Cloud platform operation and maintenance method and system

Abstract

The invention provides a cloud platform operation and maintenance method and system, and relates to the technical field of data processing, and the method comprises the steps: obtaining multi-dimensional data of a cloud platform, integrating the multi-dimensional data of the cloud platform into a high-order tensor based on a Tuck decomposition method, carrying out the non-uniform sampling of the high-order tensor, and constructing a high-order tensor structure of the cloud platform; generating a space-time multi-scale topological graph of the cloud platform based on the high-order tensor structure of the cloud platform; constructing a dynamic prediction model of the cloud platform based on the space-time multi-scale topological graph of the cloud platform; preprocessing the multi-dimensional data of the cloud platform, and sending a preprocessing result to the dynamic prediction model for prediction to obtain a state prediction result of the cloud platform in a preset time period; and obtaining a preset operation and maintenance scheme for operation and maintenance based on the state prediction result of the cloud platform in the preset time period. According to the method, the characteristics of long-range dependency and sudden fluctuation are combined, and resource scheduling and anomaly detection are automatically carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a cloud platform operation and maintenance method and system. Background Art

[0002] Currently, with the wide application of cloud computing, cloud platforms have become the key infrastructure for data storage, computing resources, and application hosting for many enterprises and individuals. Cloud platforms have greatly promoted the rapid development of the information age by providing virtualization technology, on-demand scalable computing power, and elastic storage. However, while cloud platforms provide efficient and reliable services, they also face increasingly complex operation and maintenance challenges. Existing cloud platform operation and maintenance technologies mainly rely on simple monitoring and statistical analysis methods, such as load balancing, capacity planning, and resource monitoring, and rely on manual operations or static algorithms for optimization and anomaly detection. These traditional methods often rely on fixed rules and thresholds and are difficult to adapt to the dynamic changes and load fluctuations of cloud platforms. Especially when facing complex network environments, dynamic resource consumption, and sudden load changes, problems such as resource waste, performance bottlenecks, or system crashes are likely to occur. Therefore, existing technologies are unable to cope when dealing with the complexity, dynamics, and multi-factor interactions of cloud platform operation and maintenance, and lack real-time and intelligent optimization means.

[0003] Therefore, there is an urgent need for a cloud platform operation and maintenance method and system to solve the above technical problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a cloud platform operation and maintenance method and system to solve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows: On the one hand, the present application provides a cloud platform operation and maintenance method, including: Obtaining multi-dimensional data of the cloud platform, where the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data, and abnormal event data; Integrating the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and performing non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform; Performing multi-scale topological graph construction and geometric mapping processing in the hyperbolic space on the high-order tensor structure of the cloud platform to generate a spatio-temporal multi-scale topological graph of the cloud platform; Constructing a non-linear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform, and combining it with the singular spectrum analysis method to obtain a dynamic prediction model of the cloud platform; Performing preprocessing on the multi-dimensional data of the cloud platform, and using the preprocessing result as input data to input into the dynamic prediction model of the cloud platform for prediction to obtain a state prediction result of the cloud platform in a preset time period; Based on the status prediction result of the cloud platform in a preset time period, obtain a preset operation and maintenance plan for operation and maintenance.

[0005] On the other hand, the present application also provides a cloud platform operation and maintenance system, including: An acquisition unit, configured to acquire multi-dimensional data of the cloud platform, where the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data, and abnormal event data; An integration unit, configured to integrate the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and perform non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform; A mapping unit, configured to perform multi-scale topological graph construction and geometric mapping processing in a hyperbolic space on the high-order tensor structure of the cloud platform to generate a spatio-temporal multi-scale topological graph of the cloud platform; A construction unit, configured to construct a non-linear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform, and combine it with the singular spectrum analysis method to obtain a dynamic prediction model of the cloud platform; A prediction unit, configured to preprocess the multi-dimensional data of the cloud platform, and use the preprocessing result as input data to input into the dynamic prediction model of the cloud platform for prediction to obtain a status prediction result of the cloud platform in a preset time period; An operation and maintenance unit, configured to obtain a preset operation and maintenance plan for operation and maintenance based on the status prediction result of the cloud platform in a preset time period.

[0006] The beneficial effects of the present invention are: By combining advanced algorithms such as fractional Brownian motion, non-linear delay differential equations, and multifractal detrended fluctuation analysis, the present invention can achieve accurate modeling and real-time analysis of the multi-dimensional data of the cloud platform. By inputting multi-dimensional data such as the load, resource consumption, and network bandwidth of the cloud platform into the dynamic prediction model, and combining the characteristics of long-range dependence and sudden fluctuations, the future state of the cloud platform can be accurately predicted, and resource scheduling and anomaly detection can be automatically performed. This method can dynamically adjust the anomaly detection threshold and optimization strategy according to the actual operation situation of the cloud platform, greatly improving the automation and intelligence level of the cloud platform operation and maintenance, avoiding the lag and static decision-making problems in the traditional technology, and providing more efficient, accurate, and flexible operation and maintenance support for the cloud platform.

[0007] Other features and advantages of the present invention will be described in the subsequent description, and, in part, will become apparent from the description, or will be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. Brief Description of the Drawings

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0009] Figure 1 It is a flowchart of the cloud platform operation and maintenance method of the present invention; Figure 2 It is a schematic diagram of the cloud platform operation and maintenance system of the present invention.

[0010] Markings in the figure: 701, acquisition unit; 702, integration unit; 703, mapping unit; 704, construction unit; 705, prediction unit; 706, operation and maintenance unit. Specific embodiments

[0011] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0012] It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance. Embodiment 1:

[0013] As Figure 1 shown, this embodiment provides a cloud platform operation and maintenance method, including steps S1, S2, S3, S4, S5 and S6.

[0014] Step S1, acquire multi-dimensional data of the cloud platform, and the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data and abnormal event data; It is understandable that this step not only involves real-time monitoring of various indicators of the cloud platform, but also requires ensuring the diversity and comprehensiveness of data, covering multiple dimensions including energy consumption, environmental changes, computing tasks, storage, and abnormal events. Specifically, energy consumption data reflects the energy consumption of the cloud platform when executing computing tasks, which is crucial for evaluating resource utilization efficiency and cost control; environmental change data focuses on external environmental factors such as climate change, temperature, and humidity, which affect the cooling efficiency, equipment performance, and stability of the data center of the cloud platform; computing task data involves the status, scheduling information, and load conditions of various computing tasks on the platform, reflecting the operating load and resource usage of the platform; storage data is related to the utilization of storage devices, storage bandwidth, data access frequency, etc. in the cloud platform, helping to analyze data mobility and storage bottlenecks; abnormal event data records various abnormalities that occur during the operation of the platform, such as failures, outages, resource allocation errors, etc., which are important bases for anomaly detection and fault diagnosis.

[0015] The acquisition of these multi-dimensional data not only involves obtaining static data through basic sensors or monitoring tools, but also involves continuous monitoring and real-time updating of dynamic data streams to ensure sufficient understanding of the state of the cloud platform and the ability to respond to rapidly changing workloads and environmental changes. Technically, real-time monitoring and big data analysis technologies are combined with the cloud platform architecture to form the ability to comprehensively perceive the platform. The richness and real-time nature of this information provide strong data support for subsequent anomaly detection, dynamic prediction, and adaptive optimization. The multi-dimensional data obtained through this step not only provides basic data support for subsequent data analysis and model establishment, but also provides a real-time and comprehensive data source for subsequent accurate prediction, optimization strategies, and adaptive operation and maintenance.

[0016] Step S2: Integrate the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and perform non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform. It is understandable that through Tucker decomposition and non-uniform sampling in this step, the data of the cloud platform is effectively integrated and optimally sampled, providing an accurate high-dimensional data representation for subsequent analysis and model training, while improving the efficiency of data processing and avoiding interference from redundant information to subsequent analysis. This method can efficiently capture the complex interactions between various dimensions in the cloud platform, reveal potential patterns and dependencies, and provide important data support for accurate prediction and optimization. In this step, step S2 includes step S21, step S22, step S23, and step S24.

[0017] Step S21: Represent the multi-dimensional data of the cloud platform as a product of a set of core tensors and multiple factor matrices based on the Tucker decomposition method, where each factor matrix represents the feature space of different-dimensional data, and obtain the multi-dimensional interaction features of the cloud platform. It can be understood that in this step, according to the multi-dimensional data of the cloud platform, these data are integrated into a high-order tensor. The data of each dimension (such as time, node, task type, resource usage, etc.) corresponds to an axis of the tensor. Since the multi-dimensional data of the cloud platform is usually a highly complex matrix with extremely large dimensions and scale, directly analyzing the original data would be extremely complex and costly in terms of computation. In this step, in order to effectively extract the interaction features in the data and reduce the computational complexity, the Tucker decomposition can simplify the high-dimensional data into multiple factor matrices and a core tensor. The Tucker decomposition decomposes the original multi-dimensional data into multiple factor matrices, where each factor matrix represents the feature space of different-dimensional data. For example, the energy consumption data is in one factor matrix, the computing task data is in another factor matrix, and the storage data is in another factor matrix. The core tensor reveals the interaction relationships between these dimensions and represents the interaction between different feature dimensions.

[0018] Each factor matrix represents the feature space of the data of one dimension. These matrices can be regarded as the "compressed representation" of each dimension, reducing the complexity of the original data by extracting the important features in that dimension. For example, for the energy consumption data in the cloud platform, the factor matrix contains the energy consumption patterns in different time periods; for the computing task data, the factor matrix represents the resource demand characteristics under different task types or load levels. Each factor matrix extracts the key features of the data of that dimension through matrix decomposition techniques (such as singular value decomposition, principal component analysis, etc.). During the Tucker decomposition process, these factor matrices will be multiplied by the core tensor to generate a high-order tensor, and the finally generated data representation can reveal the complex interaction features between each dimension. The core tensor is usually obtained by optimizing the product of the factor matrices and contains the complex relationships between various dimensions of the cloud platform (such as computing resources, storage, network, etc.). For example, the core tensor reveals the changing trend of energy consumption under high computing loads, or the mutual influence between task scheduling and storage bottlenecks.

[0019] The combination of the factor matrices and the core tensor obtained through the Tucker decomposition not only effectively reduces the dimension and simplifies the complexity of the data, but also can retain the key interaction features of the cloud platform while reducing the computational cost.

[0020] Step S22: Perform non-uniform sampling on the obtained multi-dimensional interaction features of the cloud platform, and screen out a data set that meets the preset conditions from the multi-dimensional interaction features. It is understandable that in this step, an appropriate sampling strategy is set according to the volatility and change rules of data in each dimension of the cloud platform. For example, when the load of the cloud platform increases sharply, the storage pressure increases, or the network bandwidth fluctuates, the system will increase the sampling frequency to capture key change information; when the load is low and the system is running smoothly, the sampling frequency can be appropriately reduced to avoid processing redundant data. Through non-uniform sampling, the cloud platform can dynamically adjust the sampling frequency according to the real-time state to ensure that the sampled data fully covers important time periods.

[0021] After non-uniform sampling, this step further performs data screening to select a data set that meets the preset conditions from the obtained multi-dimensional interaction features. These preset conditions are usually set based on the operation objectives of the cloud platform, anomaly detection requirements, and resource optimization strategies, such as threshold requirements for key metrics (load, energy consumption, etc.), or specific attention to system performance bottlenecks. Among them, these conditions can be defined based on historical data, system health status, and the requirements of specific tasks to select the most important part for subsequent analysis from the non-uniform sampling results.

[0022] Step S23: Construct a high-resolution spatio-temporal graph based on the data set that meets the preset conditions, where each functional node represents a specific function of the cloud platform, and each functional edge represents the interaction between different functional nodes; It is understandable that this step correlates the information related to each function of the cloud platform in the data set that meets the preset conditions, and each data point (such as load, energy consumption, etc.) corresponds to a specific functional node. For example, the data of computing tasks is related to the "computing node" function, and the stored data is related to the "storage node" function. Each functional node will extract corresponding metrics (such as resource consumption, performance, etc.) according to its actual role and status in the cloud platform. These nodes will serve as the basic building blocks of the graph, representing the core functional modules of the cloud platform.

[0023] In the spatio-temporal graph, each functional edge represents the interaction between different functional nodes, and these edges reveal the dependency relationships or interaction effects between each functional module. For example, the load fluctuation of computing resources affects the usage efficiency of storage resources, and the execution of computing tasks triggers bandwidth demand fluctuations, etc. This step defines the interaction edges between nodes according to the multi-dimensional data of the cloud platform (such as task scheduling, resource allocation, data flow, etc.). The weight of each edge reflects the interaction intensity and dependency between different functional nodes. The weight of the edge can be determined by calculating the similarity or dependency between two nodes, such as based on metrics such as load fluctuation, data transmission volume, and resource consumption.

[0024] Next, based on the high-frequency and real-time changing data collected in this step, the nodes and edges of the spatio-temporal graph will be updated and adjusted according to the actual system operating state at different time periods. For example, during the peak load period, the interaction edges between the computing nodes and the storage nodes will be weighted and adjusted to reflect the resource tension of the cloud platform. The high-resolution spatio-temporal graph will continuously capture the changes in the interaction relationships between the nodes of the cloud platform and update the graph structure in real time. By constructing a high-resolution spatio-temporal graph, the functional modules of the cloud platform and their interaction relationships can be accurately expressed. This graph not only reflects the usage of internal resources of the cloud platform but also can dynamically adapt to the changes in different time periods.

[0025] Step S24: Based on the tensor construction technology, the data sets in different dimensions that meet the preset conditions in the high-resolution spatio-temporal graph are merged together to form a high-order tensor structure.

[0026] It can be understood that in this step, the functional modules and their interaction relationships represented by the nodes and edges in the graph are extracted, and this data will serve as different dimensions of the tensor. Each dimension can represent a specific function, such as computing nodes, storage nodes, network nodes, etc.; each edge represents the interaction between different nodes. The nodes and edges that meet the preset conditions (for example, key nodes such as periods with large load fluctuations, storage bottlenecks, etc.) will be screened out for subsequent processing. Then, through the tensor construction technology, the feature spaces of the data in each dimension are integrated to form a high-order tensor. Each dimension can represent a functional node of the cloud platform or its state, and each tensor element stores the specific interaction or state information between the functional modules of the cloud platform. For example, a tensor contains dimensions of "computing nodes", "storage nodes", and "bandwidth usage", and each tensor element corresponds to the interaction situation or state characteristics between these functional nodes at a certain moment.

[0027] Next, this step performs normalization processing on the data in each dimension of the high-order tensor so that the data in each dimension can be compared under the same dimension. By removing outliers, normalizing the data range, etc., it is ensured that each element of the tensor can truly reflect the changes and interaction effects of the cloud platform state. After normalization, the multi-dimensional structure of the data set will be more in line with the requirements of subsequent analysis and modeling. By fusing the data in the high-resolution spatio-temporal graph into a high-order tensor structure based on the tensor construction technology, the multi-dimensional data of the cloud platform can be effectively integrated into a compact and information-rich data representation. This structure not only effectively captures the interaction relationships between the functional modules of the cloud platform but also can further explore the potential laws and patterns of the cloud platform through methods such as tensor decomposition.

[0028] Step S3: Perform multi-scale topological graph construction and geometric mapping processing in the hyperbolic space on the high-order tensor structure of the cloud platform to generate a spatio-temporal multi-scale topological graph of the cloud platform; It can be understood that through the construction of a multi-scale topological graph and the processing of hyperbolic space mapping in this step, the spatio-temporal multi-scale topological graph of the cloud platform can accurately represent each functional module and its interaction relationships. The multi-scale topological graph is optimized through hierarchical clustering and adaptive algorithms, refining the hierarchical structure of the functional modules in the cloud platform and revealing the deep interactions between nodes; the geometric mapping of the hyperbolic space further enhances the representation ability of the graph structure, capable of revealing the non-linear and non-equilibrium interaction patterns in the cloud platform. In this step, step S3 includes step S31, step S32, step S33, and step S34.

[0029] Step S31: Perform multi-scale topological graph construction processing on the high-order tensor structure of the cloud platform. Among them, decompose the data of the high-order tensor into graph structures of multiple scales. Among them, at each scale, each scale node represents different physical modules of the cloud platform, and each scale edge represents the interaction relationship between nodes, obtaining a preliminary topological structure of the cloud platform; It can be understood that in this step, data decomposition is first performed. At each scale, the data of the high-order tensor is allocated to a graph structure. At each scale, the nodes represent physical modules in the cloud platform (such as virtual machines, physical devices, etc.), and each edge represents the interaction or dependency relationship between these nodes. Through this decomposition method, the interaction relationships between the modules of the cloud platform can be observed at different levels, and subsequent analysis and modeling can be carried out according to the data characteristics at different scales. After constructing the relationships between nodes and edges at each scale, the graph structures of all scales will be integrated to form a preliminary topological structure of the cloud platform. Combining the graph structures of each scale, a hierarchical topological graph is obtained. The topological graph of each scale represents a different perspective of the cloud platform. For example, from a global perspective, it can view the macroscopic interactions between the various functional modules of the cloud platform, while from a local perspective, it can deeply analyze the detailed interactions between certain specific modules (such as virtual machines, physical devices, etc.). Through this hierarchical processing, the complex structure of the cloud platform is effectively simplified, facilitating multi-dimensional and cross-level analysis. By decomposing the high-order tensor structure of the cloud platform into topological graphs of multiple scales, the multi-level interaction relationships between specific modules of the cloud platform can be effectively revealed. The graph structure of each scale provides different levels and granularities of dynamic characteristics for the cloud platform, helping to capture complex patterns such as resource usage and load changes.

[0030] Step S32: Perform hierarchical clustering based on the preliminary topological structure of the cloud platform. Among them, perform hierarchical clustering on the scale nodes and perform hierarchical division on the scale nodes after hierarchical clustering through an adaptive algorithm to obtain a multi-level preliminary topological structure of the cloud platform; It can be understood that in this step, the scale nodes are used as inputs through a hierarchical clustering algorithm, and are divided into multiple clusters based on the interaction intensity or similarity between nodes (such as resource sharing, load fluctuations, data flow, etc.). When calculating the similarity between nodes, metrics such as cosine similarity and Pearson correlation coefficient can be used to measure the relationship strength between different nodes in terms of resource consumption, task allocation, etc. The clustering results reflect the similarity in performance and resource usage among different functional modules (such as computing modules, storage modules, network modules, etc.) in the cloud platform.

[0031] Step S33: Map and process the preliminary topological structure of the multi-level cloud platform based on the geometric mapping method in hyperbolic space to obtain its corresponding hyperbolic space mapping result; It can be understood that in this step, the geometric mapping method in hyperbolic space is first used to perform geometric transformations on the nodes and edges in the multi-level topological graph, mapping the topological structure of the cloud platform from the traditional Euclidean space to the hyperbolic space. The geometric model of the hyperbolic space is implemented through the Lorentz model. During the mapping process, the position of each scale node in the cloud platform in the hyperbolic space will be adjusted according to its importance, function, and interaction relationship in the topological graph. The distance (i.e., similarity) between nodes will reflect their functional relationship and interaction intensity, while the connection (i.e., edge) between nodes represents their resource sharing, data flow, or dependency relationship. After mapping the nodes in the multi-level topological graph to the hyperbolic space, the relative positions between nodes are adjusted using geometric methods, so that nodes with stronger interactions are mapped to closer positions in space, while nodes with weaker interactions are mapped to farther positions. This processing can not only reflect the relationship strength between nodes, but also avoid information loss due to space limitations, ensuring that the information in the topological structure is retained as completely as possible.

[0032] Step S34: Calculate the similarity and relative positions between each pair of nodes in the cloud platform based on the hyperbolic space mapping result, and perform graph optimization processing based on the similarity and relative positions between each pair of nodes in the cloud platform to obtain the spatio-temporal multi-scale topological graph of the cloud platform.

[0033] It can be understood that in this step, the Euclidean distance between each pair of scale nodes in the hyperbolic space is first calculated, and scale nodes with smaller distances have higher similarities. In addition, the similarity calculation formula can be adjusted according to the functional interaction intensity (such as load demand, resource consumption, etc.) between scale nodes, so that when calculating the distance between scale nodes, the functional and resource dependency relationships can also be taken into account. For example, the distance between the storage scale node and the computing scale node is not only the geometric distance, but also includes their resource usage relationship and data flow pattern.

[0034] Among them, the similarity calculation formula is as follows: ;

[0035] Among them, represents the similarity between scale nodes and scale node ; represents the weight factor for adjusting the influence degree of the geometric distance on the similarity, represents scale node and scale node 's Euclidean distance in the hyperbolic space, represents the weight factor for adjusting the influence degree of the functional interaction intensity on the similarity, represents scale node and scale node 's functional interaction intensity between them.

[0036] The relative positions of scale nodes are determined by geometric mapping in the hyperbolic space, and the coordinates of each scale node in the space reflect its functional relationship and interaction characteristics with other scale nodes. By calculating the relative positions between scale nodes, it is possible to further understand the priorities of each scale node in the cloud platform's resource allocation or their key roles in task scheduling. For example, the relatively short distances between the core computing scale node and the storage scale node, as well as the bandwidth scale node, indicate a strong functional coupling relationship between them.

[0037] In this step, when performing graph optimization, the shortest path algorithm is used for optimization, and the edge weights between scale nodes can be adjusted according to the similarity matrix. The edges between scale nodes with higher weights represent stronger interaction relationships. During the optimization process, the relative positions of scale nodes are also considered. The goal is to make the entire topological graph more efficient by adjusting the layout and connection methods of scale nodes. For example, clustering scale nodes with similar resource requirements together to reduce resource conflicts between scale nodes, or adjusting the data transmission path to reduce latency. Through the calculation based on the similarity and relative positions of scale nodes, combined with graph optimization processing, the spatio-temporal multi-scale topological graph of the cloud platform can accurately reflect the relationships and interaction effects between each scale node. The hyperbolic space mapping optimizes the display of the relationships between scale nodes and enhances the expressive power of the topological graph in a complex dynamic environment. The optimized spatio-temporal topological graph not only improves the efficiency of resource management and task scheduling but also provides a solid foundation for subsequent fault detection, anomaly prediction, and adaptive optimization.

[0038] Step S4: Construct a nonlinear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform, and combine it with the singular spectrum analysis method to obtain a dynamic prediction model of the cloud platform; It can be understood that in this step, by combining the non - linear delay differential equation with the singular spectrum analysis method, the dynamic prediction model of the cloud platform can capture both the complex non - linear dynamic behavior and the time - delay effect in the system. This model can not only provide accurate state prediction, but also help the cloud platform identify potential resource bottlenecks, load fluctuations and fault risks in advance, so as to perform intelligent resource scheduling and fault prevention. In this step, step S4 includes step S41, step S42 and step S43.

[0039] Step S41: Construct a non - linear delay differential equation model based on the spatio - temporal multi - scale topological map of the cloud platform. Among them, the non - linear delay differential equation model simulates the dynamic changes of the states of nodes at each scale in the cloud platform. It can be understood that when constructing the non - linear delay differential equation model in this step, each scale node is represented as a dynamic variable, and the change of the node state is described by an equation. The changes of these dynamic variables are not only affected by the current state, but also by the time - delay effect of other scale nodes. In this step, according to the multi - scale topological map of the cloud platform, the interactive influence between scale nodes is defined. For example, calculating the load of a scale node (computing device) will affect the bandwidth requirement of a storage scale node (storage device), and this influence will take a certain time to be reflected in the storage scale node. The model captures this time - delay effect by introducing a delay term, reflecting the dynamic interaction between nodes.

[0040] For each scale node, use a non - linear delay differential equation to describe the change of its state over time. For example, for a computing scale node, the equation involves the calculation of the current load, the influence of upstream tasks on its load, and the response of the downstream storage scale node to the change of its load. By introducing non - linear terms (such as power terms, exponential terms, etc.), the equation can capture the non - linear characteristics of resource requirements in the cloud platform. At the same time, through the delay term, the model can simulate the indirect influence of resource requirements or task scheduling on other parts of the system. By constructing the non - linear delay differential equation model, it can effectively simulate the dynamic changes of the states of nodes at each scale in the cloud platform, especially in terms of the interaction between nodes and the time - delay effect. This model can capture the non - linear characteristics in cloud platform resource consumption, load fluctuations and task scheduling, and reflect the interactive influence between nodes.

[0041] Among them, the non - linear delay differential equation model is as follows: ;

[0042] Among them, represents the change rate of the state of the scale node, that is, the change rate of resource demand, represents the current state of the current scale node and the delayed state and The non - linear interaction effect between... Here is the delay time, indicating that the interaction effect between nodes takes some time to manifest. This part can consider non - linear characteristics such as load fluctuations, task scheduling, etc. represents the delay state on the current node, and this part represents the self - feedback mechanism in the system.

[0043] Step S42: Perform singular spectrum decomposition on the time - series data of the spatio - temporal multi - scale topology map of the cloud platform to obtain the main dynamic patterns of the time - series data; It can be understood that in this step, the time - series data in the spatio - temporal multi - scale topology map is extracted as the input for singular spectrum decomposition. Each time - series represents the behavior pattern of a certain node or between nodes in the cloud platform within a specific time period. These series reflect load changes, bandwidth consumption, or fluctuations in computing tasks, etc. By inputting these time - series data into the singular spectrum decomposition method, the dynamic patterns of the cloud platform system at each time scale can be extracted. Through singular spectrum decomposition, the time - series data of the cloud platform will be decomposed into several main components, and each component represents an independent dynamic pattern in the system. The main dynamic patterns refer to the most representative patterns in the system that have important influences on key factors such as resource consumption, load fluctuations, and task scheduling. These patterns are periodic (such as daily or weekly load peaks) and also non - periodic (such as sudden failures or temporary load surges); these dynamic patterns include periodic changes and sudden fluctuations, which help to comprehensively understand the operating state and behavior of the cloud platform.

[0044] Step S43: Use the main dynamic patterns of the time - series data as input and construct a dynamic prediction model of the cloud platform in combination with a non - linear delay differential equation model.

[0045] It can be understood that in this step, by combining the main dynamic patterns of the time - series data with a non - linear delay differential equation model, the constructed dynamic prediction model of the cloud platform can comprehensively and accurately capture the complex dynamic interactions and delay effects between nodes at various scales of the cloud platform. This model can not only predict the resource requirements, load fluctuations, and task scheduling changes of the cloud platform, but also help the cloud platform optimize resource allocation, improve system stability, and prevent failures. In this step, Step S43 includes Step S431, Step S432, and Step S433.

[0046] Step S431: Use the main dynamic patterns of the time - series data as input data, introduce the input data into the non - linear delay differential equation model for spatio - temporal state prediction, and obtain an initial dynamic prediction model; It can be understood that the dynamic patterns extracted from the singular spectrum analysis in this step are input into the non - linear delay differential equation model. The state of each scale node in the equation (such as load, storage demand, computing resource consumption, etc.) depends on the current state and historical states, and there is a certain time delay in the interaction between nodes. For example, the load fluctuation of a computing node will affect the performance of a storage node, and this effect takes some time to manifest. The non - linear term can capture the complex non - linear feedback effects in the cloud platform (such as resource competition, task scheduling, etc.).

[0047] In the training process of the model, historical data of the cloud platform and the main dynamic patterns are used as inputs. The non - linear delay differential equation is solved by numerical solution methods (such as Euler's method, etc.) to obtain the predicted values of the future state of the cloud platform. Through repeated training and optimization, the model gradually adjusts its parameters to improve the prediction accuracy. The preliminary prediction results include the changes in the load of the computing scale nodes, the capacity requirements of the storage scale nodes, and the bandwidth of the network scale nodes, etc. These results provide a basis for resource scheduling and task scheduling in the cloud platform. Among them, the computing scale nodes represent computing devices, the storage scale nodes represent storage devices, and the network scale nodes represent network transmission devices.[[ID=**4**]] [[ID=**5**]]

[0048] Step S432: Train the initial dynamic prediction model with the time series in the preset historical data of the cloud platform and the preset historical dynamic patterns to obtain the trained initial dynamic prediction model;[[ID=**7**]] It can be understood that in this step, the time series in the historical data (such as the load data, energy consumption data, task scheduling data, etc. of each scale node of the cloud platform) are input into the initial dynamic prediction model. These time series data provide the actual state of the cloud platform at different time points, helping the model learn the dynamic changes of the cloud platform under different loads, resource demands, task scheduling, etc. Through training, the model can adjust its internal parameters to make the prediction results closer to the actual data. Combining the input of dynamic patterns, the model can more accurately understand the operation rules of the cloud platform. Dynamic patterns usually include periodic changes (such as daily load changes), trend changes (such as a gradual increase in resource demand), and sudden fluctuations (such as high load or failure states). By combining these patterns with historical data, the model can learn the long - term development trend and short - term fluctuation patterns of the cloud platform, thereby improving its prediction accuracy for future states.[[ID=**9**]] [[ID=**10**]]

[0049] Step S433: Adjust the delay parameter in the non - linear delay differential equation of the initial dynamic prediction model based on the particle swarm optimization algorithm to obtain the dynamic prediction model of the cloud platform.[[ID=**12**]] [[ID=**13**]]

[0050] It can be understood that in this step, by using the particle swarm optimization algorithm in step S433, the delay parameter in the non - linear delay differential equation of the initial dynamic prediction model is adjusted to obtain the dynamic prediction model of the cloud platform.

[0051] In the initially constructed dynamic prediction model, a non-linear delay differential equation is used to simulate the dynamic state changes of nodes at various scales in the cloud platform. The delay parameter therein is an important factor determining the response speed and interaction intensity of the system behavior. The delay parameter represents the time delay required for the interaction effect between nodes in the cloud platform. For example, the demand of a computing task for storage resources will affect the storage node after a certain period of time. Therefore, the precise setting of the delay parameter is crucial for the prediction accuracy of the model. By adjusting the delay parameter in the non-linear delay differential equation model through the particle swarm optimization algorithm, the prediction accuracy of the cloud platform dynamic prediction model can be significantly improved. The optimized model can accurately simulate the interaction and feedback effects of nodes at various scales in the cloud platform, especially the impact of time delay on resource demand and load fluctuation.

[0052] Step S5: Preprocess the multi-dimensional data of the cloud platform, and input the preprocessing result as input data into the dynamic prediction model of the cloud platform for prediction to obtain the state prediction result of the cloud platform in a preset time period; It can be understood that in this step, by combining fractional Brownian motion and multi-fractal detrended fluctuation analysis, the dynamic prediction model of the cloud platform can more accurately capture multi-scale dynamic changes such as resource demand and load fluctuation, especially showing excellent performance in dealing with long-term dependence and sudden fluctuations. The optimized input data provides more detailed dynamic features for the model, enabling the cloud platform to more accurately predict future states, optimize resource scheduling and load balancing, thereby enhancing the stability, efficiency and reliability of the overall system. In this step, step S5 includes step S51, step S52, step S53 and step S54.

[0053] Step S51: Construct a fractional Brownian motion model based on the multi-dimensional data of the cloud platform. Among them, through the fractional Brownian motion modeling technique, the multi-dimensional data of the cloud platform is modeled as a stochastic process with long-term dependence to obtain characteristic data with long-range dependence characteristics; It can be understood that in this step, the fractional Brownian motion model is used to simulate the multi-dimensional data of the cloud platform (such as computing load, storage usage, bandwidth demand, etc.), and these data show strong correlation in time. For example, the load of the computing scale node will remain high for a period of time, affecting the performance requirements of the storage scale node. By modeling these time series data as a stochastic process with long-range dependence, the fractional Brownian motion model can capture the long-term correlation between these scale nodes. When constructing the fractional Brownian motion model, the multi-dimensional data of the cloud platform is decomposed into multiple time series, and each time series represents the state change of a node. For example, the load data of the computing scale node, the storage occupancy of the storage scale node, the bandwidth usage of the network scale node, etc., and these time series are input into the fractional Brownian motion model.

[0054] Through the training of the model, the long-term dependence characteristics of each time series are obtained. Since the fractional Brownian motion model has self-similarity, it can capture the long-range dependence relationships and internal laws in these time series. The modeling technique of fractional Brownian motion quantifies the long-term dependence in the data by introducing the Hurst exponent (an index measuring long-range dependence). The value of the Hurst exponent ranges from 0 to 1, reflecting the strength of long-term dependence in the time series. For example, a Hurst exponent value close to 1 indicates strong long-term dependence, while close to 0.5 indicates strong short-term dependence. By calculating the Hurst exponent of the resource fluctuations at each scale node, the fluctuation pattern and resource consumption trend of this node in the future time period can be understood.

[0055] Step S52: Perform multifractal detrended fluctuation analysis on the multi-dimensional data of the cloud platform. Specifically, extract adaptive fluctuation scales from the fluctuations of the multi-dimensional data of the cloud platform and calculate the detrended fluctuations at each adaptive scale to obtain the fluctuation characteristics of the multi-dimensional data of the cloud platform at different scales; It can be understood that in this step, the multifractal detrended fluctuation analysis (MF-DFA) is an analysis method for dealing with non-stationary time series, which is particularly suitable for capturing data with complex fluctuation characteristics and multi-scale features. The MF-DFA method can extract the fluctuation characteristics at different scales from the time series and remove the trend component in the data, so that the analysis results can more accurately reflect the true fluctuation pattern of the system. In the cloud platform, resource consumption, load fluctuations, etc. often exhibit multi-scale fluctuation characteristics. MF-DFA can effectively capture these fluctuations and reveal the dynamic changes at different time scales. The specific operation process is as follows. MF-DFA divides the time series data into multiple windows by segmenting the time series, and removes the trend component within each window, and calculates the fluctuation characteristics at each scale. By analyzing these fluctuation characteristics, the fluctuation patterns in the cloud platform at different time scales can be revealed. The MF-DFA method is particularly suitable for dealing with data with long-range dependence and non-stationarity. The detrending process is a key step in MF-DFA. First, at each adaptive fluctuation scale, the long-term change component in the data is removed by fitting a trend (such as a linear trend or a polynomial trend). The data after detrending will only retain the fluctuation part, that is, the change in resource demand or load. This process helps to eliminate the stationary trend in the data, making the fluctuation characteristics more prominent, so that subsequent analysis and modeling can focus on the dynamic fluctuations of the system rather than its long-term trend. The steps of calculating the fluctuation characteristics (Hurst exponent) at each scale include calculating the detrended fluctuation amplitude through the MF-DFA method, and these amplitude values represent the fluctuation intensity of the cloud platform at different time periods. For example, in the short term (hour level), the load of the computing scale node shows sudden fluctuations, while in the long term (week level), the load shows a steady increase. By analyzing these fluctuation characteristics, the dynamic changes in the use of cloud platform resources can be understood more clearly.

[0056] Step S53: Feature fusion is performed on the feature data with long-range dependence characteristics and the fluctuation characteristics of the multi-dimensional data of the cloud platform at different scales to generate a plurality of composite feature vectors, and each of the composite feature vectors includes the long-term fluctuation pattern and the local sudden fluctuation pattern of the cloud platform at different time scales; Step S54: The composite feature vectors are input into the cloud platform dynamic prediction model for prediction to obtain the state prediction result of the cloud platform in a preset time period.

[0057] It can be understood that in this step, by inputting the composite feature vector into the dynamic prediction model of the cloud platform, the cloud platform can accurately predict future state changes, especially in terms of load fluctuations, resource requirements, and task scheduling. The composite feature vector provides multi-scale information on the resource requirements and load fluctuations of the cloud platform, enabling the prediction model to more accurately reflect the long-term trends and sudden fluctuations in resource requirements. Based on these prediction results, the dynamic prediction model can optimize resource allocation and task scheduling in advance, improving the adaptability, efficiency, and stability of the cloud platform in complex and dynamic environments.

[0058] Step S6: Obtain a preset operation and maintenance plan based on the state prediction result of the cloud platform in a preset time period for operation and maintenance.

[0059] It can be understood that in this step, first, according to the predicted state of the cloud platform (such as future load fluctuations, storage requirements, bandwidth consumption, etc.), the system filters out the most suitable operation and maintenance plan from a preset operation and maintenance plan library. For example, if it is predicted that a computing node will soon reach a load peak, the system will select an operation and maintenance plan to expand computing resources; if it is predicted that the storage node will face a bottleneck, a data migration or expansion plan will be selected. Each operation and maintenance plan contains a series of strategies, such as resource expansion, load balancing, fault warning, task scheduling optimization, etc. Once the most suitable operation and maintenance plan is selected, the system will perform automated operation and maintenance operations according to the specific strategies of the plan. These operations include resource scheduling, task optimization, load balancing, fault prevention, etc., so as to ensure that the cloud platform can operate efficiently and stably in the face of future state changes. By obtaining a preset operation and maintenance plan based on the state prediction result of the cloud platform in a preset time period, the cloud platform can achieve intelligent adaptive operation and maintenance, and respond to load fluctuations, resource bottlenecks, and fault risks in advance. The automatic execution and adjustment of the operation and maintenance plan can greatly improve the efficiency of resource scheduling and load balancing, while reducing human intervention and operation and maintenance costs. In this way, the cloud platform can maintain efficient and stable operation in a complex dynamic environment, ensuring the optimal utilization and reliability of system resources. Embodiment 2:

[0060] According to Figure 2 As shown, this embodiment provides a cloud platform operation and maintenance system, including: An acquisition unit 701, configured to acquire multi-dimensional data of the cloud platform, where the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data, and abnormal event data; An integration unit 702, configured to integrate the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and perform non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform; A mapping unit 703, configured to perform multi-scale topological graph construction and geometric mapping processing in a hyperbolic space on the high-order tensor structure of the cloud platform, and generate a spatio-temporal multi-scale topological graph of the cloud platform; A construction unit 704, configured to construct a non-linear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform, and combine it with a singular spectrum analysis method to obtain a dynamic prediction model of the cloud platform; A prediction unit 705, configured to preprocess multi-dimensional data of the cloud platform, and use the preprocessing result as input data to input into the dynamic prediction model of the cloud platform for prediction, so as to obtain a state prediction result of the cloud platform in a preset time period; An operation and maintenance unit 706, configured to obtain a preset operation and maintenance plan based on the state prediction result of the cloud platform in a preset time period for operation and maintenance.

[0061] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0062] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A cloud platform operation and maintenance method, characterized in that, Including: Obtain multi-dimensional data of the cloud platform, where the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data, and abnormal event data; Integrate the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and perform non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform; Perform multi-scale topological graph construction and geometric mapping processing in hyperbolic space on the high-order tensor structure of the cloud platform to generate a spatio-temporal multi-scale topological graph of the cloud platform; Construct a non-linear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform, and combine it with the singular spectrum analysis method to obtain a dynamic prediction model of the cloud platform; Preprocess the multi-dimensional data of the cloud platform, and use the preprocessing result as input data to input into the dynamic prediction model of the cloud platform for prediction to obtain a state prediction result of the cloud platform in a preset time period; Obtain a preset operation and maintenance plan based on the state prediction result of the cloud platform in a preset time period for operation and maintenance.

2. The cloud platform operation and maintenance method according to claim 1, characterized in that, Integrate the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and perform non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform, including: Represent the multi-dimensional data of the cloud platform as a product of a set of core tensors and multiple factor matrices based on the Tucker decomposition method, where each factor matrix represents the feature space of different-dimensional data, to obtain the multi-dimensional interaction features of the cloud platform; Perform non-uniform sampling on the obtained multi-dimensional interaction features of the cloud platform, and screen out a data set that meets preset conditions from the multi-dimensional interaction features; Construct a high-resolution spatio-temporal graph based on the data set that meets the preset conditions, where each functional node represents a specific function of the cloud platform, and each functional edge represents the interaction between different functional nodes; Merge the data sets that meet the preset conditions in different dimensions in the high-resolution spatio-temporal graph based on tensor construction technology to form a high-order tensor structure.

3. The cloud platform operation and maintenance method according to claim 1, wherein Perform multi-scale topological graph construction and geometric mapping processing in hyperbolic space on the high-order tensor structure of the cloud platform, including: Perform multi-scale topological graph construction processing on the high-order tensor structure of the cloud platform, where the data of the high-order tensor is decomposed into graph structures of multiple scales. Among them, at each scale, each scale node represents a different physical module of the cloud platform, and each scale edge represents the interaction relationship between nodes, to obtain a preliminary topological structure of the cloud platform; Perform hierarchical clustering based on the preliminary topological structure of the cloud platform. Among them, perform hierarchical clustering on the scale nodes, and perform hierarchical division on the scale nodes after hierarchical clustering through an adaptive algorithm to obtain a multi-level preliminary topological structure of the cloud platform; Perform mapping processing on the multi-level preliminary topological structure of the cloud platform based on the geometric mapping method in hyperbolic space to obtain its corresponding hyperbolic space mapping result; Calculate the similarity and relative positions between the nodes of the cloud platform based on the hyperbolic space mapping result, and perform graph optimization processing based on the similarity and relative positions between the nodes of the cloud platform to obtain a spatio-temporal multi-scale topological graph of the cloud platform.

4. The cloud platform operation and maintenance method according to claim 1, wherein, Construct a nonlinear delay differential equation model based on the spatio-temporal multi-scale topological map of the cloud platform and combine it with the singular spectrum analysis method, including: Construct a nonlinear delay differential equation model based on the spatio-temporal multi-scale topological map of the cloud platform, where the nonlinear delay differential equation model simulates the dynamic changes of the states of nodes at various scales in the cloud platform; Perform singular spectrum decomposition on the time series data of the spatio-temporal multi-scale topological map of the cloud platform to obtain the main dynamic patterns of the time series data; Use the main dynamic patterns of the time series data as input and combine with the nonlinear delay differential equation model to construct a dynamic prediction model for the cloud platform.

5. The cloud platform operation and maintenance method according to claim 4, wherein Use the main dynamic patterns of the time series data as input and combine with the nonlinear delay differential equation model to construct a dynamic prediction model for the cloud platform, including: Use the main dynamic patterns of the time series data as input data, introduce the input data into the nonlinear delay differential equation model for spatio-temporal state prediction, and obtain an initial dynamic prediction model; Train the initial dynamic prediction model through the time series in the preset historical data of the cloud platform and the preset historical dynamic patterns to obtain the trained initial dynamic prediction model; Adjust the delay parameters in the nonlinear delay differential equation of the initial dynamic prediction model based on the particle swarm optimization algorithm to obtain a dynamic prediction model for the cloud platform.

6. A cloud platform operation and maintenance system, characterized in that Including: An acquisition unit for acquiring multi-dimensional data of the cloud platform, where the multi-dimensional data includes energy consumption data, environmental change data, computing task data, storage data, and abnormal event data; An integration unit for integrating the multi-dimensional data of the cloud platform into a high-order tensor based on the Tucker decomposition method, and performing non-uniform sampling on the high-order tensor to construct a high-order tensor structure of the cloud platform; A mapping unit for performing multi-scale topological map construction and geometric mapping processing in the hyperbolic space on the high-order tensor structure of the cloud platform to generate a spatio-temporal multi-scale topological map of the cloud platform; A construction unit for constructing a nonlinear delay differential equation model based on the spatio-temporal multi-scale topological map of the cloud platform and combining it with the singular spectrum analysis method to obtain a dynamic prediction model for the cloud platform; A prediction unit for preprocessing the multi-dimensional data of the cloud platform and using the preprocessing result as input data to input into the dynamic prediction model of the cloud platform for prediction to obtain a state prediction result of the cloud platform in a preset time period; An operation and maintenance unit for obtaining a preset operation and maintenance plan for operation and maintenance based on the state prediction result of the cloud platform in a preset time period.

7. The cloud platform operation and maintenance system according to claim 6, wherein, The integration unit includes: A first integration subunit for representing the multi-dimensional data of the cloud platform as a product of a group of core tensors and multiple factor matrices based on the Tucker decomposition method, where each factor matrix represents the feature space of data in different dimensions, to obtain the multi-dimensional interaction features of the cloud platform; A second integration subunit for performing non-uniform sampling on the obtained multi-dimensional interaction features of the cloud platform and screening out a data set that meets the preset conditions from the multi-dimensional interaction features; The third integration subunit is used to construct a high-resolution spatio-temporal graph based on a dataset that meets preset conditions. Each functional node represents a specific function of the cloud platform, and each functional edge represents the interaction between different functional nodes; The fourth integration subunit is used to merge datasets of different dimensions that meet preset conditions in the high-resolution spatio-temporal graph based on tensor construction technology to form a high-order tensor structure.

8. The cloud platform operation and maintenance system according to claim 6, wherein The mapping unit includes: The first mapping subunit is used to perform multi-scale topological graph construction processing on the high-order tensor structure of the cloud platform. The data of the high-order tensor is decomposed into graph structures of multiple scales. At each scale, each scale node represents a physical module of a different cloud platform, and each scale edge represents the interaction relationship between nodes, obtaining a preliminary topological structure of the cloud platform; The second mapping subunit is used to perform hierarchical clustering based on the preliminary topological structure of the cloud platform. By performing hierarchical clustering on the scale nodes and hierarchically dividing the scale nodes after hierarchical clustering through an adaptive algorithm, a multi-level preliminary topological structure of the cloud platform is obtained; The third mapping subunit is used to perform mapping processing on the multi-level preliminary topological structure of the cloud platform based on the geometric mapping method in hyperbolic space to obtain its corresponding hyperbolic space mapping result; The fourth mapping subunit is used to calculate the similarity and relative positions between nodes of the cloud platform based on the hyperbolic space mapping result, and perform graph optimization processing based on the similarity and relative positions between nodes of the cloud platform to obtain a spatio-temporal multi-scale topological graph of the cloud platform.

9. The cloud platform operation and maintenance system according to claim 6, characterized in that, The construction unit includes: The first construction subunit is used to construct a non-linear delay differential equation model based on the spatio-temporal multi-scale topological graph of the cloud platform. The non-linear delay differential equation model simulates the dynamic changes of the states of each scale node in the cloud platform; The second construction subunit is used to perform singular spectrum decomposition on the time series data of the spatio-temporal multi-scale topological graph of the cloud platform to obtain the main dynamic patterns of the time series data; The third construction subunit is used to use the main dynamic patterns of the time series data as input, and combine with the non-linear delay differential equation model to construct a dynamic prediction model of the cloud platform.

10. The cloud platform operation and maintenance system according to claim 9, wherein The third construction subunit includes: The fourth construction subunit is used to use the main dynamic patterns of the time series data as input data, introduce the input data into the non-linear delay differential equation model for spatio-temporal state prediction, and obtain an initial dynamic prediction model; The fifth construction subunit is used to train the initial dynamic prediction model through the time series in the preset historical data of the cloud platform and the preset historical dynamic patterns to obtain a trained initial dynamic prediction model; The sixth construction subunit is used to adjust the delay parameter in the non-linear delay differential equation in the initial dynamic prediction model based on the particle swarm optimization algorithm to obtain a dynamic prediction model of the cloud platform.

Images