Online and offline traffic site combined screening management system based on dynamic iteration

Abstract

The invention discloses an online and offline flow site combined screening management system based on dynamic iteration, and belongs to the technical field of data processing, and the system comprises the following steps: an online flow detection module, which is used for collecting flow state data of each online channel in a preset continuous time period, carrying out the analysis to obtain a channel comprehensive access index of each online channel, and sending the channel comprehensive access index to a database; carrying out trend analysis to obtain an online flow trend type of each online channel; the offline flow detection module is used for synchronously collecting site state data of each offline flow site, analyzing to obtain a receptable index of each offline flow site, performing trend analysis based on the receptable index, dividing to obtain an offline bearing trend type of each offline flow site, and sending the offline bearing trend type to the offline flow site; the technical problem that the cooperative management effect of online and offline traffic sites is limited due to the fact that combined dynamic adjustment based on the loading capacity cannot be achieved under the traffic fluctuation and emergency conditions in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a dynamic iterative online and offline traffic venue joint screening and management system. Background Technology

[0002] Traffic venues refer to identifiable, quantifiable, and manageable carrier units capable of continuously accommodating, generating, aggregating, or releasing crowd movement within a specific timeframe. In existing online-offline integrated operation models, online traffic is typically guided to offline physical venues for actual consumption or experiences through recommendations, reservations, or marketing. Online channels serve as platforms for acquiring and converting online traffic, including but not limited to short video platforms, content recommendation platforms, search platforms, and lifestyle service platforms, such as Douyin (TikTok). To improve traffic conversion rates, existing technologies often employ traffic venue screening methods based on historical data or fixed rules to match and manage online traffic targets with offline venues. However, in actual operation, online traffic exhibits significant temporal fluctuations due to user behavior, promotional strategies, and unforeseen events, while the carrying capacity of offline venues is constrained by factors such as venue size, staffing, and operational status. These two factors have a significant dynamic multiplicative relationship. Existing technologies, in collaborative management, often use static methods for venue screening and traffic allocation, making it difficult to address the need for coordinated online-offline control in situations of sudden traffic surges or changes in status.

[0003] For example, Chinese invention patent CN120146897B discloses a method, system, and storage medium for screening traffic venues for marketing, which includes: collecting multi-dimensional data through a monitoring system and calculating quality assessment indicators; extracting user characteristics to build profiles and calculating matching degrees; predicting effects by combining business data; screening venues based on prediction and evaluation; monitoring and analyzing the screening results and dynamically adjusting to obtain an optimization plan.

[0004] The above-mentioned technology has at least the following technical problems: When managing online and offline traffic and venues, online traffic targets and offline venues are typically screened and configured using preset rules or based on historical statistics. This allows for basic operations to be maintained when traffic fluctuations are relatively stable. However, in practical applications, when online traffic fluctuates significantly due to promotional activities, changes in user behavior, or unforeseen events, this static screening method struggles to coordinate online traffic changes with the real-time capacity of offline venues. This leads to an imbalance between traffic delivery and venue capacity, making it difficult to achieve coordinated optimization of online and offline traffic and venues through multiple rounds of adjustments. Therefore, the current technology, which uses a static approach to online traffic delivery and offline venue screening, fails to achieve joint dynamic adjustments based on capacity under traffic fluctuations and unforeseen circumstances, resulting in limited effectiveness in the coordinated management of online and offline traffic and venues. Summary of the Invention

[0005] To address the limitations of existing technologies that employ static methods for online traffic allocation and offline venue selection, failing to achieve dynamic adjustments based on capacity under traffic fluctuations and emergencies, thus restricting the collaborative management of online and offline traffic and venues, this invention provides a dynamically iterative online and offline traffic and venue joint selection and management system. The technical solution is as follows: This system provides a dynamic iterative online and offline traffic venue joint screening and management system. The system includes: an online traffic detection module, used to collect traffic status data from various online channels within a preset continuous time period, analyze the comprehensive access index of each online channel, and perform trend analysis to determine the online traffic trend type of each online channel; an offline traffic detection module, used to simultaneously collect venue status data from various offline traffic venues, analyze the capacity index of each offline traffic venue, perform trend analysis based on the capacity index, and classify the offline carrying capacity trend type of each offline traffic venue; and a joint evaluation module, used to combine each online channel with each offline traffic venue to form a combined unit, based on online traffic trends... The system uses the potential type and the offline carrying capacity trend type to jointly determine the state of the combined units, thus obtaining the combination type of each combined unit. The joint prediction module directly outputs the optimal combined unit when the combination type is a stationary combined unit. Otherwise, it constructs the joint feature vector of the non-stationary combined units and inputs it into the preset machine learning prediction model, outputting the corresponding online traffic prediction value and site carrying capacity prediction value, and obtaining the imbalance degree value of each non-stationary combination. The iterative optimization module dynamically adjusts the prediction parameters and constraint parameters of the machine learning prediction model based on the imbalance degree of each combined unit, and repeatedly executes the joint prediction process in a continuous time period until the optimal combined unit that meets the preset safety constraints is obtained.

[0006] 1. The online and offline traffic venue joint screening and management system based on dynamic iteration provided by this invention synchronously collects and analyzes the trend type of the comprehensive access index of online channels and the capacity index of offline traffic venues, thereby constructing a combined unit judgment mechanism that combines online traffic trend type and offline capacity trend type. This enables dynamic identification and screening of combined units in complex scenarios such as traffic fluctuations and sudden changes, effectively solving the technical problem that the existing technology uses a static approach to online traffic placement and offline venue screening, which fails to achieve joint dynamic adjustment based on capacity under traffic fluctuations and sudden events, resulting in limited collaborative management effect of online and offline traffic venues.

[0007] 2. This invention further divides the combined unit into stationary combined units and non-stationary combined units, and constructs a joint feature vector for the non-stationary combined units. It introduces a machine learning prediction model to jointly predict the online traffic forecast and the site capacity forecast, thereby realizing a quantitative assessment of the imbalance between online demand and offline capacity in non-stationary scenarios. This improves the prediction accuracy and decision-making pertinence in the event of sudden traffic changes or fluctuations in carrying capacity.

[0008] 3. This invention dynamically adjusts the cycle length of a continuous time period by smoothing the difference based on the channel comprehensive access index, enabling the data analysis window to adaptively shrink with the speed of traffic changes. This strengthens the system's ability to perceive traffic anomalies and trend changes within a short period, thereby achieving highly timely linkage between online traffic changes and offline site conditions, and improving the stability and response efficiency of the overall joint screening and dynamic iterative adjustment process. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of the structure of the online and offline traffic venue joint screening management system based on dynamic iteration provided in the embodiments of this application; Figure 2 A system global flowchart of the online and offline traffic venue joint screening and management system based on dynamic iteration provided in the embodiments of this application; Figure 3 A flowchart of joint prediction and dynamic iteration of nonstationary combined units in a dynamic iteration-based online and offline traffic site joint screening management system provided in an embodiment of this application; Figure 4 A schematic diagram illustrating the trend of the channel comprehensive access index of the online and offline traffic venue joint screening management system based on dynamic iteration provided in this application embodiment; Figure 5 A schematic diagram of multi-curve prediction of the number of people and capacity limit in a venue based on dynamic iteration of an online and offline traffic venue joint screening management system provided in an embodiment of this application. Figure 6 This is a schematic diagram illustrating the dynamic iterative convergence of the online and offline traffic venue joint screening and management system based on dynamic iteration provided in this application embodiment. Detailed Implementation

[0011] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0012] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0013] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0014] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0015] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0016] First embodiment, such as Figure 1 As shown, Figure 1The schematic diagram of the structure of the online and offline traffic venue joint screening and management system based on dynamic iteration provided in this application embodiment is shown. The system includes the following modules: an online traffic detection module, used to collect traffic status data of each online channel within a preset continuous time period, analyze and obtain the channel comprehensive access index of each online channel, and perform trend analysis to obtain the online traffic trend type of each online channel; an offline traffic detection module, used to synchronously collect venue status data of each offline traffic venue, analyze and obtain the carrying capacity index of each offline traffic venue, perform trend analysis based on the carrying capacity index, and classify the offline carrying capacity trend type of each offline traffic venue; and a joint evaluation module, used to combine each online channel with each offline traffic venue to form a combined unit, based on the online traffic trend type and the offline... The system performs joint state determination on the load-bearing trend type of the combined units to obtain the combination type of each combined unit. The joint prediction module directly outputs the optimal combined unit when the combination type is a stationary combined unit. Otherwise, it constructs the joint feature vector of the non-stationary combined units and inputs it into the preset machine learning prediction model to output the corresponding online traffic prediction value and site carrying capacity prediction value, and obtains the imbalance degree value of each non-stationary combination. The iterative optimization module dynamically adjusts the prediction parameters and constraint parameters of the machine learning prediction model based on the imbalance degree of each combined unit, and repeatedly executes the joint prediction process in a continuous time period until the optimal combined unit that meets the preset safety constraint conditions (imbalance degree value is less than 0 and remaining carrying capacity margin processing value is greater than or equal to 0) is obtained.

[0017] In this embodiment, Figure 2The system global flowchart of the online and offline traffic venue joint screening and management system based on dynamic iteration provided in this application embodiment first collects online traffic status data of each online channel within the current continuous time period, and calculates the corresponding channel comprehensive access index based on this data. By analyzing the changes in the index within the continuous time period, the online traffic trend type of each online channel is determined. Subsequently, the venue status data of each offline traffic venue is collected synchronously, and the capacity index of each offline venue is calculated. Based on this, the capacity trend type of the offline venue is analyzed. After obtaining the online traffic trend type and the offline capacity trend type, each online channel and each offline venue are combined one by one to form multiple combination units, and the combination type of each combination unit is jointly determined based on the trend types of the two. When the determination result is a stable combination unit, the system directly outputs the corresponding optimal combination unit; when the determination result is a non-stationary combination unit, the joint feature vector of the non-stationary combination unit is further constructed and input into the machine learning prediction model to obtain the corresponding online traffic prediction value and venue capacity prediction value. Based on the above prediction results, the imbalance degree value of the combined unit is calculated, and a dynamic iterative adjustment process is performed according to the imbalance degree value. During the adjustment process, it is continuously judged whether the preset safety constraints are met, and the entire process ends when the safety constraints are met.

[0018] Furthermore, the comprehensive access index for each online channel is obtained. Specifically, traffic status data for each online channel is collected over several consecutive time periods. This data includes online page views, average dwell time, average browsing depth, and bounce rate. The traffic status data is normalized to obtain normalized values ​​for online page views, average dwell time, and average browsing depth, which are used as positive contributions. The normalized value of the bounce rate is used as a negative penalty. The positive and negative contributions are weighted and superimposed with a pre-defined comprehensive channel weight set to obtain the comprehensive access index for each online channel, representing the access intensity and user behavior quality of each channel. The comprehensive channel weight set includes weights for online page views, average dwell time, average browsing depth, and bounce rate.

[0019] In this embodiment, online pageviews can be obtained by counting the number of times all users triggered page loads within the time period through the access log system corresponding to the online channel. Average dwell time can be obtained by recording the entry and exit timestamps of each access behavior in the access log, calculating the difference between the entry and exit timestamps, and then arithmetically averaging the dwell times of all accesses within the time period. Average browsing depth represents the number of functional nodes (e.g., likes, favorites, and comments) reached by a user in a single access. This can be obtained by recording the number of functional nodes accessed within a single session in the access log, averaging the number of functional nodes across all sessions within the time period, and then calculating the average number of functional nodes per session, which is the average browsing depth. Bounce rate represents access records where only one page visit was triggered and no subsequent interaction occurred (e.g., no likes). This can be calculated by proportionally dividing the number of such visits by the total number of visits.

[0020] The comprehensive access index for each online channel is obtained as follows: A traffic status data sample set is acquired, including online pageview sample values, average dwell time sample values, average browsing depth sample values, and bounce rate sample values. The online pageview, average dwell time, and average browsing depth are divided by their respective sample values ​​to obtain normalized values ​​for online pageview, average dwell time, and average browsing depth. The bounce rate sample value is divided by the bounce rate to obtain a normalized value. The normalized values ​​for online pageview, average dwell time, and average browsing depth are then multiplied by their respective weights. These three multiplied values ​​are then summed, and the product of the normalized bounce rate value and its weight is subtracted to obtain the comprehensive access index. This process is repeated for each online channel to obtain its comprehensive access index.

[0021] It should be noted that the channel comprehensive weight set is a preset channel comprehensive weight set in the database. The weights of online pageviews, average dwell time, average browsing depth, and bounce rate in the channel comprehensive weight set are specifically obtained by statistically analyzing the traffic status data during historical online channel traffic status assessments. The specific method is as follows: obtain historical traffic status data (historical traffic status data includes historical online pageviews, historical average dwell time, historical average browsing depth, and historical bounce rate), and analyze the proportion of the trigger times of historical online pageviews, historical average dwell time, historical average browsing depth, and historical bounce rate to the total number of trigger times (the total number of trigger times is the sum of the trigger times of historical traffic status data). Use this proportion as the weight of the historical traffic status data, and combine the weights of the historical traffic status data to form the channel comprehensive weight set.

[0022] The comprehensive access index for online channels is derived by analyzing traffic data from various channels. This analysis considers the interrelationships between these parameters. For example, high online pageviews accompanied by a long average dwell time indicate sustained user interaction after entering the channel, classifying this pageview as high-quality traffic and amplifying its positive contribution. Conversely, high pageviews with short average dwell time and low average browsing depth suggest that many users are only engaging superficially, weakening the effective contribution of this pageview to the comprehensive access index. Average dwell time and average browsing depth exhibit a synergistic reinforcing relationship; a simultaneous increase in both reflects strong user content absorption and path extension capabilities within the channel, reinforcing the assessment of users' genuine interest and potential conversion intentions. A high bounce rate, even with some pageviews or dwell time, indicates insufficient user behavior stability, negatively impacting the comprehensive access index. This is particularly true in cases of high pageviews but high bounce rates, where the positive impact of bounce rate on pageviews is significantly weakened.

[0023] Furthermore, the online traffic trend types of each online channel are obtained. The specific method is as follows: based on the channel comprehensive access index within a continuous time period, the change in the channel comprehensive access index between any adjacent time periods and the comprehensive fluctuation intensity of the continuous time period are analyzed; a preset channel comprehensive access threshold set is obtained, and combined with the analysis of the change in the channel comprehensive access index between any adjacent time periods and the comprehensive fluctuation intensity of the continuous time period, the online traffic trend type of each online channel is obtained; the channel comprehensive access threshold set includes the channel comprehensive access index change threshold and the comprehensive fluctuation intensity threshold; the online traffic trend types include stable, oscillating, and abrupt changes.

[0024] In this embodiment, the change in the channel comprehensive access index between any two adjacent time periods is obtained by performing a difference processing on the channel comprehensive access index between any two adjacent time periods (subtracting the channel comprehensive access index of the previous time period from the channel comprehensive access index of the later time period) to obtain the change in the channel comprehensive access index (which can be positive or negative, indicating the direction of change; a positive number indicates an increase in the channel comprehensive access index, and a negative number indicates a decrease in the channel comprehensive access index). The comprehensive fluctuation intensity of a continuous time period is obtained by subtracting the channel comprehensive access index of the initial time period from the channel comprehensive access index of the last time period in the continuous time period.

[0025] The online traffic trend type of each online channel is obtained as follows: if the overall fluctuation intensity is less than the overall fluctuation intensity threshold and the change in the channel's overall access index between any two adjacent time periods is less than the channel's overall access index change threshold, then the online traffic trend type of the online channel is stable; if the overall fluctuation intensity is less than the overall fluctuation intensity threshold and there is a change in the channel's overall access index between two adjacent time periods that is greater than the channel's overall access index change threshold, then the online traffic trend type of the online channel is oscillating; otherwise, the online traffic trend type of the online channel is abrupt.

[0026] By jointly analyzing the changes and overall fluctuation intensity of the channel comprehensive access index over a continuous time period, online traffic trends are classified into three types: stable, oscillating, and abrupt. The stable type characterizes online access behavior with small fluctuations and controllable volatility. This type can serve as the system's baseline state, eliminating the need for additional risk corrections in subsequent online-offline joint forecasting, thus reducing model complexity. The oscillating type describes the state of frequent fluctuations in online traffic over a short period without forming a unidirectional trend. The abrupt type identifies abnormal states where the channel comprehensive access index experiences significant jumps and a simultaneous increase in fluctuation intensity within a short period. This type often corresponds to external events, concentrated exposure, or changes in channel strategies, and is a key factor leading to distortions in online demand forecasting.

[0027] Furthermore, the carrying capacity index of each offline traffic venue is obtained. The specific method is as follows: Collect venue status data for each offline traffic venue within each time period. The venue status data includes the number of people inside the venue, inflow rate, outflow rate, and average dwell time per person. Obtain the maximum capacity of each offline traffic venue and perform a difference processing with the corresponding number of people inside to obtain the remaining carrying capacity margin. Obtain a preset offline traffic venue comparison sample set and remaining carrying capacity sample values ​​from the database. The offline traffic venue comparison sample set includes inflow rate sample values, outflow rate sample values, and average dwell time per person sample values. Based on the offline traffic venue comparison sample set, normalize the venue status data to obtain processed values ​​for each venue status data (including processed values ​​for inflow rate, outflow rate, and average dwell time per person; the specific normalization method is: divide the inflow rate sample value by the inflow rate to obtain...). The outflow rate is processed by dividing the outflow rate by the outflow rate sample value to obtain the outflow rate processed value, thus completing the normalization process. The absolute difference between the average dwell time per person and the average dwell time sample value is processed to obtain the absolute difference of the average dwell time per person. This absolute difference of the average dwell time per person is divided by the average dwell time sample value to obtain the average dwell time processed value, thus completing the average dwell time normalization process. The remaining carrying capacity margin is normalized by dividing the remaining carrying capacity margin sample value by the remaining carrying capacity margin to obtain the remaining carrying capacity margin processed value. The remaining carrying capacity margin processed value and the site status data processed value are weighted based on the carrying capacity index weight set and then superimposed to obtain the carrying capacity index for each offline flow site, which is used to characterize the real-time carrying capacity level. The carrying capacity index weight set includes the inflow rate weight, outflow rate weight, average dwell time per person weight, and remaining carrying capacity margin weight. It should be noted that the remaining carrying capacity margin and the status data of each site are weighted based on the weight set of the carrying capacity index and then superimposed to obtain the carrying capacity index of each offline traffic site, which is used to characterize the real-time carrying capacity level. The specific method is as follows: the inflow rate, outflow rate, average stay time per person and the remaining carrying capacity margin are multiplied by the corresponding inflow rate weight, outflow rate weight, average stay time per person and remaining carrying capacity margin weight (negative number) respectively, and then added together to obtain the carrying capacity index of each offline traffic site.

[0028] The weight set of the carrying capacity index can be obtained from a database. Within the historical operating period, several offline traffic sample periods with known "no overload, congestion, or safety incidents" are selected as stable carrying capacity sample periods. Within each stable carrying capacity sample period, the normalized values ​​of the corresponding inflow rate, outflow rate, average dwell time per person, and remaining carrying capacity margin are simultaneously obtained. The normalized value of each parameter is arithmetically averaged over all stable carrying capacity sample periods to obtain the average contribution level of each parameter under stable carrying capacity conditions. The average contribution levels of each parameter are normalized so that their sum is 1, resulting in the initial weight values. Among them, the weight values ​​corresponding to the inflow rate, outflow rate, and average dwell time per person are set to positive values ​​to represent their positive contribution to the carrying capacity of the site. The weight value corresponding to the remaining carrying capacity margin is set to negative values ​​while keeping the numerical value unchanged to represent the inhibitory effect on the carrying capacity index when the remaining carrying capacity margin decreases. Thus, the weight set of the carrying capacity index is obtained, including the inflow rate weight, outflow rate weight, average dwell time per person weight, and remaining carrying capacity margin weight.

[0029] In this embodiment, the number of people inside the venue can be detected using an infrared beam counting device. Within each time period, the number of people entering and leaving is accumulated and corrected in real time to obtain the real-time number of people inside the venue within that time period. The inflow rate is obtained by counting the number of new people entering the venue within a preset time period and dividing this number by the corresponding time period length to obtain the number of people entering per unit time, i.e., the inflow rate, which reflects the intensity of the external flow of people converging into the venue. The outflow rate is obtained by counting the number of people leaving the venue within the same time period and dividing this number by the corresponding time period length to obtain the number of people leaving per unit time, i.e., the outflow rate, which reflects the dissipation capacity of people inside the venue. The average dwell time per person can be calculated by anonymizing and time-stamping individuals entering the venue, recording their entry and exit times, and statistically analyzing the number of people who complete the entry and exit loop within the time period to calculate the average dwell time within the venue.

[0030] The closer the number of people in the venue is to the maximum capacity, the smaller the remaining carrying capacity margin, and the lower the corresponding carrying capacity index. A high inflow rate will accelerate the consumption of carrying capacity, thereby reducing the carrying capacity index, while a high outflow rate will help release carrying capacity space and provide positive support for the carrying capacity index. An excessively long average stay time per person will slow down personnel turnover and reduce the carrying capacity that can be released per unit time, thereby inhibiting the carrying capacity index. On the other hand, an excessively short stay time usually corresponds to a high frequency of entry and exit, which may cause short-term fluctuations. Therefore, the more favorable state is that the parameters change in coordination within a reasonable range that matches the operating characteristics of the venue, so that the remaining carrying capacity margin is kept within a safe and sustainable range. Only under this condition can the carrying capacity index have practical scheduling and decision-making value.

[0031] Within the same time period, the number of people in the venue is dynamically determined by the difference between the inflow rate and the outflow rate. When the inflow rate is consistently greater than the outflow rate, the number of people in the venue will increase cumulatively. The average dwell time per person reflects the stickiness of people staying in the venue. The higher the value, the smaller the amount of venue space that can be released per unit time, thus increasing the number of people in the venue when the inflow rate remains unchanged. Conversely, when the average dwell time per person decreases, the outflow rate relatively increases, which helps to alleviate the pressure on the number of people in the venue. Therefore, the inflow rate determines the intensity of external pressure input, the outflow rate determines the internal absorption capacity, the average dwell time per person indirectly affects the number of people in the venue by adjusting the outflow rhythm, and the number of people in the venue, in turn, restricts the venue's ability to accept new flow in subsequent time periods. The four factors together constitute a dynamic balance relationship of the venue's carrying capacity.

[0032] Furthermore, the offline carrying capacity trend types of each offline traffic venue are classified. The specific method is as follows: based on the carrying capacity index within a continuous time period, and analyzing its direction and magnitude of change, if the carrying capacity index of a certain offline traffic venue remains within a preset stable range within a continuous time period, then the offline carrying capacity trend type of that offline traffic venue is determined to be stable; if the carrying capacity index of a certain offline traffic venue shows a comprehensive increasing trend within a continuous time period and the magnitude of change is greater than the magnitude threshold, then the offline carrying capacity trend type of that offline traffic venue is determined to be sudden increase; if the carrying capacity index of a certain offline traffic venue shows a comprehensive decreasing trend within a continuous time period and the magnitude of change is greater than the magnitude threshold, then the offline carrying capacity trend type of that offline traffic venue is determined to be sudden decrease.

[0033] In this embodiment, by introducing a carrying capacity trend type, "short-term fluctuations" and "structural changes" can be distinguished. When the carrying capacity index remains stable within a preset range for a long period, it is identified as a stable type. This helps avoid misjudgments caused by occasional personnel entry and exit, enabling such sites to participate in online and offline combinations as reliable carrying capacity objects, thereby reducing unnecessary prediction calculations and adjustment interventions. Conversely, when the carrying capacity index rises or falls continuously and the change exceeds a threshold, it is identified as a sudden increase or sudden decrease. This can promptly capture signals of changes in the site's internal operating status, spatial openness, or crowd behavior patterns, preventing the system from making decisions based on stable carrying capacity and thus causing risks.

[0034] Furthermore, the combination type of each combination unit is obtained. The specific method is as follows: each online channel and each offline traffic venue are combined one by one to form each combination unit (specifically, any online channel is combined with any offline traffic venue to obtain a combination unit, and so on, traversing all online channels and offline traffic venues to obtain each combination unit, wherein each combination unit includes one online channel and one offline traffic venue); the corresponding online traffic trend type and offline carrying trend type in the combination unit are obtained respectively, and the combination type of the combination unit is determined based on the combination relationship between the two; the combination type is used to characterize the potential attendance influence state and carrying matching state of the online channel in the combination unit on the offline traffic venue; if there is a combination unit whose online traffic trend type is stable and whose offline carrying trend type is stable, then the combination type of the combination unit is a stable combination unit, otherwise it is a non-stable combination unit.

[0035] In this embodiment, a distinction is made between stationary and non-stationary combined units at the combined level. The core purpose is to avoid introducing unnecessary systematic errors by using the same prediction logic under different operating conditions. When the traffic trend of online channels and the carrying capacity trend of offline venues are both in a stationary state, the supply and demand relationship structure between them is relatively fixed. The online arrival conversion relationship and the offline carrying capacity feedback relationship formed in the historical period have high continuity. At this time, a stable and reliable prediction result can be obtained by directly using the machine learning prediction model trained based on historical samples, without the need for additional repeated corrections. Conversely, when either the online or offline side exhibits abrupt changes, oscillations, or rapid changes, the original historical statistical relationship no longer fully reflects the current real operating state. If a single historical model is still used for prediction, it is easy to amplify errors or mask potential risks. Therefore, for non-stationary combined units, by constructing a joint feature vector that includes the channel comprehensive access index, the venue carrying capacity index, and their interaction relationship, the immediate impact of online changes on offline carrying capacity is explicitly incorporated into the prediction process. This allows the prediction model to perceive the dynamic adjustment of the supply and demand structure, thereby providing a more targeted prediction basis for subsequent imbalance assessment and iterative optimization.

[0036] Furthermore, a joint feature vector of non-stationary combined units is constructed. Specifically, when a combined unit is a non-stationary combined unit, the combined unit information of the non-stationary combined unit is extracted. The combined unit information includes the channel comprehensive access index, the capacity index, the online traffic trend type, and the offline carrying trend type. Based on the combined unit information, interactive features are constructed to describe the relationship between online traffic and offline venues. Thus, the combined features and the combined unit information are combined to form the joint feature vector of the non-stationary combined unit.

[0037] In this embodiment, for the combined unit determined to be non-stationary, its combined unit information is first extracted synchronously within the same time period. The combined unit information includes the channel comprehensive access index, the capacity index, the online traffic trend type, and the offline capacity trend type. The above features together constitute the basic state features of the combined unit. On the basis of the basic state features, the supply and demand tension feature is introduced. The supply and demand tension feature is specifically constructed by multiplying the channel comprehensive access index with the capacity saturation index of the offline traffic venue (the capacity saturation index is specifically: the current number of people in the venue is divided by the effective capacity limit to obtain the saturation value, the saturation value is subtracted by 1 and then the maximum value is taken by 0 to obtain the capacity saturation index) to obtain the supply and demand tension feature. This feature can reflect the objective law that "the stronger the online traffic and the closer the offline is to saturation, the higher the potential risk", thereby amplifying the overpressure trend that may be ignored in the non-stationary state. The effective capacity limit is a preset value in the database, and the value corresponding to different offline traffic venues is different.

[0038] By introducing transfer-sensitive features, the model can distinguish the differentiated impacts of "the same online growth, under different remaining capacity conditions, will have different pressures on offline operations," thereby improving its responsiveness to sudden or accelerated changes. Online traffic trend types and offline capacity trend types are introduced as discrete state features into the joint feature vector. Finally, the basic state features, supply-demand tension features, and transfer-sensitive features are systematically fused to form a non-stationary combined unit joint feature vector that can simultaneously express "state, relationship, and direction of change." This joint feature vector is then input into the machine learning prediction model for subsequent on-site demand prediction and capacity prediction.

[0039] Furthermore, the interaction features include supply and demand tension features and transfer sensitivity features. The supply and demand tension features are used to characterize the product relationship between the intensity of online channel traffic and the saturation level of offline venues, while the transfer sensitivity features are used to characterize the degree of impact of changes in online channel traffic on the remaining carrying capacity of offline venues.

[0040] In this embodiment, the specific method for constructing the supply-demand tension characteristic is as follows: the channel comprehensive access index and the capacity saturation index are multiplied to obtain the supply-demand tension characteristic used to characterize the mutual constraint relationship between online access intensity and offline carrying capacity in the combined unit. When the channel comprehensive access index is high and the capacity saturation index is non-zero, the supply-demand tension characteristic value increases, indicating that the combined unit is more likely to form carrying pressure in the subsequent time period; when the channel comprehensive access index is high but the capacity saturation index is zero, the supply-demand tension characteristic value remains within a controllable range, indicating that the combination still has carrying capacity; when the channel comprehensive access index is low, even if the capacity saturation index changes significantly, the impact on the overall tension is relatively limited.

[0041] The specific construction method of the transfer-sensitive feature is as follows: within the same time period, obtain the channel comprehensive access index and the offline traffic venue capacity index corresponding to the combined unit; based on the difference between the inflow rate and the outflow rate, dynamically correct the capacity index (dynamic correction is the update process) to obtain the corrected capacity index that reflects the changing trend of offline capacity; jointly calculate the channel comprehensive access index and the corrected capacity index (take the average value) to obtain the transfer-sensitive feature used to characterize the degree of influence of online access changes on offline capacity.

[0042] When the overall channel access index rises while the modified capacity index remains stable or rises, the transfer sensitivity characteristic indicates that the offline venue has a good ability to absorb changes in online traffic. When the overall channel access index rises while the modified capacity index falls rapidly, the transfer sensitivity characteristic increases significantly, indicating that the combination unit is highly sensitive to online changes and is prone to imbalance. When the overall channel access index itself is in a stable or declining state, the impact of the transfer sensitivity characteristic on the combination evaluation decreases accordingly.

[0043] Furthermore, the imbalance degree value of each non-stationary combination is obtained. The specific method is as follows: the joint feature vector of the non-stationary combination unit is input into a preset machine learning prediction model, and the online traffic prediction value and the site capacity prediction value of the non-stationary combination in the future prediction period are output. The online traffic prediction value is used to characterize the potential on-site demand scale that will be converted into offline on-site behavior through the online channel in the future prediction period. The site capacity prediction value is used to characterize the scale of new on-site demand that the offline traffic site can bear under the current operating status and safety constraints in the future prediction period. The difference between the online traffic prediction value and the site capacity prediction value is calculated to obtain the imbalance degree value used to characterize the non-stationary combination. The imbalance degree value of each non-stationary combination is obtained by traversing each non-stationary combination.

[0044] Furthermore, the prediction parameters and constraint parameters of the machine learning prediction model are dynamically adjusted. Specifically, within a continuous time period, the online traffic prediction value and the actual attendance value (which can be obtained by statistically analyzing the number of people who actually made online reservations and then completed offline verification) are subtracted to obtain the online attendance prediction difference. Similarly, the venue capacity prediction value and the actual venue capacity index are subtracted to obtain the venue capacity difference. A preset threshold difference set is obtained from the database, which includes the online attendance prediction difference threshold and the venue capacity difference threshold. This threshold difference set is then compared with the online attendance prediction difference and the venue capacity difference threshold, respectively. The difference in on-site acceptance is compared. If the difference between online arrival prediction and / or on-site acceptance exceeds the corresponding threshold difference set, dynamic iteration is performed; otherwise, dynamic iteration is not performed. If dynamic iteration is performed, the analysis is based on the difference between online arrival prediction and / or on-site acceptance to obtain the iteration adjustment coefficient, and the prediction parameters and constraint parameters are dynamically adjusted accordingly. The dynamic adjustment of the prediction parameters and constraint parameters includes shortening the analysis time window, shortening the prediction step size, increasing the imbalance penalty coefficient and risk constraint coefficient, and simultaneously updating the effective capacity limit of the offline traffic venue.

[0045] In this embodiment, it should be noted that the actual site capacity index can be calculated by taking the site status data of each offline traffic site that is detected in real time and inputting it into the capacity index formula.

[0046] The iterative adjustment coefficients are obtained as follows: If only the online attendance prediction difference exceeds the corresponding online attendance prediction difference threshold, the online attendance prediction difference is divided by the actual attendance value to obtain the actual attendance conversion ratio. This actual attendance conversion ratio is then matched with the database to obtain the first iterative adjustment coefficient, which is used as the iterative adjustment coefficient. If only the venue acceptance difference exceeds the corresponding venue acceptance difference threshold, the venue acceptance difference is divided by the venue's corresponding effective capacity limit to obtain the venue acceptance difference ratio. This venue acceptance difference ratio is then matched with the database to obtain the second iterative adjustment coefficient, which is used as the iterative adjustment coefficient. If both the online attendance prediction difference and the venue acceptance difference exceed the corresponding threshold difference set, the first and second iterative adjustment coefficients are averaged to obtain the iterative adjustment mean coefficient, which is then used as the iterative adjustment coefficient.

[0047] The actual attendance conversion ratio is matched with the database to obtain the first iterative adjustment coefficient. Specifically, this involves obtaining preset intervals for each actual attendance conversion ratio in the database, along with the corresponding first iterative adjustment coefficient. The actual attendance conversion ratio is then compared with each interval. If the actual attendance conversion ratio falls within a preset interval, the corresponding first iterative adjustment coefficient is used as the first iterative adjustment coefficient. Similarly, the site acceptance difference ratio is matched with the database to obtain the second iterative adjustment coefficient. This involves obtaining preset intervals for each site acceptance difference ratio in the database, along with the corresponding second iterative adjustment coefficient. The site acceptance difference ratio is then compared with each interval. If the site acceptance difference ratio falls within a preset interval, the corresponding second iterative adjustment coefficient is used as the second iterative adjustment coefficient.

[0048] like Figure 3 As shown, Figure 3 This application provides a flowchart of the non-stationary combined unit joint prediction and dynamic iteration process for an online and offline traffic venue joint screening management system based on dynamic iteration. The system first smooths the channel comprehensive access index and calculates the absolute difference between the smoothed channel comprehensive access index and the original channel comprehensive access index to obtain the smoothed channel comprehensive access index difference. Then, this difference is compared with a preset channel comprehensive access index smoothing difference threshold. When the difference exceeds the threshold, the period length of the continuous time cycle is gradually shortened according to a preset adjustment step size, and the channel comprehensive access index smoothing difference is recalculated after each adjustment until the difference decreases to within the threshold range.

[0049] The machine learning prediction model is constructed using a gradient boosting decision tree algorithm. Specifically, it uses non-stationary combined unit samples formed over multiple consecutive historical time periods as training samples. Each sample corresponds to a combined unit of online channels and offline traffic venues. Its input features are the joint feature vector constructed by the combined unit within the corresponding time period. The joint feature vector includes the channel's comprehensive access index, capacity index, online traffic trend type, offline capacity trend type, and interaction features representing the relationship between the two. The output labels include two prediction targets: online traffic prediction and venue capacity prediction. After model training, during actual operation, the joint feature vector of the non-stationary combined units, constructed in real time, is input into the gradient boosting decision tree model. The corresponding online traffic prediction and venue capacity prediction are output respectively. The difference between these two values ​​is used to calculate the imbalance value, which quantifies the degree of supply-demand mismatch of the combined unit, thus providing a quantitative basis for subsequent combination selection and dynamic iterative adjustment.

[0050] To shorten the analysis time window (i.e., the sliding window length) and the prediction step size (the prediction step size representing the future time span), the specific method is as follows: subtract the product of the current time window and prediction step size and the iterative adjustment coefficient from the current time window and prediction step size respectively to obtain the iteratively adjusted time window and prediction step size. To increase the imbalance penalty coefficient (the imbalance penalty coefficient refers to the coefficient used to weight and penalize the portion of the non-stationary combination whose imbalance exceeds the preset safety constraints) and the risk constraint coefficient (the risk constraint coefficient is specifically the coefficient of the additional constraint strength applied to the non-stationary combination unit), the specific method is as follows: divide the current imbalance penalty coefficient and risk constraint coefficient... By adding the product of the imbalance penalty coefficient, the risk constraint coefficient, and the iterative adjustment coefficient, we can obtain the iteratively adjusted imbalance penalty coefficient and risk constraint coefficient. Then, based on the adjusted prediction parameters and constraint parameters, we synchronously update the traffic status data of each online channel and the site status data of each offline traffic site, and update the effective capacity limit of the offline traffic site. We then obtain the updated online arrival prediction difference and site acceptance difference again, and compare them with the threshold difference set. If there is an online arrival prediction difference and / or site acceptance difference that is above the corresponding threshold difference set, dynamic iterative processing is performed; otherwise, dynamic iterative processing is not performed, thus completing the dynamic iterative processing.

[0051] like Figure 4 As shown, Figure 4This diagram illustrates the trend of the comprehensive access index of the online and offline traffic venue joint screening management system based on dynamic iteration, as provided in this application embodiment. The diagram uses continuous time periods as the horizontal axis and normalized online traffic prediction values ​​as the vertical axis, displaying the historical sequence of online traffic prediction values ​​for multiple online channels within continuous time periods and the extension trend of future prediction time periods. "Normalization" refers to normalizing the online traffic prediction values ​​of each online channel within each continuous time period relative to preset historical online traffic sample values ​​in the database, thereby ensuring that the online traffic prediction values ​​of different channels are on the same dimension and can be directly compared. As can be seen from the curve, different channels (only three online channels are shown, but the actual analysis is not limited to three online channels) exhibit different trajectories in historical periods: some channels show small fluctuations and remain at similar levels in most continuous time periods, reflecting a more stable online traffic trend type; some channels show significant increases followed by gradual declines in several continuous time periods, exhibiting abrupt changes. The forecast segments (dashed lines) for each channel in the figure show the direction and magnitude of the predicted online traffic value changes within the future forecast period. This is used to characterize the potential demand for offline attendance that will be converted into offline attendance behavior within the future forecast period. The current statistical time is 13:00. The length of each individual continuous time period is 10 minutes. Point A in the figure represents the online traffic value of channel C in the historical continuous time period. The sudden increase in this value is due to the traffic being driven to this online channel. The slight increase at point B is likely due to a small amount of traffic push from the channel platform (traffic push means platform traffic support). The slight decreases at points C and D may be due to the end of traffic push by the channel platform.

[0052] Figure 5This diagram illustrates the multi-curve prediction of the number of people and capacity limit of a venue in a dynamically iterative online and offline traffic venue joint screening and management system provided in this application embodiment. The diagram uses "continuous time period" as the horizontal axis and "number of people" as the vertical axis, simultaneously showing the historical curves of the number of people and the effective capacity limit of the offline traffic venue within the continuous time period, as well as the predicted curves of the number of people and the effective capacity limit for the future prediction period. As can be seen from the historical portion of the diagram, the number of people in the venue experiences phased increases and decreases within the continuous time period, reflecting the fluctuations in the carrying capacity of the offline venue at different times. The effective capacity limit remains relatively stable throughout the historical period with slight variations, representing the upper limit of the allowable carrying capacity of the offline venue within the corresponding time period. By comparing the relative distance between the number of people in the venue and the effective capacity limit, the "approaching saturation level" of the offline venue in each continuous time period can be intuitively determined: when the number of people in the venue gradually approaches the effective capacity limit curve, it indicates that the remaining carrying capacity margin of the venue is shrinking; when the difference between the two is large, it indicates that the venue still has considerable carrying capacity.

[0053] Figure 6 This diagram illustrates the dynamic iterative convergence of the online and offline traffic venue joint screening and management system based on dynamic iteration provided in this application embodiment. The diagram uses "iteration rounds (number of times)" as the horizontal axis and "error magnitude (persons)" as the vertical axis to show the changing trajectories of the online arrival prediction difference and the venue acceptance difference during multiple iterations. As can be seen from the curve trends, in the early stages of iteration, both the online arrival prediction difference and the venue acceptance difference are at relatively high levels, indicating a significant deviation between the joint prediction output and the actual arrival value and actual capacity index. As the iteration rounds progress, both the online arrival prediction difference curve and the venue acceptance difference curve show an overall downward trend, reflecting that iterative optimization gradually brings the prediction results closer to the actual situation.

[0054] In the second embodiment, based on the first embodiment, data backtracking mapping is performed on the traffic status data of each online channel and the venue status data of each offline traffic venue. Specifically, for each online channel, the average effective time lag from "online access behavior" to "offline arrival behavior" is statistically calculated based on historical data (this can be obtained by querying the online reservation ID and the corresponding offline reservation verification time in the computer backend management system, performing difference processing to obtain the time difference, and averaging this time difference to obtain the average effective time lag). The offline venue status data collected within the time period (such as the number of people in the venue, inflow rate, and capacity index) is backtracked to the online time for the combined unit at a certain time point t1 (the average effective time lag is subtracted from this time point to obtain the backtracked online time, denoted as t2). When constructing the joint feature vector of the combined unit, the following method is used: The following time alignment rules are implemented: the channel comprehensive access index at time t2 and the offline availability index at time t1 form a cross-time but logically synchronized data pair. This aims to avoid label misalignment issues such as "online traffic has already declined while offline traffic has just reached its peak" or "online traffic has just surged while offline traffic has not yet responded." Weighted smoothing is performed within a sliding window (to avoid noise). To prevent random errors caused by backtracking from a single time point, a time-weighted window centered on the average effective lag is constructed before and after the backtracking mapping time point t2. The channel comprehensive access index within the window is weighted and smoothed, with higher weights for time points closer to t2, thus forming a smoothed channel comprehensive access index. This aligns online traffic status data and offline venue status data in the causal time dimension, effectively avoiding label misalignment issues caused by differences in user arrival delays. This provides a time-consistent and logically coherent input foundation for subsequent joint prediction and dynamic iteration.

[0055] The smoothed channel comprehensive access index is compared with the channel comprehensive access index by absolute difference processing to obtain the channel comprehensive access index smoothing difference. If the channel comprehensive access index smoothing difference is greater than the preset channel comprehensive access index smoothing difference threshold, the period length of the continuous time period is gradually shortened according to the preset adjustment step size until the channel comprehensive access index smoothing difference is below the channel comprehensive access index smoothing difference threshold, and then the shortening of the period length of the continuous time period is completed.

[0056] By comparing the smoothed channel comprehensive access index with the original channel comprehensive access index, the true intensity of fluctuations in current online traffic changes that are masked by smoothing can be directly reflected. When the difference is large, it indicates that online access behavior has changed significantly in a short period of time, and a longer continuous time period will average out this change, causing the system to respond lagly to sudden increases or decreases. In this case, by gradually shortening the length of the continuous time period, the temporal resolution of data sampling and analysis can be improved, allowing the channel comprehensive access index to reflect the latest access change trends more quickly, thereby triggering subsequent joint predictions and combined adjustments in a timely manner. When the difference falls back to within the threshold, it indicates that online traffic changes have tended to stabilize, and continuing to maintain a shorter period will amplify noise. Therefore, stopping the shortening of the period length can achieve a balance between response speed and stability.

Claims

1. A dynamic iterative online and offline traffic venue joint screening and management system, characterized in that, Includes the following modules: The online traffic detection module is used to collect traffic status data of various online channels within a preset continuous time period, analyze the comprehensive access index of each online channel, and perform trend analysis to obtain the online traffic trend type of each online channel. The offline traffic detection module is used to synchronously collect site status data of each offline traffic site, analyze the carrying capacity index of each offline traffic site, perform trend analysis based on the carrying capacity index, and classify the offline carrying capacity trend type of each offline traffic site. The joint evaluation module is used to combine each online channel with each offline traffic venue to form a combined unit. Based on the online traffic trend type and the offline carrying trend type, the combined unit is jointly judged to obtain the combination type of each combined unit. The joint prediction module is used to directly output the optimal combination unit when the combination type is a stationary combination unit. Otherwise, it constructs the joint feature vector of the non-stationary combination unit and inputs it into the preset machine learning prediction model to output the corresponding online traffic prediction value and site capacity prediction value, and obtains the imbalance degree value of each non-stationary combination. The iterative optimization module is used to dynamically adjust the prediction parameters and constraint parameters of the machine learning prediction model based on the degree of imbalance of each combined unit, and repeatedly execute the joint prediction process in a continuous time period until the optimal combined unit that meets the preset safety constraints is obtained.

2. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The method for obtaining the comprehensive access index of each online channel is as follows: Within several consecutive time periods, traffic status data of various online channels are collected, including online page views, average dwell time, average browsing depth, and bounce rate. Traffic status data is normalized to obtain normalized values ​​of online page views, average dwell time, and average browsing depth, which are used as positive contribution items. The normalized value of bounce rate is used as a negative penalty item. The positive contribution items and negative penalty items are weighted and superimposed with a preset channel comprehensive weight set to obtain the channel comprehensive access index of each online channel, which is used to characterize the access intensity and user behavior quality of each online channel. The comprehensive channel weight set includes online pageview weight, average dwell time weight, average browsing depth weight, and bounce rate weight.

3. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The specific method for obtaining the online traffic trend types of each online channel is as follows: Based on the channel comprehensive access index within a continuous time period, the changes in the channel comprehensive access index between any adjacent time periods and the comprehensive fluctuation intensity within the continuous time period are analyzed. Obtain a preset set of comprehensive access thresholds for each channel, and combine the changes in the comprehensive access index between any two adjacent time periods with the comprehensive fluctuation intensity analysis over consecutive time periods to obtain the online traffic trend type of each online channel. The comprehensive channel access threshold set includes the comprehensive channel access index change threshold and the comprehensive fluctuation intensity threshold; The online traffic trend types include stable, fluctuating, and abrupt changes.

4. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The method for obtaining the carrying capacity index of each offline traffic venue is as follows: Collect site status data for each offline traffic venue in each time period. The site status data includes the number of people in the venue, inflow rate, outflow rate, and average stay time per person. Obtain the maximum capacity of each offline venue and calculate the difference between it and the number of people inside the venue to obtain the remaining capacity margin. Obtain a preset offline traffic site comparison sample set and remaining carrying capacity sample value from the database. The offline traffic site comparison sample set includes inflow rate sample value, outflow rate sample value and average dwell time sample value. Based on the offline traffic site comparison sample set, the site status data is normalized to obtain the processed value of each site status data. The remaining carrying capacity margin is normalized to obtain the processed value of the remaining carrying capacity margin. The processed value of the remaining carrying capacity margin and the processed value of each site status data are weighted based on the weight set of the carrying capacity index and then superimposed to obtain the carrying capacity index of each offline traffic site to characterize the real-time carrying capacity level. The set of weights for the carrying capacity index includes inflow rate weight, outflow rate weight, average length of stay per person weight, and remaining carrying capacity margin weight.

5. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The method for classifying offline traffic venues to obtain their offline carrying capacity trend types is as follows: Based on the carrying capacity index over a continuous time period, and by analyzing its direction and magnitude of change, when the carrying capacity index of a certain offline traffic site remains within a preset stable range over a continuous time period, the offline carrying capacity trend type of that offline traffic site is determined to be stable. When the carrying capacity index of a certain offline traffic venue shows a comprehensive increasing trend over a continuous time period and the change magnitude is greater than the change magnitude threshold, the offline carrying capacity trend type of the offline traffic venue is determined to be a sudden increase type. If the carrying capacity index of a certain offline traffic venue shows a comprehensive decline trend over a continuous time period and the change magnitude is greater than the change magnitude threshold, then the offline carrying capacity trend type of the offline traffic venue is determined to be a sudden decrease type.

6. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The specific method for obtaining the combination type of each combination unit is as follows: Each online channel is combined with each offline traffic venue to form a combined unit; Obtain the corresponding online traffic trend type and offline carrying trend type in the combined unit respectively, and determine the combination type of the combined unit based on the combination relationship between the two; The combination type is used to characterize the potential on-site impact and capacity matching status of the online channels within the combination unit on the offline traffic venue. If a certain combination unit has a stable online traffic trend type and a stable offline carrying trend type, then the combination type of the combination unit is a stable combination unit; otherwise, it is a non-stable combination unit.

7. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The specific method for constructing the joint feature vector of the non-stationary combined units is as follows: When there is a combination unit whose combination type is a non-stationary combination unit, extract the combination unit information of the non-stationary combination unit. The combination unit information includes the channel comprehensive access index, the capacity index, the online traffic trend type, and the offline carrying trend type. Based on the combined unit information, interactive features are constructed to describe the relationship between online traffic and offline site. Then, the interactive features and combined unit information are combined to form the joint feature vector of non-stationary combined units.

8. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 7, characterized in that: The interactive features include supply and demand tension features and transfer sensitivity features. The supply and demand tension features are used to characterize the product relationship between the intensity of online channel traffic and the saturation level of offline venues, while the transfer sensitivity features are used to characterize the degree of impact of changes in online channel traffic on the remaining carrying capacity of offline venues.

9. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The method for obtaining the imbalance values ​​of each non-stationary combination is as follows: The joint feature vector of the non-stationary combination unit is input into the preset machine learning prediction model, and the online traffic prediction value and site capacity prediction value of the non-stationary combination in the future prediction period are output. The online traffic forecast value is used to characterize the potential scale of on-site demand that will be converted into offline on-site behavior through the online channel within a future forecast period. The site can accept the predicted value, which is used to characterize the scale of new on-site demand that the offline traffic site can bear under the current operating status and safety constraints during the future predicted time period. The difference between the online traffic forecast and the site capacity forecast is calculated to obtain the imbalance value used to characterize the non-stationary combination. This process is repeated for each non-stationary combination to obtain the imbalance value for each non-stationary combination.

10. The online and offline traffic venue joint screening and management system based on dynamic iteration as described in claim 1, characterized in that: The prediction parameters and constraint parameters of the machine learning prediction model are dynamically adjusted, specifically by the following method: Within a continuous time period, the difference between the online traffic forecast and the actual attendance is processed to obtain the online attendance forecast difference. The difference between the site capacity forecast and the site actual capacity index is processed to obtain the site capacity difference. Obtain a preset threshold difference set from the database. The threshold difference set includes the online arrival prediction difference threshold and the site acceptance difference threshold. Compare the threshold difference set with the online arrival prediction difference and the site acceptance difference respectively. If there is an online arrival prediction difference and / or a site acceptance difference that is above the corresponding threshold difference set, then perform dynamic iteration processing; otherwise, do not perform dynamic iteration processing. If dynamic iterative processing is performed, the analysis is based on the difference between online arrival prediction and / or the difference in site acceptance to obtain the iterative adjustment coefficient, and the prediction parameters and constraint parameters are dynamically adjusted accordingly. The dynamic adjustment of prediction and constraint parameters includes shortening the analysis time window, shortening the prediction step size, increasing the imbalance penalty coefficient and risk constraint coefficient, and simultaneously updating the effective capacity limit of offline traffic venues.

Images