Advertisement space flow prediction method and system

Abstract

The invention discloses an advertisement space flow prediction method and system, and relates to the technical field of advertisement space flow prediction. The method comprises the following steps: acquiring multi-source data covering a physical environment around an advertisement space, semantic information and external association; processing the collected data by using an encryption technology, encrypting the original data at a local node, and realizing cross-domain model aggregation by using a federated learning algorithm; a data enhancement algorithm is adopted to expand data diversity, an anomaly detection algorithm is utilized to identify abnormal data points, and then a multi-dimensional feature structure is constructed and feature conversion and dimension reduction processing are performed on specific data. Through a data enhancement algorithm, causal analysis framework quantification, near-end strategy optimization and other algorithms, and through simulating flow changes under different strategies, powerful support is provided for decision making, invalid exposure is reduced, and the advertisement putting effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising space flow prediction, and in particular to an advertising space flow prediction method and system. Background Art

[0002] In the digital age, ad space traffic forecasting is crucial for precise advertising delivery, but traditional methods face numerous challenges. First, data collection and utilization are limited. Most methods rely solely on historical traffic data for the ad space itself, ignoring important factors such as surrounding environmental factors (such as foot traffic and traffic conditions) and social media buzz. This results in a single data dimension, leading to significant bias in prediction results. Furthermore, centralized data processing models, where data is uploaded to a central server, pose privacy risks. Concerned about sensitive data leaks, data holders are reluctant to share data, creating data silos, limiting multi-source data integration and preventing comprehensive data support for prediction models. Second, prediction models and decision optimization capabilities are insufficient. Traditional prediction models are often based on simple statistical analysis or linear regression, which struggles to capture the complex spatiotemporal characteristics and dynamic changes of ad space traffic. This leads to significantly increased prediction errors in the face of unexpected situations. Furthermore, existing technologies lack effective strategies for optimizing ad space resource scheduling after traffic forecasting is completed, making it impossible for advertisers to develop precise delivery plans based on forecast results, effectively utilizing advertising resources, and maximizing revenue. Summary of the Invention

[0003] The purpose of the present invention is to provide an advertisement space flow prediction method and system to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a method for predicting advertising space traffic, comprising the following steps:

[0005] Acquire multi-source data covering the physical environment around the ad space, semantic information, and external associations;

[0006] Use encryption technology to process the collected data. After encrypting the original data at the local node, use the federated learning algorithm to achieve cross-domain model aggregation.

[0007] Use data enhancement algorithms to expand data diversity, use anomaly detection algorithms to identify abnormal data points, and then construct multi-dimensional feature structures and perform feature conversion and dimensionality reduction on specific data;

[0008] Based on a causal analysis framework, we quantify the causal impact of various data factors on ad traffic, and use visualization technology to demonstrate feature contributions. We use a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad traffic, while also setting up an adaptive mechanism to ensure prediction reliability.

[0009] A proximal policy optimization algorithm is used to train the ad slot scheduling strategy, with maximizing advertiser revenue as the objective function. At the same time, a digital twin model is constructed to simulate traffic changes under different strategies, and a policy rollback mechanism is triggered based on prediction errors.

[0010] Preferably, the acquisition of multi-source data covering the physical environment surrounding the ad space, semantic information, and external associations includes:

[0011] Collect physical environment data such as traffic flow and heat distribution around the advertising space;

[0012] Parsing social media data through natural language processing models to obtain semantic label information;

[0013] Obtain city heat maps and traffic control information data.

[0014] Preferably, the encryption technology is used to process the collected data, and after the original data is encrypted at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation, including:

[0015] Deploy edge nodes locally at advertisers and use homomorphic encryption technology to encrypt original data;

[0016] After the original data is encrypted at the local node, cross-domain model aggregation is achieved through the FedProx algorithm, where the local update formula is:

[0017] ;

[0018] The global aggregation formula is:

[0019] ;

[0020] Where, Indicates the The local model is The parameters during round training, is the learning rate, It is The loss function of the local model, is a hyperparameter that controls how close the local model is to the global model. It is The global model parameters during the training round, is the number of local models participating in federated learning, It is The weight of each local model;

[0021] Among them, the FedProx algorithm adjusts the hyperparameters in the local model update formula when performing cross-domain model aggregation , which controls how close the local model is to the global model.

[0022] Preferably, the method of using a data enhancement algorithm to expand data diversity, using anomaly detection algorithms to identify abnormal data points, and then constructing a multidimensional feature structure and performing feature conversion and dimensionality reduction on specific data includes:

[0023] The StyleGAN3 algorithm is used to generate virtual traffic samples. By adjusting the potential vector in the generator, virtual traffic samples in different weather conditions and time periods are generated for data enhancement.

[0024] Use IsolationForest algorithm to detect abnormal traffic points;

[0025] Construct a spatiotemporal feature cube, and perform feature engineering on social media tag data through TF-IDF conversion and PCA dimensionality reduction.

[0026] Preferably, the causal impact of various data factors on ad space traffic is quantified based on the causal analysis framework, and feature contributions are displayed with the help of visualization technology. The fusion neural network hybrid model is used to capture the spatiotemporal characteristics and dynamic changes of ad space traffic, and an adaptive mechanism is set up to ensure the reliability of prediction, including:

[0027] Traffic forecasting is performed based on the CausalML framework and hybrid prediction model. The CausalML framework is used to quantify the causal effect of each factor on traffic, and the feature contribution is visualized through SHAP values.

[0028] The hybrid prediction model includes a spatiotemporal Transformer and a dynamic Bayesian network to capture the spatiotemporal characteristics and dynamic changes of ad slots. It also sets an adaptive gating mechanism to switch to the expert system when the prediction confidence falls below a set threshold.

[0029] The double difference method formula is used to quantify the causal effect of each factor on traffic:

[0030] ;

[0031] Where, represents the average treatment effect, and represent the flow observation values of the treatment group and the control group, respectively. is an indicator variable indicating whether it has been treated. =1 means it is processed. =0 means no processing.

[0032] Preferably, the proximal strategy optimization algorithm is used to train the advertising slot scheduling strategy, with maximizing advertiser revenue as the objective function:

[0033] The reward function formula of the proximal policy optimization algorithm is:

[0034] ;

[0035] Where, Indicates at time The reward value, 、 、 is the weight coefficient, It's time Click-through rate, It's time CPM, It's time Diversity metrics for advertising display;

[0036] The proximal strategy optimization algorithm is:

[0037] ;

[0038] Where, Indicates that in the strategy Under expectations, is the discount factor, is the time step, It's time Actions taken, It's time status.

[0039] Preferably, when the proximal strategy optimization algorithm trains the advertising slot scheduling strategy, the strategy parameters are adjusted according to the reward function, and the weight coefficient in the reward function is 、 、 Dynamically adjust based on advertisers’ business priorities and goals.

[0040] The present invention also provides an advertising space flow prediction system, comprising:

[0041] The data collection and privacy protection module is used to obtain multi-source data covering the physical environment around the ad space, semantic information, and external associations, and to process the collected data using encryption technology. After encrypting the original data at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation;

[0042] The data preprocessing module is used to expand data diversity using data enhancement algorithms, identify abnormal data points using anomaly detection algorithms, and then construct a multidimensional feature structure and perform feature conversion and dimensionality reduction on specific data;

[0043] The prediction model design module is used to quantify the causal impact of various data factors on ad traffic based on a causal analysis framework, and uses visualization technology to demonstrate feature contributions. It uses a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad traffic, and also sets up an adaptive mechanism to ensure prediction reliability.

[0044] The dynamic decision output module is used to train the ad slot scheduling strategy using the proximal policy optimization algorithm, with maximizing advertiser revenue as the objective function. At the same time, it builds a digital twin model to simulate traffic changes under different strategies and triggers the strategy rollback mechanism based on the prediction error.

[0045] The present invention further provides an electronic device, wherein the electronic device is a physical device and comprises:

[0046] a processor and a memory, wherein the memory is communicatively connected to the processor;

[0047] The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the above-mentioned advertising space traffic prediction method.

[0048] The present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the advertising space traffic prediction method as described above is implemented.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] Multimodal data collection integrates multi-source information, providing a rich data foundation for accurate predictions. Federated edge computing, combined with homomorphic encryption technology, promotes multi-source data collaboration while protecting data privacy, reducing data transmission volume and improving data processing efficiency and security.

[0051] Data augmentation and anomaly detection technologies improve data quality and diversity, enhancing the model's adaptability to complex situations. Feature engineering builds a comprehensive and efficient feature system to effectively extract and utilize data features.

[0052] Causal analysis-based methods deeply explore the causal relationship between various factors and traffic flow. Combined with hybrid prediction models, they capture complex spatiotemporal characteristics and dynamic change patterns, significantly improving prediction accuracy. The adaptive gating mechanism ensures reliable prediction results even in complex situations, enhancing system stability and reliability.

[0053] Proximal strategy optimization optimizes ad placement scheduling with the goal of maximizing advertiser revenue, taking into full consideration multiple key business indicators to achieve rational allocation and efficient utilization of advertising resources. The digital twin model provides an effective means of strategy verification and optimization. By simulating traffic changes under different strategies, it provides strong support for decision-making, reduces invalid exposure, and improves advertising effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A main flow chart of an advertising space flow prediction method provided by an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of the structure of an advertising space flow prediction system provided by an embodiment of the present invention;

[0056] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] The execution subject of the method of this embodiment is a terminal, which can be a mobile phone, tablet computer, PDA, notebook or desktop computer, etc. Of course, it can also be other devices with similar functions, and this embodiment is not limited.

[0059] See also Figure 1 The present invention provides an advertising space flow prediction method and system, the method is applied to, including:

[0060] Step 100: Acquire multi-source data covering the physical environment around the ad space, semantic information, and external associations.

[0061] Wherein, step 100 includes:

[0062] Use lidar and thermal imaging cameras to collect physical environment data such as traffic flow and heat distribution around the advertising space;

[0063] Parse social media data such as short videos using the natural language processing model ERNIE-3.0 to obtain semantic tag information;

[0064] Access city heat map API and traffic control information data.

[0065] Specifically, LiDAR and thermal imaging cameras are deployed to collect real-time pedestrian flow (accuracy of ±2 people / m2) and heat distribution (resolution of 0.1°C) around the advertising space. At the same time, the ERNIE-3.0 model is used to parse scene tags related to the advertising space on short video platforms such as Douyin and Bilibili to obtain semantic information. In addition, the city heat map API and traffic control information are connected to comprehensively collect various types of data that affect the traffic of the advertising space.

[0066] In step 200, encryption technology is used to process the collected data. After the original data is encrypted at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation.

[0067] Specifically, edge nodes are deployed locally at the advertiser, and homomorphic encryption technology is used to encrypt the original data;

[0068] After the original data is encrypted at the local node, cross-domain model aggregation is achieved through the FedProx algorithm, where the local update formula is:

[0069] ;

[0070] The global aggregation formula is:

[0071] ;

[0072] Where, Indicates the The local model is The parameters during round training, is the learning rate, It is The loss function of the local model, is a hyperparameter that controls how close the local model is to the global model. It is The global model parameters during the training round, is the number of local models participating in federated learning, It is The weight of each local model;

[0073] Among them, the FedProx algorithm adjusts the hyperparameters in the local model update formula when performing cross-domain model aggregation , which controls how close the local model is to the global model.

[0074] Specifically, edge nodes are deployed locally at advertisers, and homomorphic encryption (HE) technology is used to encrypt the original data to ensure the security of data during transmission and processing. Cross-domain model aggregation is achieved through the FedProx algorithm, which effectively reduces communication overhead while protecting data privacy, and reduces the amount of training data per round by 60%.

[0075] Step 300: Use a data enhancement algorithm to expand data diversity, use anomaly detection algorithm to identify abnormal data points, and then construct a multi-dimensional feature structure and perform feature conversion and dimensionality reduction on specific data.

[0076] Specifically, step 300 includes:

[0077] Step 310: Use the StyleGAN3 algorithm to generate virtual traffic samples. By adjusting the potential vectors in the generator, virtual traffic samples of different weather conditions and time periods are generated for data enhancement.

[0078] Step 320, using the IsolationForest algorithm to detect abnormal traffic points;

[0079] Step 330: construct a spatiotemporal feature cube and perform feature engineering on the social media tag data through TF-IDF conversion and PCA dimensionality reduction.

[0080] First, StyleGAN3 is used to generate virtual traffic samples for different weather conditions and time periods, such as simulating traffic conditions on rainy nights or early mornings during holidays, to expand data diversity. Furthermore, the IsolationForest algorithm is used to detect abnormal traffic points, allowing for timely detection of abnormalities such as traffic surges caused by unexpected parades.

[0081] Secondly, a spatiotemporal feature cube containing information such as longitude, latitude, time, thermal value, and semantic vector is constructed to comprehensively describe the spatiotemporal status of the ad space. For social media tags, they are first converted into sparse features through the TF-IDF algorithm, and then combined with the PCA algorithm to reduce the dimension to 64 dimensions for subsequent model processing.

[0082] In reality, ad traffic is affected by numerous factors, and real-world data may not capture all scenarios. Using StyleGAN3 to generate virtual traffic samples across different weather conditions and time periods effectively expands data diversity. StyleGAN3, based on the principles of generative adversarial networks (GANs), learns the distribution of real-world data through adversarial training of a generator and a discriminator. When generating virtual traffic samples, researchers can adjust the latent vectors in the generator to generate traffic data for different weather conditions (such as sunny, rainy, and snowy days) and different time periods (such as weekdays, weekend nights, and holidays). These virtual samples supplement the data for conditions that may be missing from the original data, enabling the model to learn more comprehensive traffic patterns and improve its generalization across various scenarios. For example, when training a model to predict outdoor ad traffic, generating virtual traffic samples under heavy rain allows the model to learn the patterns in the impact of inclement weather on traffic, enabling more accurate predictions in similar scenarios.

[0083] In addition, there are often outliers in the actual collected data. These outliers may be caused by various reasons, such as equipment failure, special events, etc. If they are not processed, they will interfere with the training and prediction results of the model. The IsolationForest algorithm is based on the concept of isolation forest. It isolates data points by constructing a binary tree. For normal data points, they are usually isolated at the bottom of the tree, while outliers are more likely to be isolated at the top of the tree. The algorithm determines whether it is an outlier by calculating the path length of each data point. In the processing of ad traffic data, by setting appropriate thresholds, the IsolationForest algorithm can quickly detect abnormal traffic points, such as a sudden surge in traffic for an ad spot during inactivity, or a sudden drop in traffic during regular periods. Timely discovery and processing of these outliers can ensure data quality and improve the accuracy and reliability of model predictions.

[0084] Step 400: quantify the causal impact of each data factor on ad space traffic based on the causal analysis framework, and use visualization technology to display feature contributions. Use a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad space traffic, and set up an adaptive mechanism to ensure prediction reliability.

[0085] Specifically, traffic forecasting is performed based on the CausalML framework and hybrid prediction model. The CausalML framework is used to quantify the causal effect of each factor on traffic, and the feature contribution is visualized through SHAP values.

[0086] The hybrid prediction model includes a spatiotemporal Transformer and a dynamic Bayesian network to capture the spatiotemporal characteristics and dynamic changes of ad slots. It also sets an adaptive gating mechanism to switch to the expert system when the prediction confidence falls below a set threshold.

[0087] The double difference method formula is used to quantify the causal effect of each factor on traffic:

[0088] ;

[0089] Where, represents the average treatment effect, and represent the flow observation values of the treatment group and the control group, respectively. is an indicator variable indicating whether it has been treated. =1 means it is processed. =0 means no processing.

[0090] Specifically, based on the CausalML framework, we deeply quantify the causal effects of various factors on traffic. For example, we analyze the impact of the opening of a subway line on the traffic of surrounding advertising spaces and conclude that it increases traffic by 18%. We visualize feature contributions using SHAP values to intuitively demonstrate the impact of each factor on traffic prediction. For example, the impact coefficient of "promotional activities" on traffic is +0.25.

[0091] Secondly, the spatiotemporal Transformer is used to capture the spatial dependencies between ad spaces. For example, the correlation coefficient of ad space traffic within a business district is calculated. A dynamic Bayesian network is introduced to update the weight of the impact of external events on traffic in real time. For example, during the "Gaokao" period, the traffic of ad spaces around schools dropped by 22%. An adaptive gating mechanism is set up. When the prediction confidence level is lower than 70%, it automatically switches to a rule-based expert system to ensure the reliability of the prediction.

[0092] Step 500: Use the proximal strategy optimization algorithm to train the advertising slot scheduling strategy, with maximizing advertiser revenue as the objective function. At the same time, build a digital twin model to simulate traffic changes under different strategies, and trigger the strategy rollback mechanism based on the prediction error.

[0093] Among them, the reward function formula of the proximal policy optimization algorithm is:

[0094] ;

[0095] Where, Indicates at time The reward value, 、 、 is the weight coefficient, It's time Click-through rate, It's time CPM, It's time Diversity metrics for advertising display;

[0096] The proximal strategy optimization algorithm is:

[0097] ;

[0098] This formula is used to update the policy parameters of the reinforcement learning model , where Indicates that in the strategy Under expectations, is the discount factor, is the time step, It's time Actions taken (e.g., advertising strategies), It's time status (such as ad space traffic, surrounding environment, etc.).

[0099] Among them, when the proximal strategy optimization algorithm trains the ad slot scheduling strategy, the strategy parameters are adjusted according to the reward function, and the weight coefficient in the reward function 、 、 Dynamically adjust based on advertisers’ business priorities and goals.

[0100] The construction of a digital twin model to simulate traffic changes under different strategies includes:

[0101] Based on the point cloud data collected by lidar and the thermal data collected by thermal imaging cameras, a 3D model of the area around the advertising space is constructed in combination with geographic information data. Traffic changes are simulated by inputting different advertising delivery strategy parameters.

[0102] Specifically, the Proximal Policy Optimization (PPO) algorithm is used to train ad placement scheduling strategies, with maximizing advertiser revenue as the objective function. Taking into account metrics such as cost per thousand impressions (CPM) and click-through rate (CTR), the algorithm outputs a prediction report with confidence intervals, such as "Friday evening peak traffic: 1200 ± 80 people, recommending increasing the rotation frequency," providing advertisers with a clear basis for decision-making.

[0103] Secondly, a 3D digital twin model of the ad space surrounding the ad slots was constructed to simulate traffic changes under different strategies. If the prediction error exceeded 25% for two consecutive hours, a policy rollback mechanism was automatically triggered to ensure the stability and effectiveness of the ad delivery strategy.

[0104] In this embodiment, the multimodal data collection method integrates multi-source information, comprehensively covers various factors that affect the traffic of advertising space, and provides a rich data foundation for accurate prediction. Federated edge computing, combined with homomorphic encryption technology, makes data "available but invisible," promoting multi-source data collaboration while protecting data privacy, reducing data transmission volume, and improving data processing efficiency and security. Data augmentation and anomaly detection technologies improve data quality and diversity, enhancing the model's adaptability to complex situations. Carefully designed feature engineering builds a comprehensive and efficient feature system, effectively extracting and utilizing data features, laying a solid foundation for accurate traffic forecasting. Causal analysis-based methods deeply explore the causal relationship between various factors and traffic, combined with hybrid prediction models to capture complex spatiotemporal characteristics and dynamic change patterns, significantly improving prediction accuracy. The adaptive gating mechanism ensures reliable prediction results even in complex situations, enhancing the stability and reliability of the system. Reinforcement learning strategies optimize ad scheduling with the goal of maximizing advertiser revenue, fully considering multiple key business indicators to achieve the rational allocation and efficient utilization of advertising resources. Digital twin models provide an effective means of strategy verification and optimization. By simulating traffic changes under different strategies, they provide strong support for decision-making, reduce invalid exposure, and improve advertising effectiveness.

[0105] Based on the above embodiments, Figure 2 As shown, the present invention also provides an advertisement space flow prediction system for supporting the advertisement space flow prediction method of the above embodiment, the advertisement space flow prediction system comprising:

[0106] The data collection and privacy protection module 11 is used to obtain multi-source data covering the physical environment around the ad space, semantic information, and external associations, and to process the collected data using encryption technology. After encrypting the original data at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation;

[0107] The data preprocessing module 12 is used to expand data diversity using a data enhancement algorithm, identify abnormal data points using an anomaly detection algorithm, and then construct a multidimensional feature structure and perform feature conversion and dimensionality reduction on specific data;

[0108] Prediction model design module 13, which is used to quantify the causal impact of various data factors on ad space traffic based on a causal analysis framework, demonstrate feature contributions with the help of visualization technology, and utilize a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad space traffic. It also establishes an adaptive mechanism to ensure prediction reliability.

[0109] The dynamic decision output module 14 is used to train the advertising slot scheduling strategy using the proximal strategy optimization algorithm, with maximizing advertiser revenue as the objective function. At the same time, a digital twin model is constructed to simulate traffic changes under different strategies, and a strategy rollback mechanism is triggered according to the prediction error.

[0110] In an optional embodiment, the data collection and privacy protection module 11 further includes:

[0111] The data collection unit is used to collect physical environmental data such as foot traffic and heat distribution around the advertising space; parse social media data through natural language processing models to obtain semantic label information; and obtain urban heat maps and traffic control information data;

[0112] The data fusion unit is used to deploy edge nodes locally at the advertiser and encrypt the original data using homomorphic encryption technology. After the original data is encrypted at the local node, cross-domain model aggregation is achieved through the FedProx algorithm.

[0113] In an optional embodiment, the data preprocessing module 12 further includes:

[0114] The data augmentation unit is used to generate virtual traffic samples using the StyleGAN3 algorithm. By adjusting the latent vectors in the generator, virtual traffic samples in different weather conditions and time periods are generated for data augmentation. The IsolationForest algorithm is also used to detect abnormal traffic points.

[0115] The feature engineering processing unit is used to construct a spatiotemporal feature cube and perform feature engineering processing on social media tag data through TF-IDF conversion and PCA dimensionality reduction.

[0116] In an optional embodiment, the prediction model design module 13 further includes:

[0117] The spatiotemporal causal unit is used to predict traffic based on the CausalML framework and hybrid prediction models. It uses the CausalML framework to quantify the causal effect of each factor on traffic and visualizes feature contributions through SHAP values.

[0118] The hybrid prediction unit, including the spatiotemporal Transformer and the dynamic Bayesian network, is used to capture the spatiotemporal characteristics and dynamic change patterns of ad slots, and set an adaptive gating mechanism to switch to the expert system when the prediction confidence falls below the set threshold.

[0119] In an optional embodiment, the dynamic decision output module 14 further includes:

[0120] The digital twin verification unit is used to build a 3D model of the area around the advertising space based on point cloud data collected by lidar and thermal data collected by thermal imaging cameras, combined with geographic information data, and simulate traffic changes by inputting different advertising delivery strategy parameters.

[0121] In this embodiment, the multimodal data collection method integrates multi-source information, comprehensively covers various factors that affect the traffic of advertising space, and provides a rich data foundation for accurate prediction. Federated edge computing, combined with homomorphic encryption technology, makes data "available but invisible," promoting multi-source data collaboration while protecting data privacy, reducing data transmission volume, and improving data processing efficiency and security. Data augmentation and anomaly detection technologies improve data quality and diversity, enhancing the model's adaptability to complex situations. Carefully designed feature engineering builds a comprehensive and efficient feature system, effectively extracting and utilizing data features, laying a solid foundation for accurate traffic forecasting. Causal analysis-based methods deeply explore the causal relationship between various factors and traffic, combined with hybrid prediction models to capture complex spatiotemporal characteristics and dynamic change patterns, significantly improving prediction accuracy. The adaptive gating mechanism ensures reliable prediction results even in complex situations, enhancing the stability and reliability of the system. Reinforcement learning strategies optimize ad scheduling with the goal of maximizing advertiser revenue, fully considering multiple key business indicators to achieve the rational allocation and efficient utilization of advertising resources. Digital twin models provide an effective means of strategy verification and optimization. By simulating traffic changes under different strategies, they provide strong support for decision-making, reduce invalid exposure, and improve advertising effectiveness.

[0122] Furthermore, the cache data processing device for intelligently analyzing the operating habits of investment users can run the cache data processing method for intelligently analyzing the operating habits of investment users. For specific implementation, please refer to the method embodiment, which will not be repeated here.

[0123] Based on the above embodiments, Figure 3 As shown, the present invention also provides an electronic device, the electronic device comprising:

[0124] At least one processor 22, at least one memory 21, a communication interface 23 and a communication bus 24, wherein the processor 22 is communicatively connected to the memory 21;

[0125] In this embodiment, the memory 21 can be implemented in any appropriate manner, for example, the memory 21 can be a read-only memory, a mechanical hard disk, a solid-state drive, or a USB flash drive, etc. The memory 21 is used to store at least one executable instruction executed by the processor;

[0126] In this embodiment, the processor 22 can be implemented in any appropriate manner. For example, the processor 22 can take the form of a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc.; the processor is used to execute the executable instructions to implement the cache data processing method for intelligently analyzing the operating habits of investment users as described above.

[0127] Based on the above embodiments, the present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it implements the cache data processing method for intelligently analyzing the investment user's operating habits as described above.

[0128] Those skilled in the art will appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0129] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, equipment and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0130] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or units can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or equipment, which can be electrical, mechanical or other forms.

[0131] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0132] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0133] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program instructions, such as a USB flash drive, a mobile hard disk, a read-only storage server, a random access storage server, a magnetic disk, or an optical disk.

[0134] In addition, it should be noted that the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0135] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many similar variations are possible. All variations directly derived from or associating with the present invention by those skilled in the art are intended to fall within the scope of protection of the present invention.

[0136] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting advertising space traffic, characterized in that: The following steps are involved: Acquire multi-source data covering the physical environment around the ad space, semantic information, and external associations; Use encryption technology to process the collected data. After encrypting the original data at the local node, use the federated learning algorithm to achieve cross-domain model aggregation. Use data enhancement algorithms to expand data diversity, use anomaly detection algorithms to identify abnormal data points, and then construct multi-dimensional feature structures and perform feature conversion and dimensionality reduction on specific data; Based on a causal analysis framework, we quantify the causal impact of various data factors on ad traffic, and use visualization technology to demonstrate feature contributions. We use a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad traffic, while also setting up an adaptive mechanism to ensure prediction reliability. A proximal policy optimization algorithm is used to train the ad slot scheduling strategy, with maximizing advertiser revenue as the objective function. At the same time, a digital twin model is constructed to simulate traffic changes under different strategies, and a policy rollback mechanism is triggered based on prediction errors.

2. The method according to claim 1, characterized in that The acquisition of multi-source data covering the physical environment surrounding the ad space, semantic information, and external associations includes: Collect physical environment data such as traffic flow and heat distribution around the advertising space; Parsing social media data through natural language processing models to obtain semantic label information; Obtain city heat maps and traffic control information data.

3. The method according to claim 1, characterized in that The encryption technology is used to process the collected data. After the original data is encrypted at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation, including: Deploy edge nodes locally at advertisers and use homomorphic encryption technology to encrypt original data; After the original data is encrypted at the local node, cross-domain model aggregation is achieved through the FedProx algorithm, where the local update formula is: ； The global aggregation formula is: ； Where, Indicates the The local model is Parameters during round training, is the learning rate, It is The loss function of the local model, is a hyperparameter that controls how close the local model is to the global model. It is The global model parameters during the training round, is the number of local models participating in federated learning, It is The weight of each local model; Among them, the FedProx algorithm adjusts the hyperparameters in the local model update formula when performing cross-domain model aggregation , which controls how close the local model is to the global model.

4. The method according to claim 1, wherein The data enhancement algorithm is used to expand data diversity, the anomaly detection algorithm is used to identify abnormal data points, and then a multi-dimensional feature structure is constructed and feature conversion and dimensionality reduction processing are performed on specific data, including: The StyleGAN3 algorithm is used to generate virtual traffic samples. By adjusting the potential vector in the generator, virtual traffic samples in different weather conditions and time periods are generated for data enhancement. Use IsolationForest algorithm to detect abnormal traffic points; Construct a spatiotemporal feature cube, and perform feature engineering on social media tag data through TF-IDF conversion and PCA dimensionality reduction.

5. The method according to claim 1, wherein The causal analysis framework quantifies the causal impact of various data factors on ad traffic, uses visualization technology to demonstrate feature contributions, and uses a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad traffic. At the same time, an adaptive mechanism is set up to ensure prediction reliability, including: Traffic forecasting is performed based on the CausalML framework and hybrid prediction model. The CausalML framework is used to quantify the causal effect of each factor on traffic, and the feature contribution is visualized through SHAP values. The hybrid prediction model includes a spatiotemporal Transformer and a dynamic Bayesian network to capture the spatiotemporal characteristics and dynamic changes of ad slots. It also sets an adaptive gating mechanism to switch to the expert system when the prediction confidence falls below a set threshold. The double difference method formula is used to quantify the causal effect of each factor on traffic: ； Where, represents the average treatment effect, and represent the flow observation values of the treatment group and the control group, respectively. is an indicator variable indicating whether it has been treated. =1 means it is processed. =0 means no processing.

6. The method according to claim 1, characterized in that The proximal strategy optimization algorithm is used to train the ad placement scheduling strategy, with maximizing advertiser revenue as the objective function: The reward function formula of the proximal policy optimization algorithm is: ； Where, Indicates at time The reward value, 、 、 is the weight coefficient, It's time Click-through rate, It's time CPM, It's time Diversity metrics for advertising display; The proximal strategy optimization algorithm is: ； Where, Indicates that in the strategy Under expectations, is the discount factor, is the time step, It's time Actions taken, It's time status.

7. The method according to claim 6, characterized in that When the proximal strategy optimization algorithm trains the advertising slot scheduling strategy, the strategy parameters are adjusted according to the reward function, and the weight coefficient in the reward function 、 、 Dynamically adjust based on advertisers’ business priorities and goals.

8. An advertising space traffic prediction system, characterized in that: include: The data collection and privacy protection module is used to obtain multi-source data covering the physical environment around the ad space, semantic information, and external associations, and to process the collected data using encryption technology. After encrypting the original data at the local node, a federated learning algorithm is used to achieve cross-domain model aggregation; The data preprocessing module is used to expand data diversity using data enhancement algorithms, identify abnormal data points using anomaly detection algorithms, and then construct a multidimensional feature structure and perform feature conversion and dimensionality reduction on specific data; The prediction model design module is used to quantify the causal impact of various data factors on ad traffic based on a causal analysis framework, and uses visualization technology to demonstrate feature contributions. It uses a fusion neural network hybrid model to capture the spatiotemporal characteristics and dynamic changes of ad traffic, and also sets up an adaptive mechanism to ensure prediction reliability. The dynamic decision output module is used to train the ad slot scheduling strategy using the proximal policy optimization algorithm, with maximizing advertiser revenue as the objective function. At the same time, it builds a digital twin model to simulate traffic changes under different strategies and triggers the strategy rollback mechanism based on the prediction error.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory, wherein the memory is communicatively connected to the processor; The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the advertising space traffic prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the advertising space traffic prediction method according to any one of claims 1 to 7 is implemented.

Images