Art test enrollment platform based on big data analysis

Abstract

The invention relates to the technical field of network services, and discloses a big data analysis-based art exam enrollment platform, which ensures that key businesses are preferentially processed in a fusing recovery period through request grading and priority scheduling, realizes a stable and controllable service recovery process through multi-stage flow release and detection adjustment, and improves the service recovery efficiency. Passive response is converted into active defense through flow prediction and resource allocation, comprehensive system monitoring and risk early warning are provided through capacity evaluation and anomaly recognition, and continuous optimization of a recovery strategy is realized through strategy learning. The platform is especially suitable for application scenes with obvious traffic peak characteristics and strict service timeliness requirements such as art test enrollment, a systematic technical scheme is provided for solving the problem of high-concurrency traffic management in the field of education informatization, and the stability and user experience of the platform under a high load condition are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of network service technology, specifically to an art exam admission platform based on big data analysis. Background Technology

[0002] Among the many technical means of service governance, the circuit breaker mechanism, as an important protective measure against system overload and service failures, has been widely used in various high-concurrency business scenarios. Existing art exam admission platforms generally adopt a circuit breaker model based on threshold monitoring. By monitoring service error rates, timeout rates, and other indicators, it automatically cuts off calls to faulty services when an anomaly is detected, preventing the spread of cascading failures. This mechanism includes three core components: circuit breaker triggering, state maintenance, and service detection and recovery.

[0003] The current circuit breaker recovery mechanism of the art exam admission system suffers from a crude traffic scheduling problem. During peak traffic periods such as the start of registration or the release of results, a large number of pending requests accumulate after the system triggers the circuit breaker. Existing solutions lack fine-grained traffic control strategies when detecting a half-open state, failing to release requests in batches and causing a secondary traffic surge upon recovery. Furthermore, the lack of intelligent management and priority scheduling mechanisms for the backlogged request queue makes it impossible to differentiate between business requests of varying urgency, resulting in frequent service fluctuations during recovery and difficulty in achieving a stable state.

[0004] Therefore, we proposed an art exam admission platform based on big data analysis to address the aforementioned issues. Summary of the Invention

[0005] The purpose of this invention is to provide an art exam admission platform based on big data analysis, in order to solve the problem mentioned in the background technology that the lack of a refined traffic control strategy during half-open state detection, the failure to release requests in batches and slowly, and the resulting secondary traffic surge during recovery.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an art exam admission platform based on big data analysis, characterized in that it includes a request classification module, a traffic release module, a detection and adjustment module, a traffic prediction module, a capacity assessment module, a priority scheduling module, a strategy learning module, an anomaly identification module, and a resource allocation module;

[0007] The request tiering module divides the backlog of requests during the circuit breaker period into four queues according to their urgency, and handles requests for different business scenarios respectively. Each queue is set with independent storage space and expiration time, and saves data through a persistence mechanism and monitors capacity status. When the capacity reaches a preset threshold, an alarm is triggered.

[0008] The traffic release module adopts a multi-stage recovery strategy to gradually release traffic. It starts with a small amount of traffic in the initial detection stage and gradually increases the traffic ratio according to the service response. Each stage has time requirements and quality standards. When the service performance is poor, it automatically drops back to the previous stage.

[0009] The detection and adjustment module dynamically adjusts detection parameters based on historical service performance, comprehensively evaluates multiple performance indicators, calculates service health status using a weighted approach, and allows the next recovery phase to proceed when the evaluation results meet the standards.

[0010] The traffic prediction module analyzes historical access patterns to predict traffic trends and initiates preheating preparations before peak periods arrive, including expanding system resources, loading frequently used data, and adjusting service configurations.

[0011] The capacity assessment module continuously monitors the usage of critical system resources, establishes an assessment model to calculate service carrying capacity, quantifies service status, and determines the current load level that can be tolerated.

[0012] The priority scheduling module comprehensively considers factors such as the timeliness, importance, and waiting status of requests to allocate processing order to each request, while ensuring that requests of different priorities all have a chance to be processed, and avoiding low-level requests from being stuck for a long time.

[0013] The strategy learning module records the process data of each circuit breaker recovery, analyzes the success and failure factors through the learning model, continuously optimizes the configuration of various parameters of the recovery strategy, and applies the improved solution in subsequent events.

[0014] The anomaly identification module establishes feature models for various anomalies, analyzes current traffic characteristics in real time, identifies anomaly types and assesses their severity, and takes corresponding measures based on different anomalies.

[0015] The resource allocation module manages all types of resources in a unified manner, dynamically adjusts resource allocation based on service load, transfers idle resources between services to support high-load modules, and completes resource deployment in advance based on forecast information.

[0016] Preferably, the request hierarchy mode includes a queue partitioning unit and a capacity monitoring unit;

[0017] The queue partitioning unit receives backlogged requests during the circuit breaker period, parses the service type identifier, timestamp, user identifier and other attribute information in the requests, and allocates the requests to the P0 level emergency queue, P1 level high priority queue, P2 level normal queue or P3 level low priority queue according to preset rules. It allocates independent storage space for each queue and sets expiration time parameters. It writes the request data to the preset storage medium through the persistence mechanism, outputs the hierarchical queue data to the traffic release module, receives the adjustment instructions from the priority scheduling module and updates the request arrangement position.

[0018] The capacity monitoring unit periodically collects the number of requests and space occupancy rate of each queue. When the capacity utilization rate reaches the first preset threshold, an alarm signal is generated and sent to the management terminal. The unit monitors the request waiting time and compares it with the expiration time parameter. It performs a cleanup operation on the timed-out requests and releases the storage space. It outputs the queue backlog data to the capacity assessment module and transmits the abnormal capacity growth data to the anomaly identification module.

[0019] Preferably, the flow release module includes a stage control unit and a rate control unit;

[0020] The phase control unit manages five execution phases in sequence: detection, observation, acceleration, stabilization, and full volume. It sets the traffic release ratio, duration, and quality assessment criteria for each phase. After obtaining queue data from the request classification module, it extracts and releases requests according to the parameters of the current phase. It determines the phase switching based on the health assessment data returned by the detection adjustment module. When the assessment data meets the quality criteria, it enters the next phase. When it falls below the fallback threshold, it returns to the previous phase. At the same time, it outputs the phase execution data to the strategy learning module.

[0021] The rate control unit achieves smooth traffic release through the token bucket algorithm. The token generator issues tokens to the token bucket according to the traffic ratio parameter and preset frequency. Only after requesting to obtain a token can the business processing flow be entered. If there are not enough tokens, the process will be blocked and wait. The unit continuously monitors the deviation between the actual release rate and the target rate and dynamically adjusts the token issuance frequency. It also transmits real-time release data and response time indicators to the capacity assessment module.

[0022] Preferably, the detection and adjustment module includes a parameter adjustment unit and a health assessment unit;

[0023] After obtaining operational data such as service historical success rate, response time, and number of fault recovery attempts, the parameter adjustment unit calculates a stability score. Based on the comparison between the score result and the preset threshold, it adjusts the probe window duration and the number of probe requests. The higher the service stability, the shorter the window duration and the more requests are set, and vice versa. The adjusted probe parameters are then transmitted to the stage control unit of the traffic release module to guide the probe execution.

[0024] The health assessment unit collects performance data such as request success rate, response time, connection utilization, and CPU load rate. After reading the preset weight coefficients of each indicator, it normalizes the performance data and calculates the health score by weighted summation. It obtains the quality standard and backoff threshold parameters of the current stage from the stage control unit. When the score reaches the standard, it generates a push signal; when the score is lower than the threshold, it generates a backoff signal. It transmits the signal to the stage control unit and outputs the performance data to the capacity assessment module.

[0025] Preferably, the flow prediction module includes a trend prediction unit and a preheating execution unit;

[0026] After acquiring historical access volume, traffic peaks, query volume, and other data, the trend prediction unit uses time series algorithms for cleaning, feature extraction, and pattern recognition to identify typical traffic patterns such as instantaneous peaks, continuous growth, and concentrated queries. Based on the current date and business schedule, it calculates the peak nodes that may appear in the future time window, as well as their numerical range and duration, and generates prediction results containing parameters such as peak traffic, arrival time, and duration. These results are then transmitted to the preheating execution unit, while prediction data is output to the capacity assessment module.

[0027] After receiving the traffic prediction data, the preheating execution unit calculates the preheating start time. When the system time reaches that time, the preheating steps are executed sequentially: a capacity expansion command is sent to the database module to increase the number of connection pools; a preloading command is sent to the cache module to load forms and college data; a capacity expansion request is sent to the resource module to increase the number of service instances; and an adjustment command is sent to the tiering module to reduce the resource quotas of non-core services. After execution, status data containing the execution results and timestamps of each step is generated and transmitted to the strategy learning module. At the same time, a system ready signal is sent to the stage control unit.

[0028] Preferably, the capacity assessment module includes a resource monitoring unit and a carrying capacity calculation unit;

[0029] The resource monitoring unit collects resource data such as CPU utilization, memory usage, remaining connection pool quantity, cache hit rate, and bandwidth utilization at preset cycles and records them as a time series dataset. The dataset is then transmitted to the carrying capacity calculation unit and outputs resource usage trend data to the anomaly identification module. When the resource utilization exceeds the alarm threshold, an early warning signal containing the resource type and utilization value is generated and sent to the system management terminal.

[0030] After receiving resource data, the carrying capacity calculation unit establishes a multi-dimensional evaluation model that includes computing capacity, storage capacity, and network capacity. It uses different calculation formulas to calculate the capacity value of each dimension. The normalized capacity value is then combined using a weighted summation algorithm to obtain a health score. Based on the score range, the service status is determined and the corresponding load level parameters are output and transmitted to the stage control unit. When the score is lower than the warning threshold, the type of resource with insufficient capacity is identified and output to the resource allocation module.

[0031] Preferably, the priority scheduling module includes a priority calculation unit and a scheduling execution unit;

[0032] After obtaining request data from the request classification module, the priority calculation unit extracts attributes such as business type, timestamp, and user identifier, and calculates four dimensions of parameters: timeliness, importance, waiting time, and number of retries. The scores of each dimension are obtained through time difference mapping, classification table lookup, and duration mapping, respectively. After assigning preset weight coefficients to each dimension, a weighted summation algorithm is used to calculate the comprehensive priority score. The score is written into the request record and transmitted to the scheduling execution unit.

[0033] After receiving request data with priority scores, the scheduling execution unit generates a processing queue by arranging the data in descending order of scores. It extracts requests according to the scheduling policy and allocates them to service instances. When the waiting time of a request exceeds a threshold, its score is increased. When the time remaining before the deadline is less than the threshold, the request is promoted to the highest priority. The unit processes high and low priority requests alternately according to a preset ratio, records scheduling process data, and calculates the processing volume and waiting time for each priority interval. The statistical data is then transmitted to the policy learning module and the processing rate is output to the capacity assessment module.

[0034] Preferably, the strategy learning module includes a data acquisition unit and a model optimization unit;

[0035] The data acquisition unit receives recovery process data from various relevant modules, including stage execution time, traffic ratio, success rate, response time, health score, status switching record, request processing statistics, peak resource usage, etc. It assigns a unique number to each circuit breaker event and associates it with the triggering reason, total backlog of requests and queue distribution, strategy parameter configuration, recovery result and total time. It stores the complete event data in the historical database and transmits it to the model optimization unit.

[0036] After receiving historical event data, the model optimization unit constructs a reinforcement learning model that includes a state space, action space, reward function, and policy network. It uses an online learning algorithm to update the model parameters based on the actual recovery effect, analyzes the optimal policy configuration under different scenarios, and triggers model training when the historical event reaches the learning threshold. It then generates a policy recommendation scheme that includes stage duration, traffic ratio, and quality standard, and transmits the recommendation scheme to the stage control unit and parameter adjustment unit. The unit calculates the policy improvement magnitude and outputs evaluation data to the system management terminal.

[0037] Preferably, the anomaly identification module includes a feature detection unit and a response strategy unit;

[0038] The feature detection unit obtains real-time traffic data and resource usage trends from the resource monitoring unit, receives queue capacity growth data from the capacity monitoring unit, establishes an abnormal traffic feature database to store feature vectors of preset abnormal patterns, and uses principal component analysis or cluster analysis algorithms to extract feature parameters such as the current traffic growth rate, source concentration, type distribution, and resource consumption rate and combine them into a real-time feature vector. The similarity value between the real-time feature vector and the feature vectors of each abnormal pattern in the database is calculated using a cosine similarity algorithm or an Euclidean distance algorithm. When the similarity value exceeds a preset threshold, a corresponding abnormal type identifier is generated and transmitted to the response strategy unit.

[0039] After receiving the anomaly type identifier and feature vector, the response strategy unit performs a weighted calculation based on the traffic deviation parameter, resource utilization anomaly magnitude parameter, and duration parameter to obtain an anomaly score. The score is quantified to a numerical range, and the severity level is determined according to the range. The corresponding handling scheme is matched from the anomaly handling strategy library. When the score is low, monitoring data is output to the capacity assessment module. When the score is at a medium level, a resource reservation request is sent to the resource allocation module. When the score is high, a traffic limiting command, a service circuit breaker command, or an emergency expansion request is executed. The processing execution data is recorded and transmitted to the strategy learning module, and an anomaly alarm information is sent to the system management terminal.

[0040] Preferably, the resource allocation module includes a resource allocation unit and a resource transfer unit;

[0041] The resource allocation unit obtains resource usage data and insufficient capacity indicators from the capacity assessment module, receives expansion requests from the traffic prediction module, and receives reserved and emergency expansion requests from the anomaly identification module. It maintains a resource pool containing CPU, memory, connection pool, cache, and bandwidth and records the total capacity, allocated amount, and available amount. It calculates the required resources according to the requests and extracts allocations from the resource pool. When the available amount is insufficient, it sends an application instruction to the cloud service platform to obtain new resources and updates the resource pool data. It records the allocation operation and transmits it to the resource transfer unit.

[0042] After receiving the resource pool status data, the resource transfer unit periodically monitors the resource utilization rate of each service module, identifies low-load and high-load modules, and calculates the idle and demand amounts of resources. When the idle amount meets the demand amount, resources are reclaimed from the low-load modules and redistributed to the high-load modules. After obtaining peak prediction data from the traffic prediction module, the unit triggers pre-deployment operations in advance based on the arrival time, requests additional resources and allocates them to the predicted high-load modules, records the transfer and deployment data, transmits it to the policy learning module, and outputs the available capacity to the capacity assessment module.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. This platform ensures that critical services are prioritized during circuit breaker recovery by using request tiering and priority scheduling, achieves a smooth and controllable service recovery process through multi-stage traffic release and detection adjustment, transforms passive response into proactive defense through traffic prediction and resource allocation, provides comprehensive system monitoring and risk warning through capacity assessment and anomaly identification, and achieves continuous optimization of recovery strategies through policy learning.

[0045] 2. This platform achieves several benefits, including reducing the frequency of circuit breakers, shortening recovery time, improving the success rate of critical business processing, reducing manual maintenance workload, and enhancing system adaptability. This platform is particularly suitable for application scenarios with significant traffic peaks and strict timeliness requirements, such as art exam admissions. It provides a systematic technical solution to the high-concurrency traffic management challenges in the field of educational informatization, effectively improving platform stability and user experience under high load conditions. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall three-dimensional structure of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1: Please refer to Figure 1 An art exam admission platform based on big data analysis includes a request classification module, a traffic release module, a detection and adjustment module, a traffic prediction module, a capacity assessment module, a priority scheduling module, a strategy learning module, an anomaly identification module, and a resource allocation module.

[0049] The request tiering module divides the backlog of requests during the circuit breaker period into four queues according to their urgency, and handles requests for different business scenarios respectively. Each queue is set with independent storage space and expiration time, and saves data through a persistence mechanism and monitors capacity status. When the capacity reaches a preset threshold, an alarm is triggered.

[0050] The traffic release module adopts a multi-stage recovery strategy to gradually release traffic. It starts with a small amount of traffic in the initial detection stage and gradually increases the traffic ratio according to the service response. Each stage has time requirements and quality standards. When the service performance is poor, it automatically drops back to the previous stage.

[0051] The detection and adjustment module dynamically adjusts detection parameters based on historical service performance, comprehensively evaluates multiple performance indicators, calculates service health status using a weighted approach, and allows the next recovery phase to proceed when the evaluation results meet the standards.

[0052] The traffic prediction module analyzes historical access patterns to predict traffic trends and initiates preheating preparations before peak periods arrive, including expanding system resources, loading frequently used data, and adjusting service configurations.

[0053] The capacity assessment module continuously monitors the usage of critical system resources, establishes an assessment model to calculate service carrying capacity, quantifies service status, and determines the current load level that can be tolerated.

[0054] The priority scheduling module comprehensively considers factors such as the timeliness, importance, and waiting status of requests to allocate processing order to each request, while ensuring that requests of different priorities all have a chance to be processed, and avoiding low-level requests from being stuck for a long time.

[0055] The strategy learning module records the process data of each circuit breaker recovery, analyzes the success and failure factors through the learning model, continuously optimizes the configuration of various parameters of the recovery strategy, and applies the improved solution in subsequent events.

[0056] The anomaly identification module establishes feature models for various anomalies, analyzes current traffic characteristics in real time, identifies anomaly types and assesses their severity, and takes corresponding measures based on different anomalies.

[0057] The resource allocation module manages all types of resources in a unified manner, dynamically adjusts resource allocation based on service load, transfers idle resources between services to support high-load modules, and completes resource deployment in advance based on forecast information.

[0058] In this embodiment, the request grading module divides the backlog of requests during the circuit breaker period into four-level queues based on their urgency. This enables differentiated management of requests for different business scenarios, ensuring that core business processes such as registration submission and grade inquiry are processed first, avoiding delays in critical business processes caused by a mix of requests in traditional solutions. Each queue has independent storage space and an expiration time mechanism, preventing invalid requests from occupying system resources for extended periods and ensuring data integrity after system restarts through persistence. Capacity monitoring and alarm functions help administrators promptly identify queue backlog risks, thereby achieving orderly classification and secure storage of requests, improving the timeliness of critical business processing and the efficiency of system resource utilization.

[0059] The traffic release module employs a multi-stage recovery strategy instead of the traditional one-time full recovery method. It gradually releases traffic starting with a small amount during the initial probing phase and dynamically adjusts the traffic ratio based on service response, effectively reducing the risk of service re-crash during recovery. The duration requirements and quality standards set for each stage provide clear execution guidelines for recovery. When service performance is poor, it can automatically revert to the previous stage, avoiding delays caused by manual intervention. This module smoothly and orderly releases the traffic after the circuit breaker is applied to the service instances, maximizing recovery speed while ensuring service stability, and significantly reducing the probability of secondary circuit breaker failures compared to traditional solutions.

[0060] The detection and adjustment module dynamically adjusts detection parameters based on historical service performance, overcoming the limitations of traditional fixed-parameter schemes that cannot adapt to different service states. For services with high stability, a shorter detection window is used to accelerate recovery, while for services with poor stability, the observation time is extended to reduce the risk of misjudgment. A comprehensive evaluation of multiple performance indicators and a weighted approach are used to calculate service health status, providing a more comprehensive and accurate assessment than a single indicator. When the evaluation results meet the standards, the next recovery phase can proceed, ensuring that each phase is based on sufficient data support, thereby improving the accuracy of recovery decisions and shortening the overall recovery time.

[0061] The traffic prediction module analyzes historical access patterns to predict traffic trends, enabling the system to shift from a reactive to a proactive defense. It initiates pre-emptive preparations before foreseeable peak traffic periods such as registration opening and results release, including expanding the database connection pool, preloading hotspot data, and adding service instances. This effectively avoids service disruptions caused by sudden traffic surges. The module identifies the differences between normal peak traffic and abnormal traffic, providing the system with more accurate capacity planning. It completes the tasks of traffic trend prediction and pre-deployment of resources, mitigating the risk of circuit breakers in advance and improving the system's ability to handle peak traffic. Compared to traditional solutions that wait for traffic to arrive before processing it, this significantly reduces the probability of circuit breakers occurring.

[0062] The capacity assessment module continuously monitors key system resources such as CPU utilization, memory usage, and database connection usage. It has established a multi-dimensional assessment model encompassing computing capacity, storage capacity, and network capacity, providing a more comprehensive assessment of system capacity compared to traditional single-dimensional monitoring solutions. Service status is quantified into a health score, which determines the current load level, providing a scientific basis for decision-making by the traffic release and priority scheduling modules. When insufficient resource capacity is detected, it promptly outputs warning information to the resource allocation module to trigger expansion operations. This achieves real-time monitoring of system resource usage and quantitative assessment of service capacity, realizing the goal of dynamically understanding system load status and preventing resource exhaustion-related failures.

[0063] The priority scheduling module comprehensively considers multiple factors such as request timeliness, importance, waiting status, and historical retries. It employs a weighted summation algorithm to calculate a comprehensive priority score for each request, overcoming the limitation of traditional simple first-in-first-out strategies that cannot differentiate request importance. A dynamic priority boosting mechanism ensures that requests nearing their deadlines are processed promptly, while a fairness guarantee mechanism prevents low-priority requests from lingering. This achieves reasonable sorting and fair scheduling of backlogged requests, ensuring that critical business processes are prioritized while also giving all requests a chance to be processed, significantly improving user experience and system processing efficiency.

[0064] The strategy learning module records complete information such as execution data, health scores, processing statistics, and peak resource usage during each circuit breaker recovery process. By constructing a reinforcement learning model, it analyzes the correlation between different strategy parameter configurations and recovery effectiveness, identifying key factors leading to recovery success or failure. Continuous optimization of parameters such as recovery phase division, traffic release ratio, and quality assessment standards enables the system to learn from historical experience and apply improved solutions in subsequent events. This achieves systematic accumulation of historical data and automatic optimization of strategy parameters, allowing the circuit breaker recovery strategy to evolve and improve as the system operates. Compared to traditional solutions that rely on manual parameter adjustments, this module has stronger adaptive capabilities.

[0065] The anomaly detection module has established feature models covering various abnormal situations, including normal business peaks, malicious attacks, cascading failures, and resource exhaustion. By employing principal component analysis and clustering algorithms to extract feature parameters of current traffic and comparing them with a feature database, it can accurately identify anomaly types and assess their severity. Compared to traditional solutions that rely solely on threshold alarms, it provides more intelligent anomaly detection capabilities. Based on different anomaly types and severity levels, it matches differentiated handling schemes from a policy library: triggering traffic restrictions for malicious attacks, service circuit breaking for cascading failures, and emergency capacity expansion for resource exhaustion. This achieves real-time detection and automated response to abnormal traffic, eliminating various anomalies in their early stages and reducing the need for manual intervention.

[0066] The resource allocation module maintains a unified resource pool, including CPU cores, memory, database connections, cache space, and network bandwidth, enabling centralized management and unified scheduling of various system resources. By periodically monitoring the resource utilization of each service module, it can identify low-load and high-load modules and transfer idle resources between services to support high-load modules, significantly improving resource utilization compared to traditional static resource allocation schemes. Combined with forecast information provided by the traffic prediction module, it proactively requests additional resources and deploys them to service modules predicted to experience high loads, achieving dynamic adjustment and pre-deployment of system resources. This improves the overall system capacity and response speed without increasing total resource investment.

[0067] This platform constructs a complete circuit breaker recovery and traffic management system through the coordinated operation of the aforementioned nine modules, achieving numerous improvements compared to current mainstream technologies. Traditional solutions generally suffer from problems such as a coarse circuit breaker recovery process, a lack of refined management, and an inability to dynamically adjust strategies based on actual conditions. In contrast, this platform ensures that critical services are prioritized during circuit breaker recovery through request tiering and priority scheduling, achieves a smooth and controllable service recovery process through multi-stage traffic release and detection adjustment, transforms passive response into proactive defense through traffic prediction and resource allocation, provides comprehensive system monitoring and risk warnings through capacity assessment and anomaly identification, and continuously optimizes recovery strategies through policy learning.

[0068] Through the above improvements, this platform achieves the effects of reducing the frequency of circuit breakers, shortening circuit breaker recovery time, improving the success rate of critical business processing, reducing manual maintenance workload, and enhancing system adaptability. This platform is particularly suitable for application scenarios with significant traffic peaks and strict timeliness requirements, such as art exam admissions. It provides a systematic technical solution to the high-concurrency traffic management challenges in the field of educational informatization, effectively improving the platform's stability and user experience under high load conditions.

[0069] Example 2: Please refer to Figure 1 The request tiering mode includes a queue partitioning unit and a capacity monitoring unit;

[0070] The queue partitioning unit receives backlogged requests during the circuit breaker period, parses the service type identifier, timestamp, user identifier and other attribute information in the requests, and allocates the requests to the P0 level emergency queue, P1 level high priority queue, P2 level normal queue or P3 level low priority queue according to preset rules. It allocates independent storage space for each queue and sets expiration time parameters. It writes the request data to the preset storage medium through the persistence mechanism, outputs the hierarchical queue data to the traffic release module, receives the adjustment instructions from the priority scheduling module and updates the request arrangement position.

[0071] The capacity monitoring unit periodically collects the number of requests and space occupancy rate of each queue. When the capacity utilization rate reaches the first preset threshold, an alarm signal is generated and sent to the management terminal. The unit monitors the request waiting time and compares it with the expiration time parameter. It performs a cleanup operation on the timed-out requests and releases the storage space. It outputs the queue backlog data to the capacity assessment module and transmits the abnormal capacity growth data to the anomaly identification module.

[0072] In this embodiment, the queue partitioning unit parses the service type, timestamp, user identifier, and other attribute information in the requests and accurately allocates them to four queues of different priorities according to preset rules. This achieves refined classification and management of backlogged requests, allowing urgent services such as registration submissions to enter the P0-level queue for the fastest response, while general query services enter lower-priority queues to wait for processing. This avoids the unreasonable resource allocation problem caused by all requests being mixed together in traditional solutions. By allocating independent storage space to each queue and setting expiration time parameters, the independence between requests of different priorities is ensured, and the request data is written to the storage medium through a persistence mechanism, ensuring data integrity after system restart or fault recovery.

[0073] This unit outputs the tiered queue data to the traffic release module, providing a clear data source for subsequent batch recovery. At the same time, it receives adjustment instructions from the priority scheduling module and dynamically updates the request arrangement position, enabling the priority of requests to be flexibly adjusted according to the actual situation. This completes the task of establishing a request classification system and supporting dynamic priority management, significantly improving the system's ability to handle different business scenarios and the rationality of resource allocation.

[0074] The capacity monitoring unit periodically collects the number of requests and space occupancy rates of each queue. When the capacity utilization rate reaches a preset threshold, it generates an alarm signal and sends it to the management terminal. This allows administrators to detect queue backlog risks in advance and take intervention measures such as capacity expansion or rate limiting, avoiding problems such as new requests failing to be queued and data loss due to queue capacity exhaustion. The unit monitors request waiting times and compares them with expiration time parameters, then performs cleanup operations on timed-out requests. This prevents expired and invalid requests from occupying storage space for extended periods and frees up space for newly arriving valid requests, thus completing the tasks of real-time queue capacity monitoring and automatic cleanup of expired requests.

[0075] This unit outputs queue backlog data to the capacity assessment module, providing important input for system capacity assessment. At the same time, it transmits abnormal capacity growth data to the anomaly identification module to help identify possible malicious attacks or system anomalies. This achieves the goals of preventing queue capacity risks and supporting abnormal traffic detection, thereby improving the system's self-protection capabilities and its ability to perceive abnormal situations.

[0076] The request grading module constructs a complete request grading and capacity management mechanism through the collaborative work of the queue partitioning unit and the capacity monitoring unit, achieving significant improvements compared to current mainstream technologies. Traditional solutions typically use a single queue to store all backlogged requests and cannot differentiate the urgency of different services, leading to potentially long waiting times for critical service requests. This module, however, uses a four-level queue for fine-grained classification, prioritizing urgent requests and significantly improving the response timeliness of critical services. Traditional solutions often rely on simple threshold alarms for queue capacity monitoring and lack proactive space management mechanisms, easily leading to queue overflow or wasted storage space. This module, through periodic monitoring and automatic cleanup of expired requests by the capacity monitoring unit, ensures the rational use of queue capacity and provides data support for anomaly detection. These improvements collectively achieve the effects of prioritizing critical services during circuit breaker recovery, improving storage resource utilization efficiency, and enhancing system capacity risk prevention capabilities, laying a solid foundation for subsequent traffic release and priority scheduling. It is particularly suitable for application scenarios such as art exam admissions, where the timeliness requirements of different services vary greatly.

[0077] Example 3: Please refer to Figure 1 The flow release module includes a stage control unit and a rate control unit;

[0078] The phase control unit manages five execution phases in sequence: detection, observation, acceleration, stabilization, and full volume. It sets the traffic release ratio, duration, and quality assessment criteria for each phase. After obtaining queue data from the request classification module, it extracts and releases requests according to the parameters of the current phase. It determines the phase switching based on the health assessment data returned by the detection adjustment module. When the assessment data meets the quality criteria, it enters the next phase. When it falls below the fallback threshold, it returns to the previous phase. At the same time, it outputs the phase execution data to the strategy learning module.

[0079] The rate control unit achieves smooth traffic release through the token bucket algorithm. The token generator issues tokens to the token bucket according to the traffic ratio parameter and preset frequency. Only after requesting to obtain a token can the business processing flow be entered. If there are not enough tokens, the process will be blocked and wait. The unit continuously monitors the deviation between the actual release rate and the target rate and dynamically adjusts the token issuance frequency. It also transmits real-time release data and response time indicators to the capacity assessment module.

[0080] In this embodiment, the phase control unit manages five execution phases sequentially: detection, observation, acceleration, stabilization, and full recovery. It sets clear traffic release ratios, durations, and quality assessment standards for each phase, enabling a step-by-step approach to the circuit breaker recovery process and avoiding potential service re-crash issues that might occur with traditional one-time full recovery. After obtaining queue data from the request classification module, this unit extracts and releases requests according to the current phase parameters. Based on the health assessment data returned by the detection adjustment module, it determines the phase switching direction. When the assessment data meets the quality standards, it advances to the next phase; when it falls below the fallback threshold, it reverts to the previous phase for re-observation. This bidirectional adjustment mechanism allows the recovery process to adaptively adjust based on the actual service performance. The unit outputs execution data for each phase to the policy learning module, providing historical data for subsequent policy parameter optimization. This completes the task of dividing the recovery process into controllable steps and dynamically adjusting it according to the service status, achieving the goal of shortening the recovery cycle as much as possible while ensuring service stability.

[0081] The rate control unit employs a token bucket algorithm for fine-grained control of traffic release speed. The token generator releases tokens into the token bucket according to traffic ratio parameters and a preset frequency. Requests can only enter the business processing flow after acquiring a token; if tokens are insufficient, requests will be blocked and wait. This mechanism effectively smooths the traffic release process, avoiding the impact of sudden traffic surges on backend services. This unit continuously monitors the deviation between the actual release rate and the target rate. When the actual rate is lower than the target value, the token release frequency is increased; when the actual rate is too high, the release frequency is decreased. This dynamic adjustment mechanism allows traffic release to adapt to real-time changes in service processing capacity. This unit transmits real-time release data and response time indicators to the capacity assessment module, providing accurate input information for system capacity assessment. It completes the task of precise control and adaptive adjustment of traffic release speed, achieving the goal of preventing service overload caused by traffic fluctuations and ensuring a smooth recovery process.

[0082] The traffic release module constructs a complete multi-stage smooth recovery system through the coordinated operation of the stage control unit and the rate control unit, achieving significant improvements compared to current mainstream technologies. Traditional circuit breaker recovery schemes typically employ a simple strategy of directly restoring all traffic after a fixed time delay, or a linear growth method with a fixed slope. These schemes cannot adjust the recovery pace according to the actual service load status, nor can they cope with service fluctuations during the recovery process, easily leading to a vicious cycle of triggering the circuit breaker again after recovery failure. This module adopts a five-stage progressive recovery strategy, gradually increasing the load starting from small traffic probes, combined with a two-way adjustment mechanism based on health assessment. This allows the recovery process to flexibly advance or regress according to the actual service performance. When service performance is good, the recovery speed is accelerated; when abnormal signs appear, timely degradation protection is implemented, significantly reducing the probability of secondary circuit breaker failures. Traditional traffic control solutions typically employ simple counting or time window limits, which can easily lead to sudden surges in traffic or a mismatch between the traffic release rate and service processing capacity. This module, however, utilizes a smooth traffic release mechanism implemented through the token bucket algorithm. This ensures stable traffic growth and dynamically adjusts the rate to adapt to changes in service processing capacity, preventing service overload or resource idleness caused by coarse traffic control. These technical improvements collectively enhance circuit breaker recovery success rates, shorten recovery cycles, and strengthen service stability, providing a reliable execution foundation for subsequent detection, adjustment, and capacity assessment. This approach is particularly suitable for business scenarios with significant traffic peaks and stringent service continuity requirements, such as art exam admission platforms.

[0083] Example 4: Please refer to Figure 1 The detection and adjustment module includes a parameter adjustment unit and a health assessment unit;

[0084] After obtaining operational data such as service historical success rate, response time, and number of fault recovery attempts, the parameter adjustment unit calculates a stability score. Based on the comparison between the score result and the preset threshold, it adjusts the probe window duration and the number of probe requests. The higher the service stability, the shorter the window duration and the more requests are set, and vice versa. The adjusted probe parameters are then transmitted to the stage control unit of the traffic release module to guide the probe execution.

[0085] The health assessment unit collects performance data such as request success rate, response time, connection utilization, and CPU load rate. After reading the preset weight coefficients of each indicator, it normalizes the performance data and calculates the health score by weighted summation. It obtains the quality standard and backoff threshold parameters of the current stage from the stage control unit. When the score reaches the standard, it generates a push signal; when the score is lower than the threshold, it generates a backoff signal. It transmits the signal to the stage control unit and outputs the performance data to the capacity assessment module.

[0086] In this embodiment, the parameter adjustment unit acquires operational data such as historical service success rate, response time, and number of fault recovery attempts, and calculates a stability score to achieve a quantitative evaluation of the service's historical performance. Based on the comparison between the score and preset thresholds, this unit adaptively adjusts the probe window duration and the number of probe requests. For services with higher stability, a shorter window and more requests are used to accelerate recovery; for services with lower stability, a longer window and fewer requests are used to reduce the risk of misjudgment. This overcomes the limitations of traditional fixed-parameter schemes that cannot adapt to different service states. The adjusted probe parameters are transmitted to the traffic release module to guide probe execution, completing the task of dynamically optimizing the probe strategy based on historical data and shortening the probe cycle while ensuring accuracy.

[0087] The health assessment unit collects multi-dimensional performance data, including request success rate, response time, connection utilization, and CPU load rate. After reading the preset weighting coefficients for each indicator, it normalizes the performance data and calculates a weighted sum to determine the health score. This provides a comprehensive and quantitative assessment of the service's current state, offering a more comprehensive and accurate judgment compared to traditional methods relying on a single indicator. The unit obtains the current stage's quality standards and rollback threshold parameters from the stage control unit as the basis for judgment. When the health score reaches the standard, a progress signal is generated; when the score falls below the threshold, a rollback signal is generated. This two-way decision-making mechanism based on multi-indicator comprehensive evaluation ensures sufficient data support for each stage switch, improving the scientific rigor and accuracy of stage switch decisions.

[0088] The detection and adjustment module constructs a complete adaptive detection and evaluation system through the collaborative work of the parameter adjustment unit and the health assessment unit, which is a significant improvement over the current mainstream technologies.

[0089] Traditional solutions use fixed detection parameters and a single health assessment standard, which cannot flexibly adjust detection strategies based on the historical performance and current status of different services. This module, however, dynamically optimizes the detection window and the number of requests based on historical service data, so that the detection intensity can match the actual stability level of the service. This not only speeds up the recovery of stable services but also reduces the risk of misjudgment for unstable services.

[0090] Traditional health assessment methods often rely on a single indicator, which is easily affected by fluctuations in individual indicators. This module, however, collects multi-dimensional performance data and performs weighted comprehensive evaluation, providing a more comprehensive and accurate assessment of service status. This effectively avoids misjudgments caused by abnormal single indicators, improves detection accuracy and recovery decision accuracy, and provides reliable decision support for the phased control of the traffic release module.

[0091] Example 5: Please refer to Figure 1 The flow prediction module includes a trend prediction unit and a preheating execution unit;

[0092] After acquiring historical access volume, traffic peaks, query volume, and other data, the trend prediction unit uses time series algorithms for cleaning, feature extraction, and pattern recognition to identify typical traffic patterns such as instantaneous peaks, continuous growth, and concentrated queries. Based on the current date and business schedule, it calculates the peak nodes that may appear in the future time window, as well as their numerical range and duration, and generates prediction results containing parameters such as peak traffic, arrival time, and duration. These results are then transmitted to the preheating execution unit, while prediction data is output to the capacity assessment module.

[0093] After receiving the traffic prediction data, the preheating execution unit calculates the preheating start time. When the system time reaches that time, the preheating steps are executed sequentially: a capacity expansion command is sent to the database module to increase the number of connection pools; a preloading command is sent to the cache module to load forms and college data; a capacity expansion request is sent to the resource module to increase the number of service instances; and an adjustment command is sent to the tiering module to reduce the resource quotas of non-core services. After execution, status data containing the execution results and timestamps of each step is generated and transmitted to the strategy learning module. At the same time, a system ready signal is sent to the stage control unit.

[0094] In this embodiment, the trend prediction unit acquires historical access volume, traffic peaks, query volume, and other data, and uses time series algorithms for cleaning, feature extraction, and pattern recognition to identify typical traffic patterns such as instantaneous peaks, continuous growth, and concentrated queries. Based on the current date and business schedule, this unit calculates the peak nodes, their numerical ranges, and durations that may occur within a future time window, generating prediction results containing key parameters such as peak traffic, arrival time, and duration. This allows the system to anticipate upcoming traffic peaks. The prediction results are transmitted to the preheating execution unit to guide resource deployment in advance, and simultaneously output to the capacity assessment module to provide a reference for carrying capacity assessment. This completes the tasks of traffic trend analysis and peak prediction, transforming traffic management from a passive response to an active defense.

[0095] After receiving traffic prediction data, the preheating execution unit calculates the preheating start time. When the system time reaches that time, it executes the preheating steps sequentially, including sending a scaling command to the database module to increase the number of connection pools, sending a preloading command to the cache module to load forms and school data, sending a scaling request to the resource module to increase the number of service instances, and sending an adjustment command to the tiering module to reduce the resource quotas of non-core services. This complete preheating operation ensures that the system has completed resource expansion and data preparation before the traffic peak, effectively avoiding service disruptions caused by sudden traffic surges. After execution, the generated status data is transmitted to the policy learning module, and a system ready signal is sent to the stage control unit, completing the pre-deployment of resources and system preheating preparation tasks.

[0096] The traffic prediction module constructs a complete traffic prediction and proactive defense system through the collaborative work of the trend prediction unit and the preheating execution unit, which represents a fundamental improvement over current mainstream technologies.

[0097] Traditional solutions typically employ static resource configuration or passive scaling based on real-time monitoring. These solutions only begin to respond after traffic has already reached its peak or even caused system stress, often resulting in untimely resource scaling and service outages. This module, however, predicts the arrival time and peak range of traffic peaks in advance based on historical data and business schedules, enabling the system to shift from passive response to proactive defense.

[0098] Traditional resource expansion solutions often involve simple, single-dimensional scaling. In contrast, this module systematically performs multi-dimensional warm-up operations before peak traffic arrives, such as expanding the database connection pool, preloading hot data, adding service instances, and degrading non-core services. This ensures that the system is in optimal capacity when traffic peaks arrive, significantly reducing the probability of circuit breakers being triggered by sudden traffic surges and improving the ability to handle peak traffic and service stability.

[0099] Example 6: Please refer to Figure 1 The capacity assessment module includes a resource monitoring unit and a carrying capacity calculation unit;

[0100] The resource monitoring unit collects resource data such as CPU utilization, memory usage, remaining connection pool quantity, cache hit rate, and bandwidth utilization at preset cycles and records them as a time series dataset. The dataset is then transmitted to the carrying capacity calculation unit and outputs resource usage trend data to the anomaly identification module. When the resource utilization exceeds the alarm threshold, an early warning signal containing the resource type and utilization value is generated and sent to the system management terminal.

[0101] After receiving resource data, the carrying capacity calculation unit establishes a multi-dimensional evaluation model that includes computing capacity, storage capacity, and network capacity. It uses different calculation formulas to calculate the capacity value of each dimension. The normalized capacity value is then combined using a weighted summation algorithm to obtain a health score. Based on the score range, the service status is determined and the corresponding load level parameters are output and transmitted to the stage control unit. When the score is lower than the warning threshold, the type of resource with insufficient capacity is identified and output to the resource allocation module.

[0102] In this embodiment, the resource monitoring unit collects multi-dimensional resource data such as CPU utilization, memory usage, remaining connection pool count, cache hit rate, and bandwidth utilization at preset intervals and records it as a time-series dataset, achieving comprehensive and continuous monitoring of critical system resources. The dataset is transmitted to the capacity calculation unit for capacity calculation and simultaneously outputs resource usage trend data to the anomaly identification module to help identify resource exhaustion anomaly patterns. When resource utilization exceeds the alarm threshold, an early warning signal containing the resource type and utilization value is generated and sent to the system management terminal. Through time-series data recording, historical tracking and trend analysis of resource usage are achieved, completing the tasks of real-time monitoring and anomaly early warning of system resource status. Risks are detected in a timely manner before resource exhaustion and accurate data support is provided for capacity assessment and anomaly identification.

[0103] After receiving resource data, the capacity calculation unit establishes a multi-dimensional evaluation model encompassing computing capacity, storage capacity, and network capacity. It calculates the capacity value for each dimension using different formulas, and then uses a weighted summation algorithm to synthesize the normalized capacity values ​​to obtain a health score. Compared to traditional single-dimensional monitoring methods, this provides a more comprehensive and accurate assessment of system status. Based on the score range, the unit determines the service status and outputs corresponding load level parameters, which are transmitted to the stage control unit to provide a scientific basis for the traffic release module's stage-based decision-making. When the score falls below the warning threshold, it identifies the type of resource with insufficient capacity and outputs this information to the resource allocation module to trigger expansion operations. This completes the multi-dimensional comprehensive assessment of system capacity and the quantitative determination of service status, providing reliable decision support for traffic release and resource allocation.

[0104] The capacity assessment module constructs a complete resource monitoring and capacity assessment system through the collaborative work of the resource monitoring unit and the carrying capacity calculation unit, which is a significant improvement over the current mainstream technical methods.

[0105] Traditional solutions typically use single resource metrics such as CPU utilization or memory usage for simple threshold alarms, which cannot fully reflect the true load status of the system. This can easily lead to situations where one resource is sufficient but other resources have become bottlenecks. In contrast, this module collects multi-dimensional resource data and establishes a time-series dataset, providing a more complete data foundation for system status assessment.

[0106] Traditional solutions often rely on empirical values ​​or fixed thresholds to judge system carrying capacity, lacking a scientific assessment of the combined impact of multiple resources. This module, however, establishes a multi-dimensional assessment model that includes computing capacity, storage capacity, and network capacity, and uses a weighted summation method to calculate a comprehensive health score. This enables a quantitative assessment of system carrying capacity, improving the accuracy of system load status monitoring, the accuracy of carrying capacity assessment, and the timeliness of resource bottleneck early warning.

[0107] Example 7: Please refer to Figure 1 The priority scheduling module includes a priority calculation unit and a scheduling execution unit;

[0108] After obtaining request data from the request classification module, the priority calculation unit extracts attributes such as business type, timestamp, and user identifier, and calculates four dimensions of parameters: timeliness, importance, waiting time, and number of retries. The scores of each dimension are obtained through time difference mapping, classification table lookup, and duration mapping, respectively. After assigning preset weight coefficients to each dimension, a weighted summation algorithm is used to calculate the comprehensive priority score. The score is written into the request record and transmitted to the scheduling execution unit.

[0109] After receiving request data with priority scores, the scheduling execution unit generates a processing queue by arranging the data in descending order of scores. It extracts requests according to the scheduling policy and allocates them to service instances. When the waiting time of a request exceeds a threshold, its score is increased. When the time remaining before the deadline is less than the threshold, the request is promoted to the highest priority. The unit processes high and low priority requests alternately according to a preset ratio, records scheduling process data, and calculates the processing volume and waiting time for each priority interval. The statistical data is then transmitted to the policy learning module and the processing rate is output to the capacity assessment module.

[0110] In this embodiment: After obtaining request data from the request classification module, the priority calculation unit extracts attributes such as business type, timestamp, and user identifier. It calculates four dimensions—timeliness, importance, waiting time, and number of retries—and assigns preset weight coefficients to each dimension. Then, a weighted summation algorithm is used to calculate the comprehensive priority score, achieving a multi-dimensional quantitative assessment of request importance and avoiding the problem that traditional simple first-in-first-out strategies cannot distinguish request importance. The scores for each dimension are obtained through time difference mapping, level table lookup, and duration mapping, ensuring the objectivity and accuracy of the scoring process and allowing time-sensitive services such as registration submissions to receive higher priority scores. The scores are written to the request record and transmitted to the scheduling execution unit, providing a clear basis for subsequent queuing and scheduling. This completes the comprehensive assessment and quantification of request priorities, laying the foundation for differentiated scheduling.

[0111] After receiving request data with priority scores, the scheduling execution unit arranges the requests in descending order of scores to generate a processing queue. It then extracts requests according to the scheduling policy and allocates them to service instances, ensuring that high-priority requests are processed first. When a request's waiting time exceeds a threshold, its score is increased; when the deadline is less than the threshold, it is elevated to the highest priority. This dynamic adjustment mechanism avoids the problem of low-priority requests being stuck for extended periods. The system alternates between high and low priority requests according to a preset ratio, ensuring fairness. The scheduling process data is recorded, and the processing volume and waiting time for each priority interval are statistically analyzed. This statistical data is transmitted to the policy learning module to provide a basis for subsequent optimization of the scheduling policy. The processing rate is output to the capacity assessment module to provide real-time data for assessing system capacity. This completes the task of reasonable request sorting and fair scheduling, ensuring that critical business processes are prioritized while also guaranteeing that all requests have a chance to be processed.

[0112] The priority scheduling module constructs a complete multi-dimensional priority evaluation and dynamic scheduling system through the collaborative work of the priority calculation unit and the scheduling execution unit, which is a significant improvement over the current mainstream technical means.

[0113] Traditional solutions typically employ simple first-in-first-out (FIFO) or fixed-priority queues, which fail to comprehensively consider multiple factors such as the timeliness, importance, and waiting status of requests. This can lead to critical business requests missing processing opportunities due to being placed in the back of the queue. In contrast, this module comprehensively evaluates four dimensions of parameters and uses a weighted summation algorithm to calculate priority scores, ensuring that the processing order of requests truly reflects their business importance and timeliness requirements.

[0114] Traditional solutions lack dynamic adjustment mechanisms, which can easily lead to low-priority requests being starved for extended periods. This module, however, uses a fairness guarantee mechanism that dynamically increases scores based on waiting time and alternates between high and low priority requests proportionally. This ensures that critical business processes are processed first while preventing low-priority requests from lingering, thus improving user experience and system processing efficiency. It is particularly suitable for application scenarios such as art exam admission platforms where different business processes have varying timeliness requirements and where fairness must be considered.

[0115] Example 8: Please refer to Figure 1 The strategy learning module includes a data acquisition unit and a model optimization unit;

[0116] The data acquisition unit receives recovery process data from various relevant modules, including stage execution time, traffic ratio, success rate, response time, health score, status switching record, request processing statistics, peak resource usage, etc. It assigns a unique number to each circuit breaker event and associates it with the triggering reason, total backlog of requests and queue distribution, strategy parameter configuration, recovery result and total time. It stores the complete event data in the historical database and transmits it to the model optimization unit.

[0117] After receiving historical event data, the model optimization unit constructs a reinforcement learning model that includes a state space, action space, reward function, and policy network. It uses an online learning algorithm to update the model parameters based on the actual recovery effect, analyzes the optimal policy configuration under different scenarios, and triggers model training when the historical event reaches the learning threshold. It then generates a policy recommendation scheme that includes stage duration, traffic ratio, and quality standard, and transmits the recommendation scheme to the stage control unit and parameter adjustment unit. The unit calculates the policy improvement magnitude and outputs evaluation data to the system management terminal.

[0118] In this embodiment: the resource allocation unit obtains system resource usage data and insufficient capacity indicators from the capacity assessment module, receives expansion requests from the traffic prediction module, and receives reserved and emergency expansion requests from the anomaly identification module. By maintaining a resource pool containing CPU, memory, connection pool, cache, and bandwidth, and recording the total capacity, allocated amount, and available amount, unified management of various system resources is achieved. Based on the requests, the required resources are calculated and allocated from the resource pool. When the available amount is insufficient, an application instruction is sent to the cloud service platform to obtain new resources and update the resource pool data, ensuring that the resource needs of each module are met in a timely manner. The allocation operation is recorded and transmitted to the resource transfer unit, completing the centralized allocation and dynamic expansion of resources, improving the response speed of resource acquisition and the overall resource guarantee capability of the system.

[0119] After receiving resource pool status data, the resource transfer unit periodically monitors the resource utilization of each service module, identifies low-load and high-load modules, and calculates idle and demand resources. When idle resources meet demand, resources are reclaimed from low-load modules and redistributed to high-load modules, enabling flexible resource allocation among services. After obtaining peak prediction data from the traffic prediction module, pre-deployment operations are triggered based on arrival time, requesting additional resources and allocating them to predicted high-load modules, avoiding service performance degradation due to insufficient resource preparation. Transfer and deployment data are recorded and transmitted to the policy learning module, and available capacity is output to the capacity assessment module, completing the dynamic transfer and pre-deployment of resources, improving resource utilization efficiency and the system's ability to cope with traffic fluctuations.

[0120] The resource allocation module constructs a complete unified resource management and dynamic allocation system through the collaborative work of the resource allocation unit and the resource transfer unit, which is a significant improvement over the current mainstream technical means.

[0121] Traditional solutions typically employ static resource allocation, where resource quotas for each service module are rarely adjusted after initial configuration. This can easily lead to situations where some service resources are idle while others are insufficient, resulting in low overall resource utilization. In contrast, this module periodically monitors the resource utilization of each service module and dynamically transfers idle resources between low-load and high-load modules, significantly improving resource utilization efficiency.

[0122] Traditional solutions often rely on manual prediction and capacity expansion to prepare resources for peak traffic periods, resulting in slow response times and a tendency for under- or over-preparation. This module, however, combines traffic prediction data to trigger resource pre-deployment operations in advance, completing resource application and allocation before the peak arrives. This automates and proactively prepares resources, effectively avoiding resource shortages caused by sudden traffic surges and improving the system's stability and responsiveness under high load conditions.

[0123] Example 9: Please refer to Figure 1 The anomaly identification module includes a feature detection unit and a response strategy unit;

[0124] The feature detection unit obtains real-time traffic data and resource usage trends from the resource monitoring unit, receives queue capacity growth data from the capacity monitoring unit, establishes an abnormal traffic feature database to store feature vectors of preset abnormal patterns, and uses principal component analysis or cluster analysis algorithms to extract feature parameters such as the current traffic growth rate, source concentration, type distribution, and resource consumption rate and combine them into a real-time feature vector. The similarity value between the real-time feature vector and the feature vectors of each abnormal pattern in the database is calculated using a cosine similarity algorithm or an Euclidean distance algorithm. When the similarity value exceeds a preset threshold, a corresponding abnormal type identifier is generated and transmitted to the response strategy unit.

[0125] After receiving the anomaly type identifier and feature vector, the response strategy unit performs a weighted calculation based on the traffic deviation parameter, resource utilization anomaly magnitude parameter, and duration parameter to obtain an anomaly score. The score is quantified to a numerical range, and the severity level is determined according to the range. The corresponding handling scheme is matched from the anomaly handling strategy library. When the score is low, monitoring data is output to the capacity assessment module. When the score is at a medium level, a resource reservation request is sent to the resource allocation module. When the score is high, a traffic limiting command, a service circuit breaker command, or an emergency expansion request is executed. The processing execution data is recorded and transmitted to the strategy learning module, and an anomaly alarm information is sent to the system management terminal.

[0126] In this embodiment: the feature detection unit obtains real-time traffic data and resource usage trends from the resource monitoring unit, and receives queue capacity growth data from the capacity monitoring unit. It establishes an abnormal traffic feature database to store feature vectors of preset abnormal patterns, providing a foundation for rapid comparison. Principal component analysis or clustering analysis algorithms are used to extract current traffic growth rate, source concentration, type distribution, and resource consumption rate feature parameters, which are then combined into a real-time feature vector. The similarity to each abnormal pattern in the database is calculated using a cosine similarity algorithm or Euclidean distance algorithm. When the similarity exceeds a preset threshold, a corresponding abnormal type identifier is generated and transmitted to the response strategy unit. This completes the real-time monitoring and type identification of abnormal traffic, enabling the system to accurately distinguish between normal business peaks and different abnormal types such as malicious attacks, avalanche failures, and resource exhaustion.

[0127] After receiving the anomaly type identifier and feature vector, the response strategy unit performs a weighted calculation based on parameters such as traffic deviation, resource utilization anomaly magnitude, and duration to obtain an anomaly score. The score is quantified into a numerical range, and its severity level is determined according to the range it belongs to. A corresponding handling solution is then matched from the anomaly handling strategy library. When the score is low, monitoring data is output to the capacity assessment module for observation. When the score is at a medium level, a resource reservation request is sent to the resource allocation module for prevention. When the score is high, traffic limiting commands, service circuit breaker commands, or emergency capacity expansion requests are executed to respond. This tiered response mechanism ensures that the strength of the handling measures matches the severity of the anomaly. This completes the tasks of anomaly severity assessment and differentiated handling solution execution, improving the accuracy and effectiveness of anomaly handling.

[0128] The anomaly identification module constructs a complete intelligent anomaly identification and hierarchical response system through the collaborative work of the feature detection unit and the response strategy unit, achieving many improvements compared to current mainstream technologies.

[0129] Traditional solutions typically employ a simple threshold alarm mechanism, triggering an alarm when a certain indicator exceeds a set threshold. This approach cannot distinguish between normal business peaks and abnormal traffic leading to false alarms or missed alarms. In contrast, this module establishes an abnormal traffic feature database and employs multi-dimensional feature extraction and similarity matching algorithms to accurately identify different types of abnormal patterns.

[0130] Traditional solutions often employ uniform measures for all anomalies, such as direct circuit breaking or simple rate limiting, lacking the ability to differentiate handling based on the type and severity of the anomaly. This module, however, quantifies the severity of anomalies and matches different levels of handling solutions based on the score. It forms a tiered response system from observation and monitoring to resource reservation, traffic limiting, or service circuit breaking. This system can take preventative measures in the early stages of anomalies and avoid overreacting to minor anomalies, thus improving the accuracy of anomaly identification and the precision of handling.

[0131] Example 10: Please refer to Figure 1 The resource allocation module includes a resource allocation unit and a resource transfer unit;

[0132] The resource allocation unit obtains resource usage data and insufficient capacity indicators from the capacity assessment module, receives expansion requests from the traffic prediction module, and receives reserved and emergency expansion requests from the anomaly identification module. It maintains a resource pool containing CPU, memory, connection pool, cache, and bandwidth and records the total capacity, allocated amount, and available amount. It calculates the required resources according to the requests and extracts allocations from the resource pool. When the available amount is insufficient, it sends an application instruction to the cloud service platform to obtain new resources and updates the resource pool data. It records the allocation operation and transmits it to the resource transfer unit.

[0133] After receiving the resource pool status data, the resource transfer unit periodically monitors the resource utilization rate of each service module, identifies low-load and high-load modules, and calculates the idle and demand amounts of resources. When the idle amount meets the demand amount, resources are reclaimed from the low-load modules and redistributed to the high-load modules. After obtaining peak prediction data from the traffic prediction module, the unit triggers pre-deployment operations in advance based on the arrival time, requests additional resources and allocates them to the predicted high-load modules, records the transfer and deployment data, transmits it to the policy learning module, and outputs the available capacity to the capacity assessment module.

[0134] In this embodiment: the resource allocation unit obtains resource usage data and insufficient capacity identifiers from the capacity assessment module, receives expansion requests from the traffic prediction module, and receives reserved and emergency expansion requests from the anomaly identification module. By maintaining a resource pool containing CPU, memory, connection pool, cache, and bandwidth, and recording total capacity, allocated capacity, and available capacity, it achieves unified management of various system resources. Based on the requests, it calculates the required resources and extracts allocations from the resource pool. When available capacity is insufficient, it sends an application command to the cloud service platform to obtain new resources and updates the resource pool data, ensuring that the resource needs of each module are met in a timely manner. The allocation operation is recorded and transmitted to the resource transfer unit, completing the centralized allocation and dynamic expansion of resources. Compared to the traditional method of independent resource management by each module, this significantly improves the resource acquisition response speed.

[0135] After receiving resource pool status data, the resource transfer unit periodically monitors the resource utilization of each service module, identifies low-load and high-load modules, and calculates idle and demand resources. When idle resources meet demand, resources are reclaimed from low-load modules and redistributed to high-load modules, avoiding situations where some service resources are idle while others are underutilized. After obtaining peak prediction data from the traffic prediction module, pre-deployment operations are triggered based on arrival time, requesting additional resources and allocating them to predicted high-load modules, enabling the system to prepare resources before traffic peaks arrive. Transfer and deployment data are recorded and transmitted to the policy learning module, and available capacity is output to the capacity assessment module, completing the dynamic transfer and pre-deployment of resources, improving resource utilization efficiency without increasing total resource investment.

[0136] The resource allocation module constructs a complete unified resource management and dynamic allocation system through the collaborative work of the resource allocation unit and the resource transfer unit, which is a significant improvement over the current mainstream technical means.

[0137] Traditional solutions typically employ static resource allocation, with resource quotas for each service module rarely adjusted after initial configuration. This can easily lead to situations where some service resources are idle while others are insufficient. In contrast, this module periodically monitors the resource utilization rate of each service module and dynamically transfers idle resources between low-load and high-load modules, significantly improving resource utilization efficiency.

[0138] Traditional solutions often rely on manual prediction and capacity expansion to prepare resources for peak traffic periods, resulting in slow response times and a high risk of insufficient preparation. This module, however, combines traffic prediction data to trigger resource pre-deployment operations in advance and completes resource application and allocation before the peak arrives, thus automating and proactively preparing resources. This effectively avoids resource shortages caused by sudden traffic surges and improves the system's stability and responsiveness under high load conditions.

[0139] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0140] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An art exam admission platform based on big data analysis, characterized in that... It includes a request grading module, a traffic release module, a detection and adjustment module, a traffic prediction module, a capacity assessment module, a priority scheduling module, a policy learning module, an anomaly identification module, and a resource allocation module; The request tiering module divides the backlog of requests during the circuit breaker period into four queues according to their urgency, and handles requests for different business scenarios respectively. Each queue is set with independent storage space and expiration time, and saves data through a persistence mechanism and monitors capacity status. When the capacity reaches a preset threshold, an alarm is triggered. The traffic release module adopts a multi-stage recovery strategy to gradually release traffic. It starts with a small amount of traffic in the initial detection stage and gradually increases the traffic ratio according to the service response. Each stage has time requirements and quality standards. When the service performance is poor, it automatically drops back to the previous stage. The detection and adjustment module dynamically adjusts detection parameters based on historical service performance, comprehensively evaluates multiple performance indicators, calculates service health status using a weighted approach, and allows the next recovery phase to proceed when the evaluation results meet the standards. The traffic prediction module analyzes historical access patterns to predict traffic trends and initiates preheating preparations before peak periods arrive, including expanding system resources, loading frequently used data, and adjusting service configurations. The capacity assessment module continuously monitors the usage of critical system resources, establishes an assessment model to calculate service carrying capacity, quantifies service status, and determines the current load level that can be tolerated. The priority scheduling module comprehensively considers factors such as the timeliness, importance, and waiting status of requests to allocate processing order to each request, while ensuring that requests of different priorities all have a chance to be processed, and avoiding low-level requests from being stuck for a long time. The strategy learning module records the process data of each circuit breaker recovery, analyzes the success and failure factors through the learning model, continuously optimizes the configuration of various parameters of the recovery strategy, and applies the improved solution in subsequent events. The anomaly identification module establishes feature models for various anomalies, analyzes current traffic characteristics in real time, identifies anomaly types and assesses their severity, and takes corresponding measures based on different anomalies. The resource allocation module manages all types of resources in a unified manner, dynamically adjusts resource allocation based on service load, transfers idle resources between services to support high-load modules, and completes resource deployment in advance based on forecast information.

2. The art exam admission platform based on big data analysis according to claim 1, characterized in that: The request hierarchy mode includes a queue partitioning unit and a capacity monitoring unit; The queue partitioning unit receives backlogged requests during the circuit breaker period, parses the service type identifier, timestamp, user identifier and other attribute information in the requests, and allocates the requests to the P0 level emergency queue, P1 level high priority queue, P2 level normal queue or P3 level low priority queue according to preset rules. It allocates independent storage space for each queue and sets expiration time parameters. It writes the request data to the preset storage medium through the persistence mechanism, outputs the hierarchical queue data to the traffic release module, receives the adjustment instructions from the priority scheduling module and updates the request arrangement position. The capacity monitoring unit periodically collects the number of requests and space occupancy rate of each queue. When the capacity utilization rate reaches the first preset threshold, an alarm signal is generated and sent to the management terminal. The unit monitors the request waiting time and compares it with the expiration time parameter. It performs a cleanup operation on the timed-out requests and releases the storage space. It outputs the queue backlog data to the capacity assessment module and transmits the abnormal capacity growth data to the anomaly identification module.

3. The art exam admission platform based on big data analysis according to claim 2, characterized in that: The flow release module includes a stage control unit and a rate control unit; The phase control unit manages five execution phases in sequence: detection, observation, acceleration, stabilization, and full volume. It sets the traffic release ratio, duration, and quality assessment criteria for each phase. After obtaining queue data from the request classification module, it extracts and releases requests according to the parameters of the current phase. It determines the phase switching based on the health assessment data returned by the detection adjustment module. When the assessment data meets the quality criteria, it enters the next phase. When it falls below the fallback threshold, it returns to the previous phase. At the same time, it outputs the phase execution data to the strategy learning module. The rate control unit achieves smooth traffic release through the token bucket algorithm. The token generator issues tokens to the token bucket according to the traffic ratio parameter and preset frequency. Only after requesting to obtain a token can the business processing flow be entered. If there are not enough tokens, the process will be blocked and wait. The unit continuously monitors the deviation between the actual release rate and the target rate and dynamically adjusts the token issuance frequency. It also transmits real-time release data and response time indicators to the capacity assessment module.

4. The art exam admission platform based on big data analysis according to claim 3, characterized in that: The detection and adjustment module includes a parameter adjustment unit and a health assessment unit; After obtaining operational data such as service historical success rate, response time, and number of fault recovery attempts, the parameter adjustment unit calculates a stability score. Based on the comparison between the score result and the preset threshold, it adjusts the probe window duration and the number of probe requests. The higher the service stability, the shorter the window duration and the more requests are set, and vice versa. The adjusted probe parameters are then transmitted to the stage control unit of the traffic release module to guide the probe execution. The health assessment unit collects performance data such as request success rate, response time, connection utilization, and CPU load rate. After reading the preset weight coefficients of each indicator, it normalizes the performance data and calculates the health score by weighted summation. It obtains the quality standard and backoff threshold parameters of the current stage from the stage control unit. When the score reaches the standard, it generates a push signal; when the score is lower than the threshold, it generates a backoff signal. It transmits the signal to the stage control unit and outputs the performance data to the capacity assessment module.

5. The art exam admission platform based on big data analysis according to claim 4, characterized in that: The flow prediction module includes a trend prediction unit and a preheating execution unit; After acquiring historical access volume, traffic peaks, query volume, and other data, the trend prediction unit uses time series algorithms for cleaning, feature extraction, and pattern recognition to identify typical traffic patterns such as instantaneous peaks, continuous growth, and concentrated queries. Based on the current date and business schedule, it calculates the peak nodes that may appear in the future time window, as well as their numerical range and duration, and generates prediction results containing parameters such as peak traffic, arrival time, and duration. These results are then transmitted to the preheating execution unit, while prediction data is output to the capacity assessment module. After receiving the traffic prediction data, the preheating execution unit calculates the preheating start time. When the system time reaches that time, the preheating steps are executed sequentially: a capacity expansion command is sent to the database module to increase the number of connection pools; a preloading command is sent to the cache module to load forms and college data; a capacity expansion request is sent to the resource module to increase the number of service instances; and an adjustment command is sent to the tiering module to reduce the resource quotas of non-core services. After execution, status data containing the execution results and timestamps of each step is generated and transmitted to the strategy learning module. At the same time, a system ready signal is sent to the stage control unit.

6. The art exam admission platform based on big data analysis according to claim 5, characterized in that: The capacity assessment module includes a resource monitoring unit and a carrying capacity calculation unit; The resource monitoring unit collects resource data such as CPU utilization, memory usage, remaining connection pool quantity, cache hit rate, and bandwidth utilization at preset cycles and records them as a time series dataset. The dataset is then transmitted to the carrying capacity calculation unit and outputs resource usage trend data to the anomaly identification module. When the resource utilization exceeds the alarm threshold, an early warning signal containing the resource type and utilization value is generated and sent to the system management terminal. After receiving resource data, the carrying capacity calculation unit establishes a multi-dimensional evaluation model that includes computing capacity, storage capacity, and network capacity. It uses different calculation formulas to calculate the capacity value of each dimension. The normalized capacity value is then combined using a weighted summation algorithm to obtain a health score. Based on the score range, the service status is determined and the corresponding load level parameters are output and transmitted to the stage control unit. When the score is lower than the warning threshold, the type of resource with insufficient capacity is identified and output to the resource allocation module.

7. The art exam admission platform based on big data analysis according to claim 6, characterized in that: The priority scheduling module includes a priority calculation unit and a scheduling execution unit; After obtaining request data from the request classification module, the priority calculation unit extracts attributes such as business type, timestamp, and user identifier, and calculates four dimensions of parameters: timeliness, importance, waiting time, and number of retries. The scores of each dimension are obtained through time difference mapping, classification table lookup, and duration mapping, respectively. After assigning preset weight coefficients to each dimension, a weighted summation algorithm is used to calculate the comprehensive priority score. The score is written into the request record and transmitted to the scheduling execution unit. After receiving request data with priority scores, the scheduling execution unit generates a processing queue by arranging the data in descending order of scores. It extracts requests according to the scheduling policy and allocates them to service instances. When the waiting time of a request exceeds a threshold, its score is increased. When the time remaining before the deadline is less than the threshold, the request is promoted to the highest priority. The unit processes high and low priority requests alternately according to a preset ratio, records scheduling process data, and calculates the processing volume and waiting time for each priority interval. The statistical data is then transmitted to the policy learning module and the processing rate is output to the capacity assessment module.

8. The art exam admission platform based on big data analysis according to claim 7, characterized in that: The strategy learning module includes a data acquisition unit and a model optimization unit; The data acquisition unit receives recovery process data from various relevant modules, including stage execution time, traffic ratio, success rate, response time, health score, status switching record, request processing statistics, peak resource usage, etc. It assigns a unique number to each circuit breaker event and associates it with the triggering reason, total backlog of requests and queue distribution, strategy parameter configuration, recovery result and total time. It stores the complete event data in the historical database and transmits it to the model optimization unit. After receiving historical event data, the model optimization unit constructs a reinforcement learning model that includes a state space, action space, reward function, and policy network. It uses an online learning algorithm to update the model parameters based on the actual recovery effect, analyzes the optimal policy configuration under different scenarios, and triggers model training when the historical event reaches the learning threshold. It then generates a policy recommendation scheme that includes stage duration, traffic ratio, and quality standard, and transmits the recommendation scheme to the stage control unit and parameter adjustment unit. The unit calculates the policy improvement magnitude and outputs evaluation data to the system management terminal.

9. The art exam admission platform based on big data analysis according to claim 8, characterized in that: The anomaly detection module includes a feature detection unit and a response strategy unit; The feature detection unit obtains real-time traffic data and resource usage trends from the resource monitoring unit, receives queue capacity growth data from the capacity monitoring unit, establishes an abnormal traffic feature database to store feature vectors of preset abnormal patterns, and uses principal component analysis or cluster analysis algorithms to extract feature parameters such as the current traffic growth rate, source concentration, type distribution, and resource consumption rate and combine them into a real-time feature vector. The similarity value between the real-time feature vector and the feature vectors of each abnormal pattern in the database is calculated using a cosine similarity algorithm or an Euclidean distance algorithm. When the similarity value exceeds a preset threshold, a corresponding abnormal type identifier is generated and transmitted to the response strategy unit. After receiving the anomaly type identifier and feature vector, the response strategy unit performs a weighted calculation based on the traffic deviation parameter, resource utilization anomaly magnitude parameter, and duration parameter to obtain an anomaly score. The score is quantified to a numerical range, and the severity level is determined according to the range. The corresponding handling scheme is matched from the anomaly handling strategy library. When the score is low, monitoring data is output to the capacity assessment module. When the score is at a medium level, a resource reservation request is sent to the resource allocation module. When the score is high, a traffic limiting command, a service circuit breaker command, or an emergency expansion request is executed. The processing execution data is recorded and transmitted to the strategy learning module, and an anomaly alarm information is sent to the system management terminal.

10. The art exam admission platform based on big data analysis according to claim 9, characterized in that: The resource allocation module includes a resource allocation unit and a resource transfer unit; The resource allocation unit obtains resource usage data and insufficient capacity indicators from the capacity assessment module, receives expansion requests from the traffic prediction module, and receives reserved and emergency expansion requests from the anomaly identification module. It maintains a resource pool containing CPU, memory, connection pool, cache, and bandwidth and records the total capacity, allocated amount, and available amount. It calculates the required resources according to the requests and extracts allocations from the resource pool. When the available amount is insufficient, it sends an application instruction to the cloud service platform to obtain new resources and updates the resource pool data. It records the allocation operation and transmits it to the resource transfer unit. After receiving the resource pool status data, the resource transfer unit periodically monitors the resource utilization rate of each service module, identifies low-load and high-load modules, and calculates the idle and demand amounts of resources. When the idle amount meets the demand amount, resources are reclaimed from the low-load modules and redistributed to the high-load modules. After obtaining peak prediction data from the traffic prediction module, the unit triggers pre-deployment operations in advance based on the arrival time, requests additional resources and allocates them to the predicted high-load modules, records the transfer and deployment data, transmits it to the policy learning module, and outputs the available capacity to the capacity assessment module.

Images