Intelligent ai service scheduling method and system based on cloud data fusion

Abstract

The application relates to the technical field of cloud data intelligent AI service scheduling, in particular to an intelligent AI service scheduling method and system based on cloud data fusion, which generates a service scheduling directory by performing scheduling service feature analysis on historical scheduling cloud data of a computer room, generates an initial service index based on the service scheduling directory, obtains a scheduling service index by performing weight optimization through service optimization analysis and abnormal operation data analysis, then obtains a customized scheduling atlas by performing individualized scheduling optimization based on environment data, hardware data and historical service data of each server, obtains a fusion scheduling atlas by performing environment adaptability optimization according to the customized scheduling atlas, and performs intelligent service optimization according to the fusion scheduling atlas, so that the service matching precision is improved, the server operation characteristics are matched, and the situation that resources of the server are not allocated evenly is reduced.

Description

Technical Field

[0001] This invention relates to the field of cloud data intelligent AI service scheduling technology, specifically to an intelligent AI service scheduling method and system based on cloud data fusion. Background Technology

[0002] As AI service application scenarios continue to expand, the volume of cloud-based business data from different sources and with different structures continues to grow. Most existing scheduling solutions only allocate resources to a single data source or a fixed type of AI service, failing to effectively integrate and preprocess multi-source, heterogeneous cloud data. This easily leads to data gaps and low service matching accuracy during scheduling. Furthermore, existing scheduling mechanisms cannot dynamically adjust scheduling schemes based on real-time business load and AI service operating status. During peak periods, some service resources may become overloaded and blocked, while others remain idle and wasted. This significantly reduces the overall response efficiency and operational stability of AI services, making it difficult to meet the demands of massive concurrent access from users across multiple scenarios.

[0003] To address this, a method and system for intelligent AI service scheduling based on cloud-based data fusion are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent AI service scheduling method and system based on cloud data fusion. By establishing a fusion scheduling graph, the accuracy of AI service task resource allocation is improved and the occurrence of uneven resource allocation is reduced.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Intelligent AI service scheduling methods based on cloud-based data fusion include:

[0007] Obtain historical scheduling cloud data from the data center, and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog;

[0008] Optionally, the scheduling service features include at least periodic features and service type features;

[0009] The service type semantic data of the historical scheduling cloud data is obtained based on the semantic recognition model. The service type semantic data is then clustered using a clustering algorithm to obtain multiple service type clusters. The shortest distance from the service type semantic data to the cluster center of each service type cluster is calculated using the Euclidean distance algorithm. The service type semantic data with the shortest distance is used as the service cluster label of the service type cluster. The service type features of the historical scheduling cloud data are obtained based on the service cluster label.

[0010] Based on each of the service type features, the historical scheduling cloud data with the same service type feature is obtained. The historical scheduling cloud data with the same service type feature is periodically analyzed through trend analysis to obtain the trend feature of the historical scheduling cloud data. The trend feature includes at least periodic characteristics.

[0011] Cluster analysis is performed based on the periodic characteristics to obtain periodic clusters. The shortest distance from the periodic characteristics to the cluster center of each periodic cluster is calculated using the Euclidean distance algorithm. The periodic characteristics with the shortest distance are used as the periodic cluster labels of the periodic clusters. The periodic characteristics of the service type features are obtained based on the periodic cluster labels.

[0012] The service type feature is associated with the periodic feature using initial weights to obtain the service scheduling catalog.

[0013] This method generates a service scheduling catalog by pre-extracting scheduling service features from historical cloud scheduling data. This enables rapid matching of corresponding features when new service scheduling requests arrive, significantly shortening scheduling response time and improving the overall efficiency of cloud service scheduling. Simultaneously, by using semantic recognition and clustering to obtain service type and periodic features and weighted associations, it can accurately classify the scheduling demand attributes of different services, avoiding resource mismatch and allowing for more rational allocation of data center computing power, storage, and other resources, reducing ineffective resource consumption. Furthermore, feature extraction and catalog construction are based on historical cloud data, eliminating the need to modify existing data center scheduling hardware architecture, resulting in strong adaptability and lower deployment costs.

[0014] An initial service index is obtained by performing preliminary weight association based on the server's operating data and the service scheduling directory; and an optimized service index is obtained by performing service optimization analysis on the initial service index.

[0015] Optionally, an initial service index is obtained by initially associating all the server's running data and the service type features of the service scheduling directory with initial weights. Service type semantic data of the running data is extracted based on a semantic recognition model. The service type similarity between the service type semantic data of the running data and the service type features of the service scheduling directory is calculated using a cosine similarity function. The initial weights of the initial service indexes with service type similarity less than the service similarity threshold are disconnected, and the service optimization analysis is performed to obtain the optimized service index.

[0016] This method filters out low-match weighted connections by similarity screening, which can eliminate invalid matches and further compress redundant data in the service index, resulting in lower computational load and faster response speed for subsequent scheduling and matching. At the same time, it can accurately anchor the service scheduling item with the highest match degree with the current server running status, avoid irrelevant services occupying computing resources, further improve the accuracy of scheduling and matching, and ensure the stability and efficiency of cloud service scheduling.

[0017] Based on the running data of the optimized service index, abnormal running data is analyzed to obtain abnormal running data. Based on the abnormal running data, the optimized service index is subjected to secondary weight optimization to obtain the scheduling service index.

[0018] Optionally, the running data is segmented according to the periodic characteristics of the optimized service index to obtain periodic analysis data, and the periodic analysis data is analyzed for abnormal features according to an anomaly identification algorithm to obtain abnormal data label weights. The initial weights of the optimized service index are multiplied according to the abnormal data label weights to complete the secondary weight optimization and obtain the scheduling service index.

[0019] This method amplifies the weight ratio of anomaly-related service items through secondary weight optimization, making it easier for anomaly-related scheduling items to be prioritized and retrieved during matching, reducing the waiting time for anomaly responses. At the same time, it segments data based on the periodic characteristics of runtime data for anomaly identification, which can adapt to the periodic fluctuations of cloud service operation, avoid misjudging normal periodic fluctuations as anomalies, improve the accuracy of anomaly identification and subsequent anomaly scheduling, make the scheduling service index more in line with actual anomaly scheduling needs, and ensure the timeliness and accuracy of scheduling response in service anomaly scenarios.

[0020] An initial scheduling graph is obtained by associating the scheduling service index with each server with initial weights. Environmental data, hardware data, and historical service data of each server are obtained. The historical service data and hardware data are used to perform personalized scheduling optimization on the initial scheduling graph to obtain a customized scheduling graph. Based on the environmental data, the customized scheduling graph is optimized for environmental adaptability to obtain a fused scheduling graph.

[0021] Optionally, the initial scheduling graph is obtained by associating all the running data of the scheduling service index and the number data of each server through the initial weight of the servers.

[0022] Based on the personalized service optimization function, service personalization adaptation weights are obtained using the historical service data and the hardware data. Based on these service personalization adaptation weights, the initial server weights of the initial scheduling graph are optimized for personalized scheduling to obtain a customized scheduling graph. Based on the personalized service optimization function, the service personalization adaptation weights of the customized scheduling graph are optimized for environmental adaptability using the environmental data to obtain a fused scheduling graph. The expression for the personalized service optimization function is:

[0023] ;

[0024] ;

[0025] in, The personalized adaptation weights for the service after optimization for the aforementioned environment adaptability. The average operating temperature of the server. Minimum operating temperature for the server The highest operating temperature for the server. The adjustment factor for the denominator is... This represents the number of time windows for abnormal temperatures. Let F be the time window for the i-th abnormal temperature detected by the abnormal temperature detection function F. This represents the total length of the temperature-time window. The weights for personalized adaptation of the service are defined by Cos, which is the cosine similarity algorithm. The historical service data for the j-th server, The service type characteristics of the initial scheduling graph. The required hardware parameters for the service type characteristics, Here are the hardware parameters for the j-th server. The initial weight of the server.

[0026] This method introduces multi-dimensional weight adjustments based on hardware parameters, service type, and operating environment to accurately match the scheduling needs of different AI services. It avoids uneven server resource utilization caused by a single initial weight allocation, allowing servers with hardware performance better matched to service needs to undertake more scheduling tasks. It also reduces the risk of server overheating and downtime through dynamic adaptation of operating temperature, ultimately improving the overall scheduling stability and resource utilization of cloud-based AI services.

[0027] Service scheduling is performed on the server based on the fusion scheduling graph.

[0028] Based on the above method, this application also provides an intelligent AI service scheduling system based on cloud data fusion, including:

[0029] The service scheduling catalog module is used to obtain historical scheduling cloud data from the data center and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog.

[0030] The scheduling service index module is used to obtain an initial service index by performing preliminary weight association based on the server's running data and the service scheduling directory, to perform service optimization analysis on the initial service index to obtain an optimized service index, to perform abnormal running data analysis on the running data based on the optimized service index to obtain abnormal running data, and to perform secondary weight optimization on the optimized service index based on the abnormal running data to obtain a scheduling service index.

[0031] The fusion scheduling graph module is used to associate the scheduling service index with each server with initial weights to obtain an initial scheduling graph, acquire the environmental data, hardware data and historical service data of each server, perform personalized scheduling optimization on the initial scheduling graph using the historical service data and the hardware data to obtain a customized scheduling graph, and perform environmental adaptability optimization on the customized scheduling graph based on the environmental data to obtain a fusion scheduling graph.

[0032] The intelligent scheduling module performs service scheduling on the server based on the fused scheduling map.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] 1. This application constructs a service scheduling catalog by extracting features from historical scheduling data in advance, and then gradually optimizes the weight association by combining it with real-time server operation data to obtain a more adaptable fusion scheduling map. It can effectively match the hardware conditions and operating environment of different servers, improve the scheduling adaptation accuracy under abnormal conditions, and reduce the occupation of server resources by low-matching services. It solves the problems of unreasonable scheduling allocation and uneven resource utilization under concurrent requests of multiple AI services. After feature fusion and multi-level weight optimization based on cloud data, the overall scheduling response speed can be improved, and the rationality of the scheduling results is more suitable for the actual data center operation scenario. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method of the present invention;

[0036] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention provides an intelligent AI service scheduling method and system based on cloud data fusion, the technical solution of which is as follows:

[0039] Example 1:

[0040] A cloud-based data fusion-based intelligent AI service scheduling method, referring to Figure 1 As shown, it includes.

[0041] Obtain historical scheduling cloud data from the data center, and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog;

[0042] Optionally, the scheduling service features include at least periodic features and service type features;

[0043] Service type semantic data of the historical scheduling cloud data is obtained based on a semantic recognition model (e.g., BERT model). Multiple service type clusters are obtained by clustering the service type semantic data using a clustering algorithm (e.g., DBSCAN model). Service types include, for example, image-to-image, image-to-text, AI music, etc. The shortest distance from the service type semantic data to the cluster center of each service type cluster is calculated using the Euclidean distance algorithm. The service type semantic data with the shortest distance is used as the service cluster label. The service type features of the historical scheduling cloud data are obtained based on the service cluster labels. Specifically, the Euclidean distance between the coordinates of each service type semantic data and the coordinates of each cluster center is calculated, and all cluster centers are traversed to obtain the cluster set { , ,..., Find the minimum value among them. The corresponding semantic data is labeled as the service cluster label of the cluster. Finally, after all semantic data is labeled, the corresponding service type features are extracted from the historical scheduling cloud data.

[0044] Based on each service type feature, historical scheduling cloud data with the same service type feature is obtained. Periodic analysis is then performed on the historical scheduling cloud data with the same service type feature to obtain the trend feature of the historical scheduling cloud data. The trend feature includes at least periodic characteristics. Periodic analysis, for example, is a traditional additive model in decomposition methods. , This is the original time series data. As a trend, For a period of time, The residuals can also be decomposed using STL algorithms;

[0045] Cluster analysis is performed based on the periodic characteristics to obtain periodic clusters. The shortest distance from the periodic characteristics to the cluster center of each periodic cluster is calculated using the Euclidean distance algorithm. The periodic characteristics with the shortest distance are used as the periodic cluster labels of the periodic clusters. The periodic characteristics of the service type features are obtained based on the periodic cluster labels. The calculation process is the same as the calculation process of the service type features.

[0046] The service type feature is associated with the periodic feature using an initial weight to obtain the service scheduling directory. The initial weight defaults to 1, but can be set based on expert experience and the specific scenario.

[0047] This method generates a service scheduling catalog by pre-extracting scheduling service features from historical cloud scheduling data. This enables rapid matching of corresponding features when new service scheduling requests arrive, significantly shortening scheduling response time and improving the overall efficiency of cloud service scheduling. Simultaneously, by using semantic recognition and clustering to obtain service type and periodic features and weighted associations, it can accurately classify the scheduling demand attributes of different services, avoiding resource mismatch and allowing for more rational allocation of data center computing power, storage, and other resources, reducing ineffective resource consumption. Furthermore, feature extraction and catalog construction are based on historical cloud data, eliminating the need to modify existing data center scheduling hardware architecture, resulting in strong adaptability and lower deployment costs.

[0048] An initial service index is obtained by performing preliminary weight association based on the server's operating data and the service scheduling directory. An optimized service index is then obtained by performing service optimization analysis on the initial service index. The initial weight is 1 by default, but can be set according to expert experience and the scenario.

[0049] Optionally, an initial service index is obtained by associating all the server's running data and the service type features of the service scheduling directory with initial weights. Service type semantic data of the running data is extracted based on a semantic recognition model (e.g., BERT model). The service type similarity between the service type semantic data of the running data and the service type features of the service scheduling directory is calculated using a cosine similarity function. The initial weights of the initial service indexes with service type similarity less than a service similarity threshold are disconnected, and the service optimization analysis is performed to obtain the optimized service index. The service similarity threshold is set based on expert experience, such as 70%, 60%, 55%, 50%, 40%, etc. The operation of disconnecting the initial weights is as follows: for example, the default weight is 1, and the disconnected state is 0. Disconnecting means resetting the weight to 0. The program recognizes a weight of 0 and considers the data to be disconnected.

[0050] This method filters out low-match weighted connections by similarity screening, which can eliminate invalid matches and further compress redundant data in the service index, resulting in lower computational load and faster response speed for subsequent scheduling and matching. At the same time, it can accurately anchor the service scheduling item with the highest match degree with the current server running status, avoid irrelevant services occupying computing resources, further improve the accuracy of scheduling and matching, and ensure the stability and efficiency of cloud service scheduling.

[0051] Based on the running data of the optimized service index, abnormal running data is analyzed to obtain abnormal running data. Based on the abnormal running data, the optimized service index is subjected to secondary weight optimization to obtain the scheduling service index.

[0052] Optionally, the running data is segmented according to the periodic characteristics of the optimized service index to obtain periodic analysis data, and the periodic analysis data is analyzed for abnormal features according to anomaly identification algorithms (such as LSTM, Transformer, etc.) to obtain abnormal data label weights. The initial weights of the optimized service index are multiplied according to the abnormal data label weights to complete the secondary weight optimization and obtain the scheduling service index.

[0053] In one implementation, abnormal operation data includes abnormal data of normal operation and abnormal data of non-normal operation. Abnormal data of normal operation includes anomalies such as sudden drops in operating load and short-term network fluctuations that do not affect the overall service stability. Abnormal data of non-normal operation includes anomalies such as server hardware failure, core process crash, and bandwidth exhaustion that will directly cause service interruption. The weight values ​​of the abnormal data labels corresponding to the two types of anomalies are different and can be set according to expert experience. For example, server hardware failure is set to 0.5 and sudden drops in operating load is set to 1.5. This step is mainly to separate servers with abnormal operation and reduce their weight to ensure service quality. At the same time, idle servers are also treated as anomalies and their weight is increased so that other resources can be allocated to the server, thereby improving the allocation accuracy.

[0054] This method amplifies the weight ratio of anomaly-related service items through secondary weight optimization, making it easier for anomaly-related scheduling items to be prioritized and retrieved during matching, reducing the waiting time for anomaly responses. At the same time, it segments data based on the periodic characteristics of runtime data for anomaly identification, which can adapt to the periodic fluctuations of cloud service operation, avoid misjudging normal periodic fluctuations as anomalies, improve the accuracy of anomaly identification and subsequent anomaly scheduling, make the scheduling service index more in line with actual anomaly scheduling needs, and ensure the timeliness and accuracy of scheduling response in service anomaly scenarios.

[0055] An initial scheduling graph is obtained by associating the scheduling service index with each server with initial weights. Environmental data, hardware data, and historical service data of each server are obtained. The historical service data and hardware data are used to perform personalized scheduling optimization on the initial scheduling graph to obtain a customized scheduling graph. Based on the environmental data, the customized scheduling graph is optimized for environmental adaptability to obtain a fused scheduling graph.

[0056] Optionally, the initial scheduling graph is obtained by associating all the running data of the scheduling service index and the number data of each server through the initial weight of the servers.

[0057] Based on the personalized service optimization function, service personalization adaptation weights are obtained using the historical service data and the hardware data. Based on these service personalization adaptation weights, the initial server weights of the initial scheduling graph are optimized for personalized scheduling to obtain a customized scheduling graph. Based on the personalized service optimization function, the service personalization adaptation weights of the customized scheduling graph are optimized for environmental adaptability using the environmental data to obtain a fused scheduling graph. The expression for the personalized service optimization function is:

[0058] ;

[0059] ;

[0060] in, The personalized adaptation weights for the service after optimization for the aforementioned environment adaptability. The average operating temperature of the server. Minimum operating temperature for the server The highest operating temperature for the server. This is the denominator adjustment factor, which is mainly used to prevent the denominator from being zero. The specific value can be set based on expert experience. This represents the number of time windows for abnormal temperatures. Let F be the i-th time window of the abnormal temperature identified by the abnormal temperature identification function F. The abnormal temperature identification function F is mainly a time-series abnormality identification function, such as LSTM, Transformer, etc. This represents the total length of the temperature-time window. The weights for personalized adaptation of the service are defined by Cos, which is the cosine similarity algorithm. The historical service data for the j-th server, The service type characteristics of the initial scheduling graph. The required hardware parameters for the service type characteristics, Here are the hardware parameters for the j-th server. The initial weight of the server.

[0061] This method introduces multi-dimensional weight adjustments based on hardware parameters, service type, and operating environment to accurately match the scheduling needs of different AI services. It avoids uneven server resource utilization caused by a single initial weight allocation, allowing servers with hardware performance better matched to service needs to undertake more scheduling tasks. It also reduces the risk of server overheating and downtime through dynamic adaptation of operating temperature, ultimately improving the overall scheduling stability and resource utilization of cloud-based AI services.

[0062] Server service scheduling is performed based on the fusion scheduling graph. Specifically, real-time judgment is made in conjunction with real-time data and the fusion scheduling graph. Tasks of servers with pressure exceeding the threshold are redistributed according to the graph. The threshold can be set based on expert experience, or the graph can be used as a knowledge base. The allocation scheme is regenerated by the agent in conjunction with the graph, and resources are configured according to the regenerated scheme.

[0063] Example 2:

[0064] This application also provides an intelligent AI service scheduling system based on cloud data fusion, based on the method of Embodiment 1, with reference to... Figure 2 As shown, it includes:

[0065] The service scheduling catalog module is used to obtain historical scheduling cloud data from the data center and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog.

[0066] The scheduling service index module is used to obtain an initial service index by performing preliminary weight association based on the server's running data and the service scheduling directory, to perform service optimization analysis on the initial service index to obtain an optimized service index, to perform abnormal running data analysis on the running data based on the optimized service index to obtain abnormal running data, and to perform secondary weight optimization on the optimized service index based on the abnormal running data to obtain a scheduling service index.

[0067] The fusion scheduling graph module is used to associate the scheduling service index with each server with initial weights to obtain an initial scheduling graph, acquire the environmental data, hardware data and historical service data of each server, perform personalized scheduling optimization on the initial scheduling graph using the historical service data and the hardware data to obtain a customized scheduling graph, and perform environmental adaptability optimization on the customized scheduling graph based on the environmental data to obtain a fusion scheduling graph.

[0068] The intelligent scheduling module performs service scheduling on the server based on the fused scheduling map.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for scheduling intelligent AI services based on cloud-based data fusion, characterized in that: include: Obtain historical scheduling cloud data from the data center, and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog; An initial service index is obtained by performing preliminary weight association based on the server's operating data and the service scheduling directory. An optimized service index is obtained by performing service optimization analysis on the initial service index. An abnormal operating data is obtained by performing abnormal operating data based on the operating data of the optimized service index. A scheduling service index is obtained by performing secondary weight optimization on the optimized service index based on the abnormal operating data. An initial scheduling graph is obtained by associating the scheduling service index with each server with initial weights. Environmental data, hardware data, and historical service data of each server are obtained. The historical service data and hardware data are used to perform personalized scheduling optimization on the initial scheduling graph to obtain a customized scheduling graph. Based on the environmental data, the customized scheduling graph is optimized for environmental adaptability to obtain a fused scheduling graph. Service scheduling is performed on the server based on the fusion scheduling graph.

2. The intelligent AI service scheduling method based on cloud data fusion according to claim 1, characterized in that, Obtain historical scheduling cloud data from the data center, and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog, including: The scheduling service features include at least periodic features and service type features; The service type semantic data of the historical scheduling cloud data is obtained based on the semantic recognition model. The service type semantic data is then clustered using a clustering algorithm to obtain multiple service type clusters. The shortest distance from the service type semantic data to the cluster center of each service type cluster is calculated using the Euclidean distance algorithm. The service type semantic data with the shortest distance is used as the service cluster label of the service type cluster. The service type features of the historical scheduling cloud data are obtained based on the service cluster label. Based on each of the service type features, the historical scheduling cloud data with the same service type feature is obtained. The historical scheduling cloud data with the same service type feature is periodically analyzed through trend analysis to obtain the trend feature of the historical scheduling cloud data. The trend feature includes at least periodic characteristics. Cluster analysis is performed based on the periodic characteristics to obtain periodic clusters. The shortest distance from the periodic characteristics to the cluster center of each periodic cluster is calculated using the Euclidean distance algorithm. The periodic characteristics with the shortest distance are used as the periodic cluster labels of the periodic clusters. The periodic characteristics of the service type features are obtained based on the periodic cluster labels. The service type feature is associated with the periodic feature using initial weights to obtain the service scheduling catalog.

3. The intelligent AI service scheduling method based on cloud data fusion according to claim 2, characterized in that, An initial service index is obtained by performing preliminary weight association based on the server's operating data and the service scheduling directory. An optimized service index is then obtained by performing service optimization analysis on the initial service index, including: An initial service index is obtained by associating all the server's operational data and the service type features of the service scheduling directory with initial weights. Service type semantic data of the operational data is extracted based on a semantic recognition model. The service type similarity between the service type semantic data of the operational data and the service type features of the service scheduling directory is calculated using a cosine similarity function. The initial weights of the initial service indexes with service type similarity less than the service similarity threshold are disconnected, and the service optimization analysis is performed to obtain the optimized service index.

4. The intelligent AI service scheduling method based on cloud data fusion according to claim 2, characterized in that, Based on the operational data of the optimized service index, abnormal operational data is obtained through abnormal operation data analysis. A secondary weight optimization is then performed on the optimized service index based on this abnormal operational data to obtain the scheduling service index, which includes: The running data is segmented according to the periodic characteristics of the optimized service index to obtain periodic analysis data. Anomaly feature analysis is performed on the periodic analysis data according to the anomaly identification algorithm to obtain anomaly data label weights. The initial weights of the optimized service index are multiplied according to the anomaly data label weights to complete the secondary weight optimization and obtain the scheduling service index.

5. The intelligent AI service scheduling method based on cloud data fusion according to claim 1, characterized in that, An initial scheduling graph is obtained by associating the scheduling service index with each server using initial weights. Environmental data, hardware data, and historical service data for each server are then acquired. Based on the historical service data and hardware data, the initial scheduling graph is optimized for personalized scheduling to obtain a customized scheduling graph. Finally, based on the environmental data, the customized scheduling graph is optimized for environmental adaptability to obtain a fused scheduling graph, including: The initial scheduling graph is obtained by associating all the running data of the scheduling service index and the number data of each server through the initial weight of the servers. Based on the personalized service optimization function, service personalization adaptation weights are obtained using the historical service data and the hardware data. Based on these service personalization adaptation weights, the initial server weights of the initial scheduling graph are optimized for personalized scheduling to obtain a customized scheduling graph. Based on the personalized service optimization function, the service personalization adaptation weights of the customized scheduling graph are optimized for environmental adaptability using the environmental data to obtain a fused scheduling graph. The expression for the personalized service optimization function is: ； ； in, The personalized adaptation weights for the service after optimization for the aforementioned environment adaptability. The average operating temperature of the server. Minimum operating temperature for the server The highest operating temperature for the server. The adjustment factor for the denominator is... This represents the number of time windows for abnormal temperatures. Let F be the time window for the i-th abnormal temperature detected by the abnormal temperature detection function F. This represents the total length of the temperature-time window. The weights for personalized adaptation of the service are defined by Cos, which is the cosine similarity algorithm. The historical service data for the j-th server, The service type characteristics of the initial scheduling graph. The required hardware parameters for the service type characteristics, Here are the hardware parameters for the j-th server. The initial weight of the server.

6. An intelligent AI service scheduling system based on cloud-based data fusion, characterized in that: include: The service scheduling catalog module is used to obtain historical scheduling cloud data from the data center and extract scheduling service features based on the historical scheduling cloud data to generate a service scheduling catalog. The scheduling service index module is used to obtain an initial service index by performing preliminary weight association based on the server's running data and the service scheduling directory, to perform service optimization analysis on the initial service index to obtain an optimized service index, to perform abnormal running data analysis on the running data based on the optimized service index to obtain abnormal running data, and to perform secondary weight optimization on the optimized service index based on the abnormal running data to obtain a scheduling service index. The fusion scheduling graph module is used to associate the scheduling service index with each server with initial weights to obtain an initial scheduling graph, acquire the environmental data, hardware data and historical service data of each server, perform personalized scheduling optimization on the initial scheduling graph using the historical service data and the hardware data to obtain a customized scheduling graph, and perform environmental adaptability optimization on the customized scheduling graph based on the environmental data to obtain a fusion scheduling graph. The intelligent scheduling module performs service scheduling on the server based on the fused scheduling map.

Images