A real-time budget scheduling method of an advertisement delivery system

Abstract

This invention discloses a real-time budget scheduling method for an advertising delivery system, relating to the field of dynamic budget management technology. The method includes: collecting real-time delivery traffic data; comparing the current traffic intensity with historical baseline data and time period data to generate a traffic anomaly intensity identifier and recording the corresponding timestamp; when the traffic anomaly intensity identifier meets preset conditions, performing latency detection, packet loss detection, and numerical mutation detection on the delivery feedback data to generate feedback reliability weight parameters; simultaneously obtaining the state update time of the delivery strategy and calculating the difference with the feedback reception time to obtain a lag parameter. This invention, by introducing a traffic anomaly intensity identifier and a multi-dimensional feedback detection mechanism, can achieve dynamic perception and evaluation of the quality and response lag of feedback data; under the premise that a traffic anomaly is triggered, it actively executes feedback detection operations and constructs a lag parameter by combining the strategy state response time difference.

Description

Technical Field

[0001] This invention relates to the field of dynamic budget management technology, specifically a real-time budget scheduling method for an advertising delivery system. Background Technology

[0002] In user-facing digital advertising platforms, content distribution platforms, and other real-time strategy systems, the budget allocation mechanism is usually a core part of the campaign scheduling process. Traditional budget allocation methods are mostly based on preset strategies or periodic adjustment mechanisms to complete budget delivery. Although they have a certain degree of stability, in real-time, multi-strategy concurrent competition, and significant traffic fluctuations, fixed budget models often cannot adapt to dynamically changing campaign needs. To improve the accuracy of strategy execution and the efficiency of budget utilization, some existing systems have introduced mechanisms for dynamic strategy evaluation based on feedback signals, such as dynamically adjusting strategy ranking based on metrics like click-through rate and conversion rate. However, in actual operation, feedback is often affected by factors such as network latency, data packet loss, or incomplete feedback collection, which may cause the strategy evaluation results to deviate from actual performance, thus leading to resource misallocation problems. Existing technical solutions typically lack sophisticated identification and control mechanisms to address feedback lag and abnormal traffic surges. When the data upon which the strategy relies is delayed or distorted, the system may still directly adjust the budget or ranking based on the current feedback results, potentially creating a positive feedback amplification effect that affects the fairness and effectiveness of overall resource allocation. Therefore, in high real-time, multi-feedback path systems, how to achieve dynamic correction and budget allocation control oriented towards feedback quality and response timeliness has become a problem worthy of in-depth research. Consequently, this invention proposes a real-time budget scheduling method for advertising delivery systems. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time budget scheduling method for an advertising delivery system to solve the problems mentioned in the background section.

[0004] This invention can be achieved through the following technical solution: a real-time budget scheduling method for an advertising delivery system, comprising: Step 1: Collect real-time traffic data, compare the current traffic intensity with historical baseline data and time period data, generate traffic anomaly intensity identifiers, record the corresponding timestamps, and store the traffic anomaly intensity identifiers as historical anomaly intensity identifiers in the storage unit. Step 2: When the traffic anomaly intensity indicator meets the preset conditions, perform latency detection, packet loss detection and numerical change detection on the delivery feedback data, generate feedback credibility weight parameters, and at the same time obtain the status update time of the delivery strategy and calculate the difference with the feedback reception time to obtain the lag parameter. Step 3: Based on the feedback credibility weight parameter and lag parameter, the historical performance data of the delivery strategy are jointly corrected to generate strategy effectiveness parameters with multiple constraints. Step 4: Recalculate the priority parameters of each delivery strategy based on the strategy effectiveness parameters, generate the strategy ranking results, update the execution order of each delivery strategy according to the strategy ranking results, and set the ranking adjustment parameters according to the lag parameter. Step 5: Generate the budget allocation limit for each delivery strategy based on the strategy sorting results, and limit the budget delivery amount of each delivery strategy according to the budget allocation limit. The budget control data is written to the budget scheduling channel through the main write path, and the abnormal status information is written asynchronously through the auxiliary write path. The data flow of the main write path and the auxiliary write path is kept unidirectionally separated. Step 6: Collect the actual execution data after budget allocation, and write it into the traffic anomaly intensity identification data area and the feedback credibility weight parameter data area respectively using an asymmetric writing method. The main writing path receives the master control feedback information, and the auxiliary writing path processes the abnormal status feedback information.

[0005] A further technical improvement of the present invention is that the step of generating the traffic anomaly intensity identifier in step one includes: Before collecting real-time traffic data in the current cycle, read the traffic anomaly intensity identifier generated in the previous cycle from the preset storage unit as the historical anomaly intensity value, and read the generation timestamp corresponding to the historical anomaly intensity identifier. The real-time delivery traffic data collected in the current period is processed to extract the current traffic intensity. The traffic intensity is then compared with the preset historical baseline data and time period data to calculate the current abnormal intensity value. The historical anomaly intensity value is multiplied by an attenuation coefficient, which is a real number less than 1. This attenuation coefficient is used to control the influence of historical anomalies on the current judgment. The attenuated historical anomaly intensity value is then added to the current anomaly intensity calculated value to obtain the cumulative anomaly intensity value. The cumulative value of abnormal intensity is used as the traffic abnormal intensity identifier for the current period, and the generation timestamp of the abnormal intensity identifier is recorded. The traffic abnormal intensity identifier along with the corresponding timestamp is written into the storage unit for use in subsequent periods to retrieve historical abnormal intensity values ​​and perform time-series calculations.

[0006] A further technical improvement of the present invention is that: the determination of whether the flow anomaly intensity indicator meets the preset conditions in step two includes: Set multiple abnormal trigger thresholds with different dimensions, including fluctuation intensity thresholds based on historical traffic variance, offset thresholds based on deviation from business objectives, and periodic abnormal thresholds based on time period stability. The current period's abnormal traffic intensity indicator is compared with the fluctuation intensity threshold, offset threshold, and period abnormal threshold to determine whether there are at least two dimensions of abnormal intensity exceeding the corresponding threshold. When at least two abnormal triggering conditions are met simultaneously, the traffic abnormality intensity indicator for the current period is determined to meet the preset conditions, and the detection process for delivery and feedback data is triggered.

[0007] A further technical improvement of the present invention is that the step of generating the feedback reliability weight parameter and the hysteresis parameter in step two includes: When the traffic anomaly intensity indicator meets the preset conditions, feedback delay detection, feedback packet loss detection and numerical change detection are performed on the delivery and feedback data in the current period. Feedback delay detection includes calculating the time difference between the reception time of the feedback data and the corresponding request sending time. Feedback packet loss detection includes the ratio between the number of requests that did not receive feedback within a unit time and the total number of requests. Numerical change detection includes calculating the change range of feedback data in adjacent statistical periods. Based on the feedback delay detection results, feedback packet loss detection results, and numerical change detection results, the feedback delay value, feedback packet loss rate, and change feedback ratio are determined respectively. The feedback delay value, feedback packet loss rate, and change feedback ratio are then weighted and superimposed according to a preset weight ratio to generate the feedback reliability weight parameter. Obtain the status update time of each delivery strategy in the current period, and calculate the difference between it and the reception time of the corresponding data to obtain the status response time difference of each delivery strategy. The time difference of the state response is smoothed to generate the hysteresis parameter.

[0008] A further technical improvement of the present invention is that step three, generating strategy effectiveness parameters, includes: Collect historical performance data for each campaign strategy within a preset historical period. The historical performance data includes one or more of the following metrics: conversion rate, click-through rate, user dwell time, exposure completion rate, and bounce rate. Normalize each metric separately and count the number of metrics involved in the normalization process to generate an evaluation completeness parameter. Based on the evaluation completeness parameter, an interval matching operation is performed in the preset completeness correction coefficient mapping relationship to obtain the corresponding completeness correction coefficient, and the original effect benchmark value is multiplied with the completeness correction coefficient to generate the completeness correction benchmark value; Read the feedback credibility weight parameter generated in the current period, and perform interval matching operation in the preset credibility correction coefficient mapping relationship according to the value range of the parameter to obtain the corresponding credibility correction coefficient. Multiply the integrity correction benchmark value and the credibility correction coefficient to generate the first correction result value. Read the hysteresis parameter generated in the current period, and according to the response delay interval in which the hysteresis parameter is located, perform interval matching operation in the preset hysteresis attenuation coefficient mapping relationship to obtain the corresponding hysteresis attenuation coefficient. The hysteresis attenuation coefficient and the hysteresis parameter have a monotonically decreasing relationship. For every preset interval segment that the hysteresis parameter increases, its corresponding hysteresis attenuation coefficient decreases to the next set level. The first correction result value is multiplied by the hysteresis decay coefficient to generate the second correction result value, and the second correction result value is used as the strategy effectiveness parameter for multiple constraint processing in the current period. Write the strategy effectiveness parameters into the effect parameter recording area as input parameters for calculating the priority of the delivery strategy in step four.

[0009] A further technical improvement of the present invention is that, when multiple indicators are included in the historical performance data, normalization is performed on each indicator separately, and the indicators are fused according to a preset weighting ratio to generate the original performance benchmark value. When historical performance data includes only a single indicator, the normalized value of that single indicator is used as the original performance baseline value after normalization.

[0010] A further technical improvement of the present invention is that the step of setting the sorting adjustment parameters based on the hysteresis parameter in step four includes: The lag parameter is divided into multiple lag level intervals, including the first level interval, the second level interval, and the third level interval; The first level range corresponds to the lag parameter being less than the first preset value, the second level range corresponds to the lag parameter being between the first preset value and the second preset value, and the third level range corresponds to the lag parameter being greater than the second preset value. A priority change adjustment coefficient is set for each lag level interval. The priority change adjustment coefficient is used to limit the change of the priority parameters of each deployment strategy in the current period compared with the previous period. The higher the lag level, the smaller the corresponding priority change adjustment coefficient. Obtain the lag parameter of each delivery strategy in the current period and determine its lag level range; Based on the priority change adjustment coefficient corresponding to the respective lag level interval, the priority parameters of each delivery strategy are subject to amplitude limitation.

[0011] A further technical improvement of the present invention is that step five, which limits the budget allocation based on the strategy ranking results of each allocation strategy, includes: Calculate the budget allocation coefficient for each delivery strategy. The budget allocation coefficient is generated by the priority parameter of the corresponding strategy in the strategy ranking result and the preset mapping relationship. The budget allocation cap for each campaign strategy is calculated based on the budget allocation coefficient and the total available budget in the current budget pool. Convert the budget allocation cap into the corresponding budget control instructions; Based on the feedback credibility weight parameter and lag parameter of each delivery strategy in the current period, determine whether the delivery strategy is in an abnormal state. If the abnormal state judgment condition is met, mark the strategy as an abnormal strategy. The generated budget control instructions are processed by splitting the flow. Budget control instructions corresponding to normal policies are written to the budget scheduling channel through the main write path, while budget control instructions corresponding to abnormal policies are written to the abnormal record area through the auxiliary write path. The primary write path and the secondary write path are physically isolated from each other, and the data flow is kept unidirectional to prevent abnormal data from flowing into the budget scheduling channel.

[0012] A further technical improvement of the present invention lies in: the process of asymmetric writing of the actual execution data in step six includes: Collect actual execution data for each campaign strategy within the current period. The actual execution data should include at least the budget usage, the number of request triggers, the feedback reception time, and the feedback credibility weight parameter. Extract traffic-related data fields from the actual execution data and construct traffic statistics records. Write the traffic statistics records to the traffic anomaly intensity indicator data area through the main write path or auxiliary write path according to the anomaly indicators of the corresponding strategy. Extract the return quality-related data fields from the actual execution data and form a return quality record. Write the return quality record to the return credibility weight parameter data area through the main write path or auxiliary write path according to the anomaly identifier of the corresponding strategy. The main write path adopts a synchronous write mechanism, while the auxiliary write path adopts an asynchronous write mechanism based on a cache queue. The main write path and the auxiliary write path maintain a unidirectional data isolation structure.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention introduces a traffic anomaly intensity identifier and a multi-dimensional feedback detection mechanism, enabling dynamic perception and evaluation of the quality and response lag of feedback data. Under the premise that a traffic anomaly is triggered, the feedback detection operation is actively executed, and a lag parameter is constructed by combining the policy state response time difference, providing a more reliable data foundation for subsequent policy ranking and budget allocation. Furthermore, this invention introduces three indicators—evaluation completeness parameter, feedback credibility weight parameter, and lag parameter—to jointly correct historical performance data and construct a multi-dimensional, multi-stage integrated strategy effectiveness parameter generation process. Compared to methods that evaluate strategy effectiveness based solely on current feedback or a single indicator, this invention provides a comprehensive decision-making basis that takes into account data completeness, feedback reliability, and response timeliness. On the other hand, by constructing a main write path and an auxiliary write path while maintaining physical isolation and unidirectional data flow, this invention can execute budget control instructions for normal and abnormal policies according to their states, ensuring the high availability and stability of the budget scheduling channel. At the same time, based on the asymmetric write structure of actual execution data, it further enhances the ability to record and asynchronously learn full policy feedback data, realizing a highly coupled and highly stable collaborative closed loop between policy evaluation, sorting, and budget control. Attached Figure Description

[0014] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0017] Please see Figure 1 As shown, the present invention provides a real-time budget scheduling method for an advertising delivery system, comprising: Step 1: Collect real-time delivery traffic data, compare the current traffic intensity with historical baseline data and time period data, generate a traffic anomaly intensity identifier, record the corresponding timestamp, and store the traffic anomaly intensity identifier as a historical anomaly intensity identifier in the storage unit. This step is used to achieve dynamic perception of traffic changes in the current delivery environment. By comparing the current traffic intensity with historical baseline data and time period data, abnormal traffic fluctuation trends are identified. The generated traffic anomaly intensity identifier serves as an important input signal for subsequent judgments on the reliability of feedback data and whether diagnostic processing needs to be triggered. The recorded timestamp is used for subsequent time-series judgment and status tracking in multi-period calculations.

[0018] The steps in step one to generate traffic anomaly intensity indicators include: Before collecting real-time delivery traffic data in the current period, the traffic anomaly intensity identifier generated in the previous period is read from the preset storage unit as the historical anomaly intensity value, and the generation timestamp corresponding to the historical anomaly intensity identifier is read; the real-time delivery traffic data collected in the current period is processed to extract the current traffic intensity, and the traffic intensity is compared with the preset historical baseline data and time period data respectively to calculate the current anomaly intensity value. The historical anomaly intensity value is multiplied by an attenuation coefficient, which is a real number less than 1. This attenuation coefficient is used to control the influence of historical anomalies on the current judgment. The attenuated historical anomaly intensity value is then added to the current anomaly intensity calculated value to obtain the cumulative anomaly intensity value. The cumulative value of abnormal intensity is used as the traffic abnormal intensity identifier for the current period, and the generation timestamp of the abnormal intensity identifier is recorded. The traffic abnormal intensity identifier along with the corresponding timestamp is written into the storage unit for use in subsequent periods to retrieve historical abnormal intensity values ​​and perform time-series calculations.

[0019] Specifically, before collecting real-time traffic data for the current period, the system reads the traffic anomaly intensity flag generated in the previous period from the preset storage unit, using this value as the historical anomaly intensity value, and simultaneously reads the generation timestamp corresponding to this historical anomaly intensity flag. If this is the first time this method is run, and the traffic anomaly intensity flag from the previous period does not yet exist in the storage unit, the default initial anomaly intensity value in the system initialization configuration is used by default. For example, it can be set to 0.5 as the balancing baseline value, indicating that the initial traffic state is neutral. The generation timestamp can use the current system time or the configured time as the initial time point.

[0020] The system processes the real-time delivery traffic data collected in the current period to extract the current traffic intensity within that period. In practice, traffic-related metrics such as the total number of requests, effective exposures, and number of user dwell times collected per unit time can be fused together to obtain a current traffic intensity value that characterizes the current traffic status of the delivery environment.

[0021] Because the above-mentioned indicators differ in physical meaning and numerical range, they are first normalized before merging and calculating. Specifically, based on the maximum value or historical benchmark value of each indicator within a preset statistical period, the exposure volume, request volume, and number of user dwell times are normalized to convert each indicator into dimensionless comparable values.

[0022] After normalization, the metrics are weighted and merged according to preset weights to generate the traffic intensity value for the current period. Specifically, the weight coefficient for impressions can be set to 0.4, the weight coefficient for requests can be set to 0.3, and the weight coefficient for user dwell time can be set to 0.3. These weight coefficients can be obtained through statistical analysis or regression calculation of historical campaign log data, reflecting the importance of each metric in characterizing system traffic load. Under different application scenarios, these weight coefficients can also be adjusted based on historical data analysis results.

[0023] The current traffic intensity obtained through the above method is a comprehensive indicator after unified processing, used to characterize the overall access load level of the advertising delivery system per unit time.

[0024] Subsequently, the current traffic intensity is compared with preset historical baseline data and time period data. The historical baseline data can be the average traffic intensity for the same time period within the preset period, while the time period data can be a periodic adjustment parameter reflecting the traffic variation pattern over different time periods, such as the traffic adjustment coefficient during holidays or specific events. By calculating the difference between the current traffic intensity and the above data and then performing normalization, the calculated value of the current abnormal intensity for that period can be obtained.

[0025] The previously obtained historical anomaly intensity value is multiplied by a preset attenuation coefficient, which is a real number less than 1, for example, set to 0.8, to control the influence of historical anomalies on the current judgment. This attenuation process ensures that recent anomalies have a greater impact while older anomalies gradually fade out. In this embodiment, if the anomaly intensity value of the previous period was 0.6, then after attenuation it becomes: The attenuated anomaly intensity = 0.8 × 0.6 = 0.48; Add this value to the current calculated anomaly intensity value to obtain the cumulative anomaly intensity value for this period, i.e.: Accumulated anomaly intensity value = current calculated anomaly intensity value + attenuated anomaly intensity value; The accumulated value of the anomaly intensity is used as the traffic anomaly intensity identifier for the current period, and its generation timestamp is recorded (the current data generation time can be used). The traffic anomaly intensity identifier along with the corresponding timestamp is written to a preset storage unit for subsequent periods to continue reading and participate in the time-series superposition calculation of historical anomalies. Through continuous updates, the system can generate a continuous sequence of traffic anomaly intensity identifiers periodically, forming a traffic anomaly tracking mechanism with decay memory.

[0026] Step 2: When the traffic anomaly intensity indicator meets preset conditions, delay detection, packet loss detection, and numerical mutation detection are performed on the delivery feedback data to generate feedback reliability weight parameters. Simultaneously, the status update time of the delivery strategy is obtained and its difference with the feedback reception time is calculated to obtain the lag parameter. This step is a feedback quality diagnosis process initiated when traffic anomalies are detected. It uses multiple detection methods to identify the stability of the feedback link and the reliability of the data. The generated feedback reliability weight parameters are used to measure the overall reliability of the feedback data, while the lag parameter reflects the response delay between the strategy status update and feedback reception, used to identify potential strategy deviations caused by lag. Together, they form the basis for quantitative measurement of feedback quality in strategy evaluation.

[0027] Step two determines whether the abnormal flow intensity indicator meets preset conditions, including: Set multiple abnormal trigger thresholds with different dimensions, including fluctuation intensity thresholds based on historical traffic variance, offset thresholds based on deviation from business objectives, and periodic abnormal thresholds based on time period stability. The current period's abnormal traffic intensity indicator is compared with the fluctuation intensity threshold, offset threshold, and period abnormal threshold to determine whether there are at least two dimensions of abnormal intensity exceeding the corresponding threshold. When at least two abnormal triggering conditions are met simultaneously, the traffic abnormality intensity indicator for the current period is determined to meet the preset conditions, and the detection process for delivery and feedback data is triggered.

[0028] Specifically, multiple anomaly trigger thresholds with different dimensions are set to cover various manifestations of traffic fluctuations. In this embodiment, the multiple anomaly trigger thresholds include: (1) A fluctuation intensity threshold based on historical flow variance is used to identify whether the flow anomaly intensity indicator in the current period deviates significantly from the historical normal fluctuation range. This threshold can be obtained by statistically analyzing the standard deviation of the flow anomaly intensity indicators in the previous 7 periods and multiplying it by an amplification factor, and is used as the boundary for judging the fluctuation intensity in the current period; (2) Based on the deviation of business objectives, the degree of abnormal deviation of business objective indicators such as conversion rate and click rate is first standardized to the indicator strength value with the same dimension as the traffic abnormality intensity indicator. Then, the deviation threshold is set to identify whether the current traffic deviates significantly from the business objective. For example, the degree of deviation of business objectives can be mapped to the judgment threshold of abnormal intensity value 1.6. (3) Based on the stability of time period, the periodic anomaly threshold is set by constructing a periodic baseline by analyzing the historical flow performance within a specific time period (such as nighttime, midday peak, etc.) and setting a judgment boundary after standardizing the current flow performance and the deviation of the periodic baseline. For example, if the abnormal flow during the peak period increases by more than 50% of the historical average, it can be mapped to a periodic anomaly intensity value of 1.4.

[0029] The current period's traffic anomaly intensity flag is compared with the fluctuation intensity threshold, offset threshold, and period anomaly threshold, respectively. In this embodiment, it is assumed that the traffic anomaly intensity flag generated in the current period is 1.8, and the three thresholds set by the system are the fluctuation intensity threshold of 1.5, the offset threshold of 1.6, and the period anomaly threshold of 1.4. During the comparison, it is found that the current traffic anomaly intensity flag exceeds both the fluctuation intensity threshold and the period anomaly threshold, thus meeting the trigger conditions for both dimensions.

[0030] Determine if at least two dimensions of abnormal intensity exceed the corresponding threshold. Here, "multiple dimensions" refers to three independent calculation methods: fluctuation, offset, and period, not the time dimension. The trigger judgment does not require the values ​​to be consecutive or the dimensions to be adjacent. As long as the current traffic abnormal intensity indicator simultaneously meets the threshold exceeding condition for any two dimensions, the traffic abnormal intensity indicator for that period can be considered to meet the preset trigger standard.

[0031] Finally, if the above triggering criteria are met, the subsequent delivery feedback data detection process will be initiated immediately. This process is used to detect and analyze multiple dimensions of the feedback data in the current period, such as feedback latency, feedback packet loss, and numerical mutations. It will also generate feedback credibility weight parameters and lag parameters to provide data support for the evaluation and ranking of the effectiveness of the delivery strategy in subsequent steps.

[0032] Step two, which generates the feedback reliability weight parameters and hysteresis parameters, includes: When the traffic anomaly intensity indicator meets the preset conditions, feedback delay detection, feedback packet loss detection and numerical change detection are performed on the delivery and feedback data in the current period. Feedback delay detection includes calculating the time difference between the reception time of the feedback data and the corresponding request sending time. Feedback packet loss detection includes the ratio between the number of requests that did not receive feedback within a unit time and the total number of requests. Numerical change detection includes calculating the change range of feedback data in adjacent statistical periods. Based on the feedback delay detection results, feedback packet loss detection results, and numerical change detection results, the feedback delay value, feedback packet loss rate, and change feedback ratio are determined respectively. The feedback delay value, feedback packet loss rate, and change feedback ratio are then weighted and superimposed according to a preset weight ratio to generate the feedback reliability weight parameter. Obtain the status update time of each delivery strategy in the current period, and calculate the difference between it and the reception time of the corresponding data to obtain the status response time difference of each delivery strategy. The time difference of the state response is smoothed to generate the hysteresis parameter.

[0033] Specifically, feedback latency detection, feedback packet loss detection, and numerical change detection are performed on the delivery and feedback data within the current period. Feedback latency detection is calculated by determining the time difference between the reception time of each piece of feedback data and the corresponding request sending time. For example, if the request sending time is 100 milliseconds and the feedback data reception time is 150 milliseconds, the feedback latency is 50 milliseconds. Feedback packet loss detection is calculated by the ratio of the number of requests that did not receive feedback within a unit of time to the total number of requests. For example, if 1,000 requests are sent in a statistical period and 950 requests receive feedback, the feedback packet loss rate is 0.05. Numerical change detection is calculated by determining the magnitude of change in feedback data between adjacent statistical periods. For example, if the conversion volume in the previous period was 100 and the conversion volume in this period is 150, the change feedback ratio is 0.5.

[0034] Based on the feedback delay detection results, feedback packet loss detection results, and numerical mutation detection results, the feedback delay value, feedback packet loss rate, and mutation feedback ratio are calculated respectively, and then converted into dimensionless score values ​​with unified dimensions. Specifically, the feedback delay value is mapped to a preset maximum tolerable delay interval. When the feedback delay is less than or equal to the minimum tolerable delay, the corresponding dimensionless value is zero; when the feedback delay is greater than or equal to the maximum tolerable delay, the corresponding dimensionless value is one; and when it is between the two, a linear mapping is performed according to the ratio. The feedback packet loss rate is directly used as the dimensionless score value; the mutation feedback ratio is converted into a dimensionless score value according to a preset stability grading rule. The three converted dimensionless score values ​​are weighted and superimposed according to a preset weight ratio to generate the feedback reliability weight parameter for the current period.

[0035] Obtain the status update time of each delivery strategy in the current period, and calculate the difference between it and the corresponding data reception time to obtain the status response time difference for that delivery strategy. For example, if the status update time of a certain delivery strategy is 100 milliseconds after the start of the current period, and its data reception time is 200 milliseconds after the start of the current period, then the status response time difference is 100 milliseconds.

[0036] The state response time difference is smoothed to generate a hysteresis parameter. In this embodiment, the state response time difference over multiple consecutive periods is averaged to obtain a smoothed time difference. This smoothed time difference is then converted into a dimensionless parameter based on a preset maximum tolerable delay interval: when the smoothed time difference is less than the minimum tolerable delay, the corresponding hysteresis parameter is set to zero; when it is greater than the maximum tolerable delay, the corresponding hysteresis parameter is set to one; and when it falls between these two values, it is mapped proportionally to obtain a dimensionless hysteresis parameter that can participate in the subsequent calculation of strategy effectiveness parameter correction.

[0037] Step 3: Based on the feedback credibility weight parameter and lag parameter, the historical performance data of the deployment strategy is jointly corrected to generate strategy effectiveness parameters with multiple constraints. This step is used to improve the quality of the original historical performance data. By introducing the feedback credibility weight parameter and lag parameter, a dual constraint model is constructed. The data is weighted from the perspectives of the reliability and timeliness of the feedback signal, thereby generating more stable and realistic strategy effectiveness parameters, providing an accurate and interference-resistant data foundation for subsequent priority calculation.

[0038] Step three, generating policy effectiveness parameters, includes: Historical performance data for each campaign strategy is collected within a preset historical period. This data includes one or more metrics such as conversion rate, click-through rate, user dwell time, exposure completion rate, and bounce rate. Each metric is normalized separately, and the number of metrics involved in the normalization process is counted to generate an evaluation completeness parameter. When multiple metrics are included in the historical performance data, each metric is normalized separately, and the metrics are then merged according to a preset weighting ratio to generate an original performance baseline value. When the historical performance data includes only a single metric, the normalized value of that single metric is used as the original performance baseline value.

[0039] Based on the evaluation completeness parameter, an interval matching operation is performed in the preset completeness correction coefficient mapping relationship to obtain the corresponding completeness correction coefficient, and the original effect benchmark value is multiplied with the completeness correction coefficient to generate the completeness correction benchmark value; Read the feedback credibility weight parameter generated in the current period, and perform interval matching operation in the preset credibility correction coefficient mapping relationship according to the value range of the parameter to obtain the corresponding credibility correction coefficient. Multiply the integrity correction benchmark value and the credibility correction coefficient to generate the first correction result value. Read the hysteresis parameter generated in the current period, and according to the response delay interval in which the hysteresis parameter is located, perform interval matching operation in the preset hysteresis attenuation coefficient mapping relationship to obtain the corresponding hysteresis attenuation coefficient. The hysteresis attenuation coefficient and the hysteresis parameter have a monotonically decreasing relationship. For every preset interval segment that the hysteresis parameter increases, its corresponding hysteresis attenuation coefficient decreases to the next set level. The first correction result value is multiplied by the hysteresis decay coefficient to generate the second correction result value, and the second correction result value is used as the strategy effectiveness parameter for multiple constraint processing in the current period. Write the strategy effectiveness parameters into the effect parameter recording area as input parameters for calculating the priority of the delivery strategy in step four.

[0040] Specifically, historical performance data for each campaign strategy is collected within a preset historical period. This historical performance data includes one or more metrics such as conversion rate, click-through rate, user dwell time, exposure completion rate, and bounce rate. Normalization is performed on each metric involved in the calculation, and the number of metrics involved in the normalization process is counted to generate an evaluation completeness parameter. For example, if a campaign strategy collects three metrics—conversion rate, click-through rate, and exposure completion rate—within a historical period, the evaluation completeness parameter is 3.

[0041] When historical performance data includes multiple metrics, the normalized results of each metric are merged according to a preset weighting ratio to generate the original performance baseline value. When historical performance data includes only a single metric, the normalized result of that single metric is used as the original performance baseline value. For example, if the normalized values ​​of conversion rate, click-through rate, and impression completion rate are 0.65, 0.72, and 0.84, respectively, and the preset weighting ratios are 0.4, 0.3, and 0.3, then the original performance baseline value is: 0.65×0.4+0.72×0.3+0.84×0.3=0.728.

[0042] Based on the mapping relationship between the evaluation completeness parameter and the preset completeness correction coefficient, the correction coefficient is designed using a linear or piecewise function based on the numerical range of the evaluation completeness parameter (e.g., 1 to 5) to ensure a positive correlation between the correction coefficient and the completeness parameter. The mapping relationship follows a monotonically increasing principle to ensure that the higher the data completeness, the larger the correction coefficient. For example, when the evaluation completeness parameter is 1, the corresponding completeness correction coefficient is 0.6; when the parameter is 3, the corresponding completeness correction coefficient is 0.9; and when the parameter is 5, the corresponding completeness correction coefficient is 1.0. Multiplying the original effect baseline value of 0.728 by the completeness correction coefficient of 0.9 yields the completeness correction baseline value: 0.728 × 0.9 = 0.6552.

[0043] The system reads the feedback credibility weight parameter generated in the current period and performs interval matching in the preset credibility correction coefficient mapping relationship based on the numerical range of the feedback credibility weight parameter to obtain the credibility correction coefficient. For example, when the feedback credibility weight parameter is 0.3, the matched credibility correction coefficient is 0.8. Next, the system reads the hysteresis parameter generated in the current period. The hysteresis parameter is a dimensionless parameter, for example, 0.6. Based on this parameter, the system matches the hysteresis attenuation coefficient mapping relationship to obtain a hysteresis attenuation coefficient of 0.7. Finally, the system multiplies the integrity correction benchmark value of 0.6552 with the credibility correction coefficient of 0.8 to obtain the first correction result value. 0.6552 × 0.8 = 0.52416; Multiplying the first corrected value of 0.52416 by the hysteresis decay coefficient of 0.7 yields the final strategy effectiveness parameters: 0.52416 × 0.7 = 0.36691.

[0044] The results demonstrate that the system made significant conservative corrections to data with low confidence (0.3 → correction coefficient 0.8) and high lag (0.6 → correction coefficient 0.7). The final strategy effectiveness parameter is 0.36691, showing the system's ability to suppress low-quality feedback data. Through the correction coefficients, the impact of low-confidence data was significantly reduced, while high-lag data was weakened by the system through attenuation factors, thus making the final strategy effectiveness parameter more consistent with the actual feedback situation.

[0045] Step 4: Recalculate the priority parameters of each delivery strategy based on the strategy effectiveness parameters, generate a strategy ranking result, update the execution order of each delivery strategy according to the strategy ranking result, and set the ranking adjustment parameters based on the lag parameter. This step ranks and calculates each delivery strategy based on the strategy performance reflected by the strategy effectiveness parameters and adjusts the execution priority order of the strategies. On this basis, the lag parameter is introduced to set the ranking adjustment parameters to constrain drastic fluctuations in ranking caused by data lag, ensure the stability of ranking changes, and improve the robustness of the system to feedback anomalies.

[0046] Step four, which involves setting the sorting adjustment parameters based on the lag parameter, includes: The lag parameter is divided into multiple lag level intervals, including the first level interval, the second level interval, and the third level interval; The first level range corresponds to the lag parameter being less than the first preset value, the second level range corresponds to the lag parameter being between the first preset value and the second preset value, and the third level range corresponds to the lag parameter being greater than the second preset value. A priority change adjustment coefficient is set for each lag level interval. The priority change adjustment coefficient is used to limit the change of the priority parameters of each deployment strategy in the current period compared with the previous period. The higher the lag level, the smaller the corresponding priority change adjustment coefficient. Obtain the lag parameter of each delivery strategy in the current period and determine its lag level range; Based on the priority change adjustment coefficient corresponding to the respective lag level interval, the priority parameters of each delivery strategy are subject to amplitude limitation.

[0047] Specifically, multiple lag level intervals are set for different numerical ranges of the lag parameter to classify and evaluate the timeliness of the state response of each deployment strategy in the current period. In this embodiment, the lag parameter is divided into three level intervals: the first level interval corresponds to a lag parameter less than a first preset value (e.g., the first preset value is 0.3), the second level interval corresponds to a lag parameter between the first and second preset values ​​(e.g., the first preset value is 0.3 and the second preset value is 0.6), and the third level interval corresponds to a lag parameter greater than the second preset value. The lag parameter is a dimensionless value, ranging from 0 to 1, where a larger value indicates a more delayed response.

[0048] Subsequently, a priority change adjustment coefficient is set for each lag level interval to control the adjustment rate of the strategy priority and prevent drastic fluctuations in priority parameter updates when feedback data lags significantly. In this embodiment, the priority change adjustment coefficient is set to 1.0 for the first level interval, 0.6 for the second level interval, and 0.3 for the third level interval. The smaller the adjustment coefficient, the more cautious the priority change of the strategy should be within the current period.

[0049] Secondly, obtain the lag parameter of each delivery strategy in the current period, and determine its lag level range based on the value of this parameter. Then, combine the priority parameter of the strategy in the previous period with the target priority parameter calculated based on the strategy effectiveness parameter in the current period, and calculate the final priority parameter of the current period using a smooth adjustment method. Specifically, the difference between the target priority parameter and the priority parameter of the previous period is used as the adjustment range, and multiplied by the priority change adjustment coefficient corresponding to the level range to obtain the restricted adjustment value. The final priority parameter is determined by adding this restricted adjustment value to the priority parameter of the previous period. For example, if the priority parameter of a delivery strategy in the previous period was 0.7 and the target priority parameter was 0.9, and the current lag parameter of the strategy is 0.75, belonging to the third level range, with a corresponding adjustment coefficient of 0.3, then the priority parameter of the current period is 0.7 + (0.9 - 0.7) × 0.3 = 0.76.

[0050] By adjusting the calculation methods described above, while ensuring accurate evaluation of strategy performance based on strategy effectiveness parameters, abnormal fluctuations in priority caused by feedback lag can be effectively controlled, thereby improving the stability and robustness of ranking decisions. All processed priority parameters will serve as the direct basis for budget allocation order and will be used in subsequent budget cap allocation control processes.

[0051] Step 5: Generate the budget allocation cap for each deployment strategy based on the strategy ranking results, and limit the budget allocation amount of each deployment strategy according to the budget allocation cap. Budget control data is written to the budget scheduling channel through the main write path, while abnormal status information is written asynchronously through the auxiliary write path. The data flow of the main write path and the auxiliary write path remains unidirectional. This step realizes the mapping conversion from strategy ranking results to budget control, and sets a reasonable resource boundary for each strategy by calculating the budget allocation cap. After determining the strategy status, the budget control instructions are written to the main write path or the auxiliary write path according to the results, thereby isolating the budget instructions of normal strategies and abnormal strategies at the data level, ensuring that the budget scheduling channel only receives control instructions that have been verified for reliability, and ensuring the decision purity and execution efficiency of the core loop of the system. This step is used to collect the actual execution data of each strategy after budget deployment is completed, and to identify and sort the data type and source status. By employing an asymmetric writing mechanism, feedback data is written to the corresponding data areas according to the specified paths. The main writing path synchronously updates the feedback for normal strategies, while the auxiliary writing path handles delayed feedback under abnormal conditions. This ensures the integrity of core learning data while enabling immediate isolation of abnormal data, reducing its interference with the real-time decision-making model. This further enhances the integrity and anti-interference capabilities of the data recording system, providing a full and hierarchical data foundation for subsequent feedback modeling and strategy learning.

[0052] Step five involves limiting the budget allocation based on the strategy ranking results of each campaign, including: Calculate the budget allocation coefficient for each delivery strategy. The budget allocation coefficient is generated by the priority parameter of the corresponding strategy in the strategy ranking result and the preset mapping relationship. The budget allocation cap for each campaign strategy is calculated based on the budget allocation coefficient and the total available budget in the current budget pool. Convert the budget allocation cap into the corresponding budget control instructions; Based on the feedback credibility weight parameter and lag parameter of each delivery strategy in the current period, determine whether the delivery strategy is in an abnormal state. If the abnormal state judgment condition is met, mark the strategy as an abnormal strategy. The generated budget control instructions are processed by splitting the flow. Budget control instructions corresponding to normal policies are written to the budget scheduling channel through the main write path, while budget control instructions corresponding to abnormal policies are written to the abnormal record area through the auxiliary write path. The primary write path and the secondary write path are physically isolated from each other, and the data flow is kept unidirectional to prevent abnormal data from flowing into the budget scheduling channel.

[0053] Specifically, based on the priority parameters of each delivery strategy calculated in the previous step, and combined with a set of preset priority mapping relationships, the budget allocation coefficient corresponding to each strategy is determined. The budget allocation coefficient reflects the budget acquisition weight of the delivery strategy in the current period; the higher the priority parameter, the larger the mapped budget allocation coefficient. The mapping relationship can be implemented using a linear function or a piecewise function. For example, in this embodiment, the priority parameter range is set to 0.0 to 1.0, and the budget allocation coefficient is also set accordingly in the range of 0.0 to 1.0, using a linear relationship for mapping.

[0054] Based on the total available budget in the current budget pool, and combined with the budget allocation coefficient for each campaign strategy, the upper limit of budget allocation for that strategy in the current period is calculated. The specific formula is: Upper limit of budget allocation = Total available budget × Budget allocation coefficient. For example, if the total available budget is 10,000 yuan, and the budget allocation coefficient for strategy A is 0.2, then the upper limit of budget allocation for strategy A is 2,000 yuan. This process is repeated for all strategies to allocate their upper limits. Subsequently, the upper limit of budget allocation for each campaign strategy is converted into an executable budget control instruction. This instruction records fields such as the specific strategy identifier, control period, upper limit budget value, and control target, and is used to drive the actual budget execution.

[0055] By combining the feedback credibility weight parameters and lag parameters of each delivery strategy in the current period, anomaly identification is performed. If the feedback credibility weight parameter of a certain strategy is lower than the set credibility lower limit, or the lag parameter is higher than the set lag upper limit, or both conditions are met, then the strategy is determined to be in an abnormal state and marked as an abnormal strategy. For example, if the preset credibility lower limit is 0.3 and the lag upper limit is 0.7, and the feedback credibility weight parameter of strategy B is 0.25 and the lag parameter is 0.8, then strategy B is determined to be an abnormal strategy.

[0056] Finally, based on whether the strategy is in an abnormal state, the generated budget control instructions are processed in a distributed manner. Budget control instructions corresponding to all normal strategies are written to the budget scheduling channel via the main write path for actual budget control execution; while budget control instructions corresponding to all abnormal strategies are written to the abnormal record area via the auxiliary write path for subsequent diagnosis and anomaly analysis. It is important to emphasize that the main write path and the auxiliary write path are physically isolated from each other, and the data flow remains unidirectional to ensure that control data from abnormal strategies does not flow into the budget scheduling channel, preventing abnormal strategies from erroneously occupying budget resources, thereby ensuring the stability of budget execution and system security.

[0057] Step 6: Collect the actual execution data after budget allocation, and write it into the traffic anomaly intensity identification data area and the feedback credibility weight parameter data area respectively using an asymmetric writing method. The main writing path receives the master control feedback information, and the auxiliary writing path processes the abnormal status feedback information.

[0058] Step six involves the asymmetric writing process of the actual execution data, including: Collect actual execution data for each campaign strategy within the current period. The actual execution data should include at least the budget usage, the number of request triggers, the feedback reception time, and the feedback credibility weight parameter. Extract traffic-related data fields from the actual execution data and construct traffic statistics records. Write the traffic statistics records to the traffic anomaly intensity indicator data area through the main write path or auxiliary write path according to the anomaly indicators of the corresponding strategy. Extract the return quality-related data fields from the actual execution data and form a return quality record. Write the return quality record to the return credibility weight parameter data area through the main write path or auxiliary write path according to the anomaly identifier of the corresponding strategy. The main write path adopts a synchronous write mechanism, while the auxiliary write path adopts an asynchronous write mechanism based on a cache queue. The main write path and the auxiliary write path maintain a unidirectional data isolation structure.

[0059] Specifically, the system collects actual execution data for each deployment strategy within the current period. This data includes key fields such as budget usage, request trigger count, feedback reception time, and feedback credibility weight parameters. This data provides foundational information for subsequent strategy evaluation, ensuring the system can make reasonable budget allocations based on the latest execution data.

[0060] Traffic-related data fields are extracted from the actual execution data and compiled into traffic statistics records. These records are routed based on the anomaly identifier of each delivery strategy. If the strategy is identified as a normal strategy, the traffic statistics record is synchronously written to the traffic anomaly intensity identifier data area via the main write path; if the strategy is identified as an anomalous strategy, the traffic statistics record is asynchronously written to the traffic anomaly intensity identifier data area via the auxiliary write path. This design ensures that data from both normal and anomalous strategies can ultimately be aggregated into the same data area, with a delay in writing anomalous data via an auxiliary path to avoid impacting real-time decision-making.

[0061] The return quality-related data fields are extracted from the actual execution data and used to construct return quality records. Similarly, the write path is determined based on the anomaly identifier for each strategy. For normal strategies, return quality records are synchronously written to the return credibility weight parameter data area via the main write path; while for abnormal strategies, return quality records are asynchronously written to the return credibility weight parameter data area via the auxiliary write path. This asymmetric write mechanism ensures that all return data can be stored in the return credibility weight parameter data area, and the writing of abnormal data is delayed to reduce the impact on real-time decision-making.

[0062] Ensure that the primary write path and secondary write path are physically isolated from each other, and that data flow remains unidirectional. This prevents control data for exception policies from interfering with the budget scheduling channel, avoiding the erroneous use of budget resources by exception policies, thus ensuring the rational allocation of budget resources and improving system stability and robustness.

[0063] Through the above processing, the execution data of all deployment strategies can be recorded accurately and in a timely manner, and abnormal data can be effectively isolated and delayed, ensuring that the system's decision-making process is not disturbed and improving the accuracy of the dynamic budget allocation process.

[0064] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A real-time budget scheduling method for an advertising delivery system, characterized in that, include: Step 1: Collect real-time traffic data, compare the current traffic intensity with historical baseline data and time period data, generate traffic anomaly intensity identifiers, record the corresponding timestamps, and store the traffic anomaly intensity identifiers as historical anomaly intensity identifiers in the storage unit. Step 2: When the traffic anomaly intensity indicator meets the preset conditions, perform latency detection, packet loss detection and numerical change detection on the delivery feedback data, generate feedback credibility weight parameters, and at the same time obtain the status update time of the delivery strategy and calculate the difference with the feedback reception time to obtain the lag parameter. Step 3: Based on the feedback credibility weight parameter and lag parameter, the historical performance data of the delivery strategy are jointly corrected to generate strategy effectiveness parameters with multiple constraints. Step 4: Recalculate the priority parameters of each delivery strategy based on the strategy effectiveness parameters, generate the strategy ranking results, update the execution order of each delivery strategy according to the strategy ranking results, and set the ranking adjustment parameters according to the lag parameter. Step 5: Generate the budget allocation limit for each delivery strategy based on the strategy sorting results, and limit the budget delivery amount of each delivery strategy according to the budget allocation limit. The budget control data is written to the budget scheduling channel through the main write path, and the abnormal status information is written asynchronously through the auxiliary write path. The data flow of the main write path and the auxiliary write path is kept unidirectionally separated. Step 6: Collect the actual execution data after budget allocation, and write it into the traffic anomaly intensity identification data area and the feedback credibility weight parameter data area respectively using an asymmetric writing method. The main writing path receives the master control feedback information, and the auxiliary writing path processes the abnormal status feedback information.

2. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, The steps in step one to generate traffic anomaly intensity indicators include: Before collecting real-time delivery traffic data in the current period, the traffic anomaly intensity identifier generated in the previous period is read from the preset storage unit as the historical anomaly intensity value, and the generation timestamp corresponding to the historical anomaly intensity identifier is read; the real-time delivery traffic data collected in the current period is processed to extract the current traffic intensity, and the traffic intensity is compared with the preset historical baseline data and time period data respectively to calculate the current anomaly intensity value. Multiply the historical anomaly intensity value by the attenuation coefficient, and then sum the attenuated historical anomaly intensity value with the current calculated anomaly intensity value to obtain the cumulative anomaly intensity value; The cumulative value of abnormal intensity is used as the traffic abnormal intensity identifier for the current period, and the generation timestamp of the abnormal intensity identifier is recorded. The traffic abnormal intensity identifier along with the corresponding timestamp is written into the storage unit for use in subsequent periods to retrieve historical abnormal intensity values ​​and perform time-series calculations.

3. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, Step two determines whether the abnormal flow intensity indicator meets preset conditions, including: Set multiple abnormal trigger thresholds with different dimensions, including fluctuation intensity thresholds based on historical traffic variance, offset thresholds based on deviation from business objectives, and periodic abnormal thresholds based on time period stability. The current period's abnormal traffic intensity indicator is compared with the fluctuation intensity threshold, offset threshold, and period abnormal threshold to determine whether there are at least two dimensions of abnormal intensity exceeding the corresponding threshold. When at least two abnormal triggering conditions are met simultaneously, the traffic abnormality intensity indicator for the current period is determined to meet the preset conditions, and the detection process for delivery and feedback data is triggered.

4. The real-time budget scheduling method for an advertising delivery system according to claim 3, characterized in that, Step two, which generates the feedback reliability weight parameters and hysteresis parameters, includes: When the traffic anomaly intensity indicator meets the preset conditions, feedback delay detection, feedback packet loss detection and numerical change detection are performed on the delivery and feedback data in the current period. Feedback delay detection includes calculating the time difference between the reception time of the feedback data and the corresponding request sending time. Feedback packet loss detection includes the ratio between the number of requests that did not receive feedback within a unit time and the total number of requests. Numerical change detection includes calculating the change range of feedback data in adjacent statistical periods. Based on the feedback delay detection results, feedback packet loss detection results, and numerical change detection results, the feedback delay value, feedback packet loss rate, and change feedback ratio are determined respectively. The feedback delay value, feedback packet loss rate, and change feedback ratio are then weighted and superimposed according to a preset weight ratio to generate the feedback reliability weight parameter. Obtain the status update time of each delivery strategy in the current period, and calculate the difference between it and the reception time of the corresponding data to obtain the status response time difference of each delivery strategy. The time difference of the state response is smoothed to generate the hysteresis parameter.

5. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, Step three, generating policy effectiveness parameters, includes: Collect historical performance data for each campaign strategy within a preset historical period. The historical performance data includes one or more of the following metrics: conversion rate, click-through rate, user dwell time, exposure completion rate, and bounce rate. Normalize each metric separately and count the number of metrics involved in the normalization process to generate an evaluation completeness parameter. Based on the evaluation completeness parameter, an interval matching operation is performed in the preset completeness correction coefficient mapping relationship to obtain the corresponding completeness correction coefficient, and the original effect benchmark value is multiplied with the completeness correction coefficient to generate the completeness correction benchmark value; Read the feedback credibility weight parameter generated in the current period, and perform interval matching operation in the preset credibility correction coefficient mapping relationship according to the value range of the parameter to obtain the corresponding credibility correction coefficient. Multiply the integrity correction benchmark value and the credibility correction coefficient to generate the first correction result value. Read the hysteresis parameter generated in the current period, and according to the response delay interval in which the hysteresis parameter is located, perform interval matching operation in the preset hysteresis attenuation coefficient mapping relationship to obtain the corresponding hysteresis attenuation coefficient. The hysteresis attenuation coefficient and the hysteresis parameter have a monotonically decreasing relationship. For every preset interval segment that the hysteresis parameter increases, its corresponding hysteresis attenuation coefficient decreases to the next set level. The first correction result value is multiplied by the hysteresis decay coefficient to generate the second correction result value, and the second correction result value is used as the strategy effectiveness parameter for multiple constraint processing in the current period. Write the strategy effectiveness parameters into the effect parameter recording area as input parameters for calculating the priority of the delivery strategy in step four.

6. The real-time budget scheduling method for an advertising delivery system according to claim 5, characterized in that, In historical performance data, when multiple indicators are included, each indicator is normalized separately, and then the indicators are merged according to a preset weighting ratio to generate the original performance baseline value. When historical performance data includes only a single indicator, the normalized value of that single indicator is used as the original performance baseline value after normalization.

7. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, Step four, which involves setting the sorting adjustment parameters based on the lag parameter, includes: The lag parameter is divided into multiple lag level intervals, including the first level interval, the second level interval, and the third level interval; The first level range corresponds to the lag parameter being less than the first preset value, the second level range corresponds to the lag parameter being between the first preset value and the second preset value, and the third level range corresponds to the lag parameter being greater than the second preset value. A priority change adjustment coefficient is set for each lag level interval. The priority change adjustment coefficient is used to limit the change of the priority parameters of each deployment strategy in the current period compared with the previous period. The higher the lag level, the smaller the corresponding priority change adjustment coefficient. Obtain the lag parameter of each delivery strategy in the current period and determine its lag level range; Based on the priority change adjustment coefficient corresponding to the respective lag level interval, the priority parameters of each delivery strategy are subject to amplitude limitation.

8. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, Step five involves limiting the budget allocation based on the strategy ranking results of each campaign, including: Calculate the budget allocation coefficient for each delivery strategy. The budget allocation coefficient is generated by the priority parameter of the corresponding strategy in the strategy ranking result and the preset mapping relationship. The budget allocation cap for each campaign strategy is calculated based on the budget allocation coefficient and the total available budget in the current budget pool. Convert the budget allocation cap into the corresponding budget control instructions; Based on the feedback credibility weight parameter and lag parameter of each delivery strategy in the current period, determine whether the delivery strategy is in an abnormal state. If the abnormal state judgment condition is met, mark the strategy as an abnormal strategy. The generated budget control instructions are processed by splitting the flow. Budget control instructions corresponding to normal policies are written to the budget scheduling channel through the main write path, while budget control instructions corresponding to abnormal policies are written to the abnormal record area through the auxiliary write path. The primary write path and the secondary write path are physically isolated from each other, and the data flow is kept unidirectional to prevent abnormal data from flowing into the budget scheduling channel.

9. The real-time budget scheduling method for an advertising delivery system according to claim 1, characterized in that, Step six involves the asymmetric writing process of the actual execution data, including: Collect actual execution data for each campaign strategy within the current period. The actual execution data should include at least the budget usage, the number of request triggers, the feedback reception time, and the feedback credibility weight parameter. Extract traffic-related data fields from the actual execution data and construct traffic statistics records. Write the traffic statistics records to the traffic anomaly intensity indicator data area through the main write path or auxiliary write path according to the anomaly indicators of the corresponding strategy. Extract the return quality-related data fields from the actual execution data and form a return quality record. Write the return quality record to the return credibility weight parameter data area through the main write path or auxiliary write path according to the anomaly identifier of the corresponding strategy. The main write path adopts a synchronous write mechanism, while the auxiliary write path adopts an asynchronous write mechanism based on a cache queue. The main write path and the auxiliary write path maintain a unidirectional data isolation structure.

Images