IOS application full-life-cycle management method, system and device and medium

Abstract

The invention discloses an iOS application full-life-cycle management method, system and device and a medium, and the method specifically comprises the steps that a distributed detection cluster is started to execute concurrent detection on nodes of a self-adaptive monitoring pool, and the concurrent detection comprises application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis; inputting a concurrent detection result of each node into a state analysis engine, and determining whether each node is an abnormal node or not according to a preset abnormal judgment condition; and based on the abnormal conditions of the plurality of abnormal nodes, associating the abnormal state with the advertisement putting data through a service influence evaluation model, obtaining a cost loss evaluation result, automatically generating a disposal suggestion according to the cost loss evaluation result, and executing a preset emergency scheme. According to the invention, real-time monitoring, automatic processing and efficient notification of the on-shelf state of the APP are realized, the user experience is improved, the economic loss is reduced, and a powerful guarantee is provided for popularization and operation of the iOS application.

Description

Technical Field

[0001] The present invention relates to the technical field of application management, and in particular to a method, system, device and medium for managing the entire life cycle of iOS applications. Background Art

[0002] With the rapid development of mobile internet, the promotion and operation of iOS applications (APPs) has become a focus for major companies and developers. In the operation of iOS games, app promotion often directs users to the Apple App Store for downloading. However, the Apple App Store strictly regulates the listing and removal of apps, creating the risk of app removal. Failure to promptly notify users of app removals can result in users being unable to successfully download the app, impacting user experience, advertising effectiveness, and return on investment.

[0003] Currently, the monitoring methods for iOS app listing status have the following shortcomings:

[0004] 1. Lack of real-time performance: Existing monitoring methods are mostly based on scheduled tasks or manual inspections. They cannot detect the app's listing status in real time and suffer from information lag. When an app is removed from the shelves, relevant personnel are often not notified in a timely manner, resulting in user loss and wasted advertising costs.

[0005] 2. Low automation: Most monitoring methods rely on manual inspection, which is inefficient and prone to errors. Manual inspection is not only time-consuming and labor-intensive, but can also miss important information due to negligence, affecting the accuracy and timeliness of monitoring.

[0006] 3. Incomplete notification mechanism: The lack of an effective notification mechanism prevents timely communication of app removal information to relevant personnel. Even if the monitoring system detects the app removal, if the operator is not notified in a timely manner, the problem will not be addressed in a timely manner, further exacerbating losses.

[0007] 4. Poor user experience: Users cannot download the app after clicking on the ad, resulting in a poor user experience and a negative impact on brand image. This may also create a negative impression of the brand and reduce brand loyalty.

[0008] 5. Significant economic losses: Wasted advertising costs and the loss of potential users result in unnecessary economic losses. If the app is removed from the app store during the advertising period, the advertising costs will be wasted and a large number of potential users may be lost. Summary of the Invention

[0009] The purpose of the present invention is to provide an iOS application full life cycle management method, system, device and medium. By constructing a set of efficient and automated iOS application full life cycle monitoring mechanisms, real-time monitoring, automated processing, efficient notification, improved user experience and reduced economic losses of APP listing status are achieved, providing strong guarantees for the promotion and operation of iOS applications, so as to solve at least one of the above-mentioned existing technical problems.

[0010] In a first aspect, the present invention provides a method for managing the entire lifecycle of an iOS application, the method specifically comprising:

[0011] Analyze Apple's global content delivery network architecture through CDN topology analysis algorithms, dynamically select multiple optimal monitoring nodes in the region based on node response latency, and form an initial adaptive monitoring pool;

[0012] Based on the initial adaptive monitoring pool, a digital twin model of the Apple CDN network was constructed. Using real-time network traffic data, node load status, and historical fault records, a reinforcement learning model was trained to dynamically select the optimal combination of monitoring nodes to obtain the target adaptive monitoring pool.

[0013] Launching a distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification, and user review sentiment analysis.

[0014] The concurrent detection results of each node are input into the status analysis engine, and each node is determined to be an abnormal node based on the preset abnormality judgment conditions;

[0015] Based on the abnormal conditions of several abnormal nodes, the business impact assessment model is used to associate the abnormal status with the advertising data to obtain the cost loss assessment results. According to the cost loss assessment results, disposal suggestions are automatically generated and preset emergency plans are implemented.

[0016] In a second aspect, the present invention provides an iOS application full lifecycle management system, the system specifically comprising:

[0017] The first management module is used to analyze Apple's global content delivery network architecture using a CDN topology analysis algorithm, dynamically select multiple optimal monitoring nodes in the region based on node response latency, and form an initial adaptive monitoring pool;

[0018] The second management module is used to build a digital twin model of the Apple CDN network based on the initial adaptive monitoring pool. Using real-time network traffic data, node load status, and historical fault records, the reinforcement learning model is trained to dynamically select the optimal combination of monitoring nodes to obtain the target adaptive monitoring pool.

[0019] The third management module is used to start the distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis;

[0020] The fourth management module is used to input the concurrent detection results of each node into the status analysis engine and determine whether each node is an abnormal node according to the preset abnormality judgment conditions;

[0021] The fifth management module is used to associate the abnormal status with the advertising data based on the abnormal conditions of several abnormal nodes through the business impact assessment model, obtain the cost loss assessment results, automatically generate disposal suggestions based on the cost loss assessment results and execute the preset emergency plan.

[0022] In a third aspect, the present invention provides a computer device comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, it implements the iOS application full life cycle management method as described in any one of the above methods.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for managing the entire life cycle of an iOS application as described in any one of the above methods is implemented.

[0024] Compared with the prior art, the present invention has at least one of the following technical effects:

[0025] 1. This invention builds an efficient and automated iOS application life cycle monitoring mechanism, which realizes real-time monitoring of APP listing status, automated processing, efficient notification, improved user experience and reduced economic losses, providing strong support for the promotion and operation of iOS applications.

[0026] 2. The present invention uses CDN topology analysis algorithms and digital twin models to achieve real-time monitoring and early warning of APP listing status, ensuring that operators can be informed of APP removal information in the first place.

[0027] 3. The present invention constructs a deep reinforcement learning network, dynamically selects the optimal combination of monitoring nodes, realizes the automation and intelligence of the monitoring process, reduces manual intervention, and improves monitoring efficiency and accuracy.

[0028] 4. The present invention establishes abnormal judgment conditions based on the status analysis engine and combines it with a preset notification mechanism to ensure that the APP removal information can be conveyed to relevant personnel in a timely and accurate manner so that countermeasures can be taken quickly.

[0029] 5. The present invention ensures that users can successfully download applications after clicking on advertisements through real-time monitoring and early warning, thereby improving user experience and enhancing brand image.

[0030] 6. Based on the business impact assessment model, the present invention associates abnormal status with advertising data to obtain cost loss assessment results, and automatically generates disposal suggestions and emergency plans, effectively reducing the waste of advertising costs and the risk of potential user loss caused by APP removal.

[0031] 7. The present invention constructs a network graph model based on the CDN topology analysis algorithm and dynamically selects the optimal monitoring nodes, effectively reducing monitoring delays and improving the adaptability and reliability of the initial monitoring pool.

[0032] 8. The present invention constructs a digital twin model and trains a reinforcement learning model to achieve intelligent dynamic combination of monitoring nodes in Apple's CDN network, optimize resource allocation, and improve monitoring efficiency and accuracy.

[0033] 9. The distributed detection cluster of the present invention performs concurrent detection on the nodes of the adaptive monitoring pool to achieve comprehensive analysis of application existence, code integrity, metadata compliance and user comment sentiment, thereby improving the comprehensiveness and real-time performance of application status monitoring.

[0034] 10. The present invention uses a weighted scoring model and a dynamic judgment threshold to achieve accurate assessment of the status of each node and improve the accuracy and timeliness of abnormal node identification.

[0035] 11. The present invention constructs a multi-objective loss function and uses the improved NSGA-II algorithm for optimization to generate a cost loss assessment result including the expected loss amount, optimal response time and ROI repair suggestions, providing a scientific basis for business decision-making.

[0036] 12. Based on the emergency response knowledge graph and Monte Carlo simulation, the present invention automatically generates and executes the optimal emergency plan, reduces the impact of abnormal events on the business, and improves the intelligence level of emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 This is a flowchart of a method for managing the entire lifecycle of an iOS application provided by one embodiment of the present invention;

[0039] Figure 2This is a schematic diagram of the structure of an iOS application full lifecycle management system provided by one embodiment of the present invention;

[0040] Figure 3 It is a structural diagram of a computer device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0041] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0042] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0043] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0044] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0045] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0046] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0047] In the embodiments of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A flowchart of the iOS application full lifecycle management method disclosed in the first embodiment of the present invention is shown, and is detailed as follows:

[0048] S101 analyzes Apple's global content distribution network architecture through a CDN topology analysis algorithm, dynamically selects multiple optimal monitoring nodes in the region based on node response delay, and forms an initial adaptive monitoring pool.

[0049] In this embodiment, relevant data of Apple's global content distribution network (CDN) is collected, including the geographic location information, network bandwidth, connection speed, etc. of CDN nodes. These data can be obtained through public network data sources, web crawler technology, and cooperation with Apple. The collected data is analyzed using a CDN topology analysis algorithm to construct the topology structure of Apple's global CDN. The topology analysis algorithm can use relevant algorithms in graph theory, such as the shortest path algorithm, the minimum spanning tree algorithm, etc., to determine the connection relationship and path between CDN nodes. Based on the results of the topology analysis, the architecture of Apple's global CDN is parsed, including core nodes, edge nodes, regional distribution, etc. By analyzing the architecture, the working principle and data transmission path of the CDN can be understood, providing a basis for subsequent monitoring node screening.

[0050] Multiple monitoring nodes are deployed in each area. These monitoring nodes can be servers, virtual machines, or IoT devices. The monitoring nodes periodically send requests to the Apple APPStore and measure the time delay from sending the request to receiving the response. Delay measurement can be done using network measurement tools such as ping, traceroute, etc. Based on the measured delay data, the optimal monitoring node in the area is dynamically screened. The screening criteria may include indicators such as average delay, minimum delay, and delay stability. For example, several monitoring nodes with the smallest average delay and the highest delay stability are selected as the optimal monitoring nodes. The optimal monitoring nodes screened out in each area are added to the initial adaptive monitoring pool. The initial adaptive monitoring pool is a dynamically changing collection, in which the monitoring nodes can be adjusted and updated according to actual conditions.

[0051] iOS apps that require monitoring are assigned to monitoring nodes in the initial adaptive monitoring pool. Each monitoring node is responsible for monitoring the listing status of one or more apps. Monitoring task allocation can use polling, random allocation, and other algorithms to ensure a balanced distribution of monitoring tasks. Monitoring nodes periodically send requests to the Apple App Store to check the listing status of specified apps. If the app is listed, the monitoring node records the monitoring results and proceeds to the next monitoring. If the app is removed from the App Store, the monitoring node immediately records the removal information and triggers a notification mechanism. When a monitoring node detects that an app has been removed from the App Store, it promptly communicates the removal information to relevant personnel through pre-set notification channels. Notification channels can include SMS, email, instant messaging tools, etc. At the same time, the notification information can include detailed information such as the app name, removal time, and monitoring node location, allowing relevant personnel to quickly locate the problem and take appropriate measures.

[0052] Regularly evaluate the performance of the monitoring nodes in the initial adaptive monitoring pool, using metrics such as monitoring accuracy, response latency, and stability. Automated testing tools and data analysis methods can be used for performance evaluation. Based on the performance evaluation results, dynamically adjust the monitoring nodes in the monitoring pool. If a monitoring node's performance degrades or fails, promptly remove it from the monitoring pool and replace it with a new optimal monitoring node. Simultaneously, adjust the number and distribution of monitoring nodes based on actual conditions to improve monitoring efficiency and accuracy. Continuously optimize the CDN topology analysis algorithm and monitoring node screening algorithm to improve their accuracy and efficiency. For example, machine learning algorithms can be used to analyze and model historical data to predict performance changes in CDN nodes, thereby more accurately screening the optimal monitoring nodes.

[0053] In this embodiment, a CDN topology analysis algorithm and a method for dynamically screening monitoring nodes are used to achieve real-time monitoring of iOS app listing status, significantly reducing information lag time. The entire monitoring process is automated, reducing manual intervention and improving monitoring efficiency and accuracy. An effective notification mechanism is established to promptly communicate app removal information to relevant personnel, ensuring that issues are addressed promptly. This reduces wasted advertising costs and the loss of potential users, thereby minimizing unnecessary economic losses.

[0054] S102, based on the initial adaptive monitoring pool, builds a digital twin model of the Apple CDN network. Through real-time network traffic data, node load status, and historical fault records, the reinforcement learning model is trained to dynamically select the optimal monitoring node combination to obtain the target adaptive monitoring pool.

[0055] In this embodiment, by constructing a digital twin model of Apple's CDN network and using a reinforcement learning model to dynamically select the optimal monitoring node combination, real-time optimization of the monitoring pool is achieved, thereby improving the accuracy and timeliness of iOS application listing status monitoring.

[0056] Specifically, the initial adaptive monitoring pool is formed by analyzing Apple's global content distribution network architecture through a CDN topology analysis algorithm, and dynamically screening multiple optimal monitoring nodes in the region based on the node response delay. The monitoring pool contains multiple monitoring nodes distributed in different regions, and each node has the ability to monitor the listing status of iOS applications in real time. Network traffic monitoring tools are deployed on each monitoring node in the initial adaptive monitoring pool to collect network traffic data of each node in the Apple CDN network in real time, including traffic size, traffic direction, traffic type, etc. By monitoring the system resource usage of the monitoring node itself, such as CPU utilization, memory occupancy, disk I / O, etc., the load status information of the node is obtained. A fault record database is established to collect fault information that has occurred in the past at each node in the Apple CDN network, including fault type, fault occurrence time, fault duration, etc.

[0057] A digital twin model of Apple's CDN network is constructed using collected real-time network traffic data, node load status data, and historical fault records. The digital twin model is a virtual simulation environment that closely resembles the actual CDN network, reflecting the network's operational status and performance metrics in real time. Based on the actual CDN network topology, the corresponding network topology, including core nodes, edge nodes, and regional distribution, is constructed within the digital twin model. Real-time network traffic data and node load status data are input into the digital twin model to simulate traffic distribution and node load conditions within the CDN network. Historical fault records are used to establish a fault prediction model. The digital twin model simulates the occurrence and development of faults and predicts potential future failures.

[0058] A monitoring node selection model based on deep reinforcement learning is designed. This model uses a digital twin model as the simulation environment, the combination of monitoring nodes as the action space, and monitoring performance indicators (such as monitoring accuracy, response latency, and stability) as the reward function. State representation: The network state (such as network traffic, node load, and fault conditions) in the digital twin model is used as the state input of the reinforcement learning model. Action space: The combination of monitoring nodes is defined as the action space, and each action represents the selection of a specific set of monitoring nodes for the monitoring task. Reward function: A reward function is designed based on the monitoring performance indicators. When the monitoring performance indicators reach the preset target, positive rewards are given; when the monitoring performance indicators do not meet the target, negative rewards are given.

[0059] The reinforcement learning model is trained using the digital twin model as a simulation environment. During training, the reinforcement learning model continuously tries different monitoring node combinations through interaction with the digital twin model and adjusts its own strategy based on the feedback of the reward function to learn the optimal monitoring node selection strategy. In the early stages of training, the reinforcement learning model adopts a random exploration strategy, trying various possible monitoring node combinations to accumulate experience. As training progresses, the reinforcement learning model gradually learns which monitoring node combinations can obtain higher rewards, thereby optimizing its own strategy and selecting more optimal monitoring node combinations. When the reward function value of the reinforcement learning model stabilizes and no longer improves significantly within a certain period of time, the model is considered to have converged and the training process ends.

[0060] In the actual Apple CDN network environment, the trained reinforcement learning model is deployed into the monitoring system. The monitoring system collects real-time CDN network operational status data and feeds it into the reinforcement learning model. Based on the current network status, the reinforcement learning model dynamically selects the optimal combination of monitoring nodes to form a target adaptive monitoring pool. Monitoring nodes in the target adaptive monitoring pool perform the task of monitoring the iOS app listing status according to the combination selected by the reinforcement learning model. Each monitoring node regularly sends requests to the Apple App Store to check the listing status of designated apps and feeds back the monitoring results to the monitoring system. As the Apple CDN network dynamically changes and business needs evolve, the reinforcement learning model requires continuous updating and optimization. The monitoring system regularly collects actual monitoring data, including metrics such as monitoring accuracy, response latency, and stability, and feeds it back to the reinforcement learning model. Based on this feedback, the reinforcement learning model adjusts and optimizes its strategies to adapt to the ever-changing environment.

[0061] In this embodiment, based on a digital twin model and reinforcement learning model, dynamic changes in the Apple CDN network can be perceived in real time and the optimal combination of monitoring nodes can be dynamically selected, significantly improving the real-time and accuracy of monitoring. The entire monitoring node optimization process is automated and intelligent, reducing manual intervention and improving monitoring efficiency and scientific decision-making. The reinforcement learning model can automatically adjust the monitoring node combination based on different network conditions and business needs, demonstrating strong adaptability and flexibility.

[0062] S103: Start the distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis.

[0063] In this embodiment, in order to ensure the compliance, security and user experience of the APP, it is necessary to conduct real-time and comprehensive monitoring of the APP in the IOS application store. However, traditional monitoring methods often have problems such as low detection efficiency, single detection dimension, and difficulty in coping with large-scale concurrent detection. To solve these problems, the present invention starts a distributed detection cluster and performs concurrent detection on the nodes in the adaptive monitoring pool, including application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis, to improve the efficiency, accuracy and comprehensiveness of APP monitoring.

[0064] Specifically, a distributed detection cluster architecture is designed, consisting of multiple detection nodes and a central control node. The detection nodes are responsible for executing specific detection tasks, while the central control node is responsible for task scheduling, result aggregation, and policy adjustment.

[0065] Detection nodes: High-performance servers or virtual machines are used as detection nodes. Each node is equipped with necessary software tools and libraries to support detection tasks such as application existence verification, code hash comparison, metadata compliance verification, and user review sentiment analysis.

[0066] Central control node: Deploy a powerful server as the central control node, install cluster management software and data analysis tools, and implement centralized management and monitoring of detection nodes.

[0067] Detection node deployment: Deploy detection nodes in different geographical locations and network environments to simulate real user access scenarios and improve detection accuracy and reliability.

[0068] Software configuration: Install and configure necessary software tools on each detection node, such as web crawlers, code analysis tools, natural language processing libraries, etc., to meet different detection requirements.

[0069] The central control node allocates detection tasks to each detection node based on the number of target nodes and the complexity of the detection task. Task scheduling can use polling, weighted polling, least connection number and other algorithms to achieve load balancing.

[0070] After receiving a detection task, each detection node simultaneously performs concurrent detection on the target node. Concurrent detection includes the following four aspects: 1. App Existence Verification: The detection node sends a network request to the target node to check whether the specified app exists in the app store. It analyzes the status code returned by the network request to determine the app's existence. For example, a status code of 200 indicates the app exists, while a status code of 404 indicates it does not. 2. Code Hash Comparison: If the app exists, the detection node downloads the app's installation package or code file from the target node. It uses a hash algorithm (such as MD5 or SHA-256) to calculate the hash value of the downloaded code file. The calculated hash value is compared with the pre-stored correct hash value to determine whether the code has been tampered with. 3. Metadata Compliance Verification: The detection node obtains app metadata from the target node, such as the app name, version number, developer information, and permission list. The extracted metadata is matched against pre-defined compliance rules to verify compliance with relevant laws and regulations and app store policy requirements. 4. User Review Sentiment Analysis: Web crawler technology is used to capture user reviews of the app from the target node. Use natural language processing technology to perform sentiment analysis on the captured comments to determine whether the user's evaluation of the APP is positive, negative or neutral.

[0071] After each detection node completes the detection task, it uploads the detection results to the central control node. The detection results include the application existence status, code hash comparison results, metadata compliance verification results, and user comment sentiment analysis results. The central control node summarizes and analyzes the collected detection results and generates a detailed detection report. The analysis content includes: 1. Overall detection statistics: statistics on the number of applications detected, the number of applications with problems, the distribution of problem types, etc. 2. Abnormal node identification: Based on the detection results, identify target nodes with abnormalities, such as the APP does not exist, the code has been tampered with, the metadata is non-compliant, etc. 3. User feedback analysis: Conduct an in-depth analysis of the user comment sentiment analysis results to understand users' satisfaction and needs for the APP, and provide a reference for the optimization and improvement of the APP.

[0072] Based on detection results and analysis reports, the monitoring strategy of the adaptive monitoring pool is adjusted and optimized. For example, this can include increasing the monitoring frequency of abnormal nodes and adjusting the priority of detection tasks. Based on the execution of detection tasks and resource utilization, the distributed detection cluster is optimized. For example, this can include increasing or decreasing the number of detection nodes, adjusting their deployment locations, and optimizing task scheduling algorithms.

[0073] In this embodiment, the concurrent detection capability of the distributed detection cluster is utilized to greatly improve the efficiency of APP monitoring, and a large number of APPs can be fully detected in a short period of time. By applying multiple dimensions of detection such as existence verification, code hash comparison, metadata compliance verification, and user comment sentiment analysis, the accuracy of APP monitoring is improved, and problems and risks of APP can be discovered in a timely manner. Not only does it focus on the existence status and code security of the APP, but it also analyzes the metadata compliance and user feedback of the APP, providing more comprehensive information for the operation and management of the APP. The design of the distributed detection cluster and the adaptive monitoring pool makes the present invention highly adaptable and flexible, and can be dynamically adjusted and optimized according to different monitoring needs and network environments.

[0074] S104: Input the concurrent detection results of each node into the state analysis engine, and determine whether each node is an abnormal node through the preset abnormality judgment conditions.

[0075] In this embodiment, traditional node monitoring methods are often only able to perform simple status checks on a single node, making it difficult to cope with the complex state changes of nodes in large-scale distributed systems. Furthermore, due to the lack of effective status analysis methods, it is impossible to accurately determine whether a node is an abnormal node, resulting in inefficient system maintenance and management. To address these issues, the present invention inputs the concurrent detection results of each node into a status analysis engine and uses preset abnormality judgment conditions to perform a comprehensive analysis of the node status, thereby accurately determining whether each node is an abnormal node and improving the monitoring and management efficiency of the distributed system.

[0076] Specifically, a state analysis engine architecture was designed, consisting of a data input module, an analysis and processing module, and a result output module. The data input module receives concurrent detection results uploaded by each monitoring node; the analysis and processing module analyzes and processes the detection results using pre-set anomaly determination criteria; and the result output module outputs the analysis results to the system administrator in the form of a visual interface and report files. Based on the business needs and operational experience of the distributed system, a series of anomaly determination criteria were established. These criteria can be categorized as either single-indicator or comprehensive-indicator.

[0077] The single indicator judgment condition sets one or more thresholds for each detection indicator. When a detection indicator of a target node exceeds or falls below the corresponding threshold, the node is considered to have an abnormality in that indicator. For example, if the CPU usage of a target node exceeds 80% for five consecutive minutes, the node is considered to have an abnormality in its CPU usage indicator.

[0078] Comprehensive indicator judgment criteria: This rule sets comprehensive judgment rules by comprehensively considering the relationships between multiple detection indicators. For example, if the target node's network connection status is disconnected and the service response time exceeds 10 seconds, the node is considered an abnormal node.

[0079] After completing concurrent testing, each node transmits the test results via the network to the data input module of the status analysis engine. The data input module integrates and preprocesses the received test results, classifying and summarizing the test data from different monitoring nodes according to the target node. The integrated test results are converted into a data format recognizable by the status analysis engine, such as JSON and XML. During the data format conversion process, data integrity and accuracy are ensured to avoid data loss or errors.

[0080] The analysis and processing module analyzes each detection indicator of each target node one by one according to the single indicator judgment condition. When it is found that a certain detection indicator meets the abnormal judgment condition, the abnormal situation of the node on this indicator is recorded. After completing the single indicator abnormality judgment, the analysis and processing module conducts a comprehensive analysis of multiple detection indicators of the target node according to the comprehensive indicator judgment condition. When the target node meets the comprehensive judgment rule, the node is judged to be an abnormal node, and the abnormal information is recorded in the analysis results. According to the severity of the abnormality, the nodes judged to be abnormal are divided into levels. For example, the abnormalities are divided into three levels: severe abnormalities, general abnormalities and minor abnormalities, so that the system administrator can take corresponding processing measures according to the abnormality level.

[0081] The result output module presents the anomaly determination results generated by the analysis and processing module to system administrators in a visual interface. This visual interface includes a node status distribution diagram, a list of abnormal nodes, and detailed anomaly information, allowing system administrators to intuitively understand the status of each node in the distributed system. A detailed anomaly determination report is also generated, including detailed information about the abnormal node, anomaly type, anomaly level, and occurrence time. The report can be saved and exported in formats such as PDF and Excel, enabling system administrators to conduct further analysis and processing.

[0082] In this embodiment, a state analysis engine comprehensively analyzes the concurrent detection results of each node. Combined with pre-set anomaly determination criteria, it accurately determines whether each node is an anomaly, avoiding the limitations of a single detection method. The distributed node concurrent detection system can simultaneously detect multiple nodes, greatly improving monitoring efficiency and enabling the timely detection of anomaly nodes in the distributed system. Through a visual interface and detailed anomaly determination reports, system administrators can intuitively understand the status of each node in the distributed system, facilitating management and decision-making.

[0083] S105, based on the abnormal conditions of several abnormal nodes, the abnormal conditions are associated with the advertising data through the business impact assessment model to obtain the cost loss assessment result, and disposal suggestions are automatically generated according to the cost loss assessment result and the preset emergency plan is executed.

[0084] In this embodiment, when faced with node anomalies, existing monitoring systems often lack effective means of business impact assessment and cost loss analysis, and are unable to accurately quantify the impact of abnormal nodes on the advertising delivery business. It is also difficult to automatically generate reasonable disposal suggestions and implement emergency plans based on the cost loss situation. This results in a large amount of manual analysis and decision-making work often being required when handling node anomalies, which is inefficient and prone to errors. To solve these problems, the present invention obtains cost loss assessment results by associating abnormal states with advertising delivery data, and automatically generates disposal suggestions and executes preset emergency plans based on the results. The present invention can accurately assess the cost losses of abnormal nodes to the advertising delivery business, improve the automation and efficiency of exception handling, reduce manual intervention, and reduce advertising delivery cost losses.

[0085] Specifically, data related to advertising delivery is collected, including advertising delivery plans (such as delivery time, delivery channels, delivery audiences, etc.), advertising delivery logs (such as number of ad impressions, number of clicks, number of conversions, etc.), and advertising cost data (such as cost per impression, cost per click, etc.).

[0086] Design a business impact assessment model that takes information about abnormal nodes and advertising data as input and outputs cost loss assessment results. The model can be constructed using machine learning algorithms (such as decision trees, neural networks, support vector machines, etc.) or rule-based reasoning methods. Extract features related to business impact from abnormal node monitoring data and advertising data. For example, for abnormal server nodes, features such as the duration of server downtime, the number of planned advertisements during the downtime, and the affected advertising channels can be extracted; for advertising data, features such as advertising budget, actual advertising costs, and expected conversion revenue can be extracted. Collect historical abnormal node data and corresponding advertising cost loss data as training samples to train the business impact assessment model. During the training process, continuously adjust the model parameters so that the model can accurately predict the cost loss of advertising under different abnormal node conditions.

[0087] When a new abnormal node is detected, the abnormal information and related advertising data are input into the trained business impact assessment model. Based on the input data, the business impact assessment model uses internal learning algorithms or rules to calculate the cost loss of the abnormal node to the advertising business. The cost loss assessment results can include multiple aspects, such as direct cost losses (such as wasted funds due to failed advertising) and indirect cost losses (such as potential customer loss due to poor advertising performance).

[0088] Based on the cost-loss assessment results and the characteristics of the advertising delivery business, a series of action suggestion rules are developed. For example, when the cost-loss assessment result is below a certain threshold, a "monitor and observe" suggestion is generated; when the cost-loss assessment result is within the medium range, a "adjust delivery strategy" suggestion is generated; and when the cost-loss assessment result is high, a "pause delivery and repair nodes" suggestion is generated. The cost-loss assessment results are input into the action suggestion generation module, which automatically generates corresponding action suggestions based on pre-set rules. Action suggestions can include specific steps, responsible persons, and estimated completion time.

[0089] Develop corresponding preset emergency plans for different types of abnormal nodes and cost loss situations. Emergency plans may include node repair plans (such as restarting the server, repairing the network connection, etc.), advertising adjustment plans (such as adjusting the delivery time, changing the delivery channel, etc.), and cost compensation plans (such as providing refunds or coupons to advertisers, etc.). After the disposal suggestion is generated, the system automatically triggers the execution of the preset emergency plan. Through the interface with relevant systems (such as the node management system, advertising delivery platform, financial system, etc.), the automatic execution of the emergency plan is realized. For example, the interface of the node management system is automatically called to repair the abnormal node, the delivery strategy of the advertising delivery platform is automatically adjusted, and the cost compensation information is automatically recorded in the financial system.

[0090] After the emergency plan is implemented, the repair of abnormal nodes and the recovery of advertising services are tracked and evaluated. Relevant data, such as the time it takes for nodes to resume normal operations and changes in advertising effectiveness indicators (such as number of impressions, click-through rate, conversion rate, etc.), are collected to evaluate the effectiveness of the emergency plan. Based on the results of the effectiveness evaluation, feedback is provided and optimization is made to the business impact assessment model, disposal recommendation rules, and preset emergency plans. If the model prediction is inaccurate or the emergency plan is ineffective, the model parameters, rule content, and emergency plan steps are adjusted in a timely manner to improve the performance and reliability of the system.

[0091] In this embodiment, based on the business impact assessment model, the cost loss of advertising delivery services caused by abnormal nodes can be accurately assessed, providing a scientific basis for decision-making. Automatically generating action suggestions based on the cost loss assessment results reduces the workload of manual analysis and decision-making, and improves the efficiency of exception handling. The automated execution of pre-set emergency plans enables rapid response to abnormal node situations, reducing the interruption time and cost of advertising delivery services.

[0092] In some embodiments, in step S101 above, analyzing Apple's global content delivery network architecture using a CDN topology analysis algorithm and dynamically selecting multiple optimal monitoring nodes in a region based on node response latency to form an initial adaptive monitoring pool specifically includes:

[0093] Using a topology discovery algorithm deployed in a network probe, ICMP probe packets are periodically sent to Apple CDN nodes. Based on the returned TTL value and the number of route hops, a CDN network graph model is constructed. The node set of the CDN network graph model includes multiple available CDN nodes, and the edge set of the CDN network graph model includes communication links between available CDN nodes.

[0094] For each node in the CDN network graph model, multiple HTTP HEAD requests are sent continuously, and the sending and receiving timestamps of each request are recorded to calculate the comprehensive latency score.

[0095] Based on the comprehensive latency score, all nodes are sorted using a preset node optimization function to obtain a dynamic node optimization result. The node optimization function is used to comprehensively calculate latency, availability, and geographic weight.

[0096] The top K nodes in the dynamic node selection results are selected to form the initial adaptive monitoring pool.

[0097] In this embodiment, Apple's global content delivery network (CDN) has a large number of nodes distributed around the world, which are used to distribute various contents to users quickly and stably. In order to ensure the normal operation of the CDN network and the efficient distribution of content, it is necessary to monitor the CDN nodes in real time. However, due to the large scale of the CDN network and the wide distribution of nodes, the traditional fixed monitoring node method has many problems, such as limited monitoring range, inability to adapt to network topology changes in time, inaccurate monitoring results, etc. In order to solve these problems, the present invention constructs a CDN network graph model through a topology discovery algorithm deployed in a network probe, calculates a comprehensive delay score based on the node response delay, and uses a node optimization function to sort the nodes, and finally selects the optimal node to form an initial adaptive monitoring pool. The present invention can dynamically adapt to changes in CDN network topology and improve the accuracy and efficiency of monitoring.

[0098] Specifically, network probes are deployed in multiple key regions around the world, equipped with network data collection and analysis capabilities. These probes can communicate with the Apple CDN network via a virtual private network (VPN) or direct internet access. A topology discovery algorithm based on the ICMP (Internet Control Message Protocol) is integrated into the network probes. The network probes periodically send ICMP probe packets to Apple CDN nodes, with the TTL (Time To Live) value of the probe packets set to different initial values, in order to obtain information about the data packet's transmission path within the network.

[0099] The network probe receives ICMP response packets returned from CDN nodes and extracts the TTL value and routing hop count information. It also records the timestamps of sending the probe packet and receiving the response packet to facilitate subsequent calculation of node response latency. Based on the collected TTL values and routing hop count information, a CDN network graph model is constructed. Each CDN node is represented as a node in the graph model, and the node set includes all available CDN nodes. Communication links between nodes are represented as edges in the graph model, and the edge set includes all communication links between available CDN nodes. Edge weights can be set based on factors such as routing hop count and link bandwidth.

[0100] The network probe sends multiple HTTP HEAD requests to each node in the CDN network graph model. HTTP HEAD requests are similar to GET requests, but only retrieve response header information, not the response body content, thereby reducing network traffic and transmission time. The send and receive timestamps of each HTTP HEAD request are recorded, and the response latency for each request is calculated. Statistical analysis of the response latency of these multiple requests, such as calculating the average and standard deviation, is performed to obtain a comprehensive latency score for the node. This comprehensive latency score takes into account factors such as request success rate, average response time, and response time fluctuation.

[0101] Design a node optimization function that comprehensively calculates a node's latency, availability, and geographic weight. Latency is calculated based on the node's comprehensive latency score; lower latency indicates a higher score. Availability is calculated based on the node's historical availability data; higher availability indicates a higher score. Geographic weight is calculated based on the node's geographic distance from the monitoring area; closer distances indicate higher scores. Use the node optimization function to sort all nodes in the CDN network graph model to obtain dynamic node optimization results. Based on actual needs, select the top K nodes from the dynamic node optimization results to form the initial adaptive monitoring pool. The value of K can be adjusted based on factors such as monitoring accuracy and network scale. Repeat the above steps periodically (e.g., hourly or daily) to dynamically update the nodes in the adaptive monitoring pool based on changes in the CDN network topology and node response latency. When new CDN nodes join the network or the performance of existing nodes changes, the node optimization function recalculates the node scores and adjusts the node composition of the monitoring pool.

[0102] Furthermore, the node optimization function satisfies

[0103]

[0104] in, represents the node optimization function value, α, β, γ and δ represent the adjustment coefficients, τ i Indicates the comprehensive delay score, A i represents the historical availability of the i-th node, A max represents the maximum allowable availability, G(·) represents the geographical location matching function, G(l i ,l large ) represents node l i and target area l large The spatial distance weight of N represents the number of nodes, and n represents the index of the node. Indicates the receiving timestamp of the nth detection, Indicates the timestamp of the nth detection, σ i represents the standard deviation of the delay, and λ represents the stability weight coefficient.

[0105] Apple's CDN network is monitored in real time using nodes in the initial adaptive monitoring pool. Monitoring data includes metrics such as node response latency, availability, and bandwidth utilization. This data is collected regularly using network probes and stored in a dedicated database. This collected monitoring data is analyzed to generate CDN network performance reports and anomaly warnings. Performance reports can include the overall network operating status and performance comparisons across regions. Anomaly warnings are issued promptly when abnormal conditions such as node failures and network congestion are detected, allowing operations personnel to take appropriate measures.

[0106] In this embodiment, the present invention dynamically adjusts the node composition of the adaptive monitoring pool based on changes in the CDN network topology and node response latency, ensuring accurate and effective monitoring. By employing a topology analysis algorithm and a node optimization function, optimal monitoring nodes can be quickly selected, reducing waste of monitoring resources and improving monitoring efficiency. By comprehensively considering factors such as node latency, availability, and geographic weight, monitoring results are more accurate and reliable, enabling timely detection of anomalies in the CDN network.

[0107] In some embodiments, in step S102 above, the digital twin model of the Apple CDN network is constructed based on the initial adaptive monitoring pool, and a reinforcement learning model is trained to dynamically select the optimal combination of monitoring nodes using real-time network traffic data, node load status, and historical fault records to obtain the target adaptive monitoring pool, specifically including:

[0108] By deploying data collectors on edge computing nodes, we can obtain the load rate, inbound traffic, and failure probability of Apple CDN nodes in real time and build a digital twin model containing multi-dimensional state vectors.

[0109] Construct a deep reinforcement learning network with an input layer dimension consistent with the twin state vector dimension of the digital twin model and an output layer dimension equal to the number of candidate nodes in the initial adaptive monitoring pool;

[0110] Inputting node data of the initial adaptive monitoring pool into the digital twin model to generate training samples;

[0111] The training samples are input into a deep reinforcement learning network. Combined with the reward function, the policy network parameters are updated through the double-delayed deep deterministic policy gradient algorithm to output a new node selection combination and obtain the target adaptive monitoring pool. The reward function is used to comprehensively calculate the node load penalty term, failure risk term and area coverage reward term.

[0112] In this embodiment, the present invention achieves efficient monitoring of the CDN network by constructing a digital twin model of the CDN network and using a deep reinforcement learning algorithm to dynamically select the optimal combination of monitoring nodes based on the real-time network status. Digital twin technology can build a virtual model of the CDN network that reflects the actual network status in real time; deep reinforcement learning technology can dynamically adjust the combination of monitoring nodes through interactive learning with the environment, improving monitoring efficiency and accuracy.

[0113] Specifically, data collectors are deployed at the edge computing nodes of Apple's CDN network. These data collectors have real-time data collection and transmission capabilities. The data collectors can communicate with CDN nodes through network interfaces to obtain relevant data of the nodes. The data collectors obtain data such as the load rate, inbound traffic, and failure probability of Apple's CDN nodes in real time. The load rate can be calculated by monitoring the node's CPU, memory, disk and other resource usage; the inbound traffic can be obtained through network traffic monitoring tools; and the failure probability can be predicted based on the node's historical failure records and current operating status. Based on the collected real-time data, a digital twin model containing a multi-dimensional state vector is constructed. The multi-dimensional state vector includes characteristics such as the node's load rate, inbound traffic, and failure probability. The digital twin model can be constructed using machine learning algorithms (such as decision trees, neural networks, etc.) or methods based on physical models to accurately reflect the real-time status of the CDN network.

[0114] Build a deep reinforcement learning network with an input layer dimension that matches the twin state vector of the digital twin model and an output layer dimension equal to the number of candidate nodes in the initial adaptive monitoring pool. The input layer receives the state vector output by the digital twin model, and the output layer outputs the selection probability of each candidate node. Initialize the parameters of the deep reinforcement learning network, including the weight matrix and bias vector. You can use random initialization or a pretrained model for initialization to improve network training efficiency and performance.

[0115] The node data of the initial adaptive monitoring pool is input into the digital twin model to generate training samples. The training samples include state vectors and corresponding labels (i.e., the optimal monitoring node combination). Labels can be obtained through expert experience, historical data, or simulation experiments. For example, the best performing monitoring node combination in historical monitoring data can be used as a label, or the optimal monitoring node combination can be calculated as a label by simulating different network states.

[0116] Design a reward function to comprehensively calculate the node load penalty, failure risk, and regional coverage reward. The node load penalty is calculated based on the monitoring node's load rate; higher load rates result in larger penalties. The failure risk is calculated based on the node's failure probability; higher failure probabilities result in larger penalties. The regional coverage reward is calculated based on the monitoring node's coverage of the CDN network area; the wider the coverage, the larger the reward.

[0117] Training samples are input into a deep reinforcement learning network. Combined with the reward function, the policy network parameters are updated using the double-delayed deep deterministic policy gradient (TD3) algorithm. The TD3 algorithm is an advanced deep reinforcement learning algorithm that effectively addresses the overestimation problem inherent in traditional deep deterministic policy gradient (DDPG) algorithms, improving training stability and performance. During training, the deep reinforcement learning network interacts with the virtual environment constructed by the digital twin model, continuously adjusting the monitoring node combination to maximize the cumulative reward. After multiple training iterations, the network learns the optimal monitoring node selection strategy. After training, the deep reinforcement learning network outputs a new node selection combination. Based on this output, several nodes with the highest probability are selected to form a target adaptive monitoring pool. The target adaptive monitoring pool can be adjusted based on actual needs, such as determining the number of monitoring nodes based on factors such as network scale and monitoring accuracy. Nodes in the target adaptive monitoring pool are used to monitor the Apple CDN network in real time. Simultaneously, network status data is continuously collected to update the digital twin model, and the deep reinforcement learning network is regularly retrained and optimized to adapt to dynamic changes in network status. When the network topology changes significantly or new business needs arise, the above steps can be repeated to build a new target adaptive monitoring pool.

[0118] In this embodiment, the present invention can dynamically select the optimal combination of monitoring nodes based on the real-time status of the CDN network, thereby improving the efficiency and accuracy of monitoring. Compared with the traditional fixed monitoring node combination, the present invention can better adapt to the dynamic changes in network traffic and the occurrence of node failures. By comprehensively considering factors such as node load, failure risk and regional coverage, the present invention can select a more reliable combination of monitoring nodes, reducing the problems of monitoring interruptions or inaccurate data caused by node failure or overload. By optimizing the combination of monitoring nodes, the present invention can reduce unnecessary waste of monitoring resources and reduce monitoring costs. At the same time, the high efficiency of the deep reinforcement learning algorithm also improves the operating efficiency of the monitoring system.

[0119] In some embodiments, in step S103 above, the distributed detection cluster is started to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification, and user comment sentiment analysis, specifically including:

[0120] Based on the node distribution of the adaptive monitoring pool, detection agent nodes are deployed in each geographical area, and each agent node is configured with an independent IP address and device fingerprint;

[0121] The detection proxy node concurrently sends M HTTPS HEAD requests to the target node. When a 404 status code is received for a preset number of consecutive times and the survival probability is less than the preset survival probability threshold, the target node is marked as a failed node and the application existence verification result is obtained;

[0122] If the application existence verification result passes, the application metadata is downloaded and the differential hash value is calculated. The binary difference is compared with the baseline version to determine whether a tampering alarm is triggered;

[0123] Parse the text of the application metadata, calculate the compliance score through the keyword matching engine, and determine whether there is a violation based on the compliance score;

[0124] Grab the user comment dataset associated with the target node and use the BERT model to analyze the sentiment index.

[0125] In this embodiment, the present invention deploys detection agent nodes in each geographical area, and performs concurrent detection operations such as application existence verification, code hash comparison, metadata compliance verification, and user comment sentiment analysis on the nodes in the adaptive monitoring pool, thereby achieving comprehensive monitoring and evaluation of node status.

[0126] Specifically, based on the geographical distribution of nodes in the adaptive monitoring pool, the world is divided into multiple geographical regions, such as Asia, Europe, and America. Each geographical region contains a certain number of nodes to be detected. Detection agent nodes are deployed in each geographical region, and each agent node is configured with an independent IP address and device fingerprint. The independent IP address is used to communicate with the target node, and the device fingerprint is used to identify the identity and characteristics of the agent node to prevent it from being identified and blocked by the target node. Each detection agent node is configured with the necessary software and hardware resources, including an operating system, a network communication module, a data storage module, etc. At the same time, the operating parameters of the agent node are configured, such as the number of concurrent requests, the request interval, etc., to ensure the efficient execution of the detection task.

[0127] The detection proxy node concurrently sends M HTTPS HEAD requests to the target node in the adaptive monitoring pool. The HEAD request is used to obtain the response header information of the target node without downloading the entire page content, thereby reducing network bandwidth consumption. When the detection proxy node receives a 404 status code for a preset number of consecutive times (such as 3 times), and the survival probability calculated based on historical data and current network conditions is less than the preset survival probability threshold (such as 80%), the target node is marked as a failed node. The calculation of the survival probability can use a machine learning algorithm, taking into account factors such as the node's historical fault records, network delays, and response time. Based on the marking of the failed node, the application existence verification result is generated. If the target node is not marked as a failed node, the application existence verification is considered to have passed; otherwise, the verification is considered to have failed.

[0128] If the application existence verification result passes, the detection agent node downloads the application metadata from the target node. The application metadata includes the application installation package, configuration files, log files, etc. The downloaded application metadata is processed and its differential hash value is calculated. The differential hash value is a hash algorithm used to compare the differences between two files. It can quickly detect changes in file content. The calculated differential hash value is compared with the differential hash value of the baseline version for binary difference. If the difference exceeds the preset threshold (such as 10%), it is considered that the application may have been tampered with, and a tampering alarm is triggered.

[0129] Perform text parsing on downloaded application metadata to extract key information, such as the application name, version number, developer information, and permission list. A keyword matching engine is used to match the extracted key information with pre-set compliance rules. Compliance rules include legal and regulatory requirements, industry standards, and corporate security policies. Based on the matching results, a compliance score is calculated. The compliance score is used to determine whether the target node is in violation. If the compliance score is below a pre-set threshold (e.g., 60 points), the target node is considered to have violated the regulations.

[0130] Using web crawling technology, we capture a dataset of user reviews associated with the target node. This user review data can come from app stores, social media platforms, forums, and other sources. We then use a pre-trained BERT model to perform sentiment analysis on this user review dataset. The BERT model automatically extracts semantic information from reviews and calculates a sentiment index. The sentiment index is categorized as positive, negative, or neutral, and is used to assess user satisfaction and reputation for the target node.

[0131] In this embodiment, a distributed detection cluster can concurrently test multiple nodes in the adaptive monitoring pool, greatly improving detection efficiency. By applying multi-dimensional testing such as existence verification, code hash comparison, metadata compliance verification, and user comment sentiment analysis, the present invention can accurately identify problematic nodes.

[0132] In some embodiments, in step S104, the concurrent detection results of each node are input into the state analysis engine, and whether each node is an abnormal node is determined by a preset abnormality determination condition, specifically including:

[0133] Normalize the survival probability, differential hash value, compliance score, and sentiment index to obtain a fused feature vector;

[0134] Configure differentiated weight coefficients based on node types, build a weighted scoring model based on the fused feature vectors and differentiated weight coefficients, and update the weight allocation strategy in real time through a sliding window mechanism;

[0135] Calculate the baseline anomaly threshold based on historical scoring data and combine it with the real-time scoring volatility to generate a dynamic judgment threshold;

[0136] Each node is evaluated based on the weighted scoring model to obtain the node scoring result. By comparing the node scoring result with the dynamic judgment threshold, it is determined whether each node is an abnormal node.

[0137] In this embodiment, existing node monitoring methods often only focus on a single indicator, such as the survival status of the node or data consistency, and lack the ability to comprehensively analyze multiple indicators and determine anomalies. At the same time, due to the dynamic nature and complexity of distributed systems, traditional static threshold determination methods are difficult to adapt to changes in system status, resulting in the accuracy and timeliness of anomaly detection being affected. Therefore, the present invention integrates multiple detection results such as survival probability, differential hash value, compliance score and sentiment index, configures differentiated weight coefficients based on node type, and adopts a dynamic determination threshold mechanism to achieve accurate anomaly determination of distributed nodes.

[0138] In this embodiment, the survival probability, differential hash value, compliance score and sentiment index are normalized and mapped to the interval of [0,1]. The normalized survival probability, differential hash value, compliance score and sentiment index are combined into a fused feature vector. Differentiated weight coefficients are configured according to the node type. For example, for key business nodes, a higher weight coefficient is assigned to the survival probability; for security-sensitive nodes, a higher weight coefficient is assigned to the compliance score. A weighted scoring model is constructed based on the fused feature vector and the differentiated weight coefficient. The weight allocation strategy is updated in real time using a sliding window mechanism. The size of the sliding window is set to m, and within each time window, the weight coefficient is dynamically adjusted according to the actual operation status and historical data of the node. For example, if compliance issues of a certain type of node occur frequently within a certain time window, the weight coefficient of the compliance score is appropriately increased.

[0139] The baseline anomaly threshold is calculated based on the historical scoring data. Statistical methods can be used, such as adding or subtracting a number of standard deviations from the mean to determine the baseline anomaly threshold. Assume that the historical scoring data is S1, S2, ..., S k , then the baseline abnormal threshold in, represents the mean of historical ratings, σ1 represents the standard deviation of historical ratings, and α1 represents the adjustment coefficient for the standard deviation ω1. During real-time monitoring, the volatility of node ratings within the current time window is calculated. Volatility can be measured by calculating the variance or standard deviation of rating data.

[0140] Combined with the real-time rating volatility, a dynamic judgment threshold is generated. The dynamic judgment threshold is T dynamic =T base +β1·σcurrent Among them, σ current represents the standard deviation of the node score in the current time window, and β1 represents the standard deviation σ current The adjustment coefficient is used to control the impact of real-time fluctuations on the judgment threshold.

[0141] Each node is evaluated based on the weighted scoring model to obtain the node scoring result. By comparing the node scoring result S and the dynamic judgment threshold T dynamic , determine whether each node is an abnormal node. If S>T dynamic , then the node is determined to be an abnormal node; otherwise, the node is determined to be a normal node.

[0142] In this embodiment, multiple detection results are fused and processed, and multiple aspects such as node survival status, data consistency, compliance and user feedback are comprehensively considered, thereby improving the accuracy of anomaly determination.

[0143] Differentiated weight coefficients are configured based on node types to ensure that anomaly determination is more aligned with actual business needs. For example, for critical business nodes, more attention is paid to their survival status; for security-sensitive nodes, more attention is paid to their compliance.

[0144] The dynamic threshold mechanism can adapt to the dynamic changes of distributed systems and improve the timeliness and accuracy of anomaly detection. Compared with the traditional static threshold judgment method, the false alarm rate and missed alarm rate of anomaly detection are significantly reduced.

[0145] The sliding window mechanism realizes the real-time update of the weight distribution strategy, so that the anomaly judgment model can be adaptively adjusted according to the actual operation status of the node, further improving the accuracy of anomaly judgment.

[0146] In some embodiments, in step S105, based on the abnormal conditions of the plurality of abnormal nodes, the abnormal conditions are associated with the advertising delivery data through a business impact assessment model to obtain a cost loss assessment result, specifically including:

[0147] Constructing a loss function, wherein the loss function includes a cost loss term, a user churn term, and a reputation loss term;

[0148] Based on the advertising consumption data flow, user retention rate and social media public opinion data of each abnormal node, the improved NSGA-II algorithm is used to perform multi-objective optimization on the loss function. The normal distribution disturbance term is introduced through the crossover operator to generate the Pareto optimal solution set.

[0149] Based on the Pareto optimal solution set, a cost loss assessment result is generated, including the expected loss amount, optimal response time, and ROI repair recommendations.

[0150] In this embodiment, when these nodes are abnormal, it may have a serious impact on the advertising business, such as advertising failure, user loss, brand reputation damage, etc., which in turn leads to cost losses. However, the existing technology often only considers a single factor, such as the direct cost loss caused by advertising failure, while ignoring indirect costs such as user loss and reputation loss, resulting in inaccurate evaluation results and an inability to provide effective support for business decisions. Therefore, the present invention constructs a loss function that includes cost loss terms, user loss terms, and reputation loss terms, and uses an improved NSGA-II algorithm to perform multi-objective optimization on the loss function to generate a Pareto optimal solution set, thereby obtaining a comprehensive cost loss evaluation result and providing a scientific basis for decision-making in the advertising business.

[0151] Specifically, a distributed monitoring system collects abnormal situation information from several abnormal nodes, including the type of abnormality (such as hardware failure, software vulnerability, network outage, etc.), the time of occurrence, and the duration of the abnormality. Ad consumption data for each abnormal node during and before and after the abnormality is obtained from the advertising platform's database, including ad impressions, clicks, conversions, and advertising costs. User retention data is collected within the affected area of the abnormal node. User behavior analysis systems can be used to obtain information such as user activity and login frequency before and after the abnormality occurs, and then calculate user retention rates. Web crawler technology is used to collect public opinion data related to advertising from the iOS App Store or social media platforms, including user reviews, complaints, and negative news. The collected data is cleaned to remove duplicate data, noise, and outliers. Data from different sources is integrated to construct a comprehensive dataset containing abnormal node information, advertising data, user data, and public opinion data.

[0152] The cost loss item mainly considers the direct cost loss caused by the failure of advertising, including the cost of unsuccessful advertising, the additional cost of re-advertising, etc. Assume that the number of failed advertising is N f The cost of each failed ad delivery is C f , the additional cost coefficient of re-advertising is k, then the cost loss term L c It can be expressed as: L c =N f ·C f (1+k).

[0153] The user churn item is measured based on the change in user retention rate. Assuming that the user retention rate before the anomaly is R0, the user retention rate after the anomaly is R1, and the average user lifetime value is CLV, then the user churn item L u It can be expressed as: L u =(R0-R1)·CLV·N u, where N u Indicates the number of affected users.

[0154] The reputation loss term is measured based on the degree of negativity of social media public opinion data. Natural language processing technology can be used to perform sentiment analysis on public opinion data, and the impact of negative public opinion can be quantified into a score value S1. The reputation loss term L r It can be expressed as: L r =S1·C r , where C r A coefficient representing the unit cost of reputation loss.

[0155] The cost loss term, user churn term, and reputation loss term are integrated into a comprehensive loss function L, that is, L = L c +L u +L r .

[0156] The improved NSGA-II algorithm is used to perform multi-objective optimization on the loss function. The improvement lies in the introduction of a normal distribution perturbation term in the crossover operator to increase the diversity of the population and avoid falling into a local optimal solution. The specific operation is as follows: After the traditional crossover operation, a normal distribution perturbation term is added to the generated offspring individuals. The random perturbation term ∈, that is, the new offspring individual X new =X cross +∈, where X cross is the offspring individual after the crossover operation, and σ2 is the disturbance intensity parameter. Through multiple iterative optimizations, a Pareto optimal solution set is generated. Each solution in the Pareto optimal solution set represents a trade-off between cost loss, user churn, and reputation loss. Based on the Pareto optimal solution set, the expected loss amount (calculated based on the comprehensive cost loss, user churn, and reputation loss), optimal response time (based on the time required for the response strategies corresponding to different solutions), and ROI repair suggestions (such as adjusting advertising delivery strategies, optimizing node configurations, etc.) corresponding to each solution are analyzed. The analysis results are organized into a cost loss assessment report, including the expected loss amount, optimal response time, and ROI repair suggestions, to provide a reference for business decisions.

[0157] This embodiment comprehensively considers multiple factors, including cost loss, user churn, and reputation loss, resulting in a more comprehensive and accurate cost loss assessment that truly reflects the impact of abnormal nodes on the advertising business. An improved NSGA-II algorithm is used for multi-objective optimization to generate a Pareto optimal solution set, providing multiple trade-offs for business decision-making. Decision-makers can select the optimal response strategy based on their specific circumstances. By generating optimal response timelines and ROI remediation recommendations, business teams can quickly respond to abnormal situations, take effective measures to reduce losses, and improve the stability and efficiency of the advertising business.

[0158] In some embodiments, in step S105, automatically generating a disposal suggestion based on the cost loss assessment result and executing a preset emergency plan specifically includes:

[0159] Input the cost loss assessment results of each abnormal node into the pre-built emergency response knowledge graph, calculate the TF-IDF cosine similarity with historical cases, and generate a set of candidate solutions;

[0160] A Monte Carlo simulation environment is built for abnormal nodes with major abnormal events to simulate and calculate the expected loss of each strategy in the candidate solution set;

[0161] By comparing the expected loss with the preset expected loss threshold, the strategy with the lowest operational complexity will be implemented as the emergency plan.

[0162] In this embodiment, the current handling of abnormal nodes relies primarily on manual experience, which can lead to slow response, inaccurate decision-making, and difficulty in handling complex abnormal situations. Therefore, this invention integrates cost and loss assessment results into an emergency response knowledge graph, combines them with TF-IDF cosine similarity to generate a set of candidate solutions, constructs a Monte Carlo simulation environment for major abnormal events, evaluates the expected losses of candidate solutions, and selects the optimal solution for execution, thereby achieving rapid and accurate abnormal handling.

[0163] Specifically, historical abnormal events and their corresponding disposal plans are collected to construct an emergency disposal knowledge graph. The knowledge graph uses abnormal events as nodes and disposal plans as edges. The attributes of the edges include the name of the plan, operation steps, applicable scenarios, etc. The cost loss assessment results of each abnormal node (described in text form, such as "the abnormal type is a server failure, the expected loss amount is 5,000 yuan, and the optimal response time is 2 hours") are input into the pre-built emergency disposal knowledge graph, and the text is TF-IDF (term frequency-inverse document frequency) converted with the description text of the historical cases in the knowledge graph to calculate the cosine similarity between them. Sort by cosine similarity and select the disposal plans corresponding to the top P historical cases with the highest similarity as the candidate solution set. The value of P can be adjusted according to the actual situation and is generally set to 3-5.

[0164] Set the judgment criteria for major abnormal events, such as the expected loss amount exceeds a certain threshold (such as 10,000 yuan), the optimal response time is less than a certain time (such as 1 hour), etc. For abnormal nodes that meet the judgment criteria for major abnormal events, conduct subsequent Monte Carlo simulations. For each major abnormal event, build a Monte Carlo simulation environment. The simulation environment includes the current status of the abnormal node, other relevant factors of the advertising delivery system (such as user traffic, advertising inventory, etc.), and the operation steps and impacts of each strategy in the candidate solution set. In the Monte Carlo simulation environment, the development process of abnormal events is randomly simulated multiple times (such as 1,000 times). In each simulation, the abnormal event is handled according to the operation steps of each strategy in the candidate solution set, and the final loss amount is recorded. Calculate the average loss amount of each strategy in multiple simulations as the expected loss of the strategy.

[0165] By comparing the expected loss of each strategy in the candidate set with a preset expected loss threshold (such as 8,000 yuan), the strategies with expected losses less than the threshold are screened out. Among the strategies that meet the conditions, the strategy with the most complex operation (the complexity of operation can be evaluated based on factors such as the number of operation steps and the difficulty of operation) is selected as the emergency plan for execution. The selection of the most complex operation strategy is based on the assumption that strategies with complex operations may consider more factors, have more comprehensive response capabilities, and may have better robustness in subsequent exception handling. The selected emergency plan is converted into specific operation instructions, and the emergency plan is executed through automated scripts or manual operations. During the execution process, the status of abnormal nodes and the operation of the advertising delivery system are monitored in real time, and the emergency plan is adjusted in time according to the actual situation.

[0166] After the emergency plan is executed, the execution result data is collected, including the recovery status of the abnormal node, the performance indicators of the advertising delivery system, and the actual loss amount. This execution result data is fed back to the cost loss assessment model and the emergency response knowledge graph for subsequent model optimization and knowledge graph updates. The emergency response knowledge graph is updated based on the execution results of the emergency plan. If the emergency plan successfully resolves the abnormal issue, it is added to the knowledge graph and its applicable scenarios and effectiveness are recorded. If the emergency plan fails to effectively resolve the issue, the cause is analyzed and the relevant nodes and edges in the knowledge graph are adjusted.

[0167] In this example, automated generation of response recommendations and execution of emergency plans significantly shortens exception handling response time and improves the stability of the advertising delivery system. Based on similarity calculations between cost loss assessment results and historical cases, combined with Monte Carlo simulation to estimate expected losses, the optimal emergency plan can be selected, improving decision-making accuracy. By continuously updating the emergency response knowledge graph, rich exception handling experience has been accumulated, providing better support for subsequent exception handling.

[0168] Reference Figure 2 An embodiment of the present invention provides an iOS application full lifecycle management system 2, wherein the system 2 specifically includes:

[0169] The first management module 201 is configured to analyze Apple's global content delivery network architecture using a CDN topology analysis algorithm, dynamically select multiple optimal monitoring nodes in a region based on node response delays, and form an initial adaptive monitoring pool;

[0170] The second management module 202 is used to build a digital twin model of the Apple CDN network based on the initial adaptive monitoring pool. By using real-time network traffic data, node load status, and historical fault records, the reinforcement learning model is trained to dynamically select the optimal monitoring node combination to obtain the target adaptive monitoring pool.

[0171] The third management module 203 is used to start the distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool, and the concurrent detection includes application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis;

[0172] The fourth management module 204 is used to input the concurrent detection results of each node into the status analysis engine and determine whether each node is an abnormal node according to the preset abnormality judgment conditions;

[0173] The fifth management module 205 is used to associate the abnormal status with the advertising data based on the abnormal conditions of several abnormal nodes through the business impact assessment model, obtain the cost loss assessment result, automatically generate disposal suggestions according to the cost loss assessment result and execute the preset emergency plan.

[0174] It is understandable that if Figure 1 The contents of the embodiment of the iOS application full life cycle management method shown in the figure are applicable to the embodiment of the iOS application full life cycle management system. The functions specifically implemented by the embodiment of the iOS application full life cycle management system are similar to those in the embodiment of the figure. Figure 1 The embodiment of the iOS application full life cycle management method shown is the same as that shown in FIG. Figure 1 The beneficial effects achieved by the embodiment of the iOS application full lifecycle management method shown are also the same.

[0175] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0176] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0177] Reference Figure 3 An embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, the iOS application full life cycle management method as described in any one of the above methods is implemented.

[0178] The computer device 3 may be a desktop computer, a notebook computer, a PDA, a cloud server or other computing devices. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that Figure 3 This is merely an example of the computer device 3 and does not constitute a limitation on the computer device 3 . The computer device 3 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 3 may also include input and output devices, network access devices, etc.

[0179] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0180] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the computer device 3. Furthermore, the memory 302 may include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is about to be output.

[0181] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for managing the entire life cycle of an iOS application as described in any one of the above methods is implemented.

[0182] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process of the above-mentioned method embodiment by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can at least include: any entity or device capable of carrying computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, mobile hard drive, magnetic disk, or optical disk. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0183] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0184] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0185] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0186] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for managing the entire life cycle of an iOS application, characterized in that: The method specifically includes: Analyze Apple's global content delivery network architecture through CDN topology analysis algorithms, dynamically select multiple optimal monitoring nodes in the region based on node response latency, and form an initial adaptive monitoring pool; Based on the initial adaptive monitoring pool, a digital twin model of the Apple CDN network was constructed. Using real-time network traffic data, node load status, and historical fault records, a reinforcement learning model was trained to dynamically select the optimal combination of monitoring nodes to obtain the target adaptive monitoring pool. Launching a distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification, and user review sentiment analysis. The concurrent detection results of each node are input into the status analysis engine, and each node is determined to be an abnormal node based on the preset abnormality judgment conditions; Based on the abnormal conditions of several abnormal nodes, the business impact assessment model is used to associate the abnormal status with the advertising data to obtain the cost loss assessment results. According to the cost loss assessment results, disposal suggestions are automatically generated and preset emergency plans are implemented.

2. The method according to claim 1, characterized in that The CDN topology analysis algorithm analyzes Apple's global content delivery network architecture and dynamically selects multiple optimal monitoring nodes in the region based on node response latency to form an initial adaptive monitoring pool. Specifically, it includes: Using a topology discovery algorithm deployed in a network probe, ICMP probe packets are periodically sent to Apple CDN nodes. Based on the returned TTL value and the number of route hops, a CDN network graph model is constructed. The node set of the CDN network graph model includes multiple available CDN nodes, and the edge set of the CDN network graph model includes communication links between available CDN nodes. For each node in the CDN network graph model, multiple HTTP HEAD requests are sent continuously, and the sending and receiving timestamps of each request are recorded to calculate the comprehensive latency score. Based on the comprehensive latency score, all nodes are sorted using a preset node optimization function to obtain a dynamic node optimization result. The node optimization function is used to comprehensively calculate latency, availability, and geographic weight. The top K nodes in the dynamic node selection results are selected to form the initial adaptive monitoring pool.

3. The method according to claim 1, characterized in that Based on the initial adaptive monitoring pool, a digital twin model of the Apple CDN network is constructed. Using real-time network traffic data, node load status, and historical fault records, a reinforcement learning model is trained to dynamically select the optimal combination of monitoring nodes to obtain the target adaptive monitoring pool. Specifically, the model includes: By deploying data collectors on edge computing nodes, we can obtain the load rate, inbound traffic, and failure probability of Apple CDN nodes in real time and build a digital twin model containing multi-dimensional state vectors. Construct a deep reinforcement learning network with an input layer dimension consistent with the twin state vector dimension of the digital twin model and an output layer dimension equal to the number of candidate nodes in the initial adaptive monitoring pool; Inputting node data of the initial adaptive monitoring pool into the digital twin model to generate training samples; The training samples are input into a deep reinforcement learning network. Combined with the reward function, the policy network parameters are updated through the double-delayed deep deterministic policy gradient algorithm to output a new node selection combination and obtain the target adaptive monitoring pool. The reward function is used to comprehensively calculate the node load penalty term, failure risk term and area coverage reward term.

4. The method according to claim 1, wherein The distributed detection cluster is started to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis, specifically including: Based on the node distribution of the adaptive monitoring pool, detection agent nodes are deployed in each geographical area, and each agent node is configured with an independent IP address and device fingerprint; The detection proxy node concurrently sends M HTTPS HEAD requests to the target node. When a 404 status code is received for a preset number of consecutive times and the survival probability is less than the preset survival probability threshold, the target node is marked as a failed node and the application existence verification result is obtained; If the application existence verification result passes, the application metadata is downloaded and the differential hash value is calculated. The binary difference is compared with the baseline version to determine whether a tampering alarm is triggered; Parse the text of the application metadata, calculate the compliance score through the keyword matching engine, and determine whether there is a violation based on the compliance score; Grab the user comment dataset associated with the target node and use the BERT model to analyze the sentiment index.

5. The method according to claim 4, characterized in that The concurrent detection results of each node are input into the state analysis engine, and each node is determined to be an abnormal node according to the preset abnormality judgment conditions, which specifically include: Normalize the survival probability, differential hash value, compliance score, and sentiment index to obtain a fused feature vector; Configure differentiated weight coefficients based on node types, build a weighted scoring model based on the fused feature vectors and differentiated weight coefficients, and update the weight allocation strategy in real time through a sliding window mechanism; Calculate the baseline anomaly threshold based on historical scoring data and combine it with the real-time scoring volatility to generate a dynamic judgment threshold; Each node is evaluated based on the weighted scoring model to obtain the node scoring result. By comparing the node scoring result with the dynamic judgment threshold, it is determined whether each node is an abnormal node.

6. The method according to claim 1, characterized in that Based on the abnormal conditions of several abnormal nodes, the abnormal conditions are associated with the advertising data through the business impact assessment model to obtain the cost loss assessment results, which specifically include: Constructing a loss function, wherein the loss function includes a cost loss term, a user churn term, and a reputation loss term; Based on the advertising consumption data flow, user retention rate and social media public opinion data of each abnormal node, the improved NSGA-II algorithm is used to perform multi-objective optimization on the loss function. The normal distribution disturbance term is introduced through the crossover operator to generate the Pareto optimal solution set. Based on the Pareto optimal solution set, a cost loss assessment result is generated, including the expected loss amount, optimal response time, and ROI repair recommendations.

7. The method according to claim 1, characterized in that The automatic generation of disposal suggestions and execution of preset emergency plans based on the cost loss assessment results specifically include: Input the cost loss assessment results of each abnormal node into the pre-built emergency response knowledge graph, calculate the TF-IDF cosine similarity with historical cases, and generate a set of candidate solutions; A Monte Carlo simulation environment is built for abnormal nodes with major abnormal events to simulate and calculate the expected loss of each strategy in the candidate solution set; By comparing the expected loss with the preset expected loss threshold, the strategy with the lowest operational complexity will be implemented as the emergency plan.

8. An iOS application full life cycle management system, characterized by: The system specifically includes: The first management module is used to analyze Apple's global content delivery network architecture using a CDN topology analysis algorithm, dynamically select multiple optimal monitoring nodes in the region based on node response latency, and form an initial adaptive monitoring pool; The second management module is used to build a digital twin model of the Apple CDN network based on the initial adaptive monitoring pool. Using real-time network traffic data, node load status, and historical fault records, the reinforcement learning model is trained to dynamically select the optimal combination of monitoring nodes to obtain the target adaptive monitoring pool. The third management module is used to start the distributed detection cluster to perform concurrent detection on the nodes of the target adaptive monitoring pool. The concurrent detection includes application existence verification, code hash comparison, metadata compliance verification and user comment sentiment analysis; The fourth management module is used to input the concurrent detection results of each node into the status analysis engine and determine whether each node is an abnormal node according to the preset abnormality judgment conditions; The fifth management module is used to associate the abnormal status with the advertising data based on the abnormal conditions of several abnormal nodes through the business impact assessment model, obtain the cost loss assessment results, automatically generate disposal suggestions based on the cost loss assessment results, and execute the preset emergency plan.

9. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory, which, when executed on the processor, implements the iOS application full lifecycle management method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the method for managing the entire life cycle of an iOS application is implemented as described in any one of claims 1 to 7.

Images