in the master node indicators. Specifically, the input feature map is passed through two layers of one-dimensional convolution layers in sequence, and the output results of each convolution layer are batch normalized and nonlinearly activated to obtain the local change features of each indicator of the master node within the time window. ;

[0023] Step S3113: constructing the channel adaptive threshold, the formula used is as follows:

[0024] ;

[0025] ;

[0026] Where, represents the dynamic abnormal threshold of the j-th operating indicator channel, It represents the characteristic response value of the j-th operating indicator channel at the k-th time step, which is obtained from taken out from Represents the median function, v represents the adjustable factor, from value, represents the time window length of the jth running indicator channel, N represents the total number of samples of input data, and Represent the weight matrices of the two fully connected layers, represents the global average pooling function, Indicates the final adaptive threshold of the jth operating indicator channel, Represents batch normalization operation;

[0027] Step S3114: Output of key state characteristics, specifically based on and , compare and generate compression feature response values, thereby obtaining the compression local change characteristics of the operation index channel , combined to form key state features , the formula used is as follows:

[0028] ;

[0029] Where, It represents the compression characteristic response value of the j-th operating indicator channel at the k-th time step, a represents the smoothing control parameter, which controls the compression slope, Indicates the response center adjustment parameter of the compression function, the value ;

[0030] Step S312: Node operation evolution trend features are used to extract the evolution trend features of the node's multi-dimensional operation indicators from the key features. Specifically, the key state features are sequentially input into L dilated convolutions. The dilated convolution specifically introduces a layer-by-layer decreasing expansion rate and an effective receptive field parameter V. After the dilated convolution operation is completed, the convolution output result is subjected to weight normalization, nonlinear transformation and regularization processing, and a residual connection mechanism is introduced. The result is added to the key state features to obtain the operation evolution trend features. The formula used is as follows:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] Where, represents the expansion rate of the i-th dilated convolution, represents the maximum expansion rate of the initial layer, Represents the expansion rate attenuation factor, the value , i represents the index of the current convolutional layer, represents the length of the input sequence, Represents the expansion rate reduction constant, which controls the contraction speed of the receptive field. represents the size of the convolution kernel, represents the weight normalization function, represents the regularization function, Indicates the expansion rate One-dimensional convolution operation, Indicates the The output of the dilated convolution is Represents the final convolution output result of the Lth layer, Indicates the characteristics of operation evolution trend;

[0036] Step S313: Node abnormality type identification, used to identify the abnormality type of the node, the formula used is as follows:

[0037] ;

[0038] Where, Indicates the output exception type, represents the weight matrix of the abnormal type output layer, Represents the bias parameter of the abnormal type output layer;

[0039] Step S314: Generate a node health index, which is used to obtain the node's health score. The formula used is as follows:

[0040] ;

[0041] ;

[0042] Where, The weight matrix representing the characteristics of the evolution trend of the mapping operation, The bias parameter representing the evolution trend of the mapping operation, Indicates the business layer operating status data at time t, Represents the long short-term memory network operation function, Indicates the hidden state at the previous moment, The weight matrix representing the mapping anomaly type, Indicates the bias parameter of the mapping exception type, represents the fusion features of health assessment, represents the health score of the node at time t, represents the weight matrix of the health score output layer, represents the bias parameter of the health score output layer, express Activation function;

[0043] Step S32: training a node health assessment model, specifically using the system resource usage data, network status data, service layer operation status data of the master node in the historical switching decision data and the historical master node health assessment results to train the node health assessment model to obtain a trained node health assessment model;

[0044] Step S33: Real-time health assessment of the master node, specifically using the system resource usage data, network status data and business layer operation status data of the master node in the real-time switching decision data as the input data of the trained node health assessment model to obtain the real-time health assessment result of the master node, and the real-time health assessment result of the master node includes the master node abnormality type and the health score of the master node.

[0045] Furthermore, in step S4, the adaptive dynamic intervention judgment mechanism is used to accurately judge whether the master node should enter the switching process. Specifically, the adaptive dynamic intervention judgment mechanism obtains dynamic master node switching trigger judgment parameters by improving the particle swarm optimization algorithm, and performs intervention trigger judgment on the current state of the master node according to the dynamic master node switching trigger judgment parameters, and finally obtains the master node switching result. The adaptive dynamic intervention judgment mechanism includes the following steps:

[0046] Step S41: Dynamic master node switching trigger judgment parameter definition, specifically defining dynamic master node switching trigger judgment parameters, the dynamic master node switching trigger judgment parameters include the master node health score threshold , Abnormal tolerance time and the lower bound of the score descent gradient ;

[0047] Step S42: Design a fitness function to measure the effectiveness of the dynamic master node switching trigger judgment parameters. The formula used is as follows:

[0048] ;

[0049] ;

[0050] ;

[0051] Where, represents the position of the i-th particle, represents the master node health score threshold corresponding to the i-th solution, represents the abnormal tolerance time corresponding to the i-th solution, represents the lower limit of the score descent gradient corresponding to the i-th solution, represents the adaptation error function, Indicates the downward trend of the current health score, represents the weighting factor that controls trend sensitivity, represents the fitness function, represents the miscutting risk function, represents the average service loss cost function caused by switching delay, 、 and They represent the weighting factors of the three objectives, which are used to adjust the trade-off between the risk of false switching, delay cost and state matching;

[0052] Step S43: obtaining dynamic master node switching trigger judgment parameters, specifically obtaining dynamic master node switching trigger judgment parameters by improving particle swarm optimization algorithm; including the following steps:

[0053] Step S431: Initialize the particle swarm, specifically randomly generate particle positions and velocities, and set the number of particles in the particle swarm and the maximum number of iterations ;

[0054] Step S432: Calculate the fitness value, specifically calculate the fitness value f of the particles in the particle swarm i , through the fitness function Calculate the fitness value of individual particles;

[0055] Step S433: Dynamically adaptive inertia weight, specifically based on the ratio of individual fitness of particles to the average fitness of the population, through The function dynamically adjusts the inertia weight using the following formula:

[0056] ;

[0057] Where, represents the i-th particle The inertia weight of the iteration, represents the maximum value of the inertia weight, Indicates the current iteration number, represents the i-th particle The inertia weight of the iteration, represents the average fitness of the particle swarm, Indicates a small positive number that prevents the denominator from being 0;

[0058] Step S434: Update particle parameters, specifically update particle speed and particle positions ;

[0059] Step S435: Particle position perturbation mechanism, specifically comparing the Sigmoid mapping value of the current particle position with the random number to determine whether a perturbation occurs. The formula used is as follows:

[0060] ;

[0061] Where, Indicates that the i-th particle is in the The mutation position in the iteration, Represents a random number in the range [0,1];

[0062] Step S436: Update the optimal position of the particle, specifically update the local optimal position of the individual particle and the global optimal position of the particle respectively, and obtain the local optimal position of the individual particle in the next iteration and the global optimal position of the particle in the next iteration ;

[0063] Step S437: Particle update is terminated, specifically when the particle fitness value f i When the fitness threshold is higher than the threshold and the maximum number of iterations is reached, the search is terminated and the global optimal position of the particle is obtained. The global optimal position of the particle specifically refers to the dynamic master node switching trigger judgment parameter combination;

[0064] Step S44: The intervention judgment logic is executed, specifically, according to the dynamic master node switching trigger judgment parameter combination, the switching judgment logic is applied to the current master node switching judgment. If the master node switching trigger condition is met, a switching recommendation signal is generated, and the optimal switching target is selected; the switching judgment logic includes a stable abnormality trigger logic and a rapid deterioration trigger logic. The stable abnormality trigger logic is specifically when And the abnormal state has lasted When , the switching process is triggered and a switching recommendation signal is generated, where Indicates that the abnormal type needs to be switched without causing business loss time; the rapid deterioration trigger logic is specifically when , then the switching process is triggered and a switching recommendation signal is generated, where Indicates the rate of change of health score per unit time.

[0065] Furthermore, in step S5, the optimal switching target is selected to intelligently select the optimal takeover node from multiple candidate backup nodes when the primary node needs to be switched. Specifically, the optimal backup node selection process is triggered according to the switching recommendation signal to obtain the optimal switching target result, including the following steps:

[0066] Step S51: Health score of standby nodes, specifically, the system resource usage data, network status data and business layer operation status data of each standby node in the real-time switching decision data are used as input data of the trained node health assessment model to obtain the health score of each standby node. ;

[0067] Step S52: Calculate the standby node fitness for all candidate standby nodes. Perform switching adaptability calculation, specifically based on the current abnormal type of the master node , Health score of each standby node , historical backup node switching data and network delay data from the primary node to the backup node , through the problem capacity adaptation model, the takeover capacity of each backup node is predicted to obtain the fitness result of each backup node; the problem capacity adaptation model is built based on random forest regression and trained by inputting historical switching success data;

[0068] Step S53: confirming the optimal standby node target, specifically, sorting the fitness results of each standby node from high to low according to the fitness value, selecting the highest-ranked standby node, and determining it as the optimal target standby node for the current switch, thereby obtaining the optimal switch target result.

[0069] Furthermore, in step S6, the intelligent switching is executed, specifically, according to the optimal switching target result, the most suitable backup node to take over the task is obtained. The platform automatically triggers the switching process and completes the conversion of the active and standby roles, so that the active-active switching management and control platform can achieve rapid response, seamless transition and stable recovery in fault scenarios.

[0070] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0071] (1) In view of the technical problems in the intelligent decision-making method of the traditional active-active switching control platform, such as reliance on fixed switching thresholds, inaccurate switching timing, and rough selection of backup node targets, which easily lead to misjudgment and business interruption and data loss after switching, this solution innovatively proposes an adaptive dynamic intervention judgment mechanism and designs a problem capability adaptation model; among them, the adaptive dynamic intervention judgment mechanism has a high degree of adaptability, breaking through the single-point trigger limitation of the traditional static threshold, and making multi-parameter joint judgments, which significantly improves the system's response sensitivity and judgment accuracy to different fault modes; and the problem capability adaptation model introduces a master-slave matching strategy, which can make context-aware adaptation selection based on the abnormal type of the master node and the takeover capability characteristics of the backup node, avoiding the traditional rough judgment based on static priority, and significantly improving the accuracy of backup node selection and business continuity guarantee capabilities; it realizes rapid perception, accurate judgment and high-availability switching of the abnormal state of the master node under the platform architecture, ensuring that the business system can still run continuously and stably when a fault occurs, avoiding data loss and service interruption.

[0072] (2) In response to the technical problems of existing node health assessment models being susceptible to noise interference and insufficiently sensitive to trend changes, which lead to low accuracy of node health assessment results and unstable abnormality judgment, this solution innovatively designs a channel-level adaptive abnormality threshold mechanism and introduces a dilated convolution structure; among them, the channel-level adaptive abnormality threshold mechanism dynamically calculates the abnormal boundary value of each operating indicator channel, combines the indicator fluctuation intensity and channel uncertainty factors, and achieves accurate distinction between abnormal signals and background noise to obtain key state characteristics; and the dilated convolution structure captures the trend changes of operating indicators over time by introducing multi-scale receptive fields and residual connections, significantly improving the model's ability to identify chronic degradation-type abnormalities; this solution improves the model's dual perception ability of sudden abnormalities and trend changes, enhances the stability of health scores in dynamic and complex environments, and effectively avoids the situation where abnormal jitters and trend changes are misjudged as health abnormalities, ultimately achieving high-precision health assessment of the main node's operating status, significantly enhancing the practicality and decision support capabilities of the assessment model in intelligent decision-making scenarios.

[0073] (3) In view of the technical problem that the existing optimization algorithms suitable for obtaining switching judgment parameters fall into local optimal solutions, resulting in inaccurate switching judgment parameters, this scheme adopts a dynamic inertia weight update mechanism and a position perturbation mechanism to improve the particle swarm optimization algorithm, enhance the algorithm's global search capability and ability to jump out of local extreme values, effectively enhance the optimization accuracy and stability of the switching trigger judgment parameters, ensure that the final judgment parameter group has higher environmental adaptability and strategy matching, and ultimately achieve accurate identification and dynamic control of the reasonable timing of the master node switching. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A flow chart of an intelligent decision-making method for an active-active switching management and control platform provided by the present invention;

[0075] Figure 2 Schematic diagram of the process of step S2;

[0076] Figure 3 Schematic diagram of the process of step S3;

[0077] Figure 4 Schematic diagram of the process of step S4;

[0078] Figure 5 is a flow chart of step S31;

[0079] Figure 6 is a flow chart of step S43;

[0080] Figure 7 is a schematic flow chart of step S5;

[0081] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0082] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0083] In the description of the present invention, it should be understood that terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0084] Example 1, see Figure 1 The technical solution adopted by the present invention is as follows: The present invention provides an intelligent decision-making method for a dual-active switching management and control platform, the method comprising the following steps:

[0085] Step S1: Obtaining raw data for active-active switching decision-making, specifically, collecting operating status data of the master node and multiple backup nodes in the active-active switching management and control platform to obtain the raw data for active-active switching decision-making;

[0086] Step S2: Multi-source data preprocessing is used to preprocess the raw data of the active-active switching decision, specifically performing data integrity detection, outlier detection and elimination, multi-source data time alignment, and multi-source data standardization to obtain active-active switching decision optimization data;

[0087] Step S3: Health assessment of the active-active master node is used to comprehensively evaluate the operating status of the current master node, identify whether there are any anomalies that affect system stability, and output a health score and abnormal problems. Specifically, the time series feature graph is first obtained, and the key state features are identified by combining one-dimensional convolution with the channel adaptive threshold mechanism. The node operation evolution trend features are extracted using dilated convolution, and the residual connection is used to enhance the model expression capability. Subsequently, the anomaly type is discriminated and the health score of the master node is output. The node health assessment model is constructed, and the model is trained using historical switching decision data. Finally, the real-time data is input into the trained node health assessment model to obtain the real-time health assessment results of the master node.

[0088] Step S4: An adaptive dynamic intervention judgment mechanism is used to accurately determine whether the master node should enter the switching process. Specifically, the master node switching trigger judgment parameters are first defined, and then a multi-objective fitness function is constructed. Then, an improved particle swarm optimization algorithm is introduced to obtain a dynamic master node switching trigger judgment parameter group. Finally, this parameter group is applied to the master node status judgment logic to obtain the master node switching result.

[0089] Step S5: Select the optimal switching target. This is used to intelligently select the optimal takeover node from multiple candidate backup nodes when the primary node needs to be switched. Specifically, the health score of each backup node is firstly evaluated. Then, the problem capability adaptation model is used to output the backup node fitness score. The fitness scores of all candidate nodes are sorted to obtain the optimal switching target result.

[0090] Step S6: Intelligent switching execution, specifically, according to the optimal switching target result, the platform automatically triggers the switching process and completes the conversion of the primary and backup roles.

[0091] By performing the above operations, this solution addresses technical issues in the intelligent decision-making methods of traditional active-active switching management and control platforms, such as reliance on fixed switching thresholds, inaccurate switching timing, and rough selection of backup node targets, which can easily lead to misjudgments, business interruptions, and data loss after switching. This solution innovatively proposes an adaptive dynamic intervention judgment mechanism and a problem-capability adaptation model. The adaptive dynamic intervention judgment mechanism is highly adaptive, breaking through the single-point trigger limitation of traditional static thresholds and performing multi-parameter joint judgment, significantly improving the system's response sensitivity and judgment accuracy to different failure modes. The problem-capability adaptation model introduces a primary-backup matching strategy, which can perform context-aware adaptive selection based on the anomaly type of the primary node and the takeover capability characteristics of the backup node. This avoids the traditional rough judgment based on static priority and significantly improves the accuracy of backup node selection and business continuity assurance capabilities. This platform architecture enables rapid perception, accurate judgment, and high-availability switching of primary node anomalies, ensuring continuous and stable operation of the business system in the event of a failure, avoiding data loss and service interruption.

[0092] Example 2, see Figure 1The embodiment is based on the above embodiment, and in step S1, the double-active switching decision original data is acquired, specifically, the running state data of the master node and the multiple standby nodes in the platform is collected through the double-active switching management and control platform to obtain the double-active switching decision original data; the double-active switching decision original data includes historical switching decision data and real-time switching decision data; the historical switching decision data and the real-time switching decision data both include system resource usage data, network state data and service layer running state data of each node; the historical switching decision data further includes historical master node health assessment results, historical standby node switching data and historical switching success data;

[0093] The platform includes at least one master node and multiple standby nodes, wherein the standby nodes are in an online hot standby state and keep data synchronization with the master node, meeting the double-active deployment requirement;

[0094] The system resource usage data is used to reflect the current running pressure and available performance of the node, including CPU utilization, memory usage, disk read / write rate, thread activity number and operating system load balancing state;

[0095] The network state data is used to evaluate network reliability and transmission timeliness, including average delay value, packet loss rate, network jitter and port reachability rate;

[0096] The service layer running state data is used to evaluate the stability and business execution efficiency of the system external service, including request response time, transaction success rate, HTTP status code distribution and application level error code;

[0097] The historical standby node switching data includes data synchronization quality after switching and historical switching success rate;

[0098] The historical switching success data includes switching success label, switching time consumption, business recovery time after switching, business failure rate after switching, service fluctuation after switching and main use time length after switching success.

[0099] Embodiment three, refer to Figure 1 and Figure 2 The embodiment is based on the above embodiment, and the multi-source data preprocessing is used for data preprocessing of the double-active switching decision original data, specifically, data integrity detection, abnormal value detection and elimination, multi-source data time alignment and multi-source data standardization processing are performed on the data to obtain double-active switching decision optimization data; including the following steps:

[0100] Step S21: data integrity detection, used for field level and time dimension integrity detection of the original data, specifically including field integrity detection and time alignment integrity detection;

[0101] The field integrity test specifically counts the non-empty percentage of each field in the entire sampling period to determine whether it meets the minimum availability requirement; for missing field values, linear interpolation is used to fill in the missing values ​​to maintain the continuity of the time series data;

[0102] The time alignment integrity test specifically maps the data streams from different acquisition modules to a unified reference time axis, detects whether there is time stamp loss and non-uniform sampling, and uses time interpolation to calculate and fill in the missing time points to complete the preliminary alignment;

[0103] Step S22: Outlier detection and elimination, used to identify and eliminate extreme values, incorrect sampling, or abnormal fluctuation data that may exist in the data set. Specifically, the Z-score anomaly detection method is used to calculate the standardized deviation of each field, identify data points that deviate from the mean by more than a set multiple threshold, and perform elimination operations;

[0104] Step S23: Multi-source data time alignment is used to perform formal alignment and format unification operations on data from different sources. Specifically, it constructs a unified reference time axis, performs downsampling window averaging processing on data with inconsistent sampling frequencies, and performs time zone unification and timestamp standardization operations on data deployed in different time zones and across nodes;

[0105] Step S24: multi-source data standardization processing, used to perform standardization processing on all structured numerical fields, specifically using the maximum and minimum normalization method to map the data value of each field to the [0,1] interval.

[0106] Example 4, see Figure 1 、 Figure 3 and Figure 5 This embodiment is based on the above embodiment. In step S3, the active-active master node health assessment is used to comprehensively assess the operating status of the current master node in the platform, identify whether there are anomalies that affect system stability, and output a health score and abnormal problems. Specifically, it establishes a node health assessment model, trains the node health assessment model, and conducts real-time health assessment of the master node to obtain the real-time health assessment result of the master node. The steps include:

[0107] Step S31: Establishing a node health assessment model, which includes the following steps:

[0108] Step S311: constructing key status features, which is used to obtain key status features of the master node, including the following steps:

[0109] Step S3111: Construct a feature map input, and splice the multi-dimensional data in the switching decision data collected within a fixed time window by channel to form an input feature map wherein C represents the number of node running indicator channels, and W represents the sliding window time step; the node running indicator is specifically a multi-source heterogeneous time series running indicator sequence composed of system resource usage data and network state data in the time dimension, which constitutes the node running indicator data together;

[0110] Step S3112: local change feature extraction, used for identifying local mutation points in the main node indicator, specifically for sequentially passing the input feature map through two one-dimensional convolution layers to obtain the local change features of the main node indicators in the time window ; the used formula is as follows:

[0111] ;

[0112] ;

[0113] In the formula, represents the local change feature corresponding to the jth running indicator channel extracted in the first stage, represents the local change feature corresponding to the jth running indicator channel finally, represents a batch normalization operation, represents a first one-dimensional convolution operation, represents a second one-dimensional convolution operation;

[0114] Step S3113: channel adaptive threshold construction, used for dynamically constructing an abnormality recognition threshold of each indicator channel according to the change intensity and fluctuation uncertainty of the node indicators in the time window, so as to realize enhanced expression of potential abnormal signals, and the used formula is as follows:

[0115] ;

[0116] ;

[0117] In the formula, represents the dynamic abnormal threshold of the jth running indicator channel, represents the feature response value of the jth running indicator channel at the kth time step, which is taken out from ; represents a median function, and v represents an adjustable factor, which is taken from , represents the time window length of the jth running indicator channel, and N represents the total number of samples of the input data, and respectively represent the weight matrices of two fully connected layers, represents a global average pooling function, represents the adaptive threshold of the jth running indicator channel finally;

[0118] Step S3114: Output of key state characteristics, specifically based on and , compare and generate compression feature response values, thereby obtaining the compression local change characteristics of the operating index channel , combined to form key state features , the formula used is as follows:

[0119] ;

[0120] ;

[0121] ;

[0122] Where, It represents the compression characteristic response value of the j-th operating indicator channel at the k-th time step, a represents the smoothing control parameter, which controls the compression slope, Indicates the response center adjustment parameter of the compression function, the value , represents the compressed local change characteristics of the j-th operating indicator channel, represents the compressed characteristic response value of the jth operating indicator channel at the first time step, represents the compressed characteristic response value of the jth operating indicator channel at the second time step, The compressed characteristic response value of the jth operating indicator channel at the Wth time step, Indicates key state characteristics, Indicates the compressed local change characteristics of the first operating indicator channel, Indicates the compressed local change characteristics of the second operating indicator channel, Indicates the compressed local change characteristics of the Cth operating indicator channel;

[0123] Step S312: Node operation evolution trend features are used to extract the evolution trend features of the node's multi-dimensional operation indicators from the key features. Specifically, the key state features are sequentially input into L dilated convolutions. The dilated convolution specifically introduces a layer-by-layer decreasing expansion rate and an effective receptive field parameter V. After the dilated convolution operation is completed, the convolution output result is subjected to weight normalization, nonlinear transformation and regularization processing, and a residual connection mechanism is introduced. The result is added to the key state features to obtain the operation evolution trend features. The formula used is as follows:

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] Where, represents the expansion rate of the i-th dilated convolution, represents the maximum expansion rate of the initial layer, Represents the expansion rate attenuation factor, the value , i represents the index of the current convolutional layer, represents the length of the input sequence, Represents the expansion rate reduction constant, which controls the contraction speed of the receptive field. represents the size of the convolution kernel, represents the weight normalization function, represents the regularization function, Indicates the expansion rate One-dimensional convolution operation, Indicates the The output of the dilated convolution is Represents the final convolution output result of the Lth layer, Indicates the characteristics of operation evolution trend;

[0129] Step S313: Node abnormality type identification, used to identify the abnormality type of the node, the formula used is as follows:

[0130] ;

[0131] Where, Indicates the output exception type, represents the weight matrix of the abnormal type output layer, Represents the bias parameter of the abnormal type output layer;

[0132] Step S314: Generate a node health index, which is used to obtain the node's health score. The formula used is as follows:

[0133] ;

[0134] ;

[0135] Where, The weight matrix representing the characteristics of the evolution trend of the mapping operation, The bias parameter representing the evolution trend of the mapping operation, Indicates the business layer operating status data at time t, Represents the long short-term memory network operation function, Indicates the hidden state at the previous moment, The weight matrix representing the mapping anomaly type, Indicates the bias parameter of the mapping exception type, represents the fusion features of health assessment, represents the health score of the node at time t, represents the weight matrix of the health score output layer, represents the bias parameter of the health score output layer, express Activation function;

[0136] Step S32: training a node health assessment model, specifically using the system resource usage data, network status data, service layer operation status data of the master node in the historical switching decision data and the historical master node health assessment results to train the node health assessment model to obtain a trained node health assessment model;

[0137] Step S33: Real-time health assessment of the master node, specifically using the system resource usage data, network status data and business layer operation status data of the master node in the real-time switching decision data as the input data of the trained node health assessment model to obtain the real-time health assessment result of the master node, and the real-time health assessment result of the master node includes the master node abnormality type and the health score of the master node.

[0138] By performing the above operations, this solution addresses the technical issues of existing node health assessment models being susceptible to noise interference and insufficiently sensitive to trend changes, which lead to inaccurate node health assessment results and unstable anomaly judgments. The solution innovatively designs a channel-level adaptive anomaly threshold mechanism and introduces a dilated convolution structure. The channel-level adaptive anomaly threshold mechanism dynamically calculates the anomaly boundary value of each operating indicator channel, combining indicator fluctuation intensity and channel uncertainty factors to accurately distinguish anomaly signals from background noise and obtain key state characteristics. The dilated convolution structure, by introducing multi-scale receptive fields and residual connections, captures the trend changes of operating indicators over time, significantly improving the model's ability to identify chronic degeneration anomalies. This solution improves the model's dual perception of sudden anomalies and trend changes, enhances the stability of health scores in dynamic and complex environments, and effectively avoids the misjudgment of health anomalies due to abnormal jitter and trend changes. Ultimately, it achieves high-precision health assessment of the main node's operating status, significantly enhancing the practicality and decision support capabilities of the assessment model in intelligent decision-making scenarios.

[0139] Example 5, see Figure 1 、 Figure 4 and Figure 6The embodiment is based on the above embodiment, in step S4, the adaptive dynamic intervention judgment mechanism is used to accurately judge whether the master node should enter the switching process, specifically, the dynamic master node switching trigger judgment parameter is obtained by improving the particle swarm optimization algorithm, the intervention trigger judgment of the current state of the master node is performed according to the dynamic master node switching trigger judgment parameter, and finally the master node switching result is obtained; comprising the following steps:

[0140] Step S41: Dynamic master node switching trigger judgment parameter definition, specifically, the dynamic master node switching trigger judgment parameter is defined, and the dynamic master node switching trigger judgment parameter includes a master node health score threshold , an abnormal tolerance time and a score decline gradient lower limit ; the master node health score threshold is used to judge whether the current health score is at the abnormal lower limit; the abnormal tolerance time is used to control the abnormal state duration requirement to avoid false switching caused by transient fluctuations; the score decline gradient lower limit is used to identify whether the score is in a rapidly deteriorating trend, and if it exceeds this threshold, the switching judgment process is triggered in advance;

[0141] Step S42: Design fitness function, used to measure the effectiveness of the dynamic master node switching trigger judgment parameter, the formula is as follows:

[0142] ;

[0143] ;

[0144] ;

[0145] In the formula, represents the position of the i-th particle, represents the corresponding master node health score threshold in the i-th solution, represents the corresponding abnormal tolerance time in the i-th solution, represents the corresponding score decline gradient lower limit in the i-th solution, represents the adaptive error function, represents the downward trend of the current health score, represents the weighting factor for controlling the trend sensitivity, represents the fitness function, represents the false switching risk function, represents the average business loss cost function caused by switching delay, , and They represent the weighting factors of the three objectives, which are used to adjust the trade-off between the risk of false switching, delay cost and state matching;

[0146] Step S43: obtaining dynamic master node switching trigger judgment parameters, specifically obtaining dynamic master node switching trigger judgment parameters by improving particle swarm optimization algorithm; including the following steps:

[0147] Step S431: Initialize the particle swarm, specifically randomly generate particle positions and velocities, and set the number of particles N and the maximum number of iterations in the particle swarm. ;

[0148] Step S432: Calculate the fitness value, specifically calculate the fitness value f of the particles in the particle swarm i , through the fitness function Calculate the fitness value of individual particles;

[0149] Step S433: Dynamically adaptive inertia weight, specifically based on the ratio of individual fitness of particles to the average fitness of the population, through The function dynamically adjusts the inertia weight using the following formula:

[0150] ;

[0151] Where, represents the i-th particle The inertia weight of the iteration, represents the maximum value of the inertia weight, Indicates the current iteration number, represents the i-th particle The inertia weight of the iteration, represents the average fitness of the particle swarm, Indicates a small positive number that prevents the denominator from being 0;

[0152] Step S434: Update particle parameters, specifically update particle velocity and particle position; the formula used is as follows:

[0153] ;

[0154] Where, Indicates that the i-th particle is in the Iteration speed, Indicates that the i-th particle is in the Iteration speed, represents the position of the i-th particle in the t-th iteration, represents the local optimal position of the individual particle, represents the global optimal position of the particle, and represents a random number in the range [0,1], represents the individual learning factor, which is used to control the speed at which particles move to their optimal positions. represents the group learning factor, which is used to control the speed at which particles move to the global optimal position. Indicates that the i-th particle is in the Position in iteration;

[0155] Step S435: Particle position perturbation mechanism, specifically comparing the Sigmoid mapping value of the current particle position with the random number to determine whether a perturbation occurs. The formula used is as follows:

[0156] ;

[0157] Where, Indicates that the i-th particle is in the The mutation position in the iteration, Represents a random number in the range [0,1];

[0158] Step S436: Update the optimal position of the particle, specifically update the local optimal position of the individual particle and the global optimal position of the particle respectively, and obtain the local optimal position of the individual particle in the next iteration and the global optimal position of the particle in the next iteration ;

[0159] Step S437: Particle update is terminated, specifically when the particle fitness value f i When the fitness threshold is higher than the threshold and the maximum number of iterations is reached, the search is terminated and the global optimal position of the particle is obtained. The global optimal position of the particle specifically refers to the dynamic master node switching trigger judgment parameter combination;

[0160] Step S44: The intervention judgment logic is executed, specifically, according to the dynamic master node switching trigger judgment parameter combination, the switching judgment logic is applied to the current master node switching judgment. If the master node switching trigger condition is met, a switching recommendation signal is generated, and the optimal switching target is selected; the switching judgment logic includes a stable abnormality trigger logic and a rapid deterioration trigger logic. The stable abnormality trigger logic is specifically when And the abnormal state has lasted When , the switching process is triggered and a switching recommendation signal is generated, where Indicates that the abnormal type needs to be switched without causing business loss time; the rapid deterioration trigger logic is specifically when , then the switching process is triggered and a switching recommendation signal is generated, where Indicates the rate of change of health score per unit time.

[0161] By performing the above operations, the existing optimization algorithm suitable for obtaining switching judgment parameters is used to solve the technical problem of falling into local optimal solutions, resulting in inaccurate switching judgment parameters. This solution adopts a dynamic inertia weight update mechanism and a position perturbation mechanism to improve the particle swarm optimization algorithm, enhance the algorithm's global search capability and ability to escape local extreme values, effectively enhance the optimization accuracy and stability of the switching trigger judgment parameters, ensure that the final judgment parameter group has higher environmental adaptability and strategy matching, and ultimately achieve accurate identification and dynamic control of the reasonable timing of master node switching.

[0162] Example 6, see Figure 1 and Figure 7 This embodiment is based on the above embodiment. In step S5, the optimal switching target is selected to intelligently screen the optimal takeover node from multiple candidate backup nodes when the primary node needs to be switched. Specifically, the optimal backup node selection process is triggered according to the switching recommendation signal to obtain the optimal switching target result, including the following steps:

[0163] Step S51: Health score of standby nodes, specifically, the system resource usage data, network status data and business layer operation status data of each standby node in the real-time switching decision data are used as input data of the trained node health assessment model to obtain the health score of each standby node. ;

[0164] Step S52: Calculate the standby node fitness for all candidate standby nodes. Perform switching adaptability calculation, specifically based on the current abnormal type of the master node , Health score of each standby node , historical backup node switching data and network delay data from the primary node to the backup node , through the problem capacity adaptation model, the takeover capacity of each backup node is predicted to obtain the fitness result of each backup node; the problem capacity adaptation model is built based on random forest regression and trained by inputting historical switching success data;

[0165] Step S53: confirming the optimal standby node target, specifically, sorting the fitness results of each standby node from high to low according to the fitness value, selecting the highest-ranked standby node, and determining it as the optimal target standby node for the current switch, thereby obtaining the optimal switch target result.

[0166] Example 7, see Figure 1This embodiment is based on the above embodiment. In step S6, the intelligent switching is executed, specifically, according to the optimal switching target result, the standby node that is currently most suitable for taking over the task is obtained. The platform automatically triggers the switching process and completes the conversion of the active and standby roles, so that the active-active switching management and control platform can achieve rapid response, seamless transition and stable recovery in fault scenarios.

[0167] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0168] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0169] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An intelligent decision-making method for an active-active switching management and control platform, characterized by: The method comprises the following steps: Step S1: Obtaining raw data for active-active switching decision-making, specifically, collecting operating status data of the master node and multiple backup nodes in the active-active switching management and control platform to obtain the raw data for active-active switching decision-making; Step S2: Preprocessing multi-source data, specifically optimizing the data to obtain active-active switching decision optimization data; Step S3: Health assessment of the active-active master node. Specifically, the following steps are performed: first, a time series feature graph is obtained, and then one-dimensional convolution and a channel adaptive threshold mechanism are used to identify key state features. Dilated convolution is used to extract node operation evolution trend features, and residual connections are used to enhance the model's expressiveness. The abnormality type is then identified and the master node's health score is output. A node health assessment model is then established and trained using historical switch decision data. Finally, real-time data is input into the trained node health assessment model to obtain the real-time health assessment results of the master node. Step S4: Adaptive dynamic intervention judgment mechanism, specifically, first defining the master node switching trigger judgment parameters, then constructing a multi-objective fitness function, then introducing a particle swarm optimization algorithm improved by a dynamic inertia weight update mechanism and a position perturbation mechanism to obtain dynamic master node switching trigger judgment parameters, and finally applying this parameter group to the master node state judgment logic to obtain the master node switching result; including the following steps: Dynamic master node switching trigger judgment parameter definition, specifically defining dynamic master node switching trigger judgment parameters, the definition of dynamic master node switching trigger judgment parameters includes master node health score threshold , Abnormal tolerance time and the lower bound of the score descent gradient ; Obtaining dynamic master node switching trigger judgment parameters, specifically by improving the particle swarm optimization algorithm to obtain dynamic master node switching trigger judgment parameters; The intervention judgment logic is executed, specifically, according to the dynamic master node switching trigger judgment parameter combination, the switching judgment logic is applied to the current master node switching judgment. If the master node switching trigger condition is met, a switching recommendation signal is generated, and the optimal switching target is selected; the switching judgment logic includes stable abnormality trigger logic and rapid deterioration trigger logic. The stable abnormality trigger logic is specifically when And the abnormal state has lasted When , the switching process is triggered and a switching recommendation signal is generated, where Indicates that the abnormal type needs to be switched without causing business loss time; the rapid deterioration trigger logic is specifically when , then the switching process is triggered and a switching recommendation signal is generated, where It represents the rate of change of health score per unit time; Step S5: Select the optimal switching target. Specifically, first perform a health score on each backup node, then use the problem capability adaptation model to output the backup node fitness score, sort the fitness results of all candidate nodes, and obtain the optimal switching target result. Step S6: Intelligent switching execution, specifically, according to the optimal switching target result, the platform automatically triggers the switching process and completes the conversion of the primary and backup roles.

2. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S3, the active-active master node health assessment is used to comprehensively evaluate the operating status of the current master node in the platform, identify whether there are any anomalies that affect system stability, and output a health score and anomaly issues. Specifically, the following steps are included: Step S31: Establishing a node health assessment model, which is used to establish a node health assessment model, including key state feature construction, node operation evolution trend features, node abnormality type discrimination and node health index generation; Step S32: training a node health assessment model, specifically using the system resource usage data, network status data, service layer operation status data of the master node in the historical switching decision data and the historical master node health assessment results to train the node health assessment model, thereby obtaining a trained node health assessment model; Step S33: Real-time health assessment of the master node, specifically using the system resource usage data, network status data and business layer operation status data of the master node in the real-time switching decision data as the input data of the trained node health assessment model to obtain the real-time health assessment result of the master node, which includes the master node abnormality type and the health score of the master node.

3. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S31, the establishment of the node health assessment model includes the following steps: Step S311: constructing key status features, which is used to obtain key status features of the master node, including the following steps: Step S3111: Construct a feature map input, and splice the multi-dimensional data in the switching decision data collected within a fixed time window by channel to form an input feature map , where C represents the number of node operation indicator channels, and W represents the sliding window time step; Step S3112: Local change feature extraction is used to identify local mutation points in the master node indicators. Specifically, the input feature map is passed through two layers of one-dimensional convolution layers in sequence, and the output results of each convolution layer are batch normalized and nonlinearly activated to obtain the local change features of each indicator of the master node within the time window. ; Step S3113: constructing the channel adaptive threshold, the formula used is as follows: ； ； Where, represents the dynamic abnormal threshold of the j-th operating indicator channel, It represents the characteristic response value of the j-th operating indicator channel at the k-th time step, which is obtained from taken out from Represents the median function, v represents the adjustable factor, from value, represents the time window length of the jth running indicator channel, N represents the total number of samples of input data, and Represent the weight matrices of the two fully connected layers, represents the global average pooling function, Indicates the final adaptive threshold of the jth operating indicator channel, Represents batch normalization operation; Step S3114: Output of key state characteristics, specifically based on and , compare and generate compression feature response values, thereby obtaining the compression local change characteristics of the operating index channel , combined to form key state features , the formula used is as follows: ； Where, It represents the compression characteristic response value of the j-th operating indicator channel at the k-th time step, a represents the smoothing control parameter, which controls the compression slope, Indicates the response center adjustment parameter of the compression function, the value ; Step S312: Node operation evolution trend features are used to extract the evolution trend features of the node's multi-dimensional operation indicators from the key features. Specifically, the key state features are sequentially input into L dilated convolutions. The dilated convolution specifically introduces a layer-by-layer decreasing expansion rate and an effective receptive field parameter V. After the dilated convolution operation is completed, the convolution output result is subjected to weight normalization, nonlinear transformation and regularization processing, and a residual connection mechanism is introduced. The result is added to the key state features to obtain the operation evolution trend features. The formula used is as follows: ； ； ； ； Where, represents the expansion rate of the i-th dilated convolution, represents the maximum expansion rate of the initial layer, Represents the expansion rate attenuation factor, the value , i represents the index of the current convolutional layer, represents the length of the input sequence, Represents the expansion rate reduction constant, which controls the contraction speed of the receptive field. represents the size of the convolution kernel, represents the weight normalization function, represents the regularization function, Indicates the expansion rate One-dimensional convolution operation, Indicates the The output of the dilated convolution is Represents the final convolution output result of the Lth layer, Indicates the characteristics of operation evolution trend; Step S313: Node abnormality type identification, used to identify the abnormality type of the node, the formula used is as follows: ； Where, Indicates the output exception type, represents the weight matrix of the abnormal type output layer, Represents the bias parameter of the abnormal type output layer; Step S314: Generate a node health index, which is used to obtain the node's health score. The formula used is as follows: ； ； Where, The weight matrix representing the characteristics of the evolution trend of the mapping operation, The bias parameter representing the evolution trend of the mapping operation, Indicates the business layer operating status data at time t, Represents the long short-term memory network operation function, Indicates the hidden state at the previous moment, The weight matrix representing the mapping anomaly type, Indicates the bias parameter of the mapping exception type, represents the fusion features of health assessment, represents the health score of the node at time t, represents the weight matrix of the health score output layer, represents the bias parameter of the health score output layer, express Activation function.

4. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S4, the adaptive dynamic intervention judgment mechanism further includes the following steps: Step S42: Design a fitness function to measure the effectiveness of the dynamic master node switching trigger judgment parameters. The formula used is as follows: ； ； ； Where, represents the position of the i-th particle, represents the master node health score threshold corresponding to the i-th solution, represents the abnormal tolerance time corresponding to the i-th solution, represents the lower limit of the score descent gradient corresponding to the i-th solution, represents the adaptation error function, Indicates the downward trend of the current health score, represents the weighting factor that controls trend sensitivity, represents the fitness function, represents the miscutting risk function, represents the average service loss cost function caused by switching delay, 、 and They represent the weighting factors of the three objectives, which are used to adjust the trade-off between false switching risk, delay cost and state matching.

5. The intelligent decision-making method for active-active switching management and control platform according to claim 1, characterized in that: In step S43, the dynamic master node switching trigger judgment parameter is obtained, specifically including the following steps: Step S431: Initialize the particle swarm, specifically randomly generate particle positions and velocities, and set the number of particles in the particle swarm and the maximum number of iterations ; Step S432: Calculate the fitness value, specifically calculate the fitness value f of the particles in the particle swarm i , through the fitness function Calculate the fitness value of individual particles; Step S433: Dynamically adaptive inertia weight, specifically based on the ratio of individual fitness of particles to the average fitness of the population, through The function dynamically adjusts the inertia weight using the following formula: ； Where, represents the i-th particle The inertia weight of the iteration, represents the maximum value of the inertia weight, Indicates the current iteration number, represents the i-th particle The inertia weight of the iteration, represents the average fitness of the particle swarm, Indicates a small positive number that prevents the denominator from being 0; Step S434: Update particle parameters, specifically update particle speed and particle positions ; Step S435: Particle position perturbation mechanism, specifically comparing the Sigmoid mapping value of the current particle position with the random number to determine whether a perturbation occurs. The formula used is as follows: ； Where, Indicates that the i-th particle is in the The mutation position in the iteration, Represents a random number in the range [0,1]; Step S436: Update the optimal position of the particle, specifically update the local optimal position of the individual particle and the global optimal position of the particle respectively, and obtain the local optimal position of the individual particle in the next iteration and the global optimal position of the particle in the next iteration ; Step S437: Particle update is terminated, specifically when the particle fitness value f i When the fitness threshold is higher than the threshold and the maximum number of iterations is reached, the search is terminated and the global optimal position of the particle is obtained. The global optimal position of the particle specifically refers to the dynamic master node switching trigger judgment parameter combination.

6. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S5, the optimal switching target is selected to intelligently select the optimal takeover node from multiple candidate backup nodes when the primary node needs to be switched; Specifically, according to the switching recommendation signal, triggering the optimal backup node selection process to obtain the optimal switching target result includes the following steps: Step S51: Health score of standby nodes, specifically, the system resource usage data, network status data and business layer operation status data of each standby node in the real-time switching decision data are used as input data of the trained node health assessment model to obtain the health score of each standby node. ; Step S52: Calculate the standby node fitness for all candidate standby nodes. Perform switching adaptability calculation, specifically based on the current abnormal type of the master node , Health score of each standby node , historical backup node switching data and network delay data from the primary node to the backup node , through the problem capacity adaptation model, the takeover capacity of each backup node is predicted to obtain the fitness result of each backup node; the problem capacity adaptation model is built based on random forest regression and trained by inputting historical switching success data; Step S53: confirming the optimal standby node target, specifically, sorting the fitness results of each standby node from high to low according to the fitness value, selecting the highest-ranked standby node, and determining it as the optimal target standby node for the current switch, thereby obtaining the optimal switch target result.

7. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S6, the intelligent switching is executed, specifically, according to the optimal switching target result, the most suitable backup node to take over the task is obtained. The platform automatically triggers the switching process and completes the conversion of the active and standby roles, so that the active-active switching management and control platform can achieve rapid response, seamless transition and stable recovery in fault scenarios.

8. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S1, the active-active switching decision original data is obtained, specifically by collecting the operating status data of the master node and multiple backup nodes in the platform through the active-active switching management and control platform to obtain the active-active switching decision original data; the active-active switching decision original data includes historical switching decision data and real-time switching decision data; the historical switching decision data and real-time switching decision data both include system resource usage data, network status data and business layer operating status data of each node; the historical switching decision data also includes historical master node health assessment results, historical backup node switching data and historical switching success data.

9. The intelligent decision-making method for an active-active switching management and control platform according to claim 1, characterized in that: In step S2, the multi-source data preprocessing is used to preprocess the original data of the active-active switching decision, specifically including the following steps: Step S21: Data integrity check, which is used to check the integrity of the original data at the field level and time dimension, specifically including field integrity check and time alignment integrity check; Step S22: Outlier detection and elimination, used to identify and eliminate extreme values, incorrect sampling, or abnormal fluctuation data that may exist in the data set. Specifically, the Z-score anomaly detection method is used to calculate the standardized deviation of each field, identify data points that deviate from the mean by more than a set multiple threshold, and perform elimination operations; Step S23: Multi-source data time alignment is used to perform formal alignment and format unification operations on data from different sources. Specifically, it constructs a unified reference time axis, performs downsampling window averaging processing on data with inconsistent sampling frequencies, and performs time zone unification and timestamp standardization operations on data deployed in different time zones and across nodes; Step S24: multi-source data standardization processing, used to perform standardization processing on all structured numerical fields, specifically using the maximum and minimum normalization method to map the data value of each field to the [0,1] interval.

Images