Cloud information system index acquisition method, system, equipment and medium

Abstract

The invention discloses a cloud information system index collection method, system and device and a medium, and relates to the technical field of cloud information system operation and maintaining.The method comprises the steps that information collection interfaces of a cloud system are integrated, data flow information entropy is quantized, and an entropy value sequence is constructed; an interface calling serial number is traced based on a historical log, a calling chain is formed, a directed graph topological structure is established, an influence propagation path of an entropy mutation node is identified, and a collection regulation priority of an interface cluster is determined; an acquisition strategy is formulated in combination with entropy time sequence data and a load level, a state-strategy mapping relation matrix is optimized, and a time period driving type strategy adjustment parameter set is generated; and issuing the strategies in sequence, dynamically regulating and controlling the acquisition frequency, and if state transition or fitting degree abnormity occurs, triggering strategy rollback and optimizing the learning weight. According to the method, on the premise of guaranteeing the integrity of the monitoring data of the power equipment, the utilization rate of edge computing resources is remarkably optimized, and high-timeliness and high-fidelity data support is provided for safe and stable operation of a power grid.

Description

Technical Field

[0001] The present invention relates to the field of cloud information system operation and maintenance technology, and in particular to a cloud information system indicator collection method, system, equipment and medium. Background Art

[0002] In the intelligent operation and maintenance scenarios of cloud information systems, the contradiction between the dynamic nature of the interface status of each cloud service and the static nature of traditional collection strategies is becoming increasingly prominent. Especially in ultra-large-scale distributed systems, the nonlinear, sudden, and regional characteristics of interface status changes lead to serious failure of traditional collection mechanisms based on fixed thresholds or rule engines. Taking power grid management as an example, the status changes of the power load monitoring interface are highly sudden and show significant time correlation, while the status changes of the substation's environmental monitoring interface are relatively stable. The existing collection system adopts a globally unified sampling interval and static priority strategy. In highly dynamic scenarios, the information entropy value of the data collected by some interfaces increases sharply, and the fixed sampling frequency cannot fully capture all events, resulting in the loss of key data. At the same time, for interfaces with stable status, the existing mechanism still collects data at a fixed period, generating a large amount of redundant data, increasing the real-time data analysis delay and system resource usage.

[0003] Therefore, how to build a new technology system that can quantify the information entropy characteristics of the collection interface, autonomously transform the collection strategy based on the idea of intelligent operation and maintenance, and realize the cloud information system indicator collection method for cloud computing environment is an urgent problem to be solved. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a cloud-based information system indicator collection method to solve the following problems: the existing information indicator collection method has a fixed collection frequency that cannot adapt to the dynamically changing interface status, is insufficient in capturing anomalies of high-frequency changing events, and cannot effectively distinguish between the collection requirements of high-entropy and low-entropy interfaces, as well as how to adaptively and dynamically adjust the collection strategy to improve the accuracy and efficiency of data collection.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a method for collecting indicators of a cloud information system, including: integrating cloud system information collection interfaces, quantifying the information entropy of interface data flows, and generating a real-time entropy value sequence; extracting historical logs to trace the serial numbers of interface calls, forming an interface set in the calling order, and constructing a historical interface call chain; based on the entropy value sequence value and the historical interface call chain, establishing a directed graph topology structure, identifying the influence propagation path of the entropy mutation node, and generating indicator collection control instructions; based on the entropy value time series data of each interface and the interface load level, generating a time series collection strategy table, clustering and dividing high and low entropy interfaces, and constructing an interface state-strategy mapping relationship matrix; generating time period characteristics of high and low entropy of the interface, performing time-sharing adaptation on the interface state-strategy mapping relationship matrix through a dual Q-learning algorithm, and generating a time period-driven strategy adjustment parameter set; and sending the strategy adjustment parameter set to each collection interface for collection frequency control.

[0008] As a preferred solution of the cloud information system indicator collection method described in the present invention, the generating of a real-time entropy value sequence includes integrating all information collection interfaces of the cloud system, synchronously capturing the instantaneous value of the indicators of each interface through a sliding window of predefined multiple time scales, and generating a time series original data stream with an interface identifier; performing Z-Score-based normalization processing on the original data stream, statistically analyzing the data distribution within the sliding window, and calculating the histogram frequency distribution of the data value; iteratively calculating the Shannon entropy value based on the histogram frequency distribution of the data value, quantifying the information entropy of the interface data stream by calculating the probability distribution, and generating a real-time updated entropy value sequence.

[0009] As a preferred solution of the cloud information system indicator collection method described in the present invention, the construction of the historical interface call chain includes setting a log monitoring cycle, tracing the serial number of the interface calls in the historical log of the cloud system, extracting the interface identifier corresponding to each serial number, and forming an interface set with a calling order; traversing the interface identifiers in each serial number interface set, using the union check set to merge the same elements of different sets, and constructing a historical interface call chain.

[0010] As an optimal solution of the cloud information system indicator collection method described in the present invention, the generating indicator collection and control instructions includes: constructing a directed graph topology structure based on the historical interface call chain using the FIoyd-WarshaⅡ algorithm; wherein, the node represents the interface instance, and the edge weight is set by the call frequency and delay calculation; based on the real-time calculated interface information entropy quantization value, generating a node embedding vector that integrates the topological structure and the interface entropy characteristics through the Node2Vec algorithm; monitoring the entropy mutation node based on the preset mutation ratio, starting from the entropy mutation node, traversing the topology graph based on the breadth-first search, identifying the influencing propagation path within its k-hop range, and outputting the associated node set; according to the propagation path length and the node entropy value weight, dynamically allocating the collection and control priority of the associated node set through the linear attenuation model, generating indicator collection and control instructions in sequence, and performing indicator collection and control on the interface cluster corresponding to the node set.

[0011] As a preferred solution of the cloud information system indicator collection method described in the present invention, the construction of the interface status-strategy mapping relationship matrix includes: after receiving the indicator collection control instruction, obtaining and integrating the initial system resource snapshot allocated by the corresponding interface cluster, defining the benchmark collection frequency interval parameters of each interface based on the entropy value time series data of each interface in the interface cluster and the interface load level, and generating a time series collection strategy table; using the K-means clustering algorithm to perform interface clustering on the entropy value sequence, dividing the interfaces into high-entropy interfaces and low-entropy interfaces, and constructing the interface status-strategy mapping relationship matrix in combination with the interface load rate level.

[0012] As a preferred solution of the cloud information system indicator collection method described in the present invention, the generation of a period-driven policy adjustment parameter set includes: based on the historical time series entropy value sequence of the interface cluster, predicting the future period division boundaries through the LSTM model, generating high-entropy and low-entropy period characteristics of the interface and adding time validity labels; using a dual Q-learning algorithm to perform time-sharing adaptation on the interface state-policy mapping relationship matrix, shortening the collection interval of the high-entropy interface to the upper limit of the benchmark collection frequency according to the set control step size, and extending the collection interval of the low-entropy interface to the lower limit of the benchmark collection frequency according to the set control step size, thereby generating a period-driven policy adjustment parameter set.

[0013] As an optimal solution of the cloud information system indicator collection method described in the present invention, the collection frequency control includes: sending the policy parameters in the policy adjustment parameter set to each collection interface in sequence to regulate the collection frequency of the interface; real-time monitoring of the mutation direction of the information entropy of the interface cluster during the control process, if an interface state transition occurs, that is, the high entropy interface state switches to a low entropy interface, or the low entropy interface state switches to a high entropy interface, triggering a policy rollback and redistributing the Q-learning exploration weight; if no interface state transition occurs, the dynamic fit of the fluctuation curve of the interface collection indicator and the interface historical collection indicator is adjusted after calculating the strategy through a sliding time window, and when the fit is lower than the preset quality threshold, triggering a policy rollback and redistributing the Q-learning exploration weight.

[0014] In the second aspect, the present invention provides a cloud information system indicator collection system, including: an information entropy quantification module, a call chain construction module, an entropy mutation identification module, a relationship matrix construction module, a strategy adjustment parameter module, and an acquisition frequency control module. The information entropy quantification module is used to integrate the cloud system information collection interface, quantify the information entropy of the interface data flow, and generate a real-time entropy value sequence; the call chain construction module is used to extract historical logs to trace the interface calls by serial number, form an interface set in the call order, and construct a historical interface call chain; the entropy mutation identification module is used to establish a directed graph topology structure based on the entropy value sequence value and the historical interface call chain. The structure is used to identify the impact propagation path of the entropy mutation node and generate indicator collection and control instructions; the relationship matrix construction module is used to generate a timing collection strategy table based on the entropy value time series data of each interface and the interface load level, cluster and divide the high and low entropy interfaces, and construct an interface state-strategy mapping relationship matrix; the strategy adjustment parameter module is used to generate the time period characteristics of the high and low entropy of the interface, and perform time-sharing adaptation on the interface state-strategy mapping relationship matrix through the dual Q-learning algorithm to generate a time period driven strategy adjustment parameter set; the acquisition frequency control module is used to send the strategy adjustment parameter set to each acquisition interface for acquisition frequency control.

[0015] In a third aspect, the present invention provides an electronic device, comprising:

[0016] memory and processor;

[0017] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the cloud information system indicator collection method are implemented.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the cloud information system indicator collection method.

[0019] Compared with the existing technology, the beneficial effects of the present invention are as follows: the present invention quantifies the information entropy characteristics of each interface data stream in real time, combines historical call chain analysis to build a device status association topology, accurately identifies abnormal propagation paths of high-dynamic interfaces such as high-voltage transmission lines, and prioritizes improving the acquisition accuracy and response speed of key nodes to ensure the complete capture of transient fault characteristics; for low-entropy interfaces such as environmental monitoring, based on entropy clustering and time-sharing reinforcement learning models, the acquisition frequency is adaptively reduced in the state stable stage, effectively reducing the amount of redundant data; the system realizes real-time resource expansion of high-entropy interfaces and idle resource recovery of low-entropy interfaces through dynamic strategy mapping and resource elastic allocation mechanism, and introduces a strategy rollback mechanism to quickly respond to sudden state transitions. Under the premise of ensuring the integrity of power equipment monitoring data, the present invention significantly optimizes the utilization rate of edge computing resources, providing high-efficiency and high-fidelity data support for the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 The figure is a schematic diagram of the overall process of the cloud information system indicator collection method described in one embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of a trigger policy rollback mechanism for a cloud information system indicator collection system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0024] Example 1, with reference to Figure 1 , as an embodiment of the present invention, provides a method for collecting indicators of a cloud information system, comprising:

[0025] S1: Integrate the cloud system information collection interface, quantify the information entropy of the interface data flow, and generate a real-time entropy value sequence;

[0026] S2: Extract historical logs to trace the serial numbers of interface calls, form a set of interfaces in the calling order, and build a historical interface call chain;

[0027] S3: Based on the entropy sequence value and the historical interface call chain, a directed graph topology is established to identify the impact propagation path of the entropy mutation node and generate indicator collection and control instructions;

[0028] S4: Based on the entropy time series data of each interface and the interface load level, a time series collection policy table is generated, high-entropy and low-entropy interfaces are clustered and divided, and an interface state-policy mapping relationship matrix is constructed;

[0029] S5: Generates high and low entropy time characteristics of the interface, performs time-sharing adaptation on the interface state-policy mapping relationship matrix using a dual Q-learning algorithm, and generates a time-driven policy adjustment parameter set.

[0030] S6: Send the policy adjustment parameter set to each collection interface to adjust the collection frequency.

[0031] It should be noted that the collection of information indicators can continuously monitor various components of the cloud information system, including computing resources, storage performance and network traffic. This real-time monitoring can detect abnormal situations in a timely manner and ensure the stability and reliability of the system. The collection of information indicators provides a rich data foundation and provides decision-making support information for management and technical teams. At the same time, accurate data collection and analysis can effectively save computing and storage resources and optimize the operating costs of cloud services. Therefore, the collection of information indicators plays an indispensable role in ensuring the efficient, stable, secure and compliant operation of the cloud computing environment, and is crucial to improving the performance of cloud information systems and business success.

[0032] Therefore, to address the above-mentioned problems, in step S1, data from different interfaces are integrated and the entropy of information flows is quantified in real time, ensuring effective capture of dynamic changes in interface status. In step S2, by tracing back historical logs, the system can establish call chains and analyze the dependencies between interfaces, so that the most important interfaces can be identified and prioritized during future data collection, ensuring a clear understanding of the system's operation trajectory. In step S3, a directed graph topology is constructed to visualize the relationships between interfaces and analyze entropy mutation nodes, enabling proactive management of abnormal situations. In step S4, a time-series collection strategy table is generated, and by dynamically adjusting the collection frequency, redundancy is reduced and resource allocation is optimized. In step S5, a dual Q-learning algorithm is used to generate a time-period-driven policy adjustment parameter set, ensuring accurate collection under highly dynamic conditions. In step S6, the generated policy adjustment parameters are distributed to each interface for real-time collection frequency regulation, ensuring real-time reflection of system status changes. Through the rollback mechanism, state transitions or collection indicator anomalies can be quickly responded to, further ensuring the continuity and effectiveness of data collection.

[0033] Example 2, reference Figure 1-Figure 2 , is an embodiment of the present invention. Based on the above embodiment, a method for collecting indicators of a cloud information system is provided.

[0034] In an embodiment of the present application, generating a real-time entropy value sequence in step S1 includes integrating all information collection interfaces of the cloud system, synchronously capturing the instantaneous values of the indicators of each interface through sliding windows of predefined multiple time scales, and generating a time-series original data stream with an interface identifier; performing Z-Score-based normalization processing on the original data stream, statistically analyzing the data distribution within the sliding window, and calculating the histogram frequency distribution of the data values; iteratively calculating the Shannon entropy value based on the histogram frequency distribution of the data values, quantifying the information entropy of the data stream of the interface by calculating the probability distribution, and generating a real-time updated entropy value sequence.

[0035] Specifically, all information collection interfaces of the cloud system are integrated. These interfaces can come from business nodes, virtual machines, physical servers, or other distributed computing resources of different cloud services. To ensure the collection capability at different time scales, sliding windows of multiple time scales are predefined. The window size can be dynamically adjusted according to actual needs. Each time window captures the instantaneous value of each interface's indicator within the time window through the sliding window mechanism. The instantaneous indicator value includes but is not limited to real-time temperature indicator and power load indicator.

[0036] Each data point carries the interface identifier, ensuring that subsequent analysis and processing can be performed based on the interface identifier, ultimately forming a time-series raw data stream with the interface identifier.

[0037] After data collection is completed, the raw data of each interface is subjected to Z-Score standardization;

[0038] Z-Score standardization can effectively eliminate the differences in dimensions and scales between data from different interfaces, making the data from all interfaces have the same standard. The normalized data is statistically analyzed through a sliding window to generate a histogram frequency distribution for each data point.

[0039] On this basis, the frequency distribution of the data values within the time window is calculated to reflect the central tendency and volatility of the data;

[0040] Assume that the original data of interface A is [0.1, 0.2, 0.3, 0.4, 0.5]. After Z-Score normalization, the data is [-1.41, -0.71, 0, 0.71, 1.41]. Next, calculate the frequency distribution of this data. For example, divide the data into several intervals, such as [-1, 0], [0, 1], and calculate the number of data points in each interval.

[0041] The Shannon entropy value is iteratively calculated based on the histogram frequency distribution of the data value. The information entropy of the data stream of the interface is quantified by calculating the probability distribution, and a real-time updated entropy value sequence is generated. Shannon entropy is a standard for measuring data uncertainty and information content. The calculation method is expressed as:

[0042]

[0043] Among them, H(X) is the Shannon entropy value, which is used to express the information entropy of the interface collected data, P(x i ) is the histogram frequency distribution of the i-th data value, x i is the i-th data value after normalization transformation.

[0044] In an optional embodiment, the real-time entropy value can also be generated by first performing a wavelet transform on the collected real-time indicator data when integrating the information collection interface of the cloud system. The wavelet transform can effectively decompose the different frequency components of the information and extract the key features in the data. The entropy value of each frequency component of the decomposed signal is calculated, and the entropy values are integrated to generate an instantaneous entropy value sequence.

[0045] In another optional embodiment, the real-time entropy value can also be generated by using a sliding window mechanism to adaptively sample real-time data in units of different time windows to dynamically capture the instantaneous value of the indicator of each interface; within each sliding window, the sample entropy of the real-time data of the interface is calculated, and the sample entropy can more accurately reflect the complexity of the data. Subsequently, the sample entropy value is used to generate a real-time entropy value sequence.

[0046] In the implementation manner of the present application, constructing a historical interface call chain in step S2 includes setting a log monitoring cycle, tracing the serial numbers of the interface calls in the historical logs of the cloud system, extracting the interface identifier corresponding to each serial number, and forming an interface set with a calling order; traversing the interface identifiers in each serial number interface set, using a union check set to merge the same elements of different sets, and constructing a historical interface call chain.

[0047] Specifically, set the log monitoring period, which determines the time range for monitoring and analyzing logs;

[0048] During each monitoring cycle, the system extracts interface call information from the historical logs of the cloud system;

[0049] Each interface call will record a unique serial number, which includes information such as interface identifier, call time, and interface execution status. The thread number is used as the serial number within the same cloud service, and the thread serial number carrying the initiation point service interface is used for cross-system services.

[0050] The system traces each serial number and associates it with the corresponding interface identifier by analyzing the interface call records. Specifically, the system extracts the serial number of each interface call in the log one by one, records the associated interface identifier, and organizes the interface set in the order of the serial numbers. An interface call sequence contains multiple interface identifiers, indicating the order of interface calls;

[0051] Traverse the interface set of each serial number, identify and merge the same elements in the set;

[0052] To effectively handle identical elements in an interface set, the system uses a Union-Find algorithm. The core of the Union-Find algorithm is to maintain a set of disjoint sets. When two interface identifiers belong to the same set, the system merges them into one set.

[0053] Each interface identifier is considered as an element. Through the "merge" and "find" operations of the union-find set, the system can gradually merge the same interface identifiers in different serial numbers.

[0054] Build a complete historical interface call chain. The merged result of each interface identifier will become a node, and they will be connected in sequence according to the call order to form a historical call chain.

[0055] The historical call chain shows the call sequence and association relationship of different interface identifiers in the same serial number, reflecting the call dependency between different interfaces in the system.

[0056] In an embodiment of the present application, the generation of indicator collection and control instructions in step S3 includes: constructing a directed graph topology structure based on the historical interface call chain using the FIoyd-WarshaⅡ algorithm; wherein, the node represents the interface instance, and the edge weight is set by the call frequency and delay calculation; based on the real-time calculated interface information entropy quantization value, generating a node embedding vector that integrates the topological structure and the interface entropy characteristics through the Node2Vec algorithm; monitoring the entropy mutation node based on a preset mutation ratio, starting from the entropy mutation node, traversing the topology graph based on the breadth-first search, identifying the influencing propagation path within its k-hop range, and outputting the associated node set; according to the propagation path length and the node entropy value weight, dynamically allocating the collection and control priority of the associated node set through the linear attenuation model, generating indicator collection and control instructions in sequence, and performing indicator collection and control on the interface cluster corresponding to the node set.

[0057] Specifically, a directed graph topology is constructed through the historical interface call chain;

[0058] Each interface instance is considered a node in the graph, while the relationships between interface calls are connected by directed edges;

[0059] The weight of each edge in the graph is set by the frequency and latency of the interface call through a weighted comprehensive operation;

[0060] Based on the interface information entropy quantization value calculated in real time, the system uses the Node2Vec algorithm to generate a node embedding vector that integrates the topology structure and interface entropy features;

[0061] The Node2Vec algorithm captures the structural information between nodes by performing random walks in the graph, and incorporates the entropy characteristics of the interface into the captured structural information;

[0062] The node embedding vector reflects the topological position of the node in the graph and the influence of its entropy characteristics, which can provide strong support for subsequent node analysis and decision-making;

[0063] The Node2Vec algorithm will perform multiple random walks in the graph, using the entropy feature as additional information about the node to generate an embedding vector for each node.

[0064] The generated embedding vector includes the topological location of the interface as well as entropy features;

[0065] Based on the preset mutation ratio, which is set to 30% by default, the system monitors the nodes with entropy mutations;

[0066] When the change in entropy value exceeds the preset mutation ratio, the monitoring node is marked as an entropy mutation node;

[0067] Then, starting from these entropy mutation nodes, the system uses breadth-first search (BFS) to traverse the topology graph, identify the impact propagation path within the k-hop range, and output the relevant node set;

[0068] The impact propagation path indicates the dependent data flows or other interface nodes that are affected starting from the mutation node;

[0069] The k-hop range means that starting from the mutation node, at most k layers of nodes are traversed;

[0070] The k-layer nodes to be traversed are set based on the topological complexity and the required monitoring range;

[0071] Assume that the entropy value of interface B mutates and exceeds the preset mutation rate. The system will start from interface B and perform a breadth-first search. If k = 2, the system will traverse the adjacent nodes of interface B in the graph, namely interface A and interface C, and output these nodes as the associated node set [interface A, interface B, interface C] in the influencing propagation path.

[0072] According to the propagation path length and node entropy weight, the collection priority of the associated node set is dynamically allocated through the linear attenuation model;

[0073] The longer the propagation path between nodes, the lower the priority of collection;

[0074] The higher the entropy value of a node, the higher the priority of collection;

[0075] The linear attenuation model calculates the collection priority of each node based on the path length and entropy weight. An exemplary linear attenuation model is expressed as:

[0076]

[0077] Where Q is the interface priority evaluation value, H and L are the entropy value of the node and the propagation path from the entropy mutation node to the current node, respectively; α and β are the weighting coefficient of entropy value to priority and the compensation coefficient of path length to priority, respectively;

[0078] Among them, α and β are both freely set by the Q value according to the emphasis on entropy value and path length, ensuring that the scale of the Q value calculation result can be accurately distinguished;

[0079] Based on the dynamically assigned collection priorities, the system generates indicator collection control instructions for the associated node set;

[0080] These instructions will guide the interface cluster to perform data collection and control, and give priority to collecting data from nodes with higher priority;

[0081] The purpose of establishing the collection priority order is to carry out the collection and control of the interface clusters corresponding to different interface call chains in steps, so as to prevent the system errors caused by different strategy adjustment scales between different clusters.

[0082] In an optional embodiment, the generation of indicator collection and control instructions can also be carried out after constructing a directed graph topology structure, by setting a series of entropy value thresholds, and calculating the information entropy value of each interface in real time. If the entropy value of the interface exceeds the preset threshold, the corresponding collection and control instructions are triggered; through the rule engine method, the entropy mutation is quickly responded to, and the automatic control instruction generation is realized, thereby ensuring that the collection frequency is adjusted in time under abnormal circumstances, thereby improving the accuracy and timeliness of data collection.

[0083] In another optional implementation, the generation of index collection control instructions can also introduce a genetic algorithm optimization control strategy. By analyzing historical collection results and current entropy values, the input variables are used as optimization targets. The genetic algorithm will iteratively optimize the control instructions according to the fitness evaluation value, and generate the optimal collection plan through mechanisms such as selection, crossover and mutation, thereby ensuring the efficient use of collection resources and the rationality of adjustments.

[0084] In an embodiment of the present application, constructing the interface status-strategy mapping relationship matrix in step S4 includes, after receiving the indicator collection and control instruction, obtaining and integrating the initial system resource snapshot allocated to the corresponding interface cluster, defining the baseline collection frequency interval parameters of each interface based on the entropy value timing data of each interface in the interface cluster and the interface load level, and generating a timing collection strategy table; using the K-means clustering algorithm to perform interface clustering on the entropy value sequence, dividing the interfaces into high entropy interfaces and low entropy interfaces, and constructing the interface status-strategy mapping relationship matrix in combination with the interface load rate level.

[0085] Specifically, after receiving the indicator collection and control instructions, the system interacts with the relevant resource management module interface in the cloud platform to obtain and integrate the initial system resource snapshots allocated by each interface cluster;

[0086] The initial system resource snapshot includes but is not limited to key parameters such as computing resources, storage resources, bandwidth, and response latency of each interface;

[0087] After obtaining the resource snapshot, the system calculates and defines the benchmark acquisition frequency interval for each interface based on the entropy value time series data and interface load level of each interface;

[0088] The baseline acquisition frequency interval will be adjusted according to the dynamic load of the interface and the change of entropy value;

[0089] Define the baseline acquisition frequency interval parameters for each interface. The acquisition frequency interval is adjusted based on the entropy fluctuation and load conditions of the interface.

[0090] Based on the combination of entropy fluctuation and load level, different adjustable intervals of acquisition frequency are set;

[0091] In the embodiment of the present application, the adjustable interval of the acquisition frequency is set according to the adjustable range of the current acquisition frequency compared to the upper and lower limits of the reference interval;

[0092] High entropy fluctuations combined with high loads usually require the most frequent data collection, so the adjustable collection frequency interval will be set to a higher value, such as [-0.1Hz, +0.5Hz]. Under low load conditions, even if the entropy value fluctuates greatly, the collection frequency will be lower, such as [-0.2Hz, +0.2Hz].

[0093] For interfaces with low load and small entropy fluctuations, the collection frequency is set to the lowest range of [-0.05Hz, +0.02Hz]. The upper and lower limits of the default range are set to no more than 5% of the current static collection frequency.

[0094] The specific interval is set by the operation and maintenance personnel based on the data collection volume and allocated resources of the cloud service system to which each interface belongs;

[0095] The K-means clustering algorithm is used to cluster interfaces based on the entropy value sequence, dividing the interfaces into high-entropy interfaces and low-entropy interfaces. The interface state-strategy mapping relationship matrix is constructed based on the interface load rate level.

[0096] The state-policy mapping relationship matrix is an N×W matrix;

[0097] Where N is the number of interfaces, W is the policy dimension of each interface, that is, the upper and lower limits of the collection frequency, and each row represents the collection policy corresponding to the joint status of an interface. The joint status is a combination of the interface status, load level, and interface identifier.

[0098] Interface states include high entropy and low entropy.

[0099] In an embodiment of the present application, generating a period-driven policy adjustment parameter set in step S5 includes predicting future period division boundaries through an LSTM model based on the historical time series entropy value sequence of the interface cluster, generating high-entropy and low-entropy period characteristics of the interface and attaching time validity labels; using a dual Q-learning algorithm to perform time-sharing adaptation on the interface state-policy mapping relationship matrix, shortening the collection interval of the high-entropy interface to the upper limit of the benchmark collection frequency according to the set control step size, and extending the collection interval of the low-entropy interface to the lower limit of the benchmark collection frequency according to the set control step size, thereby generating a period-driven policy adjustment parameter set.

[0100] Specifically, a historical time series entropy value sequence of the interface cluster is obtained, where the historical time series entropy value sequence includes the information entropy value of each interface within a certain period of time in the past;

[0101] Among them, the past certain time is consistent with the length of the log monitoring period and is used for the periodic change characteristics of the step sequence;

[0102] Use the long short-term memory (LSTM) network model to train and predict the entropy value sequence;

[0103] The LSTM model predicts the entropy value change trend in a certain period of time in the future by learning the temporal relationship in the historical entropy value sequence, thereby deriving the entropy value division boundary of the future period;

[0104] Based on the prediction results, the future period is divided into high entropy period and low entropy period, and a time validity label is added to each period;

[0105] Among them, the time validity tag includes but is not limited to the validity period, time period start and end time, etc.;

[0106] The dual Q-learning algorithm is used to handle the dynamic adjustment of high-entropy and low-entropy interfaces;

[0107] First, for each interface's historical time series data, a state-action space is constructed based on the interface status.

[0108] Then, dual Q-learning is used to perform time-sharing adaptation on the state-strategy mapping relationship matrix of each interface;

[0109] The interface status includes but is not limited to entropy value, load, etc.

[0110] It should be noted that during the predicted high entropy period, the interface collection interval is adjusted according to the change of the real-time entropy value of the interface. During this period, the interface collection frequency will be gradually shortened according to the preset control step until the upper limit of the benchmark collection frequency of the interface is reached;

[0111] The default increment is set to 5% of the adjustable range of the base acquisition frequency.

[0112] In an embodiment of the present application, the acquisition frequency control in step S6 includes sending the policy parameters in the policy adjustment parameter set to each acquisition interface in sequence to control the acquisition frequency of the interface; real-time monitoring of the mutation direction of the information entropy of the interface cluster during the control process, and if an interface state transition occurs, that is, the high-entropy interface state switches to a low-entropy interface, or the low-entropy interface state switches to a high-entropy interface, triggering a policy rollback and redistributing the Q-learning exploration weight; if no interface state transition occurs, then adjusting the dynamic fit of the fluctuation curve of the interface acquisition index and the interface historical acquisition index after calculating the strategy through a sliding time window, and triggering a policy rollback when the fit is lower than the preset quality threshold, and redistributing the Q-learning exploration weight.

[0113] Specifically, based on historical data analysis and LSTM prediction results, a time period-driven strategy adjustment parameter set is generated;

[0114] The policy adjustment parameter set includes but is not limited to the interface collection frequency, control step size, and collection time period division;

[0115] Based on the parameter set, the system sends the policy parameters to each collection interface in a predetermined order;

[0116] These parameters control the collection frequency of each interface in different time periods. Through dynamic adjustment, the interface collection frequency is increased during high entropy periods and reduced during low entropy periods, thereby optimizing the balance between data collection and resource usage.

[0117] A delay compensation factor is injected before the policy takes effect, and the policy switching timestamp is dynamically corrected based on the edge terminal clock synchronization error to ensure that the system-wide policy switching time difference does not exceed a single acquisition cycle;

[0118] Real-time monitoring and control of the information entropy mutation direction of the process interface cluster;

[0119] If an interface state transition occurs, that is, when a high-entropy interface switches to a low-entropy interface, or vice versa, the policy rollback is triggered and the Q-learning exploration weight is redistributed.

[0120] When the interface status transition occurs, it indicates that the adjustment strategy has affected the information characteristics of the interface data collected. The transition from high to low indicates that the data is losing information, corresponding to abnormal scenarios such as data loss or omission. The transition from low to high indicates that the data is increasing information, corresponding to abnormal scenarios such as erroneous or redundant collection.

[0121] If no interface state transition occurs, the dynamic fit between the fluctuation curve of the interface collection index and the interface historical collection index is adjusted through the sliding time window calculation strategy;

[0122] When the fit falls below the preset quality threshold, the policy rollback is triggered and the Q-learning exploration weight is redistributed;

[0123] Among them, the preset quality threshold is set by the operation and maintenance personnel based on the interface data characteristics, and the default setting is that the fitting deviation does not exceed 3%.

[0124] The dynamic fitting degree of the fluctuation curves of the interface acquisition index and the interface historical acquisition index is adjusted by the sliding time window calculation strategy, including adjusting the alignment error between the time series fluctuation curves of the interface acquisition index and the interface historical acquisition index by adopting the DTW calculation strategy, and constructing an error dynamic curve;

[0125] The sliding mean square error change rate of the error dynamic curve is calculated, the mean of the mean square error change rate is normalized, and the processing result is used as the dynamic fitting degree.

[0126] In Example 3, the above is an illustrative scheme of a method for collecting indicators of a cloud-based information system. It should be noted that the technical scheme of the system for collecting indicators of a cloud-based information system and the technical scheme of the method for collecting indicators of a cloud-based information system are of the same concept. For details not described in detail in the technical scheme of the system for collecting indicators of a cloud-based information system in this embodiment, please refer to the description of the technical scheme of the method for collecting indicators of a cloud-based information system.

[0127] This embodiment also provides a cloud information system indicator collection system, including: an information entropy quantification module, a call chain construction module, an entropy mutation identification module, a relationship matrix construction module, a strategy adjustment parameter module, and a collection frequency control module.

[0128] Among them, the information entropy quantification module is used to integrate the information collection interface of the cloud system, quantify the information entropy of the interface data flow, and generate a real-time entropy value sequence; the call chain construction module is used to extract historical logs to trace the serial numbers of interface calls, form an interface set with a call order, and construct a historical interface call chain; the entropy mutation identification module is used to establish a directed graph topology structure based on the entropy value sequence value and the historical interface call chain, identify the influence propagation path of the entropy mutation node, and generate indicator collection and control instructions; the relationship matrix construction module is used to generate a time series collection strategy table based on the entropy value time series data of each interface and the interface load level, cluster and divide high and low entropy interfaces, and construct an interface state-strategy mapping relationship matrix; the strategy adjustment parameter module is used to generate the time period characteristics of high and low entropy of the interface, and perform time-sharing adaptation on the interface state-strategy mapping relationship matrix through the dual Q-learning algorithm to generate a time period-driven strategy adjustment parameter set; the collection frequency control module is used to send the strategy adjustment parameter set to each collection interface for collection frequency control.

[0129] This embodiment also provides an electronic device suitable for collecting cloud information system indicators, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the cloud information system indicator collection method proposed in the above embodiment.

[0130] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for collecting cloud information system indicators proposed in the above embodiment is implemented.

[0131] The storage medium proposed in this embodiment and the method for implementing cloud information system indicator collection proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0132] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general hardware, and of course can also be implemented by hardware. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for collecting indicators of cloud information systems, characterized in that: include: Integrate cloud system information collection interfaces, quantify the information entropy of interface data streams, and generate real-time entropy value sequences; Extract historical logs to trace the serial numbers of interface calls, form a set of interfaces in the calling order, and build a historical interface call chain; Based on the entropy value sequence and the historical interface call chain, a directed graph topology is established to identify the impact propagation path of the entropy mutation node and generate indicator collection and control instructions; Based on the entropy time series data of each interface and the interface load level, a time series collection strategy table is generated, high-entropy and low-entropy interfaces are clustered and divided, and an interface state-strategy mapping relationship matrix is constructed; Generate high and low entropy time period characteristics of the interface, perform time-sharing adaptation of the interface state-policy mapping relationship matrix through the dual Q-learning algorithm, and generate a time period-driven policy adjustment parameter set; Send the policy adjustment parameter set to each collection interface to control the collection frequency.

2. The cloud information system indicator collection method according to claim 1, characterized in that: Generating a real-time entropy value sequence includes integrating all information collection interfaces of the cloud system, synchronously capturing the instantaneous value of the indicators of each interface through sliding windows of predefined multiple time scales, and generating a time series raw data stream with interface identifiers; Perform Z-Score normalization on the original data stream, analyze the statistical data distribution within the sliding window, and calculate the histogram frequency distribution of the data values; The Shannon entropy value is iteratively calculated based on the histogram frequency distribution of the data value. The information entropy of the data stream of the interface is quantified by calculating the probability distribution, and a real-time updated entropy value sequence is generated.

3. The cloud information system indicator collection method according to claim 2, characterized in that: The construction of the historical interface call chain includes setting a log monitoring cycle, tracing the serial numbers of interface calls in the historical logs of the cloud system, extracting the interface identifier corresponding to each serial number, and forming an interface set with a call sequence; Traverse the interface identifiers in each serial number interface set, use the union check set to merge the same elements of different sets, and build a historical interface call chain.

4. The cloud information system indicator collection method according to claim 3, characterized in that: The generating indicator collection and control instructions includes: constructing a directed graph topology structure based on the historical interface call chain using the FIoyd-Warsha II algorithm; Among them, nodes represent interface instances, and edge weights are set by call frequency and latency calculation; Based on the interface information entropy quantization value calculated in real time, the Node2Vec algorithm is used to generate a node embedding vector that integrates the topology structure and interface entropy features. Based on the preset mutation ratio, the entropy mutation nodes are monitored. Starting from the entropy mutation node, the topology graph is traversed based on breadth-first search to identify the impact propagation path within the k-hop range and output the associated node set. According to the propagation path length and the node entropy weight, the collection and control priority of the associated node set is dynamically allocated through the linear attenuation model, and the indicator collection and control instructions are generated in sequence to perform indicator collection and control on the interface cluster corresponding to the node set.

5. The cloud information system indicator collection method according to claim 4, characterized in that: The construction of the interface state-strategy mapping relationship matrix includes, after receiving the indicator collection and control instruction, obtaining and integrating the initial system resource snapshot allocated to the corresponding interface cluster, defining the benchmark collection frequency interval parameters of each interface based on the entropy value time series data of each interface in the interface cluster and the interface load level, and generating a time series collection strategy table; The K-means clustering algorithm is used to cluster the interfaces of the entropy value sequence, and the interfaces are divided into high-entropy interfaces and low-entropy interfaces. The interface state-strategy mapping relationship matrix is constructed based on the interface load rate level.

6. The method for collecting cloud information system indicators according to claim 5, wherein: The generation of the time period driven strategy adjustment parameter set includes predicting the future time period division boundaries through the LSTM model based on the historical time series entropy value sequence of the interface cluster, generating high entropy and low entropy time period features of the interface and adding time validity labels; The dual Q-learning algorithm is used to perform time-sharing adaptation on the interface state-strategy mapping relationship matrix. For high-entropy interfaces, the collection interval is shortened to the upper limit of the benchmark collection frequency according to the set control step size. For low-entropy interfaces, the collection interval is extended to the lower limit of the benchmark collection frequency according to the set control step size. This generates a time-period driven policy adjustment parameter set.

7. The cloud information system indicator collection method according to claim 6, characterized in that: The acquisition frequency control includes sending the policy parameters in the policy adjustment parameter set to each acquisition interface in sequence to control the acquisition frequency of the interface; Real-time monitoring of the information entropy mutation direction of the control process interface cluster. If an interface state transition occurs, that is, when the high-entropy interface state switches to a low-entropy interface, or when the low-entropy interface state switches to a high-entropy interface, the policy rollback is triggered and the Q-learning exploration weight is redistributed. If no interface state transition occurs, the sliding time window is used to calculate the strategy and adjust the dynamic fit of the fluctuation curve of the interface collection index and the interface historical collection index. When the fit falls below the preset quality threshold, the strategy rollback is triggered and the Q-learning exploration weight is redistributed.

8. A cloud-based information system indicator collection system, applying the method according to any one of claims 1 to 7, characterized in that: include: Information entropy quantification module, call chain construction module, entropy mutation identification module, relationship matrix construction module, strategy adjustment parameter module, acquisition frequency control module The information entropy quantification module is used to integrate the cloud system information collection interface, quantify the information entropy of the interface data flow, and generate a real-time entropy value sequence; The call chain building module is used to extract historical logs to trace the serial number of interface calls, form a set of interfaces in call order, and build a historical interface call chain; The entropy mutation identification module is used to establish a directed graph topology structure based on the entropy value sequence value and the historical interface call chain, identify the impact propagation path of the entropy mutation node, and generate indicator collection and control instructions; The relationship matrix construction module is used to generate a time series acquisition strategy table based on the entropy value time series data of each interface and the interface load level, cluster and divide high and low entropy interfaces, and construct an interface state-strategy mapping relationship matrix; The policy adjustment parameter module is used to generate the time period characteristics of high and low entropy of the interface, and perform time-sharing adaptation on the interface state-policy mapping relationship matrix through the dual Q-learning algorithm to generate a time period driven policy adjustment parameter set; The acquisition frequency control module is used to send the policy adjustment parameter set to each acquisition interface to perform acquisition frequency control.

9. An electronic device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the cloud information system indicator collection method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the cloud information system indicator collection method described in any one of claims 1 to 7.

Images