An intelligent multi-platform connector dynamic adaptation method

Abstract

The application discloses a kind of intelligent multi-platform connector dynamic adaptation methods, method includes multi-source adaptation data collection, data optimization processing, bidirectional environmental change monitoring, connector intelligent dynamic recommendation and connector real-time dynamic adaptation.The application relates to the field of data processing, specifically refers to a kind of intelligent multi-platform connector dynamic adaptation method, the scheme innovatively introduces the linkage monitoring mechanism of connector running end and target end environment, and the connector version type recommendation is separated from configuration parameter generation modeling, significantly enhance the flexibility of connector adaptation strategy;Innovatively proposed to introduce penalty term to construct improved intelligent recommendation loss function and dynamic weight adjustment mechanism method, effectively improve the accuracy and reliability of connector version recommendation result;Adaptive balancing factor, geometric migration mechanism and simplified jump search strategy are introduced to improve optimization algorithm, effectively improve the stability and accuracy of the output result of model.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data processing, and particularly discloses an intelligent multi-platform connector dynamic adaptation method. BACKGROUND

[0002] The intelligent multi-platform connector dynamic adaptation method refers to a method for intelligently analyzing and modeling the characteristics of interfaces, protocols, data formats, etc. of different platforms by using artificial intelligence technology, and dynamically optimizing the connector configuration according to real-time data flow, so as to realize automatic, efficient and stable data adaptation and interaction between multiple heterogeneous platforms in a big data environment.

[0003] However, in the traditional multi-platform connector adaptation method, there is a technical problem that only static preset parameters and one-way environment sensing are relied on for adaptation decision-making, and there is a lack of collaborative monitoring of the running end and the target end environment, thereby leading to the case that the multi-platform connector has poor stability, low compatibility and low adaptation efficiency in a dynamic heterogeneous environment; in the existing connector intelligent recommendation model, there is a technical problem that the sensitivity to high-error samples is insufficient and the weight updating strategy lacks flexibility, which easily falls into local optimum and slow training convergence, thereby leading to the case that the accuracy of the connector version type recommendation result is not high; in the existing connector parameter generation model, there is a technical problem that the hyperparameter setting of the model is unreasonable, leading to unstable and inaccurate model output results, and further leading to the case that the multi-platform connector has insufficient dynamic adaptation capability. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the application provides a kind of intelligent multi-platform connector dynamic adaptation method, for the technical problem that only relying on static preset parameter and unidirectional environment sensing is adapted in the traditional multi-platform connector adaptation method, lack of collaborative monitoring of the environment of running end and target end, thereby leading to the case that multi-platform connector is poor in stability, low in compatibility, and low in adaptation efficiency in dynamic heterogeneous environment, the application innovatively introduces the linkage monitoring mechanism of connector running end and target end environment, and separates the connector version type recommendation and configuration parameter generation modeling, respectively optimized through connector intelligent recommendation model and connector parameter generation model, to improve the real-time response of environmental change, enhance the flexibility of connector adaptation strategy, effectively realize the efficient, stable and intelligent dynamic adaptation of connector in complex multi-platform environment;For the technical problems that the existing connector intelligent recommendation model is not sensitive to high error samples and the weight updating strategy lacks flexibility, easy to fall into local optimum, slow training convergence, thereby leading to the case that the accuracy of connector version recommendation result is not high, the scheme innovatively introduces the method of introducing penalty term to build improved intelligent recommendation loss function and dynamic weight adjustment mechanism, which can significantly improve the response ability and convergence efficiency of the model to abnormal samples, enhance the stability and adaptability of the recommendation model in complex and variable scenarios, thereby effectively improving the accuracy and reliability of the connector version recommendation result, and finally realizing the intelligent recommendation capability of connector for multi-platform heterogeneous environment, significantly improving the dynamic adaptation efficiency of connector in complex operating environment and the flexibility of cross-platform deployment;For the technical problems that the existing connector parameter generation model is not reasonable in hyperparameter setting, leading to unstable and inaccurate model output, and further leading to insufficient multi-platform connector dynamic adaptation capability, the scheme innovatively introduces adaptive balance factor, geometric migration mechanism and simplified jump search strategy to improve the optimization algorithm, globally optimizes the hyperparameters in the generation model, and obtains the optimal hyperparameter combination of the model, thereby effectively improving the stability and accuracy of the output result of the connector parameter generation model, and finally realizing the efficient and intelligent adaptation capability of the connector in multi-platform environment.

[0005] The technical solutions adopted by the application are as follows: the application provides an intelligent multi-platform connector dynamic adaptation method, which comprises the following steps:

[0006] Step S1: multi-source adaptation data collection;

[0007] Step S2: data optimization processing;

[0008] Step S3: bidirectional environmental change monitoring;

[0009] Step S4: connector intelligent dynamic recommendation;

[0010] Step S5: Real-time dynamic adaptation of the connector.

[0011] Further, in step S1, the multi-source adaptation data is collected for collecting basic data required for multi-platform connector adaptation, specifically by collecting data information in the multi-platform connection management system to obtain multi-platform connector adaptation original data; the multi-platform connector adaptation original data includes historical connector adaptation data and real-time connector adaptation data, and the historical connector adaptation data and the real-time connector adaptation data both include connector running end data, connector context data and connector target end data; the historical connector adaptation data further includes connector adaptation results, connector version types and connector configuration parameters.

[0012] Further, in step S2, the data optimization processing specifically includes the following steps:

[0013] Step S21: Data cleaning, specifically missing value filling and outlier removal;

[0014] Step S22: Data format standardization processing, specifically including data format unification and numerical field standardization processing;

[0015] Step S23: Data encoding processing, specifically using one-hot encoding method to encode the category field in the original data, converting discrete text and label variables into sparse numerical vectors;

[0016] Step S24: Connector bidirectional end label specification processing, specifically mapping different platform identifications described in different source data to standard platform categories through the establishment of a platform type mapping table, and performing label specification processing on the system type field of the connector target end.

[0017] Further, in step S3, the bidirectional environment change monitoring specifically includes the following steps:

[0018] Step S31: Connector running end environment monitoring, specifically extracting key features of the connector running end data through the Pearson correlation coefficient algorithm to generate a connector running end key feature combination, for each connector running end key feature, calculating the relative change rate of the corresponding feature based on its current value and the corresponding running end reference benchmark value, and assigning a running end weight to each connector running end key feature, and finally calculating the running end total change value by multiplying the change rate of each feature with its corresponding weight and summing up;

[0019] Step S32: connector target end environment monitoring, specifically, key feature extraction of the connector target end data is performed through the Pearson correlation coefficient algorithm, a connector target end key feature combination is generated, for each connector target end key feature, based on its current value and the target end reference benchmark value, the relative change rate of the corresponding feature is calculated, and a target end weight is assigned to each connector target end key feature, and finally the total change value of the target end is calculated by multiplying the change rate of each feature by its corresponding weight and summing up;

[0020] Step S33: connector adaptation reconstruction trigger judgment, specifically, first adaptation judgment, running end environment change judgment and target end environment change judgment are sequentially performed to obtain the connector adaptation trigger result;

[0021] The first adaptation judgment, specifically, if it is the first adaptation, the connector intelligent dynamic recommendation process is directly executed, otherwise the running end environment change judgment process is continued;

[0022] The running end environment change judgment, specifically, if the running end total change value is greater than the running end threshold value, the re-adaptation operation of the connector is triggered, and the connector intelligent dynamic recommendation process is executed;

[0023] The target end environment change judgment, specifically, if the target end total change value is greater than the target end threshold value, the re-adaptation operation of the connector is triggered, and the connector intelligent dynamic recommendation process is executed.

[0024] Further, in step S4, the connector intelligent dynamic recommendation specifically includes the following steps:

[0025] Step S41: building a connector intelligent recommendation model, specifically including the following steps:

[0026] Step S411, connector data receiving layer, specifically, the connector running end key feature combination and the connector target end key feature combination are spliced to obtain a connector intelligent recommendation feature matrix;

[0027] Step S412, primary feature extraction layer, specifically, the connector intelligent recommendation feature matrix is input into a primary full connection layer, a linear weighting and bias summation operation is performed, and an activation function is introduced to realize feature nonlinear transformation to obtain recommendation primary features;

[0028] Step S413, feature importance weighting layer, specifically, the recommendation primary features are taken as input, query, key and value vectors are generated, feature importance weight coefficients are calculated through a self-attention mechanism, and then the recommendation primary features are weighted and combined to obtain recommendation weighted features;

[0029] Step S414, a deep feature fusion layer, specifically, the recommendation weighted feature is input into a second fully connected layer, and a linear weighted and biased summation operation is performed, and a The activation function realizes deep mapping of the recommendation weighted feature, and a recommendation deep feature is obtained.

[0030] Step S415, a connector configuration recommendation output layer, specifically, the recommendation deep feature is input into an output layer, a linear combination operation with a weight matrix and a bias term is performed, and a Softmax activation function is introduced, and a connector version type recommendation result is obtained.

[0031] Step S42: connector intelligent recommendation model training, specifically including the following steps:

[0032] Step S421, designing an intelligent recommendation loss function, specifically, an improved intelligent recommendation loss function is constructed by introducing a penalty term, and the formula is as follows:

[0033] ;

[0034] In the formula, represents the intelligent recommendation loss function value, represents the connector version type recommendation result of the i th sample, represents the connector version type real result of the i th sample, represents an error tolerance threshold, represents a penalty weight coefficient, I(·) represents an indicator function, which is 1 when the error exceeds the threshold, otherwise 0, and N represents the number of samples;

[0035] Step S422, model weight update, specifically, a dynamic weight adjustment mechanism is uniformly applied to the weights of each layer of the model, the learning rate and the weight gradient are adaptively adjusted according to the error change trend of the current round, and the weights of each layer of the model are updated; the formula is as follows:

[0036] ;

[0037] ;

[0038] In the formula, represents the i th iterative weight change, represents a learning rate adjustment factor, represents the current learning rate, represents the current momentum coefficient, represents the weight gradient, represents the i th iterative weight change, represents the error change in the i th iteration, representing the first iterative model weight values, representing the first iterative model weight values;

[0039] Step S423, model training termination, specifically, in each round of iterative training, based on the connector running end key feature combination and the connector target end key feature combination generated through the historical connector adaptation data as the model training data, the weight parameters of all layers of the model are updated through the design of the intelligent recommendation loss function and the model weight update, the iterative correction value of each layer weight is obtained, and the intelligent recommendation loss function value under the current training round is calculated, if the loss value converges to the preset error termination threshold or the model training iteration number reaches the maximum training number, the model iterative training is stopped, and the optimized connector intelligent recommendation model is obtained, otherwise the iterative training is continued;

[0040] Step S43, connector version type real-time recommendation, specifically, through the real-time connector adaptation data, the connector running end key feature combination and the connector target end key feature combination generated are used as the input data of the optimized connector intelligent recommendation model, and the currently recommended connector version type is obtained.

[0041] Further, in step S5, the connector real-time dynamic adaptation, specifically including the following steps:

[0042] Step S51, constructing and training a connector parameter generation model, specifically, establishing a connector parameter generation model based on the Transformer architecture, and based on the connector context data and the connector version type in the historical connector adaptation data as the training input data, the generation model is trained to obtain the trained connector parameter generation model;

[0043] Step S52, model hyperparameter optimization, specifically, using an improved optimization algorithm to globally optimize the connector parameter generation model hyperparameters to obtain the optimal hyperparameter combination of the connector parameter generation model, and finally obtaining the optimized connector parameter generation model, including the following steps:

[0044] Step S521: initialize the search population individuals, specifically, encode the connector parameter generation model hyperparameters into search individual position vectors, and generate N search individual position vectors through random initialization, each individual code represents a candidate parameter combination, and constitutes an initial search population;

[0045] Step S522: calculate the individual fitness value, specifically, calculate the search individual fitness value in the population, and the performance of the connector parameter generation model established based on the search individual position is used as the search individual fitness value;

[0046] Step S523: Calculate the search regulation parameters, specifically by calculating the adaptive balance factor based on the exponential decay function and combining it with the geometric migration mechanism to calculate the individual's movement distance; the formula used is as follows:

[0047] ;

[0048] ;

[0049] In the formula, Indicates the first The adaptive balance factor of the generation population Indicates the current iteration number. Indicates the maximum number of iterations. This represents the adaptive balance factor for initializing the population, with a value range of... between, Indicates that the i-th individual is in the first place. The distance traveled by the generation Indicates the movement control factor, which is in Random numbers within a range that follow a normal distribution. This indicates the position of the optimal individual in the current iteration. Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population;

[0050] Step S524: Position update during the multi-directional exploration phase, specifically when... At that time, a multi-directional search position update strategy is adopted, and the formula used is as follows:

[0051] ;

[0052] In the formula, Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population, where j represents the index of the dimension. , and They all said Random numbers uniformly distributed within a range Indicates that a random individual is in the 1st... Random dimension position in the population, Indicates that the i-th individual is in the first place. Random dimension position in the population, Indicates the random dimension. Represents even numbers, Indicates odd numbers;

[0053] Step S525: Local stage position update, specifically when At this time, a simplified jump search strategy is used for local position updates; the formula used is as follows:

[0054] ;

[0055] In the formula, and They all said Random numbers uniformly distributed within a range Indicates that a random individual is in the 1st... The j-th dimension position in the population, Indicates the jump step length. Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population;

[0056] Step S526: Obtain the optimal position. Specifically, after each iteration, the fitness values ​​of all individuals in the current population are re-evaluated. If the fitness of an individual's position is better than the current global optimal individual position, then the global optimal individual position is updated using that individual position.

[0057] Step S527: The search iteration terminates, specifically when the fitness value of the search individual is... When the fitness threshold is exceeded or the maximum number of iterations is reached, the search is terminated and the globally optimal individual position is obtained. The globally optimal individual position specifically refers to the optimal combination of hyperparameters of the connector parameter generation model.

[0058] Step S528: Model performance optimization, specifically, based on the optimal hyperparameter combination, adjusting the hyperparameters of the trained connector parameter generation model to obtain the optimal connector parameter generation model;

[0059] Step S53: Real-time generation of connector parameters, specifically, using the connector context data in the real-time connector adaptation data and the currently recommended connector version type as input data for the optimal connector parameter generation model to obtain the connector running configuration parameters;

[0060] Step S54: Connector dynamic adaptation, specifically, based on the currently recommended connector version type and connector running configuration parameters, dynamically load and initialize the recommended version type connector component, execute running configuration parameter injection, and complete the connection establishment process with the target system and platform, thereby realizing the intelligent and dynamic adaptation and deployment of connectors in multi-platform heterogeneous environments.

[0061] The beneficial effects achieved by the present invention using the above solution are as follows:

[0062] (1) In view of the technical problems in the traditional multi-platform connector adaptation method that only rely on static preset parameters and one-way environment sensing for adaptation decision-making, lack of collaborative monitoring of the running end and target end environment, resulting in poor stability, low compatibility and low adaptation efficiency of the multi-platform connector in the dynamic heterogeneous environment, the application innovatively introduces a linkage monitoring mechanism of the connector running end and target end environment, and separates the connector version type recommendation and configuration parameter generation modeling, and optimizes independently through the connector intelligent recommendation model and the connector parameter generation model, thereby improving the real-time response of environmental changes, enhancing the flexibility of the connector adaptation strategy, and effectively realizing the efficient, stable and intelligent dynamic adaptation of the connector in the complex multi-platform environment.

[0063] (2) In view of the technical problems in the existing connector intelligent recommendation model that the sensitivity to high error samples is insufficient and the weight updating strategy lacks flexibility, which is easy to fall into local optimum and slow training convergence, thereby resulting in low accuracy of the connector version type recommendation result, the scheme innovatively introduces a method of introducing a penalty term to construct an improved intelligent recommendation loss function and a dynamic weight adjustment mechanism, which can significantly improve the response ability and convergence efficiency of the model to abnormal samples, enhance the stability and adaptability of the recommendation model in complex and variable scenarios, thereby effectively improving the accuracy and reliability of the connector version recommendation result, and finally realizing the intelligent recommendation capability of the connector for the multi-platform heterogeneous environment, significantly improving the dynamic adaptation efficiency of the connector in the complex operating environment and the flexibility of cross-platform deployment.

[0064] (3) In view of the technical problems in the existing connector parameter generation model that the hyperparameters of the model are not set reasonably, resulting in unstable and inaccurate model output, and further leading to insufficient dynamic adaptation capability of the multi-platform connector, the scheme innovatively introduces an adaptive balance factor, a geometric migration mechanism and a simplified jump search strategy to improve the optimization algorithm, globally optimizes the hyperparameters in the generation model, and obtains the optimal parameter combination of the model, thereby effectively improving the stability and accuracy of the output results of the connector parameter generation model, and finally realizing the efficient and intelligent adaptation capability of the connector in the multi-platform environment. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A flowchart of an intelligent multi-platform connector dynamic adaptation method provided by the application;

[0066] Figure 2 A flowchart of step S3;

[0067] Figure 3 A flowchart of step S4;

[0068] Figure 4Flowchart for step S41;

[0069] Figure 5 Flowchart for step S42;

[0070] Figure 6 Flowchart for step S5;

[0071] Figure 7 Flowchart for step S52;

[0072] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application together with the embodiments thereof, and explain the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0073] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0074] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the system or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0075] Embodiment one, refer to Figure 1 The technical solutions adopted by the present application are as follows: The present application provides an intelligent multi-platform connector dynamic adaptation method, which comprises the following steps:

[0076] Step S1: Multi-source adaptation data collection, through data acquisition operation, obtain multi-platform connector adaptation original data;

[0077] Step S2: Data optimization processing, through data cleaning, data format standard processing, data encoding processing and connector bidirectional end label specification processing, obtain multi-platform connector adaptation optimized data;

[0078] Step S3: Bidirectional environmental change monitoring, by monitoring the connector running end environment and the connector target end environment respectively, generating running end total change value and target end total change value respectively, and sequentially performing first adaptation judgment, running end environment change judgment and target end environment change judgment, obtaining connector adaptation trigger result, and deciding whether to execute connector intelligent dynamic recommendation process according to the connector adaptation trigger result;

[0079] Step S4: Connector intelligent dynamic recommendation, by constructing a connector intelligent recommendation model including a connector data receiving layer, a primary feature extraction layer, a feature importance weighting layer, a deep feature fusion layer and a connector configuration recommendation output layer, combining historical data, training and optimizing the model by designing an intelligent recommendation loss function and model weight update, and finally inputting the real-time running end and target end environment feature combination into the optimized recommendation model to obtain the current recommended connector version type;

[0080] Step S5: Connector real-time dynamic adaptation, by constructing and training a connector parameter generation model based on the Transformer architecture, and through the improved optimization algorithm for model hyperparameter optimization, combining real-time connector context data and recommended version type to dynamically generate connector running configuration parameters, realizing intelligent connection establishment and dynamic adaptation with the target platform.

[0081] By performing the above operations, in view of the technical problems in the traditional multi-platform connector adaptation method that only rely on static preset parameters and one-way environmental perception for adaptation decision, lack of collaborative monitoring of running end and target end environment, which leads to poor stability, low compatibility and low adaptation efficiency of multi-platform connector in dynamic heterogeneous environment, the application innovatively introduces a linkage monitoring mechanism of connector running end and target end environment, and separates the connector version type recommendation and configuration parameter generation modeling, and optimizes independently through the connector intelligent recommendation model and the connector parameter generation model respectively, thereby improving the real-time response of environmental changes, enhancing the flexibility of the connector adaptation strategy, and effectively realizing the efficient, stable and intelligent dynamic adaptation of the connector in the complex multi-platform environment.

[0082] Embodiment two, refer to Figure 1The embodiment is based on the above embodiment, in step S1, the multi-source adaptation data is collected, which is used to collect the basic data required for multi-platform connector adaptation, specifically, data information is collected through the multi-platform connection management system to obtain multi-platform connector adaptation original data; the multi-platform connector adaptation original data includes historical connector adaptation data and real-time connector adaptation data, and the historical connector adaptation data and the real-time connector adaptation data both include connector running end data, connector context data and connector target end data; the historical connector adaptation data further includes connector adaptation results, connector version types and connector configuration parameters; the connector running end data is used for connector version type selection and includes running platform types, operating system versions, processor architectures, device performance parameters, network types and current system resource states, the connector context data is used for connector parameter configuration generation and includes connection success rates, packet loss rates, delays, reconnection times, message load indicators, quality of service requirements, connection keeping time lengths and system scheduling states, and the connector target end data is used for connector version type selection and includes target system types, communication protocol types and versions, authentication mechanisms, interface response capabilities, target system vendors and version requirements; the multi-platform specifically refers to a plurality of heterogeneous platform environments in which the connector can be deployed and run, including mobile operating system platforms, desktop or server operating system platforms, embedded system platforms, containerized running environments, edge computing nodes and Web browser environments; and the connector target end specifically refers to a target system or platform to which the connector needs to access, dock or communicate.

[0083] Embodiment three, refer to Figure 1 The embodiment is based on the above embodiment, in step S2, the data optimization processing is performed, specifically, the original data is optimized to obtain multi-platform connector adaptation optimized data; including the following steps:

[0084] Step S21: data cleaning, which is used to improve the quality and availability of the original adaptation data, specifically, missing value filling and outlier removal are performed on the data;

[0085] The missing value filling is used to keep the number of samples complete, specifically, the missing numerical value type field in the original data is completed by using the mean filling method; and the outlier removal is specifically based on the Z-Score method to identify and remove data records deviating from the statistical feature threshold;

[0086] Step S22: data format standard processing, which is used to unify the numerical scale and field format in the multi-source data, specifically, data format unification and numerical field standardization processing are included;

[0087] The data format is unified for structural alignment of heterogeneous data fields from different platforms or systems, specifically unified field type definition, and conversion of fields in different units to standard units;

[0088] The numerical field standardization processing specifically refers to processing the numerical type field by using the minimum-maximum normalization method, so that all numerical characteristics are uniformly mapped to the interval [0, 1];

[0089] Step S23: data encoding processing, for converting non-numerical information in the original data into a vector format acceptable to the model, specifically using a one-hot encoding method to encode the category field in the original data, and converting discrete text and label variables into sparse numerical vectors;

[0090] Step S24: connector bidirectional end label specification processing, for semantic unification of the connector running end platform identification field and the target end system field, specifically by establishing a platform type mapping table to uniformly map different platform identifications in different sources to standard platform categories, and performing label specification processing on the system type field of the connector target end.

[0091] Embodiment four, see Figure 1 and Figure 2 This embodiment is based on the above-mentioned embodiments, in step S3, the bidirectional environment change monitoring is used to monitor the dynamic changes of the connector running end and target end environment in real time, to ensure that the key state changes can be captured in time during the running process, so as to trigger the re-recommendation and dynamic adaptation mechanism of the connector, specifically connector running end environment monitoring, connector target end environment monitoring and connector dynamic adaptation triggering, to obtain the connector adaptation triggering result, including the following steps:

[0092] Step S31: connector running end environment monitoring, for real-time evaluation of the change intensity of the connector running end environment, specifically by using the Pearson correlation coefficient algorithm to extract key features of the connector running end data, to generate a combination of key features of the connector running end, for each connector running end key feature, based on its current value and the corresponding running end reference value, the relative change rate of the corresponding feature is calculated, and each connector running end key feature is assigned a running end weight, finally the change rate of each feature is multiplied by its corresponding weight and summed to calculate the total change value of the running end;

[0093] The running end reference value specifically refers to the median value of the running end key feature in the historical stable connection state;

[0094] The running end weight assignment method specifically refers to assigning weights to each feature by combining the actual influence degree of each key feature on the connector adaptation;

[0095] Step S32: connector target end environment monitoring, for real-time evaluation of the intensity of the change of the connector target end environment, specifically, key feature extraction is performed on the connector target end data through the Pearson correlation coefficient algorithm, a combination of key features of the connector target end is generated, for each key feature of the connector target end, based on the current value and the reference benchmark value of the target end, the relative change rate of the corresponding feature is calculated, and each key feature of the connector target end is assigned a target end weight, and finally the change rate of each feature is multiplied by the corresponding weight and summed to obtain the total change value of the target end;

[0096] The target end reference benchmark value specifically refers to the median value of the feature recorded in the past connection success state of the target end key feature;

[0097] The target end weight assignment method specifically refers to assigning a weight to each feature according to the influence degree of each target end feature on the connection stability and communication quality using a weighted scoring method;

[0098] Step S33: connector adaptation reconstruction trigger judgment, for judging whether the trigger condition of re-adaptation is reached according to the respective change degree of the running end and the target end environment during the operation of the connector, specifically, first adaptation judgment, running end environment change judgment and target end environment change judgment are sequentially performed to obtain the connector adaptation trigger result;

[0099] The first adaptation judgment is used to judge whether the current connector is in the first adaptation state, specifically, if it is the first adaptation, the running end environment change judgment and the target end environment change judgment processes are skipped, and the connector intelligent dynamic recommendation process is directly executed, otherwise the running end environment change judgment process is continued;

[0100] The running end environment change judgment specifically refers to if the running end total change value is greater than the running end threshold, it is determined that the current connector configuration is no longer suitable for the running end environment, then the re-adaptation operation of the connector is triggered, and the connector intelligent dynamic recommendation process is executed, otherwise the target end environment change judgment is continued;

[0101] The target end environment change judgment specifically refers to if the target end total change value is greater than the target end threshold, it is determined that the current connector configuration is no longer suitable for the target end environment, then the re-adaptation operation of the connector is triggered, and the connector intelligent dynamic recommendation process is executed, otherwise the current connector configuration remains unchanged.

[0102] Embodiment five, refer to Figure 1 , Figure 3 , Figure 4 and Figure 5 This embodiment is based on the above-mentioned embodiments, in step S4, the connector intelligent dynamic recommendation is used to realize intelligent recommendation of the connector version type, specifically including the following steps:

[0103] Step S41: constructing a connector intelligent recommendation model, specifically including the following steps:

[0104] Step S411, a connector data receiving layer, configured to receive a connector key feature combination generated by bidirectional environment change monitoring, specifically to splice a connector running end key feature combination and a connector target end key feature combination to obtain a connector intelligent recommendation feature matrix; the formula used is as follows:

[0105] ;

[0106] In the formula, represents the connector intelligent recommendation feature matrix, represents the connector running end key feature combination, represents the connector target end key feature combination;

[0107] Step S412, a primary feature extraction layer, configured to perform nonlinear conversion and basic feature extraction on the multi-source input running end and target end feature data to construct a preliminary mapping relationship between the input and the output, specifically to input the connector intelligent recommendation feature matrix into a primary full connection layer, perform linear weighting and bias summation operation, and introduce an activation function to realize feature nonlinear transformation to obtain a recommendation primary feature;

[0108] ;

[0109] In the formula, represents the recommendation primary feature, represents a weight matrix of the primary feature extraction layer, represents a bias parameter of the primary feature extraction layer;

[0110] Step S413, a feature importance weighting layer, configured to automatically identify and strengthen key feature dimensions that have a greater impact on connector configuration decisions among connector multi-dimensional adaptive features, specifically to input the recommendation primary feature, generate a query, key and value vector, and calculate a feature importance weight coefficient through a self-attention mechanism, and then weight and combine the recommendation primary feature to obtain a recommendation weighted feature; the formula used is as follows:

[0111] ;

[0112] ;

[0113] In the formula, represents the recommendation weighted feature, represents a query vector, represents a key vector, represents a value vector, a weight matrix representing trainable keys, a weight matrix representing trainable queries, a weight matrix representing trainable values, a dimension representing a key vector, a transpose operation;

[0114] Step S414, a deep feature fusion layer, is configured to further model high-order semantics and fuse deep features of the important features weighted by the attention mechanism, so as to extract potential connector configuration rules. Specifically, the recommendation weighted features are input into a second fully connected layer, and a linear weighting and bias summation operation is performed, and a An activation function is used to realize deep mapping of the recommendation weighted features, and a recommendation deep feature is obtained. The formula is as follows:

[0115] ;

[0116] In the formula, recommendation deep feature, weight matrix of the second fully connected layer, bias parameter of the second fully connected layer;

[0117] Step S415, a connector configuration recommendation output layer, is configured to generate a connector version recommendation result based on the high-order semantic features extracted by the deep feature fusion layer. Specifically, the recommendation deep feature is input into the output layer, and a linear combination operation is performed with the weight matrix and the bias term, and a Softmax activation function is introduced to obtain a connector version type recommendation result.

[0118] ;

[0119] In the formula, connector version type recommendation result, weight matrix of the output layer, bias parameter of the output layer;

[0120] Step S42: connector intelligent recommendation model training, is configured to optimize the parameters of the connector intelligent recommendation model through training, so that the model can sufficiently learn the relationship between historical connector adaptation behaviors and real-time running environments. Specifically, the following steps are included:

[0121] Step S421, designing an intelligent recommendation loss function, is configured to improve the sensitivity of the model to high-error samples and enhance the fitting performance. Specifically, a penalty term is introduced to construct an improved intelligent recommendation loss function. The formula is as follows:

[0122] ;

[0123] In the formula, This represents the value of the intelligent recommendation loss function. This represents the connector version type recommendation result for the i-th sample. This represents the true result of the connector version type for the i-th sample. Indicates the error tolerance threshold. represents the penalty weight coefficient, I(·) represents the indicator function, which is 1 when the error exceeds the threshold and 0 otherwise, and N represents the number of samples;

[0124] Step S422, model weight update, is used to adaptively adjust the network weights based on the model error results during each training round to optimize the model parameters; specifically, a dynamic weight adjustment mechanism is uniformly applied to the weights of each layer of the model, adaptively adjusting the learning rate and weight gradient according to the error change trend of the current round, and updating the weight parameters of each layer of the model; the formula used is as follows:

[0125] ;

[0126] ;

[0127] In the formula, Indicates the first Iterative weight change This represents the learning rate adjustment factor. This represents the current learning rate. Indicates the current momentum coefficient. Represents the weight gradient. Indicates the first Iterative weight change Indicates the first The amount of error change during iteration Indicates the first Iterative model weights, Indicates the first The weight values ​​of the next iteration model;

[0128] Step S423, model training terminates, used to determine whether the termination condition is met during model training. Specifically, in each round of iterative training, based on the key feature combination of the connector running end and the key feature combination of the connector target end generated by the historical connector adaptation data, the model training data is used. The weight parameters of all layers of the model are updated through the designed intelligent recommendation loss function and model weight update to obtain the iterative correction value of the weight of each layer. The intelligent recommendation loss function value under the current training round is calculated. If the loss value converges to the preset error termination threshold or the number of model training iterations reaches the maximum number of training iterations, the model iterative training is stopped and the optimized connector intelligent recommendation model is obtained. Otherwise, iterative training continues.

[0129] In step S43, the connector version type real-time recommendation is performed, specifically, the connector running end key feature combination and the connector target end key feature combination generated by the real-time connector adaptation data are taken as input data of the optimized connector intelligent recommendation model, and a currently recommended connector version type is obtained.

[0130] By performing the above operations, the technical problems of the existing connector intelligent recommendation model, such as lack of sensitivity to high error samples and lack of flexibility in weight update strategy, which easily fall into local optimum and slow training convergence, thereby leading to low accuracy of the connector version type recommendation result, are solved. The method of introducing a penalty term to construct an improved intelligent recommendation loss function and a dynamic weight adjustment mechanism is proposed, which can significantly improve the response ability and convergence efficiency of the model to abnormal samples, enhance the stability and adaptability of the recommendation model in complex and variable scenarios, thereby effectively improving the accuracy and reliability of the connector version recommendation result, and finally realizing the intelligent recommendation capability of the connector for the multi-platform heterogeneous environment, and significantly improving the dynamic adaptation efficiency of the connector in the complex running environment and the flexibility of cross-platform deployment.

[0131] Embodiment six, refer to Figure 1 、 Figure 6 and Figure 7 This embodiment is based on the above-mentioned embodiments. In step S5, the connector real-time dynamic adaptation is performed, specifically including the following steps:

[0132] In step S51, a connector parameter generation model is constructed and trained, which is used to construct a neural network model for dynamically generating connector configuration parameters, and complete the training process of the model. Specifically, a connector parameter generation model is established based on the Transformer architecture, and the model training is performed based on the connector context data and the connector version type in the historical connector adaptation data as training input data, to obtain the trained connector parameter generation model.

[0133] The model training process specifically includes training the generation model in a supervised learning manner, taking the actual configuration parameters of the connector as the supervision label, taking the mean square error loss function as the error measurement index, and iteratively updating the weight parameters in the model network through the back propagation algorithm, gradually reducing the prediction error, gradually reducing the error between the prediction output and the real configuration, and realizing the optimal adjustment of the generation model weight.

[0134] In step S52, the model hyperparameters of the generation model are optimized, which is used to improve the model performance of the connector parameter generation model. Specifically, an improved optimization algorithm is used to globally optimize the hyperparameters of the connector parameter generation model, to obtain the optimal hyperparameter combination of the connector parameter generation model, and finally obtain the optimized connector parameter generation model, including the following steps:

[0135] Step S521: Initialize the search population individuals. Specifically, the hyperparameters of the connector parameter generation model are encoded into search individual position vectors, and N search individual position vectors are generated through random initialization. Each individual is encoded to represent a candidate parameter combination, forming the initial search population.

[0136] Step S522: Calculate the fitness value of an individual, specifically by calculating the fitness value of the individual being searched in the population. The performance of the connector parameter generation model based on the location of the search individual is used as the fitness value of the search individual.

[0137] Step S523: Calculate the search regulation parameters, specifically by calculating the adaptive balance factor based on the exponential decay function and combining it with the geometric migration mechanism to calculate the individual's movement distance; the formula used is as follows:

[0138] ;

[0139] ;

[0140] In the formula, Indicates the first The adaptive balance factor of the generation population Indicates the current iteration number. Indicates the maximum number of iterations. This represents the adaptive balance factor for initializing the population, with a value range of... between, Indicates that the i-th individual is in the first place. The distance traveled by the generation Indicates the movement control factor, which is in Random numbers within a range that follow a normal distribution. This indicates the position of the optimal individual in the current iteration. Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population;

[0141] Step S524: Position update during the multi-directional exploration phase, specifically when... At that time, a multi-directional search position update strategy is adopted, and the formula used is as follows:

[0142] ;

[0143] In the formula, Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population, where j represents the index of the dimension. , and They all said Random numbers uniformly distributed within a range represents a random dimension position of a random individual in the jth generation population, represents a random dimension position of a random individual in the jth generation population, represents a random dimension position of the ith individual in the jth generation population, represents a random dimension position of the ith individual in the jth generation population, represents a random dimension, represents an even number, represents an odd number;

[0144] Step S525: local stage position update, specifically, when , a simplified jump search strategy is used for local position update; the formula used is as follows:

[0145] ;

[0146] ;

[0147] In the formula, rand and rand represent random numbers uniformly distributed in the range of and , rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population, rand represents a random individual in the jth dimension position of the jth generation population,

[0148] Step S526: obtaining the optimal position, specifically, after each round of iteration, the fitness values of all individuals in the current population are re-evaluated, and if the fitness of the position of an individual is better than that of the current global optimal individual position, the global optimal individual position is updated with the position of the individual;

[0149] Step S527: search iteration termination, specifically, when the fitness value of the search individual is higher than the fitness threshold and the maximum number of iterations is reached, the search is terminated and the global optimal individual position is obtained, and the global optimal individual position specifically refers to the optimal hyperparameter combination of the connector parameter generation model;

[0150] Step S528: model performance optimization, specifically, based on the optimal hyperparameter combination, the hyperparameters of the trained connector parameter generation model are adjusted to obtain an optimal connector parameter generation model;

[0151] Step S53: connector parameter real-time generation, for generating corresponding running configuration parameters on the basis of the recommended connector version type in combination with the current connection environment, specifically, taking the connector context data in the real-time connector adaptation data and the currently recommended connector version type as input data of an optimal connector parameter generation model to obtain the connector running configuration parameters;

[0152] The connector running configuration parameters include a QoS level, a maximum reconnection number, a reconnection interval time, a connection keep-alive time, an authentication mode, a communication protocol type, a message format type, and a maximum number of concurrent connections.

[0153] Step S54: connector dynamic adaptation, for realizing dynamic deployment and connection establishment of the connector based on the recommendation result and the generated parameters, specifically, based on the currently recommended connector version type and the connector running configuration parameters, dynamically loading and initializing the connector components of the recommended version type, performing running configuration parameter injection, completing the connection establishment process with the target system and platform, and thus realizing intelligent and dynamic adaptive deployment of the connector in a multi-platform heterogeneous environment.

[0154] By performing the above operations, in view of the technical problem that the hyperparameter settings in the existing connector parameter generation model are unreasonable, resulting in unstable and insufficiently accurate model output results, and further resulting in insufficient multi-platform connector dynamic adaptation capability, the scheme innovatively introduces an adaptive balance factor, a geometric migration mechanism, and a simplified jump search strategy to improve and optimize the algorithm, globally optimizes the hyperparameters in the generation model, and obtains the optimal parameter combination of the model, thereby effectively improving the stability and accuracy of the output results of the connector parameter generation model, and finally realizing efficient and intelligent adaptation capability of the connector in a multi-platform environment.

[0155] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0156] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, alternatives, and variations can be made thereto without departing from the principles and spirit of the application.

[0157] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. An intelligent multi-platform connector dynamic adaptation method, characterized in that: The method includes the following steps: Step S1: Multi-source adaptation data collection. Through data acquisition operations, obtain the raw data for multi-platform connector adaptation. Step S2: Data optimization processing. Through data cleaning, data format standardization processing, data encoding processing, and connector bidirectional end label specification processing, multi-platform connector adaptation optimization data is obtained. Step S3: Bidirectional environmental change monitoring. By monitoring the environment at the connector's operating end and the environment at the connector's target end respectively, the total change value at the operating end and the total change value at the target end are generated. The initial adaptation determination, the change judgment at the operating end environment, and the change judgment at the target end environment are performed in sequence to obtain the connector adaptation trigger result. Based on the connector adaptation trigger result, it is determined whether to execute the intelligent dynamic recommendation process for the connector. Specifically, if the initial adaptation is successful, the intelligent dynamic recommendation process for the connector is executed directly; otherwise, the change judgment process at the operating end environment continues. Step S4: Connector intelligent dynamic recommendation. By constructing a connector intelligent recommendation model, and training the connector intelligent recommendation model by introducing a penalized intelligent recommendation loss function and a dynamic weight adjustment mechanism for model weight update, the real-time data is finally input into the optimized recommendation model to obtain the current recommended connector version type. Step S5: connector real-time dynamic adaptation, by constructing and training a connector parameter generation model based on a Transformer architecture, and performing generation model hyperparameter optimization to obtain an optimal hyperparameter combination, finally obtaining an optimized connector parameter generation model, combining real-time connector context data and recommended version type to dynamically generate connector running configuration parameters, and realizing intelligent connection establishment and dynamic adaptation with the target platform; in the generation model hyperparameter optimization process, based on an adaptive balance factor When , a multi-directional search position updating strategy is adopted, and when , a simplified jump search strategy is adopted for local position updating to obtain the updated position of the individual in the next generation population, wherein, represents the current iteration number, represents the maximum iteration number, represents an adaptive balance factor of the initialized population, and the value range is .

2. The intelligent multi-platform connector dynamic adaptation method of claim 1, wherein: In step S3, the bidirectional environmental change monitoring specifically includes the following steps: Step S31: Connector operating end environment monitoring, specifically, extracting key features from connector operating end data using the Pearson correlation coefficient algorithm to generate a combination of key features for connector operating end. For each key feature of connector operating end, the relative rate of change of the corresponding feature is calculated based on its current value and the corresponding operating end reference value. Operating end weights are assigned to each key feature of connector operating end. Finally, the total change value of the operating end is calculated by multiplying the rate of change of each feature with its corresponding weight and summing the results. Step S32: Connector target end environment monitoring, specifically, extracting key features from connector target end data using the Pearson correlation coefficient algorithm to generate a combination of key features for connector target end. For each key feature of connector target end, the relative rate of change of the corresponding feature is calculated based on its current value and the target end reference value. Target end weights are assigned to each key feature of connector target end. Finally, the total change value of the target end is calculated by multiplying the rate of change of each feature with its corresponding weight and summing the results. Step S33: Connector adaptation and reconstruction trigger judgment, specifically, the first adaptation judgment, the runtime environment change judgment, and the target environment change judgment are performed in sequence to obtain the connector adaptation trigger result.

3. The intelligent multi-platform connector dynamic adaptation method of claim 1, wherein: In step S4, the intelligent dynamic recommendation of connectors specifically includes the following steps: Step S41: Construct a connector intelligent recommendation model, which specifically includes the following steps: Step S411, the connector data receiving layer, specifically involves combining the key feature combination of the connector operating end with the key feature combination of the connector target end to obtain the connector intelligent recommendation feature matrix; Step S412, a primary feature extraction layer, specifically inputting the connector intelligent recommendation feature matrix into a primary full connection layer, performing linear weighting and bias summation operation, introducing an activation function to realize feature nonlinear transformation, and obtaining recommended primary features; Step S413, a feature importance weighting layer, specifically taking the recommended primary features as input, generating query, key and value vectors, calculating feature importance weight coefficients through a self-attention mechanism, and then weighting and combining the recommended primary features to obtain recommended weighted features; Step S414, a deep feature fusion layer, specifically, inputting the recommendation weighted feature into a second fully connected layer, performing linear weighting and bias summation operation, and introducing The activation function realizes deep mapping of the recommendation weighted feature, and obtains a recommendation deep feature. Step S415, a connector configuration recommendation output layer, specifically inputting the recommended deep features into the output layer, performing linear combination operation with the weight matrix and the bias term, and introducing a Softmax activation function to obtain the connector version type recommendation result; Step S42: training of the connector intelligent recommendation model; Step S43, real-time recommendation of the connector version type, specifically taking the generated connector running end key feature combination and the connector target end key feature combination as input data of the optimized connector intelligent recommendation model through real-time connector adaptation data to obtain the currently recommended connector version type.

4. The intelligent multi-platform connector dynamic adaptation method of claim 3, wherein: In step S42, the connector intelligent recommendation model training specifically includes the following steps: Step S421, designing an intelligent recommendation loss function, specifically introducing a penalty term to construct an improved intelligent recommendation loss function, and the formula used is as follows: ； In the formula, represents the intelligent recommendation loss function value, represents the connector version type recommendation result of the i th sample, represents the connector version type true result of the i th sample, represents the error tolerance threshold, represents the penalty weight coefficient, represents an indicator function, 1 if the error exceeds the threshold, otherwise 0, and N represents the number of samples. Step S422, model weight updating, specifically uniformly applying a dynamic weight adjustment mechanism to the weights of each layer of the model, adaptively adjusting the learning rate and weight gradient according to the error change trend of the current round, and updating the weight parameters of each layer of the model; the formula used is as follows: ； ； wherein denotes the iteration denotes the iteration denotes a learning rate adjustment factor, denotes the current learning rate, denotes the current momentum coefficient, denotes the weight gradient, denotes the iteration denotes the iteration denotes the iteration denotes the error change in the iteration, denotes the iteration denotes the iteration model weight value, denotes the iteration denotes the iteration model weight value; Step S423, model training termination, specifically in each round of iterative training, taking the generated connector running end key feature combination and the connector target end key feature combination as model training data based on historical connector adaptation data, updating the weight parameters of all layers of the model through the designed intelligent recommendation loss function and model weight updating, obtaining the iterative correction value of each layer weight, and calculating the intelligent recommendation loss function value under the current training round, if the intelligent recommendation loss function value converges to a preset error termination threshold or the model training iteration number reaches a maximum training number, the model iterative training is stopped, and an optimized connector intelligent recommendation model is obtained, otherwise the iterative training is continued.

5. The intelligent multi-platform connector dynamic adaptation method of claim 1, wherein: In step S5, the connector real-time dynamic adaptation specifically includes the following steps: Step S51, constructing and training a connector parameter generation model, specifically establishing a connector parameter generation model based on the Transformer architecture, and taking the connector context data and the connector version type in the historical connector adaptation data as training input data to train the generation model, and obtaining the trained connector parameter generation model; Step S52, generation model hyperparameter optimization; Step S53: connector parameter real-time generation, specifically, taking the connector context data in the real-time connector adaptation data and the currently recommended connector version type as input data of the optimal connector parameter generation model, to obtain the connector running configuration parameter; Step S54: connector dynamic adaptation, specifically, based on the currently recommended connector version type and the connector running configuration parameter, dynamically loading and initializing the connector components of the recommended version type, performing running configuration parameter injection, completing the connection establishment process with the target system and platform, and thus realizing intelligent and dynamic adaptation deployment of the connector in a multi-platform heterogeneous environment.

6. The intelligent multi-platform connector dynamic adaptation method of claim 5, wherein: In step S52, the generation model hyperparameter optimization specifically includes the following steps: Step S521: initialize search population individuals, specifically, encode the connector parameter generation model hyperparameters into search individual position vectors, and generate N search individual position vectors through random initialization, each individual code representing a candidate parameter combination, to form an initial search population; Step S522: calculate the individual fitness value, specifically, calculate the search individual fitness value in the population ; the performance of the connector parameter generation model established based on the search individual position is taken as the search individual fitness value; Step S523: calculate search control parameters, specifically, calculate an adaptive balance factor based on an exponential decay function, and calculate individual movement distance based on a geometric migration mechanism; the formula used is as follows: ； wherein, represents the moving distance of the i-th individual in the j-th dimension in the g-th generation, represents the moving control factor, which is a random number obeying normal distribution in the range of [0, 1], represents the optimal individual position in the current iteration, represents the moving distance of the i-th individual in the j-th dimension in the g-th generation, represents the j-th dimension position of the i-th individual in the j-th dimension in the g-th generation, represents the j-th dimension position of the i-th individual in the j-th dimension in the g-th generation.​ Step S524: Multidirectional exploration phase position update. Specifically, when the multidirectional search position update strategy is used, the formula is as follows: ； In the formula, Indicates that the i-th individual is in the first place. The position of the j-th dimension in the population, where j represents the index of the dimension. , and They all said Random numbers uniformly distributed within a range Indicates that a random individual is in the th... Random dimension position in the population, Indicates that the i-th individual is in the first place. Random dimension position in the population, Indicates the random dimension. Represents even numbers, Indicates odd numbers; Step S525: Local stage position update, specifically, when the simplified jump search strategy is used for local position update; the formula used is as follows: ； ； In the formula, and They all said Random numbers uniformly distributed within a range Indicates that a random individual is in the th... The j-th dimension position in the population, Indicates the jump step length. Indicates that the i-th individual is in the first place. The j-th dimension position in the population, and Represents a random number that follows a normal distribution; Step S526: obtain the optimal position, specifically, after each iteration, re-evaluate the fitness values of all individuals in the current population, if the fitness of a certain individual position is better than that of the current global optimal individual position, update the global optimal individual position with the individual position; Step S527: search iteration termination, specifically, when the search individual fitness value higher than the fitness threshold value and the maximum iteration number is reached, the search is terminated and the global optimal individual position is obtained, and the global optimal individual position specifically refers to the optimal hyperparameter combination of the connector parameter generation model; Step S528: model performance optimization, specifically, based on the optimal hyperparameter combination, adjust the hyperparameters of the trained connector parameter generation model, to obtain the optimal connector parameter generation model.

7. The intelligent multi-platform connector dynamic adaptation method of claim 1, wherein: In step S1, the multi-source adaptation data collection, the multi-platform connector adaptation raw data includes historical connector adaptation data and real-time connector adaptation data, the historical connector adaptation data and the real-time connector adaptation data both include connector running end data, connector context data and connector target end data; the historical connector adaptation data further includes connector adaptation results, connector version types and connector configuration parameters.

8. The intelligent multi-platform connector dynamic adaptation method of claim 1, wherein: In step S2, the data optimization processing specifically includes the following steps: Step S21: data cleaning, specifically, missing value filling and outlier removal; Step S22: data format standardization processing, specifically, including data format unification and numerical field standardization processing; Step S23: data encoding processing, specifically, using one-hot encoding method to encode the category fields in the original data, to convert discrete text and label variables into sparse numerical vectors; Step S24: connector bidirectional end label specification processing, specifically, through establishing a platform type mapping table, mapping different platform identifications described in different source data into standard platform categories, and performing label specification processing on the system type field of the connector target end.

Images