Dynamic resource scheduling method for website load prediction

Abstract

The invention relates to the technical field of website resource scheduling, and discloses a website load prediction-oriented dynamic resource scheduling method, which comprises the following steps of: obtaining a predicted load value in a future time period output by a load prediction model and a confidence coefficient score for quantifying uncertainty; comparing the confidence score with a high-confidence threshold, and judging whether the prediction is high-confidence prediction; when the confidence coefficient is high, generating a full capacity expansion instruction to schedule all expected traffic to a new capacity expansion node, and when the confidence coefficient is not high, calculating a traffic migration proportion according to the confidence coefficient, and gradually migrating the traffic by adopting a progressive strategy; monitoring performance indexes and service response quality of the new nodes in real time in the migration process, and judging whether the running state and the service capability of the new nodes reach the standard or not; and if the standard is reached, adjusting the migration proportion or full migration in the next period, and if the standard is not reached, pausing / rollback migration and feeding back abnormal information to the prediction model for correction. According to the invention, dynamic traffic scheduling based on confidence is realized, and prediction verification and service stability are balanced.

Description

Technical Field

[0001] This invention relates to the field of resource scheduling technology in cloud computing and website service management, and in particular to a dynamic resource scheduling method for website load prediction. Background Technology

[0002] In the internet service sector, website operations face the challenge of fluctuating user access demands. Achieving efficient dynamic resource scheduling is crucial for ensuring stable service operation and controlling operating costs. The key to dynamic resource scheduling lies in accurately predicting future website load changes and adjusting the configuration and allocation of server and other hardware resources in advance based on these predictions.

[0003] However, load forecasting itself is inherently uncertain due to factors such as data noise, business scenario complexity, and sudden traffic surges. Existing technologies, when performing resource scheduling, often treat load forecasts as deterministic instructions for resource adjustments, completely ignoring the differences in the reliability of the forecast results. For example, a load forecasting model's output of "future load growth of 50%" might be based on clear historical business patterns, or it might stem from data noise or random factors. However, existing scheduling strategies treat these two different confidence levels of forecasts equally, employing the same scaling or traffic scheduling methods for both.

[0004] This scheduling method presents significant technical problems: when the prediction confidence is low, if the system expands its node capacity on a large scale based on this unreliable prediction signal, the newly launched server cluster may instantly carry all the expected traffic. If the prediction deviation is large or the new nodes have unknown hardware or software vulnerabilities, it can easily cause widespread service failures, leading to website response delays, request failures, or even system crashes. On the other hand, if an overly conservative scheduling strategy is adopted due to concerns about the uncertainty of the prediction, service performance may decline due to insufficient resources when the actual load surges, affecting user experience. It may also result in low utilization of existing resources, leading to resource waste.

[0005] The core technical challenges of existing technologies lie mainly in two aspects: First, the lack of an effective method for measuring the uncertainty of prediction results makes it impossible to accurately quantify the reliability of each model prediction; second, the failure to deeply integrate the measurement of prediction uncertainty with the dynamic scheduling process, and the lack of a confidence-based buffering and probing mechanism in traditional scheduling processes, makes scheduling actions rigid and carry high business risks in low-confidence intervals. For example, when the model predicts a sudden surge in load whose reliability is difficult to determine, the scheduling system can only choose to expand all capacity or remain inactive, both of which are accompanied by significant business risks.

[0006] Therefore, how to design a mechanism that can quantify the uncertainty of load forecasting and intelligently and smoothly guide business traffic to newly expanded resource nodes based on the forecast confidence level, and achieve a balance between verifying the accuracy of forecasting and ensuring service stability, has become a key technical problem that current dynamic resource scheduling methods for website load forecasting urgently need to solve. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a dynamic resource scheduling method for website load prediction. By introducing a confidence score for load prediction, it enables a differentiated traffic scheduling strategy based on confidence, solving the technical problems of existing scheduling ignoring prediction uncertainty and rigid strategies. This method balances the risks and efficiency of resource scheduling, ensuring website service stability while improving resource utilization.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: A dynamic resource scheduling method for website load prediction, the method comprising: Obtain the predicted load values ​​for future time periods and their corresponding confidence scores from the load forecasting model. The confidence scores are used to quantify the uncertainty of the forecast results. The confidence score is compared with a preset high confidence threshold to determine whether the predicted load value is a high confidence prediction. If the predicted load value is a high-confidence prediction, a full expansion instruction is generated, which is used to instruct all expected growth in business traffic to be scheduled to the new expansion resource node. If the predicted load value is not a high-confidence prediction, then the traffic migration ratio is calculated based on the confidence score. The traffic migration ratio is used to determine the proportion of business traffic that migrates from the existing resource node to the newly expanded resource node. A gradual traffic scheduling strategy is adopted, and according to the traffic migration ratio, the corresponding proportion of online service traffic is gradually routed from the existing resource nodes to the newly expanded resource nodes within a preset time window; During the gradual traffic scheduling process, the performance indicators of the newly expanded resource nodes and the service response quality corresponding to the migrated traffic are monitored in real time. Based on the performance metrics and service response quality data obtained from real-time monitoring, determine whether the newly expanded resource nodes are operating stably and whether their service capabilities meet expectations. If the newly expanded resource node is operating stably and its service capacity meets expectations, then in the next scheduling cycle, the traffic migration ratio will be adjusted or a full traffic migration will be performed based on the updated load prediction results and confidence level. If the newly expanded resource node is unstable or its service capacity does not meet expectations, the traffic migration operation will be paused or rolled back, and the abnormal status information of the node will be fed back to the load prediction model so that the prediction model can be corrected in the future.

[0009] Furthermore, obtaining the predicted load values ​​for future time periods and their corresponding confidence scores output by the load prediction model specifically includes: The load prediction model is used to obtain the predicted load data for future time periods and extract the corresponding confidence scores to form preliminary prediction results. Based on the predicted load data, the load change trend in different time periods is analyzed to determine the time distribution characteristics of load fluctuations. If the time distribution characteristics of load fluctuations exceed the preset threshold range, the data in the abnormal time period is deeply mined to determine whether there are potential abnormal patterns. The uncertainty of the prediction results is quantitatively characterized by confidence scores to obtain the reliability assessment results of the predicted load data for each time period; based on the reliability assessment results, the data of the time periods with low confidence scores are screened, the corresponding load prediction values ​​are extracted, and the key prediction intervals are determined. For the key prediction intervals, a comparative analysis is conducted using historical load data to determine the degree of fit between the prediction results and the actual load trends. Based on the comparative analysis results, a load prediction optimization and adjustment strategy for future periods is generated, and the final prediction output is determined.

[0010] Furthermore, the confidence score is compared with a preset high confidence threshold to determine whether the predicted load value is a high confidence prediction, specifically including: Obtain the predicted load values ​​and their corresponding confidence scores for multiple future time periods; compare the confidence scores for each time period with a preset high confidence threshold: if the confidence score is higher than the high confidence threshold, mark the predicted load value for the corresponding time period as a high-reliability prediction; if the confidence score is lower than the high confidence threshold, mark the predicted load value for the corresponding time period as a low-reliability prediction. For periods marked as low-reliability predictions, obtain the corresponding historical load data for the same period; based on the volatility characteristics of the historical load data, correct the predicted load values ​​for low-reliability periods; and generate the final comprehensive load prediction sequence based on the corrected predicted load values ​​and the high-reliability prediction values.

[0011] Furthermore, if the predicted load value is a high-confidence prediction, a full capacity expansion instruction is generated, specifically including: Obtain the predicted load sequence and its corresponding confidence sequence, and compare each confidence score in the confidence sequence with a preset high confidence threshold; if the confidence score at a certain time point is higher than the high confidence threshold, then the predicted load value corresponding to that time point is determined to be a high reliable load value. Based on the high-reliability load value, aggregate calculations are performed to determine the expected total service traffic; a full scheduling instruction is generated through the resource scheduling model to schedule the expected total service traffic to the newly expanded resource node; the full scheduling instruction is executed to complete the migration of service traffic from the original resource node to the newly expanded resource node.

[0012] Furthermore, if the predicted load value is not a high-confidence prediction, the traffic migration ratio is calculated based on the confidence score, specifically including: Obtain the predicted load data and corresponding confidence scores. By comparing them with the preset confidence standard, filter the predicted load values ​​that do not meet the high confidence condition to form a set of substandard loads. For the set of substandard loads, determine the traffic migration ratio corresponding to each predicted load value by means of score extrapolation based on the difference between the confidence score and the preset confidence standard. A service traffic allocation scheme is generated based on the traffic migration ratio. Corresponding traffic data is extracted from the service traffic carried by the existing resource nodes according to the ratio to form a traffic set to be transferred. The traffic set to be transferred is gradually scheduled to the newly expanded resource nodes through the resource scheduling unit to complete the traffic migration scheduling. After the business traffic distribution data is obtained after the scheduling is completed, it is determined whether the traffic distribution of each resource node is balanced. If there is an uneven traffic distribution, the traffic redistribution mechanism is activated to dynamically adjust the node traffic until the distribution is balanced. The traffic migration ratio, scheduling rate and traffic redistribution process are completely retained through the log recording unit to form a scheduling log and determine the final resource node allocation result.

[0013] Furthermore, the adoption of a gradual traffic scheduling strategy, which, according to the traffic migration ratio, gradually routes a corresponding proportion of online service traffic from existing resource nodes to newly expanded resource nodes within a preset time window, specifically includes: Acquire online business traffic data, collect real-time distribution information of existing resource node business traffic through the traffic monitoring unit, and determine the initial benchmark for traffic migration; based on the initial benchmark and traffic migration ratio, calculate the share of business traffic to be migrated in each stage within a preset time window according to preset traffic control rules, and generate a phased traffic transfer plan. The phased transfer plan is implemented in a gradual manner, and the routing switch is carried out step by step in each time window to gradually redirect business traffic from existing resource nodes to newly expanded resource nodes, and the completion status of traffic transfer in each stage is verified. The node switching monitoring mechanism collects service traffic distribution data in real time after the route switching. If an imbalance in traffic distribution between existing resource nodes and newly expanded resource nodes is detected, the traffic control adjustment mechanism is triggered to make the traffic distribution more balanced. Based on the balanced traffic allocation results, the online service operation status is continuously monitored. If an abnormal traffic change is detected, dynamic routing adjustment is performed through the traffic control unit to determine the final traffic allocation scheme. The log recording unit fully preserves the traffic migration process, routing switching operations, and traffic adjustment actions, forming a timestamped business traffic transfer trajectory and building a traceable scheduling archive.

[0014] Furthermore, during the gradual traffic scheduling process, the performance metrics of the newly expanded resource nodes and the service response quality corresponding to the migrated traffic are monitored in real time, specifically including: By collecting operational status data of newly expanded resource nodes through preset monitoring probes, the node processing capacity and stability indicators are obtained to determine the real-time operational status of the nodes. Based on the node operational status, the service response time and business request success rate indicators of the migration traffic are extracted through the data analysis unit to form a service response quality assessment result. If the service response time exceeds the preset threshold, the traffic allocation unit will dynamically adjust the migrated traffic to optimize the traffic allocation status. Based on the adjusted traffic allocation results, the differences in service operation performance between existing resource nodes and newly expanded resource nodes are continuously compared to generate performance-differentiated distribution data. If the service performance of a certain resource node is lower than the preset standard, the standby node activation mechanism is triggered, and some service traffic is routed to the standby node to form an updated traffic scheduling scheme. The log recording unit records the entire scheduling process, performance anomalies, and adjustment actions, generating a traceable scheduling operation trajectory.

[0015] Furthermore, the step of determining whether the newly expanded resource node's operating status is stable and whether its service capabilities meet expectations based on performance indicators and service response quality data obtained from real-time monitoring specifically includes: The real-time acquisition unit continuously acquires performance indicators and service response quality data of newly expanded resource nodes to form real-time node operating status information; the data analysis unit performs comprehensive comparative analysis of the above multi-dimensional indicators to generate preliminary assessment results of node operating stability. If the evaluation result is lower than the preset stability standard, the information processing unit will locate the time period of the anomaly and the related business type, and the traffic analysis unit will perform in-depth analysis of the load distribution characteristics of the corresponding resource nodes. If the analysis results indicate that the node load exceeds the preset threshold, the traffic scheduling system will be used to divert and schedule the service traffic of that node to determine the traffic distribution pattern after diversion. After the traffic split is completed, the changes in node performance indicators are continuously tracked. The entire process of the traffic split operation is retained through the log recording unit to form a complete operation trajectory. The scheduling logs are classified and archived through the data storage system to build a historical dataset for subsequent analysis.

[0016] Furthermore, the process of adjusting the traffic migration ratio or performing a full traffic migration specifically includes: The status monitoring module continuously collects information on the operating status and service capabilities of resource nodes to form a record of node health status; the information processing unit integrates and analyzes multi-dimensional operating data to determine whether the current service capabilities of the node meet the preset standards. If the node's service capacity does not meet expectations, the load prediction data for that node is obtained through the load prediction system to identify potential traffic overload risks; the reliability of the prediction results is evaluated by the confidence analysis unit, and the prediction fluctuation characteristics are extracted to form a basis for traffic migration and adjustment. Based on the aforementioned adjustment criteria, the traffic migration ratio of each node in the next scheduling cycle is dynamically calculated by the scheduling cycle management unit to form a traffic allocation scheme. If the traffic migration ratio exceeds the preset safety threshold, the traffic allocation system will perform step-by-step migration to achieve node load balancing. The data update unit records the changes in node load status and operating indicators before and after migration, generates a complete scheduling cycle execution log, and archives and stores it.

[0017] Furthermore, the process of pausing or rolling back the traffic migration operation and feeding back the abnormal status information of the node to the load prediction model specifically includes: The status monitoring module obtains the real-time operating indicators of the newly expanded resource nodes and compares them with the preset health threshold. If the operating indicators are lower than the health threshold, the node is determined to be in an abnormal operating state. Based on the abnormal state, a traffic migration interruption command is generated, and the scheduling module stops allocating new traffic to the node. Collect load time-series data and performance time-series data within a preset time window of abnormal nodes, and use a time series decomposition algorithm to decompose the load series into trend components and residual components to extract abnormal fluctuation characteristics. The abnormal fluctuation characteristics and corresponding performance sequences are used as model correction samples to update the internal feature weight parameters of the load prediction model. Based on the updated load prediction model, the load prediction values ​​and risk levels for subsequent scheduling cycles are recalculated to form the basis for the corrected scheduling decisions.

[0018] The technical effects and advantages of this invention are as follows: 1. The dynamic resource scheduling method for website load prediction provided in this application, by introducing a confidence score for load prediction, achieves a quantitative measurement of the uncertainty of prediction results, solves the technical problem that existing technologies cannot distinguish the reliability of predictions, provides a data foundation for differentiated resource scheduling, and fundamentally avoids the scheduling risks caused by blindly accepting low-confidence predictions. Based on the confidence-based differentiated scheduling strategy, full-scale expansion scheduling is performed when the confidence prediction is high, improving resource scheduling efficiency; when the confidence prediction is not high, a confidence-driven traffic migration ratio calculation and a gradual scheduling strategy are adopted to achieve gradual traffic migration. While testing the carrying capacity of new expansion nodes, the service failure risk of a one-time large-scale traffic migration is effectively avoided, achieving a balance between prediction accuracy verification and service stability.

[0019] 2. The dynamic resource scheduling method for website load prediction provided in this application sets up a real-time monitoring and dynamic adjustment mechanism throughout the gradual scheduling process. It continuously monitors the performance indicators and service response quality of newly expanded nodes, and makes timely dynamic adjustments in response to uneven traffic, substandard node performance, sudden traffic changes, etc., and even activates backup nodes, thereby further ensuring the continuity and stability of website services.

[0020] 3. The dynamic resource scheduling method for website load prediction provided in this application establishes a closed-loop feedback mechanism for scheduling results, feeds back the abnormal status information of newly expanded nodes to the load prediction model, and continuously corrects the prediction model by updating the internal weight parameters of the model, thereby improving the accuracy of subsequent load prediction and forming a closed-loop management of "prediction-scheduling-monitoring-feedback-correction", realizing the self-optimization and intelligence of resource scheduling. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of the dynamic resource scheduling method for website load prediction according to the present invention; Figure 2 This is a flowchart illustrating the real-time monitoring and dynamic adjustment process during the progressive traffic scheduling of this invention. Figure 3 This is a schematic diagram of the full capacity expansion scheduling process during high-confidence prediction in this invention. Detailed Implementation

[0022] 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.

[0023] like Figure 1 As shown in the figure, the dynamic resource scheduling method for website load prediction provided by this embodiment of the invention specifically includes the following steps: Step S1: Obtain the predicted load values ​​and corresponding confidence scores for future time periods output by the load forecasting model. This step specifically includes: obtaining predicted load data for future time periods through the load forecasting model and extracting the corresponding confidence scores (between 0 and 1, with higher scores indicating stronger prediction reliability) to obtain preliminary prediction results; analyzing the changing trends of the obtained predicted load data in different time periods to determine the temporal distribution characteristics of load fluctuations; if the temporal distribution characteristics of load fluctuations exceed a preset threshold range, conducting in-depth analysis of the data within abnormal time periods to determine if there are potential abnormal patterns (such as scheduled batch processing tasks, potential malicious traffic attacks, etc.); quantifying the uncertainty of the prediction results using confidence scores to obtain a reliability assessment of the predicted load data for each time period; filtering out time periods with lower confidence scores based on the reliability assessment results, obtaining their corresponding load prediction values, and determining the prediction intervals that require focused attention; comparing and analyzing the prediction intervals with historical load data to determine the degree of consistency between the prediction results and the actual trends; and generating optimization and adjustment strategies for load forecasting in future time periods based on the comparative analysis results to determine the final prediction output.

[0024] Specifically, in the scenario of server load management in cloud computing data centers, the system obtains processor utilization prediction data for the next 24 hours through a load prediction model. While the model outputs the predicted load percentage for each hour, it also outputs a confidence score ranging from 0 to 1. The confidence score represents the degree of confidence of the model in the corresponding prediction result, providing basic data support for subsequent resource scheduling decisions.

[0025] For example, after acquiring the forecast data, the system analyzes the load fluctuation characteristics of each time period and defines the formula for calculating the load fluctuation amplitude as follows: ;in, Let be the load fluctuation amplitude in time period t. Let t be the predicted load value for time period t. This represents the average load value for the (t-1)th time period and the k preceding time periods (k=3 in this example). If a sudden increase in the predicted processor load from the normal 20% to 85% is detected between 2 AM and 4 AM, the calculated value is... The magnitude of this drastic fluctuation exceeds the preset 30% normal fluctuation threshold. ,satisfy If the abnormal period is detected, in-depth analysis will be performed to identify potential abnormal patterns, including but not limited to scheduling anomalies in timed big data batch processing tasks and potential malicious traffic attacks, so as to achieve early warning of abnormal load and avoid server downtime due to sudden high load.

[0026] Furthermore, the system performs quantitative analysis of prediction uncertainty based on confidence scores, and defines the formula for calculating confidence bias as follows: ;in, Let be the confidence bias for time period t. The confidence baseline threshold, This represents the actual confidence score for time period t. If the predicted load from 2 PM to 4 PM is 60%, but its confidence score is only 0.4, the standard baseline threshold... Calculated If the confidence level is significantly higher than the allowable deviation (0.2 in this example), then this low-confidence period is selected as a key focus interval. Low confidence indicates that the business traffic characteristics during this period are complex and have weak regularity. By selecting the above intervals, the system can avoid making incorrect resource scaling operations based on unreliable predictions.

[0027] For the key focus period of 2 PM to 4 PM identified above, historical load data from the past month was retrieved for comparative analysis, and the prediction deviation rate was calculated using the following formula: ;in, To predict the deviation rate, Let t be the predicted load value for time period t. This represents the historical average load value for time period t. If historical data indicates that the load during this period is prone to surge to 90% due to online promotional activities (…), then… ), while the current forecast is only 60% ( ), calculated ,when If the predicted result deviates from the actual trend, a load prediction optimization and adjustment strategy is generated based on the comparison result. The resource reservation capacity for that period is increased to a level that adapts to 90% load, correcting the model prediction deviation and ensuring that the final output prediction result can guarantee the stable operation of the business.

[0028] Step S2 involves comparing the confidence score with a preset high confidence threshold to determine whether the predicted load value is a high-confidence prediction. This step specifically includes: obtaining predicted load values ​​and their corresponding confidence scores for multiple future time periods; comparing the confidence score for each time period with a preset high confidence threshold (e.g., 0.85); if the confidence score is higher than the high confidence threshold, the predicted load value for that time period is marked as a high-reliability prediction; if the confidence score is lower than the high confidence threshold, the predicted load value for that time period is marked as a low-reliability prediction; for time periods marked as low-reliability predictions, obtaining their corresponding historical load data for the same period; using the volatility characteristics of the historical load data for the same period to correct the predicted load values ​​for the low-reliability time periods; and generating the final comprehensive load prediction sequence based on the corrected predicted load values ​​and the high-reliability prediction values.

[0029] Specifically, in the scenario of server load management in cloud computing data centers, the system continuously acquires processor load prediction data for multiple future time periods and confidence scores used to quantify the uncertainty of predictions; the system presets a high confidence threshold (such as 0.85) to classify the reliability of the prediction results.

[0030] For example, the system obtains the load forecast sequence for the next 12 hours and compares it time by time; if the predicted load is 70% from 9:00 AM to 10:00 AM, the confidence score is 0.92 ( If the predicted load value for that period is higher than the 0.85 threshold, it is marked as a high-reliability prediction, indicating that the model has fully captured the business characteristics of that period, and the prediction result can be directly accepted; if the predicted load for 8 PM to 9 PM is 45%, the confidence score is only 0.65. If the predicted load value for that period is below the threshold, it is marked as a low-reliability prediction.

[0031] For periods marked as having low reliability forecasts, the system activates a historical data backtracking mechanism to retrieve corresponding historical load data for the same period. For example, for the low reliability interval from 8 PM to 9 PM, the system extracts the actual load records for the corresponding period on the same workday over the past four weeks. The system does not simply use historical averages, but rather analyzes the volatility characteristics of historical load data for the same period. The historical load volatility calculation formula is as follows:

[0032] in, Historical load volatility n The number of historical data samples. Let i be the load value of the i-th historical sample. This represents the historical load sample mean. If the historical average load for this period is low, but the volatility is high... If the volatility variance is significantly large, then this high volatility characteristic is the main factor leading to low confidence in the model prediction.

[0033] The system corrects the predicted load values ​​for periods of low reliability based on the extracted volatility characteristics. The correction formula is as follows: ;in, This is the corrected predicted load value. For fluctuation correction coefficient (in this example) Based on historical fluctuation limits, the original forecast of 45% has been adjusted to... (assuming historical volatility) This process covers potential traffic surges. After correcting all low-reliability periods, the corrected predicted load values ​​are time-series concatenated with the high-reliability predicted values ​​to form a complete comprehensive load prediction sequence, providing a unified data benchmark for global resource scheduling in the data center.

[0034] Step S3: If the predicted load value is a high-confidence prediction, a full expansion instruction is generated. This full expansion instruction is used to instruct all expected increases in service traffic to be scheduled to the newly expanded resource nodes. Figure 3 As shown, this step specifically includes: obtaining the predicted load sequence and its corresponding confidence sequence; comparing each element in the confidence sequence with a preset high confidence threshold (e.g., 0.90); if the confidence at a certain time point is higher than the threshold, then the predicted load value at that time point is determined to be a high-reliability load value; determining the expected total traffic volume based on the sum of the high-reliability load values; using a resource scheduling model to generate a full scheduling instruction that directs the expected total traffic volume to the newly expanded resource node; and executing the full scheduling instruction to complete the migration of traffic volume from the original node to the newly expanded node.

[0035] Specifically, in the scenario of expanding and scheduling cloud computing data center clusters, the system obtains the predicted load sequence and corresponding confidence sequence of future business cycles in real time, and presets a high confidence threshold (such as 0.90) to filter high-deterministic prediction data.

[0036] For example, based on comparison, the predicted load for the period from 10:00 AM to 11:00 AM is 8000 requests per second. The confidence score is 0.95. If the predicted load value is higher than the preset threshold, the predicted load value for that period is a high-reliability load value, indicating that the surge in business traffic during that period has a high degree of certainty.

[0037] The system aggregates the high-reliability load values ​​obtained from the screening to determine the total expected business traffic. The calculation formula is as follows: ;in, Let T be the expected total traffic volume, and T be the set of highly reliable prediction periods. Let t be the high-reliability load value for time period t. For example, the high-reliability load from 10:00 AM to 11:00 AM is 8000 requests per second. The predicted load from 11:00 to 12:00 was 9000 requests per second with a confidence score of 0.94, which is also considered a high-reliability load. After cumulative calculation, The highly certain expected total traffic volume during this business cycle is 17,000 requests per second. Traffic volume assessment based on high-confidence data can eliminate interference from low-confidence fluctuations, providing reliable data for resource scheduling.

[0038] After determining the total expected traffic volume, the system generates a full scheduling instruction through its built-in resource scheduling model. Given the large scale and high certainty of this traffic volume, to avoid resource contention and performance bottlenecks on existing resource nodes, the entire expected traffic volume is scheduled to the newly expanded resource nodes that have completed preheating.

[0039] For example, the resource scheduling model generates a full scheduling instruction containing the target network address pool and traffic weight allocation strategy, and sends it to the global load balancing device. After the instruction is executed, the system updates the underlying routing configuration rules, smoothly migrating the business traffic of 17,000 requests per second from the original high-load node to the newly expanded node, thereby releasing the resource pressure of the original cluster.

[0040] Step S4: If the predicted load value is not a high-confidence prediction, calculate the traffic migration ratio based on the confidence score. The traffic migration ratio is used to determine the proportion of service traffic migrated from existing resource nodes to newly expanded resource nodes. This step specifically includes: acquiring predicted load data and corresponding confidence scores; comparing with preset load standards to determine which load values ​​do not meet the high-confidence requirement, thus obtaining a set of non-compliant loads; for the non-compliant load set, using a score extrapolation method, calculating the difference between the confidence score corresponding to each load value and the preset standard, and determining the traffic migration ratio (the larger the difference, the lower the migration ratio); generating a service traffic allocation scheme based on the traffic migration ratio, extracting traffic data that meets the ratio from existing resource nodes to obtain a set of traffic to be transferred; using a resource node scheduling tool, gradually guiding the set of traffic to be transferred to the newly expanded resource nodes to complete the node transfer process; acquiring service traffic distribution data after node transfer, determining whether there is an uneven traffic proportion, and if an uneven phenomenon is detected, triggering a traffic redistribution mechanism to adjust to a balanced state; using a recording tool to save detailed logs of traffic migration and node transfer, and determining the final resource node allocation result.

[0041] Specifically, in cloud computing cluster resource scheduling scenarios, the system monitors predicted load data in real time and categorizes predicted loads with confidence scores below a preset threshold (such as 0.90) into the substandard load set.

[0042] For example, the predicted request volume between 2 PM and 3 PM is 5000 requests per second. The confidence score is 0.70. The confidence difference is lower than the preset threshold; the system calculates the confidence difference as follows: Based on this difference, the dynamic traffic migration ratio is determined using a fractional calculation method. The formula for calculating the migration ratio is: ;in, r For the traffic migration ratio, Maximum migration ratio (in this example) ), The maximum allowable confidence difference (in this example) Substituting the data, we can calculate... In this example, the traffic migration rate is 30%.

[0043] Based on the aforementioned 30% traffic migration ratio, a business traffic allocation plan is generated, and the formula for calculating the traffic to be migrated is as follows: ;in, For traffic to be transferred, Let r be the predicted load value for this period, and r be the traffic migration ratio. Substituting the data yields... The system extracts 1,500 requests per second from the business traffic corresponding to the existing resource nodes as a set of traffic to be transferred; and modifies the underlying routing weights through the resource scheduling unit to gradually schedule the traffic to be transferred to the newly expanded resource nodes, so as to reduce the risk of traffic fluctuations caused by low confidence prediction.

[0044] After traffic migration is complete, the system collects traffic distribution data from each resource node in real time to monitor the cluster's operating status. The formula for calculating traffic distribution balance is defined as follows: ;in, For the balance of traffic distribution, This represents the maximum load value for each node. This represents the minimum load value for each node. This represents the average load value for each node. If the resource utilization rate of the newly expanded node reaches 85%, while the utilization rate of the original resource nodes is only 40%, the calculated value is... ,when If a significant traffic imbalance is detected, a traffic redistribution mechanism is immediately triggered to dynamically adjust routing configurations and traffic distribution strategies until the load on each resource node tends to be balanced.

[0045] The system records the entire process, including traffic migration ratio, step-by-step scheduling rate, and traffic redistribution operations, through a log recording unit, forming a standardized scheduling log archive, which is used to determine the final resource node allocation result.

[0046] Step S5: Employing a progressive traffic scheduling strategy, according to the traffic migration ratio, within a preset time window, a corresponding proportion of online service traffic is gradually routed from existing resource nodes to newly expanded resource nodes. This step specifically includes: acquiring online service traffic data; collecting real-time data using traffic monitoring tools based on the current distribution of existing resource nodes and service traffic to determine the initial baseline for traffic migration; calculating the share of service traffic to be transferred at each stage within the time window based on the initial baseline and the proportional allocation requirements, using preset traffic control rules to obtain a phased transfer plan; employing a progressive scheduling strategy for the phased transfer plan, gradually executing routing and transfer operations within each time window to guide service traffic from existing nodes to the new expansion point, determining the completion status of each stage's transfer; and acquiring data through a node switching monitoring mechanism. After the routing transfer, if an uneven distribution of traffic between existing nodes and new expansion points is detected, traffic control adjustments are triggered to obtain a balanced traffic distribution. Based on the balanced traffic distribution, the operation of online services among resource nodes is continuously monitored to determine if there are any sudden traffic changes. If an anomaly is detected, dynamic routing adjustments are performed using traffic control tools to determine the final traffic allocation scheme. Log recording tools are used to save detailed operation records of traffic migration and node switching. For each routing transfer and traffic control adjustment, a complete business traffic transfer trajectory is generated, resulting in a long-term traceable scheduling archive.

[0047] Specifically, the system uses traffic monitoring tools to collect the number of concurrent connections of existing resource nodes at the second level, and currently handles 8000 requests per second. ) is set as the initial baseline for traffic migration. Based on preset traffic control rules, a total of 4000 requests (to be migrated) are selected. The time window is divided into m time windows, and the migration traffic calculation formula for each time window is: In this example, m=4, substituting the data yields... This forms a phased migration plan that migrates 1,000 requests per minute.

[0048] Within the first time window, the scheduling unit modifies the gateway routing configuration, routing 1000 requests to the newly expanded resource node and verifying the connection establishment status. If, during the scheduling process, the utilization rate of the newly expanded resource node rises to 90% and the utilization rate of the existing resource node drops to 20%, the traffic distribution balance is calculated. If uneven traffic distribution occurs, a traffic rollback adjustment will be triggered, migrating some traffic back to existing nodes to restore balance.

[0049] If online service traffic suddenly surges during the scheduling period, the formula for determining traffic surge is defined as follows: ;in, This refers to the sudden change in flow rate. This is a real-time traffic value. To predict flow rates. When When a sudden traffic surge occurs, the system determines it and performs dynamic traffic allocation, scheduling the new traffic according to the remaining capacity of the nodes. All migration data, routing adjustments, and anomaly handling records are persistently stored in structured logs, forming a complete and traceable scheduling trajectory.

[0050] Step S6: During the progressive traffic scheduling process, the performance indicators of the newly expanded resource nodes and the service response quality corresponding to the migrated traffic are monitored in real time. For example... Figure 2 As shown, this step specifically includes: acquiring the operational status data of the newly expanded resource nodes; collecting the node's processing capacity (CPU utilization, network throughput, etc.) and stability indicators (service process survival status, etc.) at fixed intervals (e.g., 10 seconds) using preset monitoring tools (such as monitoring probes) to determine the node's real-time operational status; using data analysis methods to extract service response time and success rate information of the migration traffic based on the node's real-time operational status, obtaining the service response quality assessment results; if the response time exceeds a preset threshold, dynamically adjusting the migration traffic using traffic allocation tools to determine the adjusted traffic distribution status; continuously monitoring the differences in business performance between the resource nodes and the newly expanded points based on the adjusted traffic distribution status, obtaining information on nodes with inconsistent performance through comparative analysis, and obtaining differentiated distribution data; if the business performance of a node is found to be lower than a preset standard based on the differentiated distribution data, triggering the activation mechanism of the backup node, routing some traffic to the backup node, and determining a new traffic scheduling scheme; and using a log recording tool to save detailed data of each scheduling process, generating a complete scheduling operation trajectory, and obtaining traceable business records.

[0051] Specifically, the system collects operational metrics such as CPU utilization, network throughput, and service process liveness status of newly expanded resource nodes every 10 seconds using monitoring probes. Based on gateway access logs, it extracts end-to-end response latency and business success rate of migration traffic, such as an average response time of 300 milliseconds and a request success rate of 99.5%, to generate a service quality assessment report.

[0052] If the average response time of a node is detected to reach 650 milliseconds, exceeding the preset threshold of 500 milliseconds ( ,satisfy If a node's processing capacity is deemed to be bottlenecked, the migration traffic is dynamically adjusted through the traffic allocation unit. The adjusted migration traffic is calculated using the following formula: ;in, For the adjusted migration traffic, For the migration traffic before adjustment, For real-time response time, This is the response time threshold. (In this example...) Substituting the data yields In actual engineering, the number of requests per second is rounded up to 800 (fine-tuned based on the remaining processing capacity of the nodes), reducing the migration traffic from 2000 requests per second to 800 requests per second.

[0053] If a significant performance difference is consistently detected between the newly added node and existing nodes, and the performance of the new node remains below a preset benchmark, the backup resource pool is activated, and traffic carried by the abnormal node is routed to the backup node. The entire process, including the time of anomaly detection, traffic adjustment ratio, and node switching instructions, is recorded as structured data for subsequent business auditing and review.

[0054] Step S7: Based on the performance indicators and service response quality data obtained from real-time monitoring, determine whether the newly expanded resource node is operating stably and whether its service capabilities meet expectations. This step specifically includes: continuously collecting performance metrics (memory fragmentation rate, disk I / O wait time, concurrent connections, etc.) and response quality data of newly expanded resource nodes through a real-time monitoring system to obtain real-time status information of node operation; using data analysis tools to comprehensively compare performance metrics and response quality based on real-time status information to determine the preliminary assessment results of node operation stability; if the preliminary assessment results show that the operation stability is lower than the preset standard, the information processing module extracts specific response quality anomaly data to determine the specific time period and business type of the anomaly; based on the specific time period and business type of the anomaly, the traffic analysis tool is called to perform in-depth analysis of the load of relevant resource nodes to obtain detailed characteristics of load distribution; based on the detailed characteristics of load distribution, if it is found that the load of a certain node exceeds the preset threshold, the traffic scheduling system is used to divert the business traffic of that node to determine the traffic allocation status after diversion; based on the traffic allocation status after diversion, the performance metrics of the newly expanded resource nodes are continuously tracked, and the log data of each diversion operation is saved using a recording tool to obtain complete operation trajectory information; based on the complete operation trajectory information, the data storage system is used to classify and archive it to obtain historical data available for subsequent analysis.

[0055] Specifically, the system uses a sliding window approach to collect real-time statistics on node memory fragmentation rate, disk I / O latency, and application-layer concurrent connections via underlying data acquisition components. The data analysis unit comprehensively evaluates underlying hardware metrics and long-tail request latency from the business layer. The formula for calculating the node operational stability comprehensive evaluation index is as follows: ; Wherein, S is the comprehensive stability evaluation index (the value ranges from 0 to 1, and the closer it is to 1, the better the stability). The weights for CPU utilization, disk I / O latency, and concurrent connections are respectively (in this example). ), These are the preset thresholds for each indicator. These are the real-time values ​​for each indicator. If the 99th percentile request latency reaches 800 milliseconds and the number of concurrent connections approaches the upper limit, the node's operational stability is deemed substandard.

[0056] The information processing unit further identifies the abnormally concentrated time periods and business types, such as large file uploads causing a continuous increase in memory usage between 8 PM and 9 PM. After confirming that a node's load has exceeded its limit, the traffic scheduling system lowers the traffic allocation weight for the abnormal nodes, redirecting excess traffic to lower-loaded nodes within the cluster. The traffic diversion calculation formula is as follows:

[0057] in, For traffic that needs to be diverted, The current load of the abnormal node. This serves as the node load threshold. The system continuously tracks node performance recovery and archives the threshold trigger time, weight adjustment ratio, and scheduling actions in a structured manner, forming historical data that can be used for model iteration and operational review.

[0058] Step S8: If the newly expanded resource node is running stably and its service capacity meets expectations, then in the next scheduling cycle, the traffic migration ratio will be adjusted or a full traffic migration will be performed based on the updated load prediction results and confidence level. This step specifically includes: continuously collecting operational status data and service capability information (such as CPU instruction cycle consumption rate, network bandwidth throughput, and service processing rate) of resource nodes through the status monitoring module to obtain real-time node health status records; classifying and organizing the operational status data using information processing tools based on the node health status records to determine whether the current service capability of each node meets the expected standards; if the current service capability of a node is lower than the expected standard, obtaining the load prediction data of that node through the load prediction system to determine whether there is a potential risk of traffic overload; extracting key fluctuation information from the prediction results using the confidence level analysis tool based on the load prediction data to obtain the preliminary adjustment basis for traffic migration; calling the scheduling cycle management module based on the preliminary adjustment basis for traffic migration to dynamically calculate the traffic migration ratio of each node in the next cycle and determine the specific allocation scheme; if the traffic migration ratio of a node exceeds a preset threshold (such as 20%), implementing a step-by-step migration operation through the traffic allocation system to obtain the node load balance status after migration; and recording the node load balance status and operational status changes after migration through the data update module to obtain a complete scheduling cycle execution log.

[0059] Specifically, the status monitoring module collects indicators such as node CPU utilization and network bandwidth throughput in real time to form a node health profile. The information processing unit compares the service processing rate with a preset baseline; if the actual performance is lower than the baseline, the service capacity is deemed unqualified. The load forecasting system combines historical data to extrapolate future load trends, identifies overload risks caused by traffic surges, and outputs a predictive reliability assessment result through confidence analysis.

[0060] The scheduling cycle management unit calculates the target traffic migration ratio based on the above information. If the ratio exceeds a preset safety threshold (e.g., ...), the unit will determine the migration method. If so, a multi-stage progressive migration will be performed, and the formula for calculating the migration ratio in each stage is as follows: ;in, The migration percentage at each stage, The target migration ratio is represented by k, and the number of migration stages is represented by k. Throughout the migration process, node performance changes, queue lengths, and scheduling times are all recorded in a structured manner, forming a standardized scheduling cycle execution log.

[0061] Step S9: If the newly expanded resource node's operating status is unstable or its service capacity does not meet expectations, the traffic migration operation is paused or rolled back, and the abnormal status information of the node is fed back to the load prediction model to enable subsequent prediction model correction. This step specifically includes: obtaining the real-time operating indicators and preset health thresholds of the new node through the status monitoring module; if the operating indicators are lower than the health threshold, the node is determined to be in an abnormal state; generating a migration interruption command based on the abnormal state, and the scheduling module stops allocating new traffic to the node; collecting load sequences (such as the number of concurrent requests) and performance sequences (such as the business processing success rate) within a specific time window from the abnormal node; processing the load sequences using a time series decomposition algorithm to separate the abnormal fluctuations of the trend component and residual component; using the abnormal fluctuation data and the corresponding performance sequence as correction data to update the internal weight parameters of the prediction model; and recalculating the load value and risk indicators for future scheduling cycles based on the updated prediction model.

[0062] Specifically, the status monitoring module monitors indicators such as disk I / O latency and memory swapping frequency of the newly expanded node in real time. If the disk I / O latency exceeds the preset 20 millisecond threshold ( The time spiked to 55 milliseconds. ,satisfy If the node is found to be faulty, the scheduling module will immediately perform traffic isolation and stop sending new service traffic to that node.

[0063] The system extracts the load and performance sequences within the abnormal time window, removes the trend component using a time series decomposition algorithm, and locates sudden abnormal fluctuations in the residuals. The decomposition formula is as follows: ;in, Let be the load value at time t. As a trend component, For seasonal portions, The residual components (abnormal fluctuations are mainly reflected in the residual components) are used as training samples to input the load prediction model, optimize the model's perception weights of resource bottlenecks, and improve the model's prediction sensitivity.

[0064] Based on the updated model, the load level and risk indicators of subsequent scheduling cycles are re-inferred, and the corrected load prediction results are output, providing more accurate data support for subsequent resource scheduling and capacity planning.

[0065] This invention obtains the future load value and its confidence score output by the load prediction model to determine whether the prediction is a high-confidence prediction. If so, it directly executes full capacity expansion scheduling; otherwise, it dynamically calculates the traffic migration ratio based on the confidence score and adopts a gradual traffic scheduling strategy to gradually migrate business traffic. Simultaneously, it monitors the performance and service quality of the newly expanded resource nodes in real time to ensure stable operation. If an anomaly occurs, the migration is paused or rolled back, and feedback is sent to the prediction model for correction. This confidence-based dynamic traffic ratio adjustment and gradual scheduling mechanism reduces the scheduling risks caused by uncertain predictions while ensuring resource utilization efficiency and service stability, ultimately achieving intelligent and precise management of website load prediction and resource scheduling.

[0066] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A dynamic resource scheduling method for website load prediction, characterized by: The method includes: Obtain the predicted load values ​​for future time periods and their corresponding confidence scores from the load forecasting model. The confidence scores are used to quantify the uncertainty of the forecast results. The confidence score is compared with a preset high confidence threshold to determine whether the predicted load value is a high confidence prediction. If the predicted load value is a high-confidence prediction, a full expansion instruction is generated, which is used to instruct all expected growth in business traffic to be scheduled to the new expansion resource node. If the predicted load value is not a high-confidence prediction, then the traffic migration ratio is calculated based on the confidence score. The traffic migration ratio is used to determine the proportion of business traffic that migrates from the existing resource node to the newly expanded resource node. A gradual traffic scheduling strategy is adopted, and according to the traffic migration ratio, the corresponding proportion of online service traffic is gradually routed from the existing resource nodes to the newly expanded resource nodes within a preset time window; During the gradual traffic scheduling process, the performance indicators of the newly expanded resource nodes and the service response quality corresponding to the migrated traffic are monitored in real time. Based on the performance metrics and service response quality data obtained from real-time monitoring, determine whether the newly expanded resource nodes are operating stably and whether their service capabilities meet expectations. If the newly expanded resource node is operating stably and its service capacity meets expectations, then in the next scheduling cycle, the traffic migration ratio will be adjusted or a full traffic migration will be performed based on the updated load prediction results and confidence level. If the newly expanded resource node is unstable or its service capacity does not meet expectations, the traffic migration operation will be paused or rolled back, and the abnormal status information of the node will be fed back to the load prediction model so that the prediction model can be corrected in the future.

2. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The acquisition of the predicted load values ​​for future time periods and their corresponding confidence scores from the load forecasting model specifically includes: The load prediction model is used to obtain the predicted load data for future time periods and extract the corresponding confidence scores to form preliminary prediction results. Based on the predicted load data, the load change trend in different time periods is analyzed to determine the time distribution characteristics of load fluctuations. If the time distribution characteristics of load fluctuations exceed the preset threshold range, the data in the abnormal time period is deeply mined to determine whether there are potential abnormal patterns. The uncertainty of the prediction results is quantitatively characterized by confidence scores to obtain the reliability assessment results of the predicted load data for each time period; based on the reliability assessment results, the data of the time periods with low confidence scores are screened, the corresponding load prediction values ​​are extracted, and the key prediction intervals are determined. For the key prediction intervals, a comparative analysis is conducted using historical load data to determine the degree of fit between the prediction results and the actual load trends. Based on the comparative analysis results, a load prediction optimization and adjustment strategy for future periods is generated, and the final prediction output is determined.

3. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The confidence score is compared with a preset high confidence threshold to determine whether the predicted load value is a high confidence prediction, specifically including: Obtain the predicted load values ​​and their corresponding confidence scores for multiple future time periods; compare the confidence scores for each time period with a preset high confidence threshold: if the confidence score is higher than the high confidence threshold, mark the predicted load value for the corresponding time period as a high-reliability prediction; if the confidence score is lower than the high confidence threshold, mark the predicted load value for the corresponding time period as a low-reliability prediction. For periods marked as low-reliability predictions, obtain the corresponding historical load data for the same period; based on the volatility characteristics of the historical load data, correct the predicted load values ​​for low-reliability periods; and generate the final comprehensive load prediction sequence based on the corrected predicted load values ​​and the high-reliability prediction values.

4. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, If the predicted load value is a high-confidence prediction, then a full capacity expansion instruction is generated, specifically including: Obtain the predicted load sequence and its corresponding confidence sequence, and compare each confidence score in the confidence sequence with a preset high confidence threshold; if the confidence score at a certain time point is higher than the high confidence threshold, then the predicted load value corresponding to that time point is determined to be a high reliable load value. Based on the high-reliability load value, aggregate calculations are performed to determine the expected total service traffic; a full scheduling instruction is generated through the resource scheduling model to schedule the expected total service traffic to the newly expanded resource node; the full scheduling instruction is executed to complete the migration of service traffic from the original resource node to the newly expanded resource node.

5. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, If the predicted load value is not a high-confidence prediction, then the traffic migration ratio is calculated based on the confidence score, specifically including: Obtain the predicted load data and corresponding confidence scores. By comparing them with the preset confidence standard, filter the predicted load values ​​that do not meet the high confidence condition to form a set of substandard loads. For the set of substandard loads, determine the traffic migration ratio corresponding to each predicted load value by means of score extrapolation based on the difference between the confidence score and the preset confidence standard. A service traffic allocation scheme is generated based on the traffic migration ratio. Corresponding traffic data is extracted from the service traffic carried by the existing resource nodes according to the ratio to form a traffic set to be transferred. The traffic set to be transferred is gradually scheduled to the newly expanded resource nodes through the resource scheduling unit to complete the traffic migration scheduling. After the business traffic distribution data is obtained after the scheduling is completed, it is determined whether the traffic distribution of each resource node is balanced. If there is an uneven traffic distribution, the traffic redistribution mechanism is activated to dynamically adjust the node traffic until the distribution is balanced. The traffic migration ratio, scheduling rate and traffic redistribution process are completely retained through the log recording unit to form a scheduling log and determine the final resource node allocation result.

6. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The progressive traffic scheduling strategy, according to the traffic migration ratio, gradually routes a corresponding proportion of online service traffic from existing resource nodes to newly expanded resource nodes within a preset time window, specifically including: Acquire online business traffic data, collect real-time distribution information of existing resource node business traffic through the traffic monitoring unit, and determine the initial benchmark for traffic migration; based on the initial benchmark and traffic migration ratio, calculate the share of business traffic to be migrated in each stage within a preset time window according to preset traffic control rules, and generate a phased traffic transfer plan. The phased transfer plan is implemented in a gradual manner, and the routing switch is carried out step by step in each time window to gradually redirect business traffic from existing resource nodes to newly expanded resource nodes, and the completion status of traffic transfer in each stage is verified. The node switching monitoring mechanism collects service traffic distribution data in real time after the route switching. If an imbalance in traffic distribution between existing resource nodes and newly expanded resource nodes is detected, the traffic control adjustment mechanism is triggered to make the traffic distribution more balanced. Based on the balanced traffic allocation results, the online service operation status is continuously monitored. If an abnormal traffic change is detected, dynamic routing adjustment is performed through the traffic control unit to determine the final traffic allocation scheme. The log recording unit fully preserves the traffic migration process, routing switching operations, and traffic adjustment actions, forming a timestamped business traffic transfer trajectory and building a traceable scheduling archive.

7. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, During the gradual traffic scheduling process, the performance metrics of the newly expanded resource nodes and the service response quality corresponding to the migrated traffic are monitored in real time, specifically including: By collecting operational status data of newly expanded resource nodes through preset monitoring probes, the node processing capacity and stability indicators are obtained to determine the real-time operational status of the nodes. Based on the node operational status, the service response time and business request success rate indicators of the migration traffic are extracted through the data analysis unit to form a service response quality assessment result. If the service response time exceeds the preset threshold, the traffic allocation unit will dynamically adjust the migrated traffic to optimize the traffic allocation status. Based on the adjusted traffic allocation results, the differences in service operation performance between existing resource nodes and newly expanded resource nodes are continuously compared to generate performance-differentiated distribution data. If the service performance of a certain resource node is lower than the preset standard, the standby node activation mechanism is triggered, and some service traffic is routed to the standby node to form an updated traffic scheduling scheme. The log recording unit records the entire scheduling process, performance anomalies, and adjustment actions, generating a traceable scheduling operation trajectory.

8. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The process of determining whether the newly expanded resource nodes are stable and whether their service capabilities meet expectations based on performance metrics and service response quality data obtained from real-time monitoring specifically includes: The real-time acquisition unit continuously acquires performance indicators and service response quality data of newly expanded resource nodes to form real-time node operating status information; the data analysis unit performs comprehensive comparative analysis of the above multi-dimensional indicators to generate preliminary assessment results of node operating stability. If the evaluation result is lower than the preset stability standard, the information processing unit will locate the time period of the anomaly and the related business type, and the traffic analysis unit will perform in-depth analysis of the load distribution characteristics of the corresponding resource nodes. If the analysis results indicate that the node load exceeds the preset threshold, the traffic scheduling system will be used to divert and schedule the service traffic of that node to determine the traffic distribution pattern after diversion. After the traffic split is completed, the changes in node performance indicators are continuously tracked. The entire process of the traffic split operation is retained through the log recording unit to form a complete operation trajectory. The scheduling logs are classified and archived through the data storage system to build a historical dataset for subsequent analysis.

9. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The process of adjusting the traffic migration ratio or performing a full traffic migration specifically includes: The status monitoring module continuously collects information on the operating status and service capabilities of resource nodes to form a record of node health status; the information processing unit integrates and analyzes multi-dimensional operating data to determine whether the current service capabilities of the node meet the preset standards. If the node's service capacity does not meet expectations, the load prediction data for that node is obtained through the load prediction system to identify potential traffic overload risks; the reliability of the prediction results is evaluated by the confidence analysis unit, and the prediction fluctuation characteristics are extracted to form a basis for traffic migration and adjustment. Based on the aforementioned adjustment criteria, the traffic migration ratio of each node in the next scheduling cycle is dynamically calculated by the scheduling cycle management unit to form a traffic allocation scheme. If the traffic migration ratio exceeds the preset safety threshold, the traffic allocation system will perform step-by-step migration to achieve node load balancing. The data update unit records the changes in node load status and operating indicators before and after migration, generates a complete scheduling cycle execution log, and archives and stores it.

10. The dynamic resource scheduling method for website load prediction according to claim 1, characterized in that, The process of pausing or rolling back the traffic migration operation and feeding back the abnormal status information of the node to the load prediction model specifically includes: The status monitoring module obtains the real-time operating indicators of the newly expanded resource nodes and compares them with the preset health threshold. If the operating indicators are lower than the health threshold, the node is determined to be in an abnormal operating state. Based on the abnormal state, a traffic migration interruption command is generated, and the scheduling module stops allocating new traffic to the node. Collect load time-series data and performance time-series data within a preset time window of abnormal nodes, and use a time series decomposition algorithm to decompose the load series into trend components and residual components to extract abnormal fluctuation characteristics. The abnormal fluctuation characteristics and corresponding performance sequences are used as model correction samples to update the internal feature weight parameters of the load prediction model. Based on the updated load prediction model, the load prediction values ​​and risk levels for subsequent scheduling cycles are recalculated to form the basis for the corrected scheduling decisions.

Images