Operation method of power grid online channel based on multi-dimensional fusion identity and flywheel fission

Abstract

The invention provides a power grid online channel operation method based on multi-dimensional fusion identity and flywheel fission, and the method comprises the steps: obtaining the multi-dimensional identity information of a user, carrying out the fusion processing of the multi-dimensional identity information, and forming a unified user identity identifier; establishing a user tag system containing industry exclusive tags, and carrying out real-time circling selection on users based on the user tag system and a preset circling selection rule to generate a target user group; and initiating a marketing activity to the target user group. Executing the marketing activity on the basis of a multi-level cache mechanism, performing real-time acquisition, analysis and visual display on full-life-cycle data of the marketing activity, and performing iterative optimization on a strategy of the marketing activity according to an analysis result; therefore, the user identification accuracy can be improved, the marketing strategy can be dynamically optimized, the fission conversion efficiency can be improved, and the system response capability can be enhanced.

Description

Technical Field

[0001] This invention relates to the field of digital operation technology for power grid enterprises, and in particular to an operation method for online channels of power grids based on multi-dimensional fusion identity and flywheel fission. Background Technology

[0002] With the widespread adoption of online service portals such as provincial power grid companies' apps, mini-programs, and official accounts, traditional user operation methods are facing numerous challenges.

[0003] In existing technologies, user identity information is scattered across multiple systems (such as the State Grid website, marketing systems, and third-party login systems), lacking a unified identity fusion mechanism. This leads to inaccurate user identification and wasted marketing resources. Furthermore, existing operational methods typically employ static user segmentation and push strategies, failing to achieve precise targeting based on real-time behavior and scenarios. This results in high new user acquisition costs, low user activity, and poor conversion rates.

[0004] On the other hand, existing systems often suffer from problems such as system response delays, difficulties in fund reconciliation, and weak risk control capabilities in high-concurrency marketing campaigns, making it difficult to support the digital operation needs of provincial-level scale. At the same time, due to the lack of a unified strategy engine and multi-level caching mechanism, campaign launch cycles are long, iteration efficiency is low, and it is unable to quickly respond to market changes and user needs.

[0005] Therefore, how to improve the accuracy of user identification, dynamically optimize marketing strategies, enhance the efficiency of viral conversion, and strengthen the system's responsiveness are technical problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0006] This invention provides an operation method for online channels in the power grid based on multi-dimensional fusion identity and flywheel fission, which can improve user identification accuracy, dynamically optimize marketing strategies, improve fission conversion efficiency and enhance system responsiveness.

[0007] On one hand, this invention provides an operation method for an online channel in a power grid based on multi-dimensional fusion identity and flywheel fission, comprising: The user's multidimensional identity information is obtained, and the multidimensional identity information is fused to form a unified user identity identifier. Establish a user tagging system that includes industry-specific tags, and based on the user tagging system and preset selection rules, select users in real time to generate target user groups; A marketing campaign is launched to the target user group; wherein the marketing campaign is used to convert at least one target user in the target user group into an online user, and to enable the online user to share tasks based on a flywheel fission model, so as to convert the sharing user into an online user; the flywheel fission model includes dynamically adjusted task thresholds and dynamically adjusted reward strategies; Based on a multi-level caching mechanism, the marketing campaign is executed, and the full lifecycle data of the marketing campaign is collected, analyzed, and visualized in real time. The strategy of the marketing campaign is iteratively optimized based on the analysis results. The multi-level caching mechanism includes edge node caching, central node caching, and persistent storage caching.

[0008] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the operation method of the power grid online channel based on multidimensional fusion identity and flywheel fission as described above.

[0009] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the operation method of the power grid online channel based on multidimensional fusion identity and flywheel fission as described above.

[0010] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the operation method of the power grid online channel based on multidimensional fusion identity and flywheel fission as described above.

[0011] The present invention provides an operation method for online channels of the power grid based on multi-dimensional fusion identity and flywheel fission. By integrating multi-dimensional identity information to establish a unified identifier, target users are selected in real time and dynamic fission strategies are implemented. Combined with a multi-level caching mechanism, efficient operation is achieved. This can improve the accuracy of user identification, dynamically optimize marketing strategies, improve fission conversion efficiency, and enhance system responsiveness. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the operation method of an online channel on the power grid based on multi-dimensional fused identity and flywheel fission provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the online channel operation system based on multi-dimensional fusion identity and flywheel fission provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0015] In existing technologies, power grid companies face challenges in online channel operations, including fragmented user identities, rigid marketing strategies, and insufficient system responsiveness. Traditional methods rely on static user segmentation and fixed push strategies, making precise targeting difficult. High-concurrency scenarios often result in data delays and difficulties in reconciliation. Furthermore, the lack of dynamic adjustment capabilities in referral mechanisms leads to low conversion rates. For example, when a provincial power grid company launched an electricity discount campaign, 30% of its marketing resources were wasted due to incorrect user identification, and system crashes during peak periods resulted in lost orders.

[0016] To address the aforementioned issues, research revealed that the core problem lies in the lack of an identity fusion mechanism and insufficient dynamic strategy adjustment capabilities. Analysis showed that user behavior data is scattered across different systems, forming information silos, and the static tagging system cannot reflect real-time changes in demand. To solve the identity fusion problem, a cross-system data association method was proposed; a real-time selection algorithm was designed to meet dynamic strategy requirements; and a multi-level caching architecture was constructed to improve system stability. After multiple experimental verifications, a complete technical solution integrating identity management, intelligent grouping, dynamic fission, and high concurrency support was finally formed.

[0017] Therefore, this application proposes an operational method that involves acquiring multi-dimensional user identity information, integrating and processing it to form a unified identifier, establishing an industry-specific tag system to select target user groups in real time, launching marketing campaigns to convert target users into online users and sharing them through a flywheel fission model, executing campaigns based on a multi-level caching mechanism, and collecting and analyzing data in real time for iterative optimization.

[0018] Specifically, Figure 1 This is a flowchart illustrating the operation method of an online channel on the power grid based on multi-dimensional fusion identity and flywheel fission, as provided in an embodiment of the present invention.

[0019] like Figure 1As shown in the embodiments of the present invention, the execution subject of the operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission can be an electronic device, and the method mainly includes the following steps: 101. Obtain the user's multi-dimensional identity information, and perform fusion processing on the multi-dimensional identity information to form a unified user identity identifier; 102. Establish a user tagging system that includes industry-specific tags. Based on the user tagging system and preset selection rules, select users in real time to generate target user groups. 103. Launch a marketing campaign to the target user group; wherein the marketing campaign is used to convert at least one target user in the target user group into an online user, and to enable the online user to share tasks based on the flywheel fission model, so as to convert the sharing user into an online user; wherein the flywheel fission model includes dynamically adjusted task thresholds and dynamically adjusted reward strategies; 104. Based on a multi-level caching mechanism, the marketing campaign is executed, and the full lifecycle data of the marketing campaign is collected, analyzed, and visualized in real time. The strategy of the marketing campaign is iteratively optimized based on the analysis results. The multi-level caching mechanism includes edge node caching, central node caching, and persistent storage caching.

[0020] In a specific implementation, multi-dimensional identity information fusion refers to associating user data from different systems through hash algorithms and confidence assessments. For example, it involves bidirectional matching between the account number system and the mobile phone number system, generating a unique identity identifier when the matching confidence exceeds a preset threshold. The user tagging system uses clustering algorithms to extract features from historical behavioral data, forming a dynamic tag library containing industry attributes. For example, it generates over 80 tags, such as peak-valley electricity price sensitivity tags, photovoltaic user tags, and charging pile potential value tags, based on electricity consumption behavior analysis. The flywheel fission model dynamically adjusts task thresholds and reward strategies, for example, based on the number of participants, task completion volume, and the proportion of high-risk users. A multi-level caching mechanism employs edge computing nodes working collaboratively with the central server, for example, deploying a Redis cluster in a provincial power grid data center to achieve high-speed access to activity data.

[0021] Specifically, the system first integrates basic user information (such as account number and mobile phone number), electricity usage records (such as payment records and peak / valley electricity usage data), and equipment information (such as associated equipment identifiers) through data cleaning and feature extraction modules. A distributed computing framework is then used to complete identity fusion. For example, a hash algorithm is employed to perform cross-system association on the preprocessed multi-dimensional identity information (such as bidirectional matching between the account number system and the mobile phone number system). Simultaneously, a confidence assessment model is used to calculate the confidence level of the association results. When the confidence level exceeds a preset threshold (confidence level > 0.92), a unique user identity identifier is generated, achieving the normalization and fusion of user data from different systems.

[0022] Subsequently, clustering algorithms are used to extract features from users' historical behavioral data (such as electricity consumption habits and channel access records) to construct a dynamic user tag library containing exclusive tags for the power industry. These exclusive tags include more than 80 types, such as peak-valley price difference sensitivity, photovoltaic user identification, and charging pile potential value. The tag library is dynamically updated based on real-time data streams. Based on the dynamic user tag library and preset selection rules, a rule engine is used to filter users in real time to generate target user groups that meet the needs of marketing activities. The response delay of the filtering process does not exceed 10 minutes.

[0023] The marketing campaign execution phase can be based on a multi-level caching mechanism. This mechanism employs an architecture where edge computing nodes and a central server work together. Specifically, it includes: edge node caching (storing static campaign resources such as posters and rule descriptions), central node caching (deploying a Redis cluster in the provincial power grid data center to achieve high-speed access to dynamic campaign data, such as user participation status and real-time task parameters), and persistent storage caching (storing all historical campaign data, such as user behavior logs and financial statements). This three-level caching design can increase the system's peak request capacity by more than three times. A streaming computing platform can be used to collect and analyze the entire lifecycle data of the marketing campaign in real time, displaying conversion funnels (such as conversion rates at each stage of registration, binding, payment, and sharing) and abnormal indicators (such as the proportion of risky users and data discrepancy rate) through a visual dashboard. The strategy optimization module automatically generates marketing campaign strategy adjustment suggestions based on the analysis results, enabling iterative optimization of task thresholds, reward strategies, and push content.

[0024] This solution achieves precise incentives through a dynamic adjustment mechanism; furthermore, its multi-level caching design significantly enhances its peak request handling capacity. Through these technical solutions, this application effectively addresses the resource waste caused by inaccurate user identification, enabling precise marketing based on real-time behavioral data. The dynamic adjustment mechanism significantly improves viral conversion efficiency, and the multi-level caching architecture ensures system stability under high concurrency scenarios, providing reliable technical support for the online operation of provincial power grid enterprises.

[0025] In some embodiments, this application further proposes a marketing campaign based on a target user group, comprising the following steps: matching marketing content to different target users based on the tags of each target user in the target user group; acquiring external scenario data, and if it is determined that the external scenario data meets preset push conditions, selecting a push channel based on the channel preference data of each target user; and pushing the respective marketing content to each target user through the push channel.

[0026] In a specific implementation process, an external data interface can be called through the strategy engine to obtain external scene data. The external data interface includes a meteorological API, and the external scene data includes meteorological warning information. It is determined whether the external scene data meets a preset trigger condition, which is to obtain a high temperature orange warning issued by a provincial meteorological observatory. When the external scene data meets the preset trigger condition, the current time or a preset time is determined as the push timing.

[0027] In a specific implementation process, user channel preference data can be obtained from user identity identifiers. The channel preference data includes the user's commonly used push channels, including WeChat service notifications and SMS. The user group type can be determined based on the user's age tag, which is obtained from the user's multidimensional identity information or behavioral data. For users belonging to the elderly user group, SMS is selected as the push channel, and for users belonging to the young user group, WeChat service notifications are selected as the push channel.

[0028] This solution integrates industry-specific tags with real-time scenario data to establish a dynamic triggering mechanism, achieving precise matching between marketing campaigns and user needs. Simultaneously, a tiered push strategy based on age group segmentation and channel preference data effectively addresses the low reach rate of elderly users encountered in traditional methods. Furthermore, through the aforementioned technical solutions, this application enables dynamic adaptation of marketing content to user needs, improving campaign response rates and conversion efficiency. By combining a real-time scenario data triggering mechanism, it ensures that marketing campaigns reach target users at the optimal time. A differentiated channel selection strategy based on user group characteristics significantly improves the reach to users of different age groups, reducing the waste of marketing resources.

[0029] In a specific implementation process, when matching marketing content to different target users based on the tags of each target user in the target user group, the demand type of each target user can be determined according to the tags of each target user and divided into multiple user groups with the same demand type. For different target users in any user group with the same demand type, the marketing content for each target user in the user group is matched according to the preset matching rules. The marketing content for each target user is obtained by traversing all user groups with the same demand type. The preset matching rules are to select a preset proportion of sample users from the user group, wherein the user group is clustered into at least 3 subgroups. The dimensions of subgroup clustering include user age, channel preference, and historical marketing activity participation. Sample users are selected from each subgroup to form the total sample set according to the stratified sampling rules. This includes a total sample size of 5%-10% of the user group size, with the sample size for each subgroup equal to the total sample size multiplied by the proportion of that subgroup in the user group, excluding users who have participated in similar marketing tests within the past 30 days; pushing multiple candidate content to sample users and collecting the click-through rate and open rate of each candidate content; when the click-through rate of at least one candidate content is greater than a preset click-through rate threshold or the open rate is greater than a preset open rate threshold, at least one candidate content is used as the marketing content for that user group; if the conditions are not met, the performance data of the failed content is recorded and the reasons for inefficiency are analyzed; using content generation tools, based on the results of the reason analysis, the failed content is iteratively adjusted or multiple sets of differentiated marketing content are regenerated; the effectiveness of the adjusted content or newly generated content is verified again, and when the verification data meets the standards, it is pushed to the target user group.

[0030] Among them, the user group with similar needs refers to the set of users with the same core needs characteristics, filtered through a user tagging system. Specifically, clustering algorithms combined with business rules can be used for segmentation. For example, users with the tag combination of "high potential for charging piles + age 25-35" can be grouped into the same group. This segmentation method can accurately identify common user needs, providing a foundation for subsequent differentiated marketing.

[0031] Subgroup clustering refers to further subdividing users into smaller subgroups within a user group with similar needs. This can be achieved using the K-means algorithm combined with factors such as user age and channel preferences. This step helps avoid marketing strategy biases caused by differences within user groups and improves the representativeness of the sample selection.

[0032] Stratified sampling, in particular, involves allocating sample sizes according to subgroup proportions and excluding users who have recently participated in similar tests. This can be achieved using a proportional allocation algorithm combined with a user participation record database. This rule ensures that the sample covers the characteristics of each subgroup while avoiding interference from repeated testing.

[0033] Effectiveness verification refers to evaluating the effectiveness of marketing content through gray-scale testing, which can be achieved using an A / B testing framework combined with a real-time data monitoring system. This verification mechanism can quickly identify high-conversion content and reduce the waste of resources from ineffective push notifications.

[0034] Specifically, the implementation process begins by dividing target users into multiple user groups with similar needs based on a user tagging system. For example, in the context of promoting charging pile services, users tagged with "high charging pile potential" are grouped into the same user group. Within each user group, sub-clusters are then formed based on dimensions such as age and channel preference, creating at least three subgroups. A test group is formed by sampling users from each subgroup using stratified sampling rules, excluding users who have recently participated in similar activities to avoid data contamination. Multiple candidate marketing content, such as charging discount packages A, B, and C, are pushed to the test group, and the click-through rate and open rate of each package are monitored in real time. If a package reaches a preset threshold, it is selected directly; otherwise, the reasons for inefficiency are analyzed, and an optimized version is generated using AIGC tools. For example, if the copy lacks appeal, a new version including a dynamic electricity price comparison chart can be automatically generated. The optimized content undergoes further small-scale validation, and only after confirming the effectiveness is met can it be officially pushed to the entire user group.

[0035] This solution employs a combined sample selection mechanism of subgroup clustering and stratified sampling to accurately capture the differences in characteristics among segmented groups. Combined with dynamic verification and iterative optimization processes, it effectively improves the matching accuracy between marketing content and user needs. Furthermore, through the aforementioned technical solution, this application addresses the problems of low marketing content matching accuracy and high invalid push rates in existing technologies caused by differences in characteristics within user groups. The combination of stratified sampling and dynamic verification mechanisms can accurately identify the differences in content preferences among different segmented groups while ensuring testing efficiency. Simultaneously, the use of automated content generation and iterative optimization technologies significantly shortens the adjustment cycle of marketing strategies, effectively reducing operating costs and improving user conversion rates.

[0036] Furthermore, the aforementioned "pushing multiple candidate contents to the sample users" may include obtaining historical participation behavior data of existing users corresponding to any user group with the same demand type; determining the content of interest of the existing users based on the historical participation behavior data; selecting the multiple candidate contents from the content of interest of the existing users; and pushing the multiple candidate contents to the sample users.

[0037] Among them, historical participation behavior data refers to the operation records generated by existing users in the past participation in marketing activities. Specifically, it can be realized by user clicks, dwell time, number of shares, and completion or abandonment behavior logs, which are used to explore users' true preferences.

[0038] Atomized parsing refers to breaking down user behavior into the smallest analyzable units. This can be achieved by using natural language processing techniques to segment logs, semantically annotate them, and store them in a structured manner, facilitating accurate identification of behavioral features. Behavioral atoms refer to the smallest semantic units of user behavior. This can be achieved by using encoding rules to combine and annotate operation types, object characteristics, scene parameters, and emotional feedback, forming a quantifiable behavioral gene.

[0039] Among them, the user behavior gene map refers to the network of associations of user behavior features. Specifically, it can be implemented by storing multidimensional relationships between users, behaviors, and content using a graph database, and by mining implicit interest associations through graph algorithms.

[0040] Among them, the dynamic interest model refers to the feature set that reflects the real-time changes in users' interests. Specifically, it can be achieved by using time series analysis to detect abrupt changes in interest and combining it with machine learning models for dynamic updates, thus ensuring the timeliness of interest capture.

[0041] Specifically, when determining candidate marketing content, the system first retrieves historical engagement data for the target user group and decomposes it into behavioral atoms containing operation type, object characteristics, scene parameters, and emotional feedback through atomized parsing. A user behavior gene map is constructed based on a graph neural network to identify correlation patterns between different behavioral atoms. Entropy calculations detect user interest mutation points, triggering interest model reconstruction and generating a dynamic interest model reflecting current preferences. A generative adversarial network is then used to intelligently recombine the content gene library, outputting candidate content that matches the dynamic interests. For example, when it detects that a user group frequently clicks on electricity discount content during high-temperature warning scenarios, the system automatically generates a combination of content including tiered electricity pricing explanations and red envelope incentives as a candidate.

[0042] This solution achieves precise modeling of user interests through behavioral atomic analysis and genomic mapping. Combined with a generative adversarial network (GAN) content reorganization mechanism, it effectively solves the problem of mismatch between candidate content and real-time user needs. Furthermore, through the aforementioned technical solution, this application can intelligently filter candidate content based on user behavioral genes, improving the matching accuracy between marketing content and target user needs. By dynamically capturing changes in interests, it avoids push notification failures caused by shifts in user preferences. Utilizing a generative content reorganization mechanism, it automatically generates diverse candidate solutions while ensuring compliance, significantly reducing manual planning costs and improving testing efficiency.

[0043] Furthermore, the aforementioned "determining the content of interest of existing users based on the historical participation behavior data" includes: performing atomized parsing on the historical participation behavior data to obtain behavioral atoms containing operation type, object characteristics, scene parameters, and emotional feedback; constructing a ternary relationship graph of existing users, behavioral characteristics, and business content based on the behavioral atoms, mining implicit associations between behavioral atoms through graph neural networks to generate a user behavior gene graph; calculating entropy values ​​on the user behavior gene graph, detecting interest mutation points in the user behavior sequence, triggering dynamic reconstruction of the user interest model based on the interest mutation points to obtain a dynamic interest model reflecting the current interest characteristics of existing users; constructing a content gene library containing topic genes, form genes, and incentive genes, using the dynamic interest model as input parameters through a generative adversarial network to recombine gene fragments in the content gene library to generate personalized content matching the user's current interest characteristics as the content of interest of the existing users.

[0044] Atomized parsing refers to the process of breaking down user behavior data into its smallest behavioral units and encoding them. Specifically, natural language processing techniques can be used to perform semantic segmentation and structuring of log data, associating operation types such as clicks, dwell times, and shares with their corresponding business themes and time scenarios. This behavior atomic encoding mechanism enables the standardized expression of user behavior characteristics in the power industry, providing a unified data foundation for subsequent graph construction.

[0045] Among them, the ternary relationship graph refers to a graph data structure that represents the multidimensional relationships between users, behavioral characteristics, and business content. Specifically, it can be stored and computed using the Neo4j graph database. By using the embedding learning algorithm of graph neural networks, commonalities in user behavior patterns in different scenarios can be mined. For example, it can be found that photovoltaic users pay more attention to content related to electricity fee bonuses after the release of electricity price policies, providing implicit association support for the construction of interest models.

[0046] The dynamic interest model reconstruction refers to the calculation process of updating interest features in real time based on the abrupt changes in user behavior sequences. Specifically, a sliding window mechanism combined with KL divergence can be used to detect abrupt change points in interest. When it is detected that a user's dwell time on the charging pile service page exceeds twice the historical average for three consecutive times, the model reconstruction mechanism is triggered to update the user's preference intensity parameters, ensuring that the interest model remains synchronized with the user's current needs.

[0047] Content gene pool recombination refers to a content generation method based on generative adversarial networks, specifically using the StyleGAN framework to optimize the combination of gene fragments. For example, combining charging station discount theme genes with video tutorial format genes and points reward incentive genes can generate personalized content that includes charging station usage tutorial videos and points redemption rules, achieving precise matching between marketing materials and user interest characteristics.

[0048] Specifically, this method first performs atomic parsing of user historical behavior logs, transforming user clicks on the electricity bill payment page and dwell time on the photovoltaic service channel into behavioral atoms with timestamps and scene tags. Next, a graph neural network is used to analyze the co-occurrence relationships between these behavioral atoms, revealing a periodic focus on electricity bill bonus content by elderly users during hot weather. When entropy calculation detects that a user has frequently visited the charging pile service page in the morning for three consecutive days, the interest model reconstruction module is immediately triggered to update the user's scene adaptation requirements. Finally, a generative adversarial network is used to match the updated interest model with a content gene library, generating personalized content including charging pile discount videos and points reward rules. After A / B testing to verify the content's effectiveness, it is pushed to the target users.

[0049] This solution constructs a dynamically updated user behavior gene map, enabling the identification of implicit interest associations that are difficult to discover using traditional methods. For example, it reveals an increased level of interest among elderly users in home electricity safety content during the time their children are home. Furthermore, while existing content generation technologies largely rely on manual template combinations, this solution employs generative adversarial networks to intelligently recombine content elements, automatically generating marketing materials that align with real-time changes in user interests. Through these technical solutions, this application addresses the technical problems of coarse-grained behavioral feature extraction and insufficient latent association mining in traditional user interest analysis, achieving dynamic capture and precise modeling of user interest features. This method effectively improves the matching accuracy between marketing content and user needs, reduces resource waste caused by ineffective content pushes, and ensures the timeliness of content recommendations through real-time updated interest models, providing more personalized services tailored to the actual needs of users in the power industry.

[0050] Furthermore, the aforementioned "using the dynamic interest model as input parameters through a generative adversarial network to recombine gene fragments in the content gene library to generate personalized content matching the user's current interest characteristics" includes: decomposing business content into independently recombinable gene fragments, wherein the theme gene includes electricity bill payment, photovoltaic services, charging pile services, smart grid and electricity safety; the form gene includes graphic and text display, video explanation, H5 interaction and live Q&A; the incentive gene includes electricity bill red envelopes, discount coupons, free service time and points rewards; inputting the dynamic interest model into the generator of the generative adversarial network, so that the generator selects matching gene fragments from the content gene library based on the current interest characteristics for targeted combination to generate initial candidate content; evaluating the matching degree between the initial candidate content and the current interest characteristics through the discriminator of the adversarial network, and iteratively optimizing the generator parameters based on the evaluation results until personalized candidate content that meets the preset matching degree threshold is generated; wherein, the matching degree threshold is dynamically adjusted according to the intensity of the user's interest mutation, wherein the more significant the interest mutation, the lower the threshold is set to retain more exploratory content.

[0051] In a specific implementation, generative adversarial networks refer to a deep learning framework that includes a generator and a discriminator. Specifically, it can be implemented using a generator based on the Transformer architecture and a discriminator based on a convolutional neural network. The generator is responsible for recombining content gene fragments according to user interest features, and the discriminator is responsible for evaluating the matching degree between the generated content and user needs.

[0052] Among them, the content gene library refers to breaking down business content into gene units that can be independently combined. Specifically, knowledge graph technology can be used to structurally decompose power business content to form a gene fragment library with three dimensions: theme, form, and incentive. Each gene fragment carries a unique metadata tag for the power industry.

[0053] Among them, the dynamic interest model refers to the feature vector that reflects the real-time changes in users' interests. Specifically, it can be constructed using a time-series-based user behavior encoder and attention mechanism, and the model can be dynamically updated by capturing the interest mutation points in the user behavior sequence.

[0054] Among them, the dynamic adjustment of the matching threshold refers to the adaptive adjustment of the content selection criteria based on the stability of user interests. Specifically, a sliding window can be used to count the variance of user interest fluctuations. When the variance exceeds the preset range, the matching threshold is automatically reduced, allowing more exploratory content to be generated to capture potential interests.

[0055] Specifically, the working principle of this technical solution is as follows: First, the content of power business is decomposed into gene fragments in three dimensions: theme, form, and incentive, forming a structured content gene library. When personalized marketing content needs to be generated, the dynamically updated interest model is input into the generator module of the generative adversarial network. Based on the user's current interest characteristics, the generator selects relevant gene fragments from the gene library and combines them. For example, the "photovoltaic service" theme gene is combined with the "video explanation" form gene and the "free service duration" incentive gene to generate initial candidate content. The discriminator module then evaluates the degree of matching between the generated content and the interest characteristics by analyzing the user's historical behavior patterns. When the matching degree does not meet the standard, the system automatically adjusts the generator parameters and recombines the gene fragments until personalized content that meets the dynamic threshold requirements is generated. When the user's interests change significantly, the system allows the generation of more diverse candidate content by lowering the matching degree threshold, avoiding content rigidity caused by the lag in the interest model.

[0056] This solution ensures that generated content complies with industry standards by constructing a content gene library specifically for the power industry; it employs a dynamic threshold adjustment mechanism to maintain content relevance while preserving room for exploration; and it combines adversarial training to effectively balance content innovation with user demand matching. Through these technical solutions, this application enables intelligent content generation in power industry scenarios, significantly improving the accuracy of matching marketing content with real-time user needs. The gene recombination mechanism ensures content compliance, preventing the generation of content that does not comply with power business rules. The dynamic threshold adjustment strategy maintains core user engagement while effectively mining the interests and preferences of potential user groups, improving the coverage of marketing activities to diverse user groups. The adversarial training mechanism of the generative adversarial network ensures continuous optimization of the content generation process, continuously improving the quality and efficiency of content production.

[0057] Furthermore, the aforementioned "selecting multiple candidate contents from the existing users' content of interest" includes: constructing a dynamic scene factor library, which includes time dimension factors, spatial dimension factors, and social dimension factors; calculating the matching degree between the personalized candidate contents and the dynamic scene factors in the dynamic scene factor library based on a scene and content matching degree algorithm, and filtering out multiple candidate contents that are suitable for the current scene.

[0058] In a specific implementation, the dynamic scenario factor library refers to a database used to store influencing factors that are in real-time related to the user's environment. This can be implemented using a distributed database combined with a streaming computing framework, dynamically updating factor status by accessing external data sources such as meteorological data, geographic location information, and policy announcements in real time. Time-dimensional factors include periodic or sudden time tags such as holidays, seasons, and peak electricity consumption periods; for example, automatically marking the summer high-temperature period as "peak summer season." Spatial-dimensional factors include the user's geographic location, power grid load level, and charging pile distribution density; for example, prioritizing charging discount content for users in areas with dense charging piles. Social-dimensional factors include policy and regulatory changes, social hot topics, and industry dynamics; for example, when a local government releases a new energy vehicle subsidy policy, the matching weight of related marketing content is increased.

[0059] Scene and content matching algorithms refer to computational models used to quantify the suitability of candidate content with the current scene. Specifically, they can be implemented using a vector space model based on cosine similarity or a multimodal matching model based on deep learning. This involves calculating the similarity between the feature vectors of candidate content and the encoded vectors of dynamic scene factors, outputting a matching score. For example, when a "yellow rainstorm warning" scene is detected, the algorithm automatically increases the matching weight of content such as "electricity safety reminders," ensuring that the pushed content highly matches the user's real-time needs.

[0060] Specifically, the dynamic scene factor library obtains the latest scene information in real time by accessing external data interfaces, such as real-time weather data returned by meteorological APIs, policy documents released by public data platforms, and geofencing information provided by LBS services. These factors are standardized and stored in the factor library, and dynamically weighted according to preset rules. When it is necessary to filter candidate content, the scene and content matching algorithm extracts the currently effective scene factors from the factor library, aligns them with the metadata of the candidate content, and calculates the matching score of each content in different dimensions. For example, for the "winter cold wave" scene, the algorithm will prioritize filtering content containing themes such as "electric heating discounts" and "freezing prevention electricity usage guidelines," while also performing spatial dimension weighting calculations based on whether the user's area belongs to a heating renovation pilot area. Finally, the system selects the top N candidate contents based on the comprehensive matching score ranking results and puts them into the push queue, ensuring that marketing activities keep pace with real-time scene changes.

[0061] This solution constructs a dynamic scene factor library and introduces a matching algorithm to automatically perceive the spatiotemporal characteristics and social environment changes of the user's current scenario, enabling dynamic optimization of content strategies. For example, traditional methods might continue to push regular marketing content during sudden meteorological disasters, while this solution can immediately switch to emergency power guidance information, significantly improving the practicality of the content and the perceived value to users. Furthermore, through the above technical solution, this application effectively solves the problem of delayed content delivery caused by the lack of scene perception in existing technologies, improving the accuracy of matching candidate content with real-time scenarios. By dynamically matching changes in the external environment, it avoids the waste of resources caused by pushing inefficient content in incompatible scenarios, while enhancing users' willingness to interact and conversion efficiency in specific scenarios, providing real-time and scenario-based content supply guarantees for high-concurrency marketing activities.

[0062] In some embodiments, this application further proposes a process for adjusting task thresholds and reward strategies, including setting an initial task threshold and an initial reward amount; acquiring first participation data of the marketing activity in real time, the first participation data including the number of participants, task completion volume, and the proportion of risky users; wherein, risky users refer to participants who have engaged in fraudulent activities, have fake account bindings, have abnormal device IPs, or have historical violation records, and their data is invalid participation data that needs to be removed in real time; adjusting the initial task threshold based on the first participation data to obtain a dynamically adjusted task threshold; wherein, the adjustment amounts corresponding to the number of participants, task completion volume, and the proportion of risky users are weighted and rounded to adjust the task threshold; and calculating the reward amount using a decreasing function, the decreasing function being... ,in Where t is the initial reward amount, and t is the time after the event starts. The adjustment coefficient λ is adjusted based on the first participating data.

[0063] The task threshold refers to the minimum indicator value that users need to achieve to complete the viral task. Specifically, it can be implemented using a dynamic calculation model based on real-time participation data. By adjusting the weighted calculation of the number of participants, the amount of tasks completed, and the proportion of risky users, the threshold can be dynamically optimized according to the activity's performance.

[0064] The reward strategy refers to the incentive distribution rules that are dynamically adjusted based on time decay and activity results. Specifically, it can be implemented by using an exponential decay function combined with a risk factor weighting algorithm. By adjusting the coefficient λ, the reward decay rate can be controlled, balancing activity costs and user participation.

[0065] Among them, the first participation data refers to the set of core indicators of marketing activities collected in real time. Specifically, it can be achieved through data collection by tracking points and linkage with the risk control engine to remove user data with abnormal behavior in real time and ensure data validity.

[0066] Specifically, during the execution of the marketing campaign, the initial task threshold is set to a fixed value, such as a daily sharing task requirement. Once the campaign begins, data on the number of participants, task completion rate, and the percentage of high-risk users are collected in real time. For example, if the percentage of high-risk users exceeds 5%, a threshold adjustment mechanism is triggered. A comprehensive adjustment coefficient is calculated by weighting the adjustments of these three metrics according to preset weights, such as 40% for participants, 35% for task completion, and 25% for high-risk user percentage. The initial task threshold is then adjusted in integer multiples based on this coefficient; for example, if the comprehensive adjustment coefficient is +1.2, the task threshold is increased by 10%. Simultaneously, the reward amount decreases exponentially over time. For example, if the initial reward amount is 10 yuan, the adjustment coefficient λ dynamically changes based on real-time participation data. When participation is high, the value of λ is appropriately increased to accelerate reward decay and balance campaign costs.

[0067] This solution establishes a dynamic adjustment model based on multi-dimensional participation data to achieve synergistic optimization of task thresholds and reward strategies, effectively addressing the resource waste caused by rigid strategies in traditional methods. Simultaneously, by removing risky user data in real time, it avoids interference from abnormal behavior in strategy adjustments, improving the accuracy of activity performance evaluation.

[0068] Through the aforementioned technical solution, this application can dynamically adjust operational strategies based on the real-time performance of the campaign, improving user engagement and conversion efficiency while controlling costs. By establishing a risk user identification mechanism, it effectively prevents resource depletion caused by malicious behavior and ensures the safety of campaign funds. The application of the dynamic adjustment model makes the operational strategy adaptive, enabling it to quickly respond to changes in campaign needs at different stages and improve the overall effectiveness of marketing activities.

[0069] In some embodiments, this application further proposes dividing all online users into a control group and an experimental group, wherein the control group users share tasks using the task threshold and reward strategy of the previous time step, and the experimental group users share tasks using the task threshold and reward strategy of the current time step; the second participation data of the control group and the third participation data of the experimental group are obtained and compared to obtain a comparison result; if the comparison result indicates that the experimental group has not met the target, the task threshold and reward strategy of the previous time step are rolled back; if the target is met, the task threshold and reward strategy of the current time step are maintained.

[0070] The control group refers to the user group using historical strategy parameters, specifically implemented by bucketing user IDs using a hash algorithm, used to compare the effects of the adjusted strategy on the experimental group. The experimental group refers to the user group using the newly adjusted strategy parameters, specifically implemented using a dynamic traffic allocation algorithm, used to verify the effectiveness of the new strategy. The second set of data refers to the set of behavioral indicators of the control group users under the historical strategy, specifically acquired in real-time through a data collection system, including dimensions such as activity participation rate, task conversion rate, and the proportion of high-risk users. The third set of data refers to the set of behavioral indicators of the experimental group users under the new strategy, specifically collected through a distributed log system, used to quantify the changes in the effect after the strategy adjustment. The comparison result refers to the quantitative evaluation of the differences between the two sets of data using a confidence calculation model, specifically using a weighted summation algorithm combined with a preset threshold to determine the superiority or inferiority of the strategies.

[0071] Specifically, during the execution of online marketing activities on the power grid, when adjustments to task sharing thresholds and reward strategies are needed, a group comparison mechanism is used for strategy verification to avoid potential risks associated with a full rollover. First, online users are randomly divided into a control group and an experimental group using user ID hashing to ensure a balanced distribution of user profiles between the two groups. The control group continues to use the previous task thresholds and reward strategies, while the experimental group applies the newly adjusted parameter configurations. Subsequently, a real-time data acquisition system synchronously acquires key indicators such as the number of participants, task completion rate, and percentage of high-risk users for both groups, and inputs them into a confidence calculation model for dynamic evaluation. This model uses the percentage of high-risk users as a negative indicator and the activity participation rate and task conversion rate as positive indicators, generating a comprehensive confidence score through weighted summation. When the experimental group's score is higher than or equal to that of the control group, the new strategy is deemed effective and fully implemented; if it is lower than that of the control group, a strategy rollback mechanism is triggered, restoring the parameter configuration to its previous state.

[0072] In some specific implementations, user grouping can employ a consistent hashing algorithm to ensure group stability. For example, user IDs can be mapped to a circular hash space and the control and experimental groups can be proportionally split. The weighting coefficients in the confidence calculation model can be dynamically adjusted according to business objectives. For instance, in risk-sensitive scenarios, the weighting coefficient for the proportion of risky users can be increased to 0.6, and in user activation scenarios, the weighting coefficient for task conversion rate can be increased to 0.7. The strategy rollback mechanism can be configured with a buffer period; for example, automatic rollback can be triggered when the confidence level of the experimental group is lower than that of the control group for three consecutive statistical periods.

[0073] Compared to existing technologies, current marketing systems typically employ a full-scale switch when adjusting strategies, lacking a phased verification mechanism, which can easily lead to large-scale user churn due to strategy failure. Traditional methods rely on human experience to judge strategy effectiveness, resulting in subjective bias and high response latency. This solution establishes a group comparison mechanism and an automated evaluation model to achieve closed-loop verification of strategy adjustments, effectively reducing trial-and-error costs. Furthermore, through the above technical solutions, this application can verify the effectiveness of strategy adjustments in real time, rapidly iterating and optimizing strategy parameters while ensuring the stability of online activities. This method avoids the full spread of ineffective strategies through an automated comparison mechanism, improves the accuracy of strategy optimization through a quantitative evaluation model, and enhances the system's fault tolerance through a dynamic rollback mechanism, thereby comprehensively improving the security and conversion efficiency of online marketing activities on the power grid.

[0074] In some embodiments, this application further proposes to obtain and compare the second participation data of the control group and the third participation data of the experimental group to obtain a comparison result, including determining the first participation rate and the first conversion rate of the experimental group, and determining the second participation rate and the second conversion rate of the control group; weighted summing of the risk user ratio, the first participation rate and the first conversion rate of the experimental group to obtain a first confidence level; weighted summing of the risk user ratio, the second participation rate and the second conversion rate of the control group to obtain a second confidence level; when the first confidence level is greater than or equal to the second confidence level, it is determined that the experimental group has met the target, otherwise it is rolled back to the task threshold and reward strategy of the previous moment.

[0075] The control group refers to users who shared tasks using the previous task threshold and reward strategy. This can be achieved through random or stratified sampling, providing a benchmark for strategy adjustments. The experimental group refers to users who shared tasks using the current task threshold and reward strategy. This is achieved by dividing users into groups in real-time and applying the new strategy, validating the effectiveness of the new strategy. Confidence level is a quantitative indicator that comprehensively evaluates the strategy's effectiveness. It can be achieved by weighting participation rate, conversion rate, and the proportion of at-risk users, objectively reflecting the actual benefits of strategy adjustments. The proportion of at-risk users refers to the percentage of users with invalid participation behavior, which can be achieved by real-time monitoring of user device information, IP addresses, and historical behavior data, eliminating data noise interference in strategy evaluation.

[0076] Specifically, online users were divided into a control group and an experimental group, and each group implemented different task-sharing strategies. Within a preset time window, participation rate, conversion rate, and the percentage of at-risk users were collected for both groups. Participation rate reflects user responsiveness, conversion rate reflects task completion effectiveness, and the percentage of at-risk users reflects the safety of strategy implementation. A weighted summation method was used to combine positive and negative indicators into a confidence score; for example, participation rate and conversion rate were given positive weights, while the percentage of at-risk users was given a negative weight. If the confidence score of the experimental group was not lower than that of the control group, it indicated that the new strategy improved effectiveness without significantly increasing risk, and could continue to be implemented; otherwise, it was necessary to revert to the original strategy to avoid expanding the negative impact.

[0077] This solution dynamically divides the control and experimental groups and introduces a confidence-based quantitative evaluation mechanism to accurately identify the risk-reward ratio of strategy adjustments, avoiding the continued execution of ineffective strategies. Through the above technical solution, this application can dynamically verify the effectiveness of strategy adjustments, continuously optimize task thresholds and reward strategies while ensuring activity security, and effectively improve user participation quality and activity resource utilization.

[0078] In a specific implementation process, the "electricity bill bonus" activity for peak summer demand will be used as an example to illustrate one implementation process of the present invention. The steps listed below are only a relatively simple application scenario, which only involves the preceding related content: Step 1: On T-7, the strategy engine calls the meteorological API interface. When the provincial meteorological observatory issues an orange high-temperature warning, it automatically selects 1.8 million users who have "paid more than 200 yuan in electricity bills in the past 30 days and have not registered with the State Grid online service". Seven days prior to the official launch date of the event (Day T) (T-7 days) are reserved for subsequent poster design and testing to avoid process delays. The strategy engine proactively calls the official API interface of the provincial meteorological department to obtain high-temperature warning information in real time: when the provincial meteorological observatory issues an orange high-temperature warning, user selection is initiated (an orange warning indicates that the temperature is consistently ≥35℃, electricity demand is surging, and users are most sensitive to "electricity bill bonuses," which aligns with the "scenario-based incentive" logic). If no warning is issued, selection is delayed to ensure a strong correlation between the event and user needs.

[0079] Step 2: On T-5, the poster factory generates 20 creative designs, conducts a 3-hour grayscale test, and selects 4 designs with a CTR > 3.5%. Five days before T (T-5 days), after selecting users and before the official push, time is reserved for testing and optimizing materials to avoid failure to reach users due to insufficient material appeal.

[0080] The "Event Factory" built-in poster factory automatically generates 20 sets of creative posters, covering design dimensions: Contextual elements: High-temperature scene (sun, air conditioner, thermometer) + electricity bill red envelope (red envelope icon, amount); Copywriting differentiation: such as "35℃ savior! Electricity bill red envelopes help you save money" and "High temperatures lead to increased electricity consumption, exclusive red envelopes are ready" to avoid aesthetic fatigue from monotonous copywriting; Visual styles: minimalist, cartoon, formal, etc., suitable for users of different age groups (such as elderly users who prefer large fonts and high-contrast posters).

[0081] The advantage of the poster factory lies in "low cost + high efficiency": traditional design of 20 posters takes 3-5 days, while AIGC only takes 2 hours and does not require additional design manpower, which meets the goal of "low cost implementation".

[0082] Grayscale testing: 3-hour rapid verification of results: Test scope: 1% (18,000 people) of the 1.8 million selected users were randomly selected as the gray-scale test sample, divided into 20 groups (900 people in each group), and one set of posters was pushed to each group; Test duration: 3 hours (short-term testing allows for quick feedback on results, avoiding delays in the campaign's progress, while also covering peak user activity periods within a short timeframe). Selection criteria: Calculate the click-through rate (CTR = clicks / impressions) for each poster set, and select 4 poster sets with CTR > 3.5%: This threshold is set based on the historical operation data of the provincial power grid. When the posters with CTR > 3.5% are pushed to the full audience in the future, it can be ensured that the overall click-through rate is not less than 3%, which is much higher than the average level of 1.2% of traditional push. Final materials: 4 sets of high CTR posters will be used for full push on day T, covering different user preferences and maximizing reach.

[0083] Step 3: At 8:00 AM on day T, a WeChat service notification will be sent saying, "Today's temperature is 35℃, your exclusive electricity bill bonus has been credited to your account." The campaign officially launches at 8:00 AM on the day of T. This is a peak time for user activity (such as checking mobile phones before work or during breakfast), and the reach rate is 25% higher than during inactive times.

[0084] Step 4: After clicking, the user is redirected to the mini-program to complete registration → bind account number → pay bill → share task; Step 5: The fission task adopts a "3-7-15" step model, and the dynamic threshold is adjusted in real time by the risk control decision tree; This involves using a tiered "3-7-15" task-based referral system to encourage existing users to share, achieving low-cost customer acquisition through referrals. For example, inviting 3 people to register on the State Grid website earns a 5 yuan reward. Inviting 7 people to bind their account number earns a 15 yuan reward. Inviting 15 people to make a single payment of ≥50 yuan earns a 40 yuan reward.

[0085] Step 6: The robot reconciles accounts daily at 4:00 AM; anomalies exceeding 0.1% trigger an alarm. Step 7: On T+30, the Operations Dashboard automatically generates a debriefing report and saves the strategy template to the "Event Factory" template library.

[0086] Based on the same general inventive concept, this invention also protects an operating system for an online power grid channel based on multidimensional fusion identity and flywheel fission. The operating system for an online power grid channel based on multidimensional fusion identity and flywheel fission provided by this invention will be described below. The operating system for an online power grid channel based on multidimensional fusion identity and flywheel fission described below and the operating method for an online power grid channel based on multidimensional fusion identity and flywheel fission described above can be referred to in correspondence.

[0087] Figure 2 This is a schematic diagram of the structure of an operation system for an online power grid channel based on multi-dimensional fused identity and flywheel fission, as provided in an embodiment of the present invention. Figure 2 As shown, the online channel operation system based on multi-dimensional fusion identity and flywheel fission in this embodiment includes an acquisition module 21, an establishment module 22, an initiation module 23, and an execution module 24.

[0088] The acquisition module 21 is used to acquire the user's multidimensional identity information, and to perform fusion processing on the multidimensional identity information to form a unified user identity identifier. Module 22 is used to establish a user tag system containing industry-specific tags, and to select users in real time based on the user tag system and preset selection rules to generate a target user group; Initiation module 23 is used to initiate a marketing campaign to the target user group; wherein, the marketing campaign is used to convert at least one target user in the target user group into an online user, and to enable the online user to share tasks based on the flywheel fission model, so as to convert the sharing user into an online user; the flywheel fission model includes dynamically adjusted task thresholds and dynamically adjusted reward strategies; The execution module 24 is used to execute the marketing campaign based on a multi-level caching mechanism, and to collect, analyze and visualize the full lifecycle data of the marketing campaign in real time, and to iteratively optimize the strategy of the marketing campaign based on the analysis results; wherein, the multi-level caching mechanism includes edge node caching, central node caching and persistent storage caching.

[0089] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute an operation method for power grid online channels based on multi-dimensional fused identity and flywheel fission.

[0090] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission provided by the above methods.

[0092] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an operation method for the power grid online channel based on multidimensional fused identity and flywheel fission provided by the above methods.

[0093] It should be noted that all relevant information that may be involved in the various embodiments of the present invention is processed in strict accordance with the requirements of laws and regulations, following the principles of legality, legitimacy, and necessity, based on the reasonable purpose of the business scenario, and is information that users actively provide or generate during the use of the product / service, as well as information obtained with user authorization.

[0094] The information processed by this invention may vary depending on the specific product / service scenario and should be based on the specific scenario in which the user uses the product / service. This may involve user account information, device information, or other related information. This invention will treat the relevant information and its processing with the utmost diligence.

[0095] This invention places great importance on the security of related information and has adopted reasonable and feasible security protection measures that comply with industry standards to protect related information and prevent unauthorized access, public disclosure, use, modification, damage or loss of related information.

[0096] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0097] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An operational method for online power grid channels based on multi-dimensional fusion identity and flywheel fission, characterized in that, include: The user's multidimensional identity information is obtained, and the multidimensional identity information is fused to form a unified user identity identifier. Establish a user tagging system that includes industry-specific tags, and based on the user tagging system and preset selection rules, select users in real time to generate target user groups; A marketing campaign is launched to the target user group; wherein the marketing campaign is used to convert at least one target user in the target user group into an online user, and to enable the online user to share tasks based on a flywheel fission model, so as to convert the sharing user into an online user; the flywheel fission model includes dynamically adjusted task thresholds and dynamically adjusted reward strategies; Based on a multi-level caching mechanism, the marketing campaign is executed, and the full lifecycle data of the marketing campaign is collected, analyzed, and visualized in real time. The strategy of the marketing campaign is iteratively optimized based on the analysis results. The multi-level caching mechanism includes edge node caching, central node caching, and persistent storage caching.

2. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 1, characterized in that, Marketing campaigns launched based on the target user group include: Based on the tags of each target user in the target user group, marketing content is matched to different target users. Acquire external scene data, and if it is determined that the external scene data meets the preset push conditions, select a push channel based on the channel preference data of each target user; The marketing content is pushed to each target user through the aforementioned push channel.

3. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 2, characterized in that, Based on the tags of each target user in the target user group, marketing content is matched to different target users, including: Based on the tags of each target user, the demand type of each target user is determined and divided into multiple user groups with the same demand type; For different target users in any user group with the same demand type, the marketing content of each target user in the user group with the same demand type is matched according to the preset matching rules. All user groups with the same demand type are traversed to obtain the marketing content of each target user. The preset matching rule is as follows: Select a preset proportion of sample users from any of the user groups with the same demand type; Multiple candidate content items are pushed to the sample users, and the click-through rate and open rate of each candidate content item are collected. When the click-through rate of at least one candidate content is greater than a preset click-through rate threshold and / or the open rate is greater than a preset open rate threshold, the at least one candidate content will be used as marketing content of the same demand type.

4. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 3, characterized in that, Multiple candidate content items are pushed to the sample users, including: Obtain historical participation behavior data of existing users corresponding to any user group with the same demand type; Based on the historical participation behavior data, determine the content of interest of the existing users; Select multiple candidate contents from the existing users' content of interest, and push multiple candidate contents to the sample users.

5. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 4, characterized in that, Based on the historical participation behavior data, determine the content of interest of the existing users, including: The historical participation behavior data is atomically parsed to obtain behavioral atoms that include operation type, object characteristics, scene parameters and emotional feedback. Based on the aforementioned behavioral atoms, a ternary relationship graph of existing users, behavioral features, and business content is constructed. The implicit associations between behavioral atoms are mined through graph neural networks to generate a user behavior gene graph. Entropy values ​​are calculated on the user behavior gene map, interest mutation points are detected in the user behavior sequence, and dynamic reconstruction of the user interest model is triggered based on the interest mutation points to obtain a dynamic interest model that reflects the current interest characteristics of existing users. A content gene library containing theme genes, form genes, and incentive genes is constructed. The dynamic interest model is used as input parameters through a generative adversarial network to recombine gene fragments in the content gene library, generating personalized content that matches the user's current interest characteristics as the existing user's content of interest.

6. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 5, characterized in that, By using the dynamic interest model as input parameters through a generative adversarial network, gene fragments in the content gene library are recombined to generate personalized content that matches the user's current interest features, including: Break down business content into gene fragments that can be independently recombined; The dynamic interest model is input into the generator of the generative adversarial network, so that the generator can select matching gene fragments from the content gene library based on the current interest features and combine them in a targeted manner to generate initial candidate content. The discriminator of the adversarial network evaluates the matching degree between the initial candidate content and the current interest features, and iteratively optimizes the generator parameters based on the evaluation results until personalized candidate content that meets the preset matching degree threshold is generated; wherein, the matching degree threshold is dynamically adjusted according to the intensity of the user's interest change.

7. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 5, characterized in that, Selecting the multiple candidate contents from the existing users' content of interest includes: Construct a dynamic scene factor library, which includes time dimension factors, spatial dimension factors and social dimension factors; The matching degree algorithm of scene and content is used to calculate the matching degree between the personalized candidate content and the dynamic scene factor in the dynamic scene factor library, and to filter out multiple candidate contents that are suitable for the current scene.

8. The method for operating a power grid online channel based on multi-dimensional fusion identity and flywheel fission according to any one of claims 1-7, characterized in that, The adjustment process for the task threshold and the reward strategy includes: Set the initial task threshold and initial reward amount; The first participation data of the marketing campaign is acquired in real time, including the number of participants, the amount of tasks completed, and the proportion of high-risk users. The initial task threshold is adjusted based on the first participation data to obtain the dynamically adjusted task threshold; The reward amount is calculated using a decreasing function, wherein the decreasing function is: ,in The initial reward amount is t, where t is the time since the event started. The adjustment coefficient λ is adjusted based on the first participating data.

9. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 8, characterized in that, Also includes: All online users were divided into a control group and an experimental group. The control group users shared tasks using the task threshold and reward strategy from the previous time step, while the experimental group users shared tasks using the task threshold and reward strategy from the current time step. Obtain the second participation data of the control group and the third participation data of the experimental group, and compare them to obtain the comparison results; If the comparison result indicates that the experimental group has not met the target, roll back to the task threshold and reward strategy of the previous time step; If the comparison result indicates that the experimental group has met the target, the task threshold and reward strategy at the current moment shall be maintained.

10. The operation method of the power grid online channel based on multi-dimensional fusion identity and flywheel fission according to claim 9, characterized in that, Obtain the second participation data of the control group and the third participation data of the experimental group, and compare them to obtain the comparison results, including: Determine the first participation rate corresponding to the number of participants in the experimental group and the first conversion rate corresponding to the amount of tasks completed in the experimental group; and determine the second participation rate corresponding to the number of participants in the control group and the second conversion rate corresponding to the amount of tasks completed in the control group. The first confidence level of the experimental group is obtained by weighted summing of the proportion of risky users, the first participation rate, and the first conversion rate; wherein, the proportion of risky users in the experimental group is a negative deduction value, and the first participation rate and the first conversion rate are positive contribution values. The second confidence level of the experimental group is obtained by weighted summing of the proportion of at-risk users in the control group, the second participation rate, and the second conversion rate; wherein, the proportion of at-risk users in the control group is a negative deduction value, and the second participation rate and the second conversion rate are positive contribution values; If the first confidence level is greater than or equal to the second confidence level, the comparison results of the experimental group meeting the standard are determined; If the first confidence level is less than the second confidence level, it is determined that the experimental group has not met the comparison results.

Images