Local life service integration method and device and terminal equipment

Abstract

The embodiment of the invention provides a local life service integration method and device and terminal equipment. The method comprises the following steps: integrating a plurality of heterogeneous local life service providers into a gas station service platform to obtain a local life service platform, and collecting real-time service data in real time; non-oil sales fluctuation of the gas station is predicted based on the real-time service data, and the order quantity and the delivery cycle of upstream and downstream suppliers are dynamically adjusted according to a prediction result; constructing a multi-dimensional data analysis model, predicting user traffic and various service demands in different time periods every day, and configuring a corresponding local life service time window based on a prediction result; dynamically updating callable local life service resources and dynamically deploying gas station human resources based on the order quantity, the delivery cycle and the local life service resource configuration; and generating a personalized service list of the user by using a collaborative filtering algorithm and a reinforcement learning algorithm, interacting with the user in an interaction interface, and providing corresponding local life service resources for the user.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method and device for integrating local life services and a terminal device. Background Art

[0002] In the context of the booming development of Internet technology and the significant acceleration of people's life rhythms, the market demand for local life service integration platforms has witnessed an explosive growth. Integrating diverse services such as catering, express delivery, and shopping onto the same platform not only meets users' pursuit of a convenient and efficient life but also greatly improves life efficiency. Among them, the concept of integrating local life services into a gas station platform, as an innovative direction in the industry, aims to break the inherent single-energy supply model of gas stations and promote their transformation into comprehensive life service hubs.

[0003] However, existing technologies generally remain at the preliminary concepts of data governance and resource aggregation and lack effective technical support for in-depth collaborative scheduling among service providers. As each local life service provider adopts different data formats and interface protocols during the business development process, for example, the order systems of catering merchants and the logistics systems of express delivery service providers have significant differences in data architectures, which poses numerous obstacles for gas station platforms in integrating heterogeneous service providers. Poor data interaction makes it difficult to ensure service coherence and timeliness. When users use integrated services across service providers, they often encounter problems such as order delays and information errors. For example, when a service provider experiences resource shortages, such as a catering merchant running out of dishes due to a sudden influx of customers or an express delivery service provider being saturated during a shopping festival, existing technologies lack the ability to quickly switch to other candidate service providers, directly resulting in service interruptions and user dissatisfaction. In scenarios of sudden increases in user traffic, such as during holidays or specific promotional activities, the platform is unable to promptly adjust and expand service types, making it difficult to meet users' diverse immediate needs, resulting in limited platform service capabilities and an increased risk of user loss. At the same time, taking inventory and distribution management as an example, in the traditional mode, gas stations rely on the experience of managers to predict fluctuations in non-oil product sales and lack scientific data models and algorithm support. Statistics show that problems such as inventory backlogs or shortages caused by inaccurate predictions increase the operating costs of enterprises. Due to the inability to predict sales, inventory backlogs or shortages occur frequently, seriously affecting supply chain efficiency and cost control.

[0004] Therefore, in view of the above technical problems, there is an urgent need to design a new technical solution to solve at least one of the above technical problems. Summary of the Invention

[0005] The main objective of the embodiments of this application is to provide a local life service integration method, device, and terminal device, aiming to solve at least one of the technical problems of data heterogeneity, blocked interaction, lack of service provider switching mechanism, and insufficient supply chain management ability between traditional gas station platforms and local life service providers.

[0006] In a first aspect, the embodiments of this application provide a local life service integration method, including:

[0007] Integrate multiple heterogeneous local life service providers into the gas station service platform to obtain a local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and collect the real-time service data of each local life service provider in real time;

[0008] Predict the non-oil product sales volume fluctuations of the gas station based on the real-time service data, and dynamically adjust the order quantities and delivery cycles of upstream and downstream suppliers according to the prediction results;

[0009] Construct a multi-dimensional data analysis model for the gas station scenario, integrate the historical user traffic data of the gas station, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predict the user traffic and various service demand quantities at different time periods of each day, and configure the corresponding local life service time windows based on the prediction results to dynamically optimize the local life service resource allocation of the gas station at different time periods;

[0010] Dynamically update the local life service resources that can be called by the interactive interface in the local life service platform based on the order quantities and delivery cycles of upstream and downstream suppliers and the local life service resource allocation at different time periods, and dynamically allocate the human resources of the gas station;

[0011] For the users in the local life service platform, use collaborative filtering algorithms and reinforcement learning algorithms to generate personalized service lists for the users, and interact with the users in the interactive interface based on the personalized service lists to provide the corresponding local life service resources for the users.

[0012] In a second aspect, the embodiments of this application provide a local life service integration device, including:

[0013] A collection module, which is used to integrate multiple heterogeneous local life service providers into the gas station service platform to obtain a local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and collect the real-time service data of each local life service provider in real time;

[0014] A configuration module, which is used to predict the fluctuation of non-oil product sales volume of gas stations based on the real-time service data, and dynamically adjust the order quantity and delivery cycle of upstream and downstream suppliers according to the prediction results; construct a multi-dimensional data analysis model for gas station scenarios, integrate the historical user traffic data of gas stations, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predict the user traffic and the demand for various services at different time periods of each day, and configure the corresponding local life service time windows based on the prediction results to dynamically optimize the allocation of local life service resources in gas stations at different time periods; based on the order quantity and delivery cycle of upstream and downstream suppliers, and the allocation of local life service resources at different time periods, dynamically update the local life service resources that can be called by the interactive interface in the local life service platform, and dynamically allocate the human resources of gas stations.

[0015] An interaction module, which is used for users in the local life service platform to generate a personalized service list for users by using collaborative filtering algorithms and reinforcement learning algorithms, and interact with users in the interactive interface based on the personalized service list to provide corresponding local life service resources for users.

[0016] In a third aspect, an embodiment of the present application further provides a terminal device, which includes a processor and a memory for storing a computer program; the processor is used to execute the computer program and implement the local life service integration method according to the first aspect or any embodiment of the present application when executing the computer program.

[0017] An embodiment of the present application provides a local life service integration method, device, and terminal device. In this method, by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, multiple heterogeneous local life service providers are integrated into a gas station service platform to obtain a local life service platform, and the real-time service data of each local life service provider is collected in real time; based on the real-time service data, the non-oil product sales volume fluctuation of the gas station is predicted, and the order quantity and delivery cycle of upstream and downstream suppliers are dynamically adjusted according to the prediction result; a multi-dimensional data analysis model for the gas station scenario is constructed, integrating the historical user traffic data of the gas station, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predicting the user traffic and various service demand quantities at different time periods of each day, and configuring corresponding local life service time windows based on the prediction result to dynamically optimize the local life service resource allocation of the gas station at different time periods; based on the order quantity and delivery cycle of upstream and downstream suppliers, and the local life service resource allocation at different time periods, the local life service resources that can be called by the interactive interface in the local life service platform are dynamically updated, and the human resources of the gas station are dynamically allocated; for the users in the local life service platform, a personalized service list of the user is generated by using collaborative filtering algorithms and reinforcement learning algorithms, and based on the personalized service list, an interaction is carried out with the user in the interactive interface to provide the corresponding local life service resources for the user. Through the local life service integration method, the efficient integration of multiple heterogeneous local life service providers is realized, and it has a powerful dynamic resource allocation and flexible supply network management ability. By designing an integrated service device and optimizing the collaborative work of the terminal device and the platform, a convenient, efficient, and personalized local life service experience is provided for users, and the overall service ability in this area is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic flowchart of a local life service integration method provided by an embodiment of the present application;

[0019] Figure 2 It is a schematic block diagram of the module structure of a local life service integration device provided by an embodiment of the present application;

[0020] Figure 3 It is a schematic block diagram of the structure of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In view of the technical problems existing in the traditional tower crane control method, an embodiment of the present application proposes a local life service integration method, device, and terminal device.

[0022] The embodiments of the present application provide a method, apparatus, and terminal device for integrating local life services. Among them, the method for integrating local life services can be applied to a terminal device, which can be a console in a gas station, a gas station management system, or a mobile terminal communicating with the management system, such as electronic devices like mobile phones, virtual reality devices, tablets, laptops, desktop computers, wearable devices, etc. The terminal device can be a server connected to the gas station management system or a server cluster. The above connection methods can be implemented through hardware circuits or through communication modules.

[0023] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other. Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for integrating local life services provided by an embodiment of the present application.

[0024] As Figure 1 shown, the method for integrating local life services includes the following steps:

[0025] Step S101: Integrate multiple heterogeneous local life service providers into the gas station service platform to obtain a local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and collect real-time service data of each local life service provider in real time.

[0026] In the embodiments of the present application, the real-time service data at least includes: sales volume of various types of goods, consumption of various local life resources, operation conditions of various local life service providers, geographical location of the gas station, types of incoming vehicles, incoming traffic volume, queuing time, surrounding traffic conditions, working status of in-station personnel, and complaint situations at various time periods.

[0027] Exemplarily, in step S101, a standardized interface protocol can be formulated by using RESTful API and / or gRPC technology, and the data exchange format can be unified by using JSON Schema or Apache Avro.

[0028] Specifically, in step S101, to enable different local life service providers to access, a standardized interface protocol can be built through RESTful API and gRPC technologies, and the unified data exchange format can be achieved with the help of JSON Schema or Apache Avro. Specifically, for lightweight service interaction scenarios such as catering and retail, RESTful API is adopted. Based on the HTTP / HTTPS protocol, the OpenAPI 3.0 is used to define the standardized interface document, clarify the endpoint path, request method, and the structures of the request body and response body, and the OAuth 2.0 client credentials mode is used for security authentication. A unique API key is assigned to each service provider to ensure the security of data interaction. For example, for the interface to obtain the inventory of catering merchants, the inventory data can be obtained through a specific GET request and the corresponding response body structure. For service communication scenarios with high throughput and low latency such as express delivery and logistics, the gRPC technology is selected. Protocol Buffers is used to define the service interface and message types, and multi-language client / server codes are generated to meet the access requirements of service providers with different technical architectures.

[0029] In terms of unifying the data exchange format, JSON Schema is used to define exclusive data models for various services such as catering, retail, and express delivery to achieve data verification. For old service provider systems that do not support JSON output, with the help of ETL tools such as Apache NiFi, data in formats such as CSV / XML is mapped into JSON format that conforms to the JSON Schema specification. For big data volume scenarios such as high-frequency collected traffic flow and equipment energy consumption, Apache Avro is adopted. Its binary format is used to reduce storage and transmission overhead, and through the compatible mode design, new fields are allowed to be added to ensure the compatibility of historical data and new data.

[0030] To ensure the stable access of different service providers, a perfect compatibility design and exception handling mechanism also need to be established. Through the interface gateway as the unified entry, it supports the coexistence of multiple versions of interfaces, realizes protocol conversion, traffic control, and request routing. A unified error code system is defined, and automatic retry is performed for temporary failures to ensure the stability of data interaction. At the same time, a dual-mode data synchronization mechanism of timed polling and event-driven is adopted. Reasonable collection frequencies are set for different types of data to achieve real-time data updates, and through the unique identifier design and incremental synchronization algorithm, the transmission of invalid data is reduced, and the data processing efficiency is improved, so as to achieve the efficient and stable access of various local life service providers.

[0031] Furthermore, a resource update strategy is established based on the message queue (such as Kafka) combined with the version control mechanism.

[0032] Specifically, a resource update strategy is established based on the message queue Kafka combined with a version control mechanism, which can be achieved by deeply integrating the high throughput and low latency characteristics of Kafka with the precise management ability of version control. In actual operation, each local life service provider encapsulates resource change information (such as product listing, delisting, service time adjustment, etc.) into messages and sends them to the corresponding topics in Kafka. With its partition parallel processing ability, Kafka can quickly receive and process a large number of messages, ensuring that resource change information is transmitted to the gas station service platform at almost real-time speed. At the same time, a version control mechanism is introduced to assign a unique version number to each resource update operation, and this version number is transmitted along with the message. When the platform receives a message, it first verifies the version number in the message. If the version number is higher than the currently recorded version, the resource update operation is executed, and the local version number is updated to the latest value; if the version number is lower than or equal to the current version, the message is determined to be a duplicate or invalid message and is discarded to avoid incorrect or duplicate updates of resources.

[0033] The innovation of this strategy lies in breaking through the limitations of traditional resource update methods in terms of real-time performance and accuracy. The introduction of Kafka makes resource updates no longer restricted by network latency and data volume. Even in high-concurrency scenarios, it can ensure rapid information transmission, greatly improving the real-time performance of resource updates; while the version control mechanism adds a precise valve to resource updates, ensuring that each update operation is based on the latest and valid information, avoiding problems such as message duplication or out-of-order caused by network fluctuations, and guaranteeing the accuracy and consistency of resource updates. The combination of the two forms an efficient and reliable resource update system, providing a solid guarantee for the stable operation of the local life service integration platform, and also offering a new solution idea for the real-time processing of multi-source heterogeneous data.

[0034] The collection of real-time service data is the data foundation for realizing the integration of gas station services. The following introduces some data collection methods in step S101 in combination with various data types:

[0035] For the sales volume of various products and the consumption situation of local life resources, transaction records can be captured in real-time through the POS terminal data interface deployed in gas station convenience stores and the systems of cooperative service providers, and combined with RFID tags or intelligent shelf sensors to automatically identify changes in the inbound and outbound and display status of products. For example, when the stock of coffee drinks on the shelf is lower than the threshold, the sensor automatically triggers the upload of replenishment reminder data.

[0036] Regarding the operation status of service providers (such as service load and response speed), a lightweight data collection SDK can be embedded in the service provider's business system. The service process status (such as the number of idle / busy threads) is reported through a timed heartbeat mechanism, and a distributed tracing tool (such as OpenTelemetry) is used to monitor the time consumption of the service call chain. For example, after a catering service provider receives an order, the system automatically records the time interval from order reception to meal preparation completion and pushes it to the platform.

[0037] The collection of geographical location data relies on GPS positioning modules or Beidou satellite navigation terminals deployed at gas stations and service provider stores. Combining with LBS (Location-Based Service) technology to calibrate geographical coordinates in real time, and at the same time accessing the surrounding road network data through a GIS (Geographic Information System). For example, the longitude and latitude of the intersection where the gas station is located and the location distribution of service providers such as catering and car washing within 3 kilometers around can be obtained in real time.

[0038] The collection of the type of incoming vehicles, traffic flow, and queuing time is achieved through license plate recognition cameras and inductive loops deployed at the entrances and exits of gas stations. The camera uses computer vision algorithms to identify vehicle types (such as sedans, SUVs, trucks) and count the traffic flow. The inductive loop is linked with the queue length sensor to calculate the queuing duration of vehicles. For example, when it is detected that there are more than 5 queuing vehicles, the system automatically generates congestion warning data.

[0039] The surrounding traffic conditions can be obtained by docking with the open API of the traffic management department or third-party map service interfaces (such as real-time traffic conditions of Gaode / Baidu Maps), including information such as road congestion index, accident warnings, and temporary controls. For example, after the platform synchronizes the congestion data of a main road in real time, it automatically pushes a detour suggestion to the incoming vehicles.

[0040] The working status of in-station personnel is achieved through intelligent work badges or video surveillance analysis: The work badge is built-in with an acceleration sensor and a Bluetooth beacon to collect the employee's movement trajectory and working hours in real time (such as cashier desk duty, shelf arrangement, etc.). Video surveillance uses AI behavior analysis algorithms to identify the implementation of employee service specifications (such as whether to wear a mask, whether to respond to customer inquiries in a timely manner).

[0041] Complaint situations are collected in real time through the online complaint form and voice message-to-text function of the platform user-side APP. Combining with natural language processing (NLP) technology to perform sentiment analysis and keyword extraction on the complaint content. For example, it automatically identifies high-frequency problems such as "poor service attitude" and "long waiting time" and triggers the workflow to be transferred to the corresponding department.

[0042] The above multi-source data collection method performs preprocessing (such as data cleaning and format conversion) through edge computing nodes, and then transmits it to the platform data center in real time through the Kafka message queue to ensure the timeliness (second-level delay), accuracy, and integrity of data collection, providing underlying data support for subsequent service integration, intelligent scheduling, and user portrait construction.

[0043] Step S102: Predict the non-oil product sales volume fluctuation of the gas station based on the real-time service data, and dynamically adjust the order quantity and delivery cycle of upstream and downstream suppliers according to the prediction results.

[0044] It can be understood that step S102 realizes non-oil product sales volume prediction and supply chain optimization by constructing a three-dimensional dynamic decision-making model and introducing the Copula function.

[0045] As an optional example, in step S102, historical non-oil product sales volume data and associated geographical locations, surrounding business ecosystems, emergencies, seasonal data, and weather data are used to determine the optimal combination weights through Bayesian optimization, and a three-dimensional dynamic decision-making model including sales volume prediction, inventory status, and delivery capacity is constructed; a virtual mirror of the supply chain corresponding to the real-time service data is established; the three-dimensional dynamic decision-making model is used to adaptively construct multiple ordering strategies matching the virtual mirror of the supply chain, and dynamic simulation is performed on each ordering strategy to obtain the dynamic simulation results under each ordering strategy; the dynamic simulation results include: predicted sales volume fluctuation value, predicted inventory fluctuation value, and predicted delivery load value; the Copula function is introduced as a connection function, and by constructing a non-parametric joint distribution model, the tail correlation between the sales volume fluctuation and supply delay phenomena in the virtual mirror of the supply chain is identified to obtain the tail risk characteristics under each ordering strategy; based on the tail risk characteristics, the target ordering strategy with the inventory fluctuation reaching the safety inventory level is selected and the target ordering strategy is executed to balance the order quantity and delivery cycle of upstream and downstream suppliers.

[0046] Specifically, integrate historical non-oil product sales volume data (such as monthly sales volume of convenience store goods) and associated geographical locations (such as the density of office buildings around the gas station), surrounding business ecosystems (such as the number of catering businesses within 3 kilometers), emergencies (such as business district promotion activities), seasonal data (such as the peak sales volume of cold drinks in summer), and weather data (such as the decrease in car wash service demand on rainy days), and use the Bayesian optimization algorithm to automatically search and determine the optimal combination weights of each influencing factor (such as the seasonal factor weight is 30% and the weather factor weight is 15%) to form a three-dimensional dynamic decision-making model including sales volume prediction, inventory status, and delivery capacity.

[0047] It can be noted that the Bayesian optimization algorithm efficiently finds the optimal solution through the iterative update of the probability model and the acquisition function. It first constructs a Gaussian process model based on the initial sampling points (such as randomly selected weather-season-promotion weight combinations). This model regards the objective function (such as sales prediction error) as the joint distribution of infinitely many random variables and captures the similarity between input points through the covariance function. After each sampling, the algorithm updates the posterior distribution according to Bayes' theorem and incorporates the new observed data into the model. The acquisition function (such as the expected improvement EI) balances exploration (sampling in unknown regions) and exploitation (digging deeper in known high-value regions) and finds the next most promising sampling point through numerical optimization. For example, in sales prediction, the algorithm continuously adjusts the weights of factors such as weather, season, and promotion, making the prediction error gradually converge to the minimum value. The entire process does not require derivative information of the objective function and is suitable for dealing with complex nonlinear optimization problems.

[0048] Meanwhile, a virtual mirror of the supply chain is constructed based on real-time service data (such as the current inventory level and the remaining capacity of suppliers). Through digital twin technology, the operation of the supply chain under different ordering strategies (such as increasing the order quantity by 20% and shortening the delivery cycle to 2 days, or maintaining the regular order quantity but extending it to 5 days) is simulated, generating predicted values of sales fluctuations (such as predicting a 12% increase in the sales volume of potato chips next week), inventory fluctuations (such as the inventory peak of 80 pieces when the safety inventory threshold is 50 pieces), and delivery load (such as the maximum daily delivery volume limit of the third-party logistics).

[0049] On this basis, the Copula function is introduced as the connection function, and a joint distribution model of sales fluctuations and supply delays is constructed through a non-parametric method to identify the tail correlation between the two in extreme cases (such as the probability that the supply delay exceeds 48 hours when the sales volume suddenly increases by 30%), and to quantify the tail risk characteristics of each ordering strategy (such as the probability of out-of-stock risk under a certain strategy being 5%).

[0050] Further optionally, when constructing the joint distribution model of sales volume fluctuations and supply delays, step S102 uses the Copula function as a connection tool. First, the marginal distribution fitting is performed on the sales volume fluctuation sequence (such as the daily sales change rate) and the supply delay sequence (such as the difference between the actual arrival time and the planned time) in the historical data respectively. Non-parametric kernel density estimation or parametric distributions (such as normal distribution, t-distribution) can be selected. Then, a Copula function (such as Clayton, Gumbel or Frank Copula) is introduced to describe the dependence structure between the two. The Copula function can connect the marginal distributions into a joint distribution and does not depend on the specific distribution form of the variables, which is particularly suitable for capturing non-linear and asymmetric correlations. After determining the parameters of the Copula function through maximum likelihood estimation or Bayesian inference, the joint distribution model of sales volume fluctuations and supply delays can be constructed. This model can not only describe the average correlation between variables, but also quantify the occurrence probability of extreme events (such as serious supply delays when sales volume surges) through tail dependence parameters (such as upper and lower tail correlation coefficients), providing an accurate mathematical tool for supply chain risk assessment. Finally, the tail risk characteristics (such as the probability of out-of-stock) under different ordering strategies can be calculated using this model to support dynamic decision-making.

[0051] Furthermore, according to the correlation characteristics, candidate Copula function families are screened. The elliptical family (such as Gaussian, t-Copula) is suitable for symmetric correlation scenarios. The t-Copula can capture tail correlations but assumes symmetric tails. The Archimedean family (such as Clayton, Gumbel, Frank) can flexibly express asymmetric dependence structures. Among them, Clayton is suitable for lower tail correlation, Gumbel is suitable for upper tail correlation, and Frank is suitable for no obvious tail correlation. Extreme value Copulas (such as Galambos, Hüsler-Reiss) are specifically used to model extreme event dependencies and are suitable for risk assessment scenarios. For example, if the data shows strong upper tail correlation and no lower tail correlation, the Gumbel Copula is the preferred choice. If there is correlation in both the upper and lower tails, the t-Copula is more appropriate.

[0052] Furthermore, the Copula parameters are estimated using the maximum likelihood estimation (MLE) or the inference function method (IFM). For example, for the Gumbel Copula, the parameter θ ∈ [1, ∞), and the larger the θ, the stronger the upper tail dependence. Compare the differences in the cumulative distribution functions between the empirical Copula and the theoretical Copula. The Cramer-von Mises test is more sensitive to tail differences and is suitable for risk assessment scenarios. The AIC / BIC information criterion balances goodness of fit and model complexity, and selects the model with the smallest value. For example, after fitting the Gumbel, Clayton, and t-Copula simultaneously, if the AIC value of the t-Copula is the lowest and the p-value of the K-S test > 0.05, it indicates the best fitting effect.

[0053] Furthermore, for the selected candidate Copulas, verify whether the tail dependence conforms to the reality through the following methods. Estimate the tail correlation coefficient and compare the theoretical tail correlation coefficient with the empirical value. For example, calculate the upper tail correlation coefficient of the Gumbel Copula using an empirical formula. Or, generate a large number of samples based on the fitted Copula and observe whether the occurrence frequency of extreme events (such as sales volume growth > 30% and supply delay > 48 hours) is consistent with historical data. If the deviation between theory and reality is large, it is necessary to reselect the Copula function family.

[0054] Furthermore, in combination with the requirements of supply chain management, evaluate the practicality of the Copula model. If it is necessary to focus on the out-of-stock risk (lower tail event), then the Clayton Copula is preferred; if it is necessary to guard against overstock (upper tail event), then the Gumbel Copula is more suitable. For real-time decision-making scenarios, the Archimedean Copula with low computational complexity (such as the Frank Copula) is preferred over the t-Copula that requires multi-dimensional integration. If more variables (such as price fluctuations, weather anomalies) need to be incorporated in the future, then select the Copula family that can be extended to multiple variables (such as the vine Copula).

[0055] Finally, the platform screens out the target ordering strategy with inventory fluctuations meeting the safety inventory level (such as the inventory always being higher than the threshold) based on the tail risk characteristics. For example, when the sales volume forecast is increasing and the supply delay risk is high, automatically trigger the strategy of "increasing the safety inventory by 20% + activating the alternative supplier", dynamically balancing the order quantity and delivery cycle of upstream and downstream suppliers, and achieving the balance between supply chain elasticity and cost. The entire process upgrades the traditional empirical ordering mode to an intelligent prediction-driven dynamic decision-making through data-driven model simulation and risk quantification, increasing the inventory turnover rate by 25% and reducing the out-of-stock rate by 30%.

[0056] Step S103: Build a multi-dimensional data analysis model for gas station scenarios, integrate the historical user traffic data of gas stations, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predict the user traffic and the demand for various services at different time periods of each day, and configure the corresponding local life service time windows based on the prediction results to dynamically optimize the allocation of local life service resources at gas stations in different time periods.

[0057] Step S103 realizes the dynamic allocation of local life service resources at gas stations by building a multi-dimensional data analysis model. The core lies in integrating multi-dimensional data of time and space, environment, and society to predict user traffic and service demand.

[0058] The multi-dimensional data analysis model is based on historical user traffic data (such as the traffic flow of vehicles entering the station at each time period in the past year) and local life service demand data (such as convenience store product purchase records, car wash service orders), and integrates external variables such as surrounding traffic flow (such as the real-time congestion index of the main road), weather changes (such as temperature, precipitation probability), and community activities (such as business district promotions, school holidays) to form multi-modal input features. The model architecture uses a spatio-temporal graph neural network (STGNN) to model the spatial associations between gas stations and surrounding service facilities through a graph structure (such as the distribution of restaurants and parking lots within 3 kilometers), and uses a gated recurrent unit (GRU) to capture the periodicity and trend in the time series (such as a sharp increase in car wash demand during the weekday morning rush hour).

[0059] In the training stage of the multi-dimensional data analysis model, the attention mechanism is used to automatically learn the importance weights of each dimension of data (such as the weight of the car wash service increases by 20% on rainy days), and the hyperparameters are determined through Bayesian optimization (such as the learning rate of 0.001 and the hidden layer dimension of 64).

[0060] In terms of the implementation process, first, data preprocessing is carried out. For missing values (such as the loss of traffic data caused by sensor failures), Kalman filter interpolation is used, and for outliers (such as a sudden increase in traffic flow caused by an accident), Z-score detection and smoothing processing are performed. Subsequently, a spatio-temporal feature matrix is constructed, the geographical coordinates are converted into spatial embedding vectors, and the timestamps are encoded as periodic features. When the model makes predictions, it combines real-time data (such as current weather warnings, traffic control information) to generate the probability distributions of user traffic and service demand for each hourly segment in the next 24 hours (such as the predicted car wash demand from 18:00 to 20:00 tomorrow is 70% ± 5%).

[0061] Finally, the service time window is dynamically configured based on the prediction results. For example, during high-demand periods (such as weekend afternoons), additional car wash stations are added and the business hours of convenience stores are extended. During low-demand periods (such as late at night), manpower input is reduced and an automatic replenishment system is activated. By continuously learning the dynamic association between user behavior patterns and environmental variables, this model achieves an accurate match between service resource allocation and actual demand, increasing the non-oil business income of gas stations while reducing the cost of idle resources.

[0062] Step S104: Dynamically update the local life service resources that can be called by the interactive interface in the local life service platform and dynamically allocate the human resources of gas stations based on the order quantities and delivery cycles of upstream and downstream suppliers, as well as the local life service resource allocation in different time periods.

[0063] The core of step S104 lies in constructing a dynamic resource linkage mechanism. By integrating the order delivery data at the supply chain end and the resource allocation requirements at the service end, the collaborative optimization of the local life service platform and gas station operations is achieved. Its configuration principle is based on the principles of supply-demand balance and efficiency priority. Taking the order quantity and delivery cycle of the supply chain as the basis of resource supply, combined with the service demand prediction in different time periods, a dynamic mapping relationship between resource supply and user demand is established. When the upstream supplier changes the delivery cycle due to order quantity adjustment (such as shortening to 2 days during peak periods), or the downstream user demand changes resulting in service resource allocation adjustment (such as adding car wash stations on weekends), through real-time data interaction and algorithm calculation, the resource list that can be called by the service platform is automatically adjusted, and the human resource allocation is optimized synchronously.

[0064] In the specific implementation process, first, receive the order quantity, delivery cycle, and service resource allocation data output by steps S102 and S103, and convert them into computable resource status parameters, such as available commodity inventory, the number of service stations, estimated arrival time, etc. Subsequently, associate these parameters with the interactive interface of the local life service platform through a data mapping algorithm. When the inventory is lower than the threshold, automatically hide the purchase entrance of the corresponding commodity, or update the reservation status of car wash services in real-time according to the occupancy of service stations.

[0065] In terms of human resource allocation, establish a matching model between personnel skills and service requirements. Combining the scheduling rules and real-time workload prediction, automatically generate a human resource scheduling plan. For example, when it is predicted that the passenger flow in the convenience store increases in the afternoon and the replenishment demand rises due to the delay in delivery by a certain supplier, the system will preferentially allocate employees with replenishment experience to the convenience store and push task instructions to employees through mobile terminals. Throughout the process, the system continuously monitors the dynamic changes in resource status and service demand, and uses the event-driven mechanism to trigger resource updates and personnel allocation to ensure that the service supply and demand of gas stations remain dynamically balanced in different time periods, improving operational efficiency and user satisfaction.

[0066] Furthermore, the specific method for dynamically allocating gas station human resources is mainly based on real-time demand forecasting, personnel skill matching, and intelligent scheduling algorithms, and achieves precise docking of human resources and service requirements by integrating multi-dimensional data. First, establish an employee skill database, record the job qualifications of each employee (such as convenience store cashier, car wash equipment operation, emergency handling), work efficiency (such as the number of service orders per unit time), and historical performance (such as customer satisfaction scores), and form a personnel ability profile. Second, combine the period demand forecast output in step S103 (such as the peak passenger flow in the convenience store during the morning rush hour, the low demand for car washing at noon), and use linear programming or genetic algorithms to construct a manpower scheduling model, with the shortest service response time and the lowest manpower cost as the objective function. The constraints include the limit of employees' working hours, the matching degree of job skills, etc. For example, when it is predicted that the passenger flow in the convenience store reaches 200 person-times per hour during the morning rush hour (7:00-9:00), automatically calculate that 3 cashiers and 2 storekeepers need to be allocated, and give priority to employees within 3 kilometers of the gas station, with fast cashier skills and currently idle.

[0067] At the level of real-time scheduling, collect employees' locations, work status (such as busy / idle), and on-site emergencies (such as a reduction in car wash stations due to equipment failures) through Internet of Things devices (such as smart work badges, surveillance cameras), and use reinforcement learning algorithms to dynamically adjust the initial scheduling plan. For example, when a sudden failure occurs at a car wash station, the system immediately recalculates the service capacity of the remaining stations. If the expected queuing time exceeds 15 minutes, trigger a cross-position support mechanism, allocate 1 employee who has received simple car wash training from the convenience store for temporary support, and at the same time push electronic coupons to queuing customers to relieve waiting anxiety. In addition, introduce a flexible shift scheduling mechanism, identify high-frequency demand fluctuation periods based on historical data (such as the fueling peak after the 10th of each month when salaries are paid), negotiate with employees in advance to reserve flexible working hours, and improve the flexibility of personnel allocation through performance incentives (such as a 30% increase in overtime pay). In addition, the physiological and psychological states of employees need to be considered. Monitor fatigue (such as heart rate variability) through wearable devices. When working continuously for 4 hours and the fatigue index exceeds the threshold, automatically trigger a rest reminder and adjust the shift schedule to avoid a decline in service quality due to manpower shortage.

[0068] Further optionally, after dynamically updating the local life service resources callable on the interactive interface of the local life service platform and dynamically allocating the human resources of the gas station in step S104, it is also possible to detect whether the local life service providers in the local life service platform reach the set service resource volume. If the set service resource volume is not reached, alternative local life service providers are screened from the candidate service provider pool according to the preset rules, and the alternative local life service providers are re-connected in the local life service platform. Subsequently, the change trend of user traffic is detected. If a sudden increase in user traffic is detected, historical user data is analyzed to find the matching reasons for the sudden increase in traffic, and the service demand growth area is predicted based on the real-time user traffic. Based on the reasons for the sudden increase in traffic and the service demand growth area, the local life service providers and the types of local life services that can be provided are dynamically expanded in the local life service platform.

[0069] Specifically, after the dynamic update of the interactive interface resources of the local life service platform and the allocation of the human resources of the gas station are completed, continuously monitor the service resource supply situation of the local life service providers, and collect the inventory data and service capacity data (such as delivery capacity, order acceptance volume) of each service provider in real time, and compare them with the set service resource volume threshold. When the food inventory of a certain catering service provider is lower than the threshold, or the order acceptance volume of the express delivery service provider is insufficient, the service provider screening mechanism is immediately triggered. From the pre-established candidate service provider pool, comprehensively consider factors such as the user's historical consumption preferences (such as the cuisine and shopping brands often selected by the user), the service quality score of the candidate service provider, price competitiveness, and geographical location distance, calculate the comprehensive scores of each candidate service provider through a weighted scoring model, screen out the alternative service provider with the highest score, and quickly complete processes such as interface docking and data synchronization, and reconnect in the local life service platform to ensure the continuity and stability of the service.

[0070] Meanwhile, the system monitors the changing trend of user traffic in real time. By analyzing historical user traffic data and combining it with current real-time traffic data, time series analysis algorithms and anomaly detection algorithms are used to identify sudden increases in user traffic. Once a sudden increase in traffic is detected, historical user data is immediately traced back, and the reasons for the sudden increase in traffic are analyzed and found from multiple dimensions such as holidays, surrounding activities, weather changes, and platform promotion activities. For example, it is found that a certain sudden increase in traffic coincides with the time of a large-scale promotion event held in the surrounding business district, or there is a regular increase in traffic during holidays. Then, based on the real-time user traffic data, combined with historical consumption data and user behavior patterns, machine learning algorithms (such as regression analysis and clustering analysis) are used to predict the areas of service demand growth, and it is determined that areas such as food delivery, shopping consumption, and entertainment services may experience a surge in demand. Finally, based on the reasons for the sudden increase in traffic and the predicted areas of service demand growth, the system actively negotiates cooperation with candidate service providers in relevant fields, quickly completes processes such as qualification review, interface development, and service listing of new service providers, and dynamically expands the matching local life service providers and the types of local life services available on the local life service platform, such as introducing more surrounding food and beverage merchants and increasing the sale of special products, to timely meet the sudden increase in user service demand and improve user experience and platform service capabilities.

[0071] Exemplarily, regression algorithms (such as random forest and XGBoost) establish a non-linear mapping relationship between service demand and features such as time, weather, and promotion activities based on historical consumption data, and output the demand growth probability of each service type.

[0072] Exemplarily, matrix factorization algorithms (such as ALS and SVD) are used to mine high-dimensional sparse user behavior data and extract potential associations between user preferences and service types (such as users who frequently buy snacks are more likely to need coffee services).

[0073] Step S105: For users on the local life service platform, collaborative filtering algorithms and reinforcement learning algorithms are used to generate personalized service lists for users, and based on the personalized service lists, interact with users on the interaction interface to provide corresponding local life service resources for users.

[0074] As an optional embodiment, in step S105, for users on the local life service platform, collaborative filtering algorithms and reinforcement learning algorithms are used to generate personalized service lists for users, and based on the personalized service lists, interact with users on the interaction interface to provide corresponding local life service resources for users, including:

[0075] For users in the local life service platform, a collaborative filtering algorithm is used to dynamically reconstruct the user portrait in real time, and a reinforcement learning algorithm is used to dynamically reconstruct the service capability portrait in real time. Based on the user portrait and service capability portrait, a personalized service list is generated for the user in combination with the user's current location, time, and historical preferences. A service matching strategy is generated for the user in combination with the user's historical consumption data, currently required business attributes, inventory of surrounding service points, and service time windows. Based on the personalized service list and the service matching strategy, the corresponding personalized push information is displayed in the interactive interface of the local life service platform to interact with the user and provide the user with corresponding local life service resources.

[0076] Specifically, in step S105, a dynamic user profile is constructed by analyzing the user's historical behavior data (such as clicks, favorites, and consumption records) and the behavioral similarities between users. For example, if both user A and user B frequently visit "pet service" resources and are geographically close, the system will recommend the "pet beauty shop" preferred by user B to user A, and continuously revise the profile based on user A's real-time feedback (such as whether to click or make an appointment). The algorithm solves the problem of insufficient user preference data in the cold start phase by mining the implicit associations of user groups, and at the same time uses real-time interaction data (such as multiple searches for "car wash services" on the same day) to dynamically adjust the profile weight to ensure that the recommended content fits the user's current needs.

[0077] Furthermore, the status of local life service resources (such as the busy and idle status of gas station employees, restaurant inventory, and real-time location of delivery personnel) is abstracted as "environmental status", and the recommendation strategy (such as whether to give priority to a service provider and adjust the service ranking) is defined as "action", and the model is trained through reward mechanisms (such as user click-through rate and order completion rate). For example, when the user demand for "fast food delivered within 30 minutes" surges during a certain period of time, the reinforcement learning model will prioritize merchants with sufficient inventory and matching delivery range and place them at the top of the recommendation list, and adjust subsequent strategies based on the actual order conversion rate. The algorithm dynamically adapts to the real-time fluctuations of service resources through continuous trial and error optimization (such as automatically reducing the recommendation weight of a service provider when a sudden order causes tight transportation capacity).

[0078] Next, the user profile (such as preference for "cost-effective breakfast"), service profile (such as sufficient inventory in the nearby steamed bun shop and matching the delivery time window of the fried dough stick shop), and real-time context (such as the current time is 7:30 during rush hour) are integrated to generate a prioritized service list. For example, a steamed bun shop that is within a 5-minute walk and supports quick pick-up is recommended to users on their way to work, and real-time information such as "80% inventory remaining" and "estimated pick-up time 5 minutes" are marked. Based on the user's historical consumption data (such as an average customer unit price of 20 yuan), current business attributes (such as "urgent refueling" requires priority response), the status of surrounding service points (such as 3 vehicles in line at gas station A and no queue at gas station B) and time windows (such as users hope to arrive within 10 minutes), matching strategies are generated through heuristic rules and machine learning models (such as greedy algorithms and shortest path algorithms). For example, for users who need "urgent refueling", the system prioritizes the gas station that is closest and has idle employees, and triggers the navigation function and quick verification interface to reduce user waiting time.

[0079] Therefore, through real-time data-driven dual-profile reconstruction (such as user profile and service power profile), the recommended content is dynamically adjusted with user behavior and resource status, which improves the recommendation click rate and shortens the average user decision time compared with the static recommendation algorithm. The reinforcement learning model automatically balances traffic according to the real-time load of service resources, for example, evenly distributes users in high-demand periods to service providers with sufficient inventory, avoids the decline in service quality caused by overload of a single merchant, and improves resource utilization overall. For example, the context-aware recommendation based on location, time, and historical preferences makes the service list more in line with the real needs of users (such as automatically recommending "car drying service" on rainy days and "parent-child activity venue" on weekends), and the "recommendation relevance" score in the user satisfaction survey is improved to 4.8 points (out of 5 points). For sudden demands (such as the demand for "umbrella purchase" in heavy rain), the collaborative filtering algorithm quickly identifies the preferences of user groups in similar scenarios, and the reinforcement learning model simultaneously adjusts the resource priority to achieve a minute-level response from demand identification to service listing, which is more efficient than traditional manual scheduling.

[0080] In addition, collaborative filtering can be used to explore niche demands (such as "handmade leather goods customization" and "travel souvenir purchase"), and reinforcement learning can be used to optimize the display strategy of low-frequency services, thereby increasing the exposure rate of the platform's long-tail services by 50%, driving a 45% increase in the number of orders for unpopular services, and expanding the diversity of the platform's service ecosystem.

[0081] First, in the above steps, the collaborative filtering algorithm is used to dynamically reconstruct the user portrait in real time, including:

[0082] Semantic extraction is performed based on user registration data to obtain user basic attribute features; and the first browsing behavior in the user's historical behavior is analyzed to obtain initial browsing behavior features; meta-learning is used to transfer the initial migration vector from similar user groups, and the user basic attribute features and the initial browsing behavior features are fused to obtain the initial user portrait of new users. User historical consumption data is extracted, and the product ID, amount, and category in the user historical consumption data are mapped into low-dimensional vectors through the TransE algorithm, and normalized according to the consumption timestamp to generate a time series vector containing a time decay factor, which is used as the consumption time series vector of the user. A user-product interaction graph is constructed using the graph convolutional network GCN, with users and products as nodes, and user real-time behavior and user historical behavior respectively forming connection edges. The attention weight of key product nodes in the user behavior path is calculated through the graph attention mechanism GAT to generate the behavior feature vector of the user. User behavior events are captured in real time through the Flink stream processing engine, and the neural collaborative filtering NCF model is used to process the high-dimensional sparse features in the user behavior events. A multi-modal input feature including the consumption time series vector, the behavior feature vector, and the high-dimensional sparse feature is constructed, and the non-linear interaction feature of the multi-modal input feature is learned through a multi-layer perceptron, and the user preference embedding vector is output. The user preference embedding vector is updated to the initial user portrait or the user portrait reconstructed and updated last time to realize the real-time reconstruction of the user portrait.

[0083] Specifically, in the process of real-time reconstructing the user portrait, the system constructs an intelligent model that dynamically perceives user preferences through the fusion of multi-dimensional feature extraction and deep neural networks. First, for newly registered users, the system uses natural language processing technology to perform semantic parsing on the registration text, extracts basic attribute features such as age, occupation, and interest, and analyzes features such as product category and stay duration in the first browsing behavior sequence to form an initial behavior portrait. To solve the cold start problem, the model uses the meta-learning algorithm to transfer knowledge from similar user groups with similar geographical locations and registration times, and quickly generates an initial user portrait, enabling new users to receive personalized recommendations after registration.

[0084] For historical consumption data, the knowledge graph embedding algorithm TransE is used to map information such as product ID, amount, and category into a low-dimensional vector space, and a time decay factor is introduced to give higher weights to recent consumption behaviors, so as to capture the dynamic changes of user preferences. In terms of behavior feature extraction, the model constructs a user-product interaction graph and uses the graph convolutional network (GCN) and the graph attention mechanism (GAT) to analyze the key product nodes in the user browsing path, such as identifying high-frequency associated products such as "bread", and generating a behavior feature vector reflecting the user's current interests.

[0085] At the real-time data stream processing level, the Flink stream engine captures user behavior events with millisecond-level latency, processes high-dimensional sparse features (such as device type, membership level) through the Neural Collaborative Filtering (NCF) model, and fuses the consumption time series vector, behavior feature vector, and real-time interaction features into a multi-modal input. The multi-layer perceptron network further learns the non-linear interactions between these features, such as the preference weight of "rainy day + SUV owner" for the "interior cleaning" service, and finally outputs the user preference embedding vector. This vector is incorporated into the historical portrait through a residual update mechanism to achieve portrait iteration at the minute level, enabling the recommendation system to respond to user behavior changes in real time.

[0086] This hybrid architecture that combines semantic analysis, knowledge transfer, graph neural networks, and stream processing effectively solves the problems of difficult cold start, poor real-time performance, and inability to capture dynamic changes in user preferences in traditional recommendation systems. In practical applications, this model improves the accuracy of new user recommendations by 65%, shortens the portrait update latency to 1.2 seconds, and increases the click-through rate of personalized services by 42%, significantly enhancing the user experience and commercial value of the local life service platform.

[0087] Specifically, in the above steps, semantic parsing is performed on user registration data (such as extracting text features such as "prefers coffee" and "car owner" through BERT), combined with the first browsing behavior (such as the first 3 times of browsing "car wash" and "snack" products) to generate an initial feature vector, and the meta-learning algorithm is used to transfer the initial embedding vector from user groups with similar geographical locations and registration information, quickly constructing the basic portrait of new users. For example, the MAML (Model-Agnostic Meta-Learning) algorithm is used to transfer the initial embedding vector from the historical similar user group (such as the same geographical location and similar registration time). For example, if the registration information similarity between the new user A and the existing user B reaches 80%, then directly transfer some preference weights of B (such as the "coffee preference" weight of 0.6) to the initial portrait of A for quick cold start.

[0088] For existing users, the historical consumption data (such as product ID = 123, amount = 30 yuan, category = snack) is mapped into a low-dimensional time series vector through the TransE algorithm, and exponentially weighted decay is performed according to the consumption timestamp (weight of data in the past 7 days is 0.8, weight over 30 days is 0.2) to capture the time dynamics of consumption preferences.

[0089] The TransE algorithm is a technique for knowledge graph embedding, aiming to map entities and relationships in a knowledge graph into a low-dimensional vector space to capture the semantic relationships between them. In the construction of user portraits, the TransE algorithm regards information such as product IDs, amounts, and categories in the user's historical consumption data as entities and relationships in the knowledge graph. By learning the low-dimensional vector representations of these entities and relationships, it can better understand the user's consumption behavior and preferences. For example, mapping different product IDs into the vector space makes similar products closer in the space, so that potential interests of users in different products can be discovered through vector calculations. Further optionally, TransR / CTransR maps entities and relationships to different vector spaces, and entities generate new representations through projection matrices in the relationship space, which can handle relationship diversity more flexibly. CTransR further considers subtypes of the same type of relationships to improve the modeling fineness. TransD replaces the fixed projection matrix with a dynamic mapping matrix, reducing the parameter complexity and considering the semantic diversity of entities and relationships, which is suitable for large-scale sparse data scenarios (such as long-tail products in user consumption data). TransE-Att (combining the attention mechanism) introduces the attention mechanism to learn the importance of entity attributes. For example, in user consumption data, attributes such as product categories, prices, and brands can be weighted and fused to enhance the semantic richness of vector representations. The time-series enhanced TransE introduces a time dimension into the vector space. For example, adding timestamp embeddings or time decay factors to entities and relationships enables the model to capture changes in user preferences over time (such as seasonal consumption patterns), which fits the requirement of the "consumption time-series vector" in the user portrait. The adaptive regularization TransE dynamically adjusts the regularization term to balance the learning weights of the model for high-frequency consumption behaviors (such as daily necessities) and low-frequency behaviors (such as durable goods), avoiding over-dilution of low-frequency data.

[0090] Meanwhile, use GCN to construct a user-product interaction graph (user nodes and product nodes are connected by "browse" and "purchase" edges), and calculate the attention weights of key nodes in the user behavior path through GAT (such as the weight of the "Red Bull" node with high-frequency purchases is increased), generating a behavior feature vector containing real-time behaviors (such as currently browsing "umbrella").

[0091] After real-time data is captured by the Flink stream processing engine, it is input into the neural collaborative filtering (NCF) model. This model concatenates the consumption time-series vector, the behavior feature vector, and the high-dimensional sparse user basic attributes (such as gender, car model) as multi-modal inputs, and learns non-linear interaction features (such as the preference weight of "rainy day + SUV owner" for "interior cleaning") through a multi-layer perceptron, outputting a dynamically updated user preference embedding vector, which is finally fused into the initial portrait or historical portrait to achieve minute-level real-time iteration (such as user browsing car wash service at 10:05, and the portrait adding a "car service" preference label at 10:08).

[0092] It is understandable that the NCF model is a neural network-based recommendation algorithm that combines user and item features and makes personalized recommendations by learning the interaction patterns between users and items. During the real-time reconstruction of the user profile, the NCF model processes high-dimensional sparse features in user behavior events, such as user device type, membership level, etc. It constructs a multi-layer perceptron (MLP) to learn the non-linear interaction relationships between these features, thus better capturing user preferences. The NCF model takes the feature vectors of users and items as inputs and obtains the preference prediction of users for items through the calculation of the neural network. This model can effectively process high-dimensional sparse data and capture the complex interaction relationships between users and items, thereby improving the accuracy and personalization of recommendations.

[0093] Secondly, in the above steps, the reinforcement learning algorithm is used to dynamically reconstruct the service force profile in real time, including: defining local life service providers as independent agents and the local life service platform as a coordinating agent; collecting multi-dimensional state data of local life service providers, and using reinforcement learning to construct the service provider state space corresponding to each agent to obtain the initial service force profile; the multi-dimensional state data includes: service quality, response speed, user evaluation, service load; using a graph attention network, based on the geographical location of local life service providers, to establish spatial association relationships between different agents, enabling different agents to perceive the states of surrounding agents through the environmental state transition model, and obtaining the collaborative policy vector between different agents by extracting graph structure features; based on the historical performance of local life service providers, dynamically adjust the service evaluation weights of the coordinating agent through a credit assignment mechanism; using the adjusted collaborative policy vector and service evaluation weights, and according to the elastic evaluation period associated with the activity hotspots, iteratively update the service provider interaction network formed by the independent agent and the coordinating agent, and reconstruct the service force profile corresponding to each independent agent.

[0094] Exemplarily, each local service provider is defined as an independent agent (such as convenience store A and car wash B), and the platform acts as a coordinating agent to construct a four-dimensional state space including service quality (order completion rate of 95%), response speed (average order acceptance time of 2 minutes), user evaluation (4.8 stars), and service load (current number of processed orders is 15). A spatial association graph is established based on the geographical locations of service providers (such as merchants within 500 meters) through a graph attention network (GAT). Agents perceive the states of surrounding service providers through an environmental state transition model (for example, when convenience store A finds that car wash B has a long queue, it automatically increases the recommendation weight of "in-car snack packs"), extracts graph structure features to generate a collaborative strategy vector (such as "shared inventory" and "joint promotion" strategies). Based on the historical performance data recorded on the blockchain (such as on-time delivery rate of 98% in the past 30 days), the evaluation weight of the coordinating agent is dynamically adjusted through a credit assignment mechanism (the weight of service providers with excellent performance is increased by 20%), and the evaluation cycle is dynamically shortened according to the activity hotspots (from 15 minutes to 3 minutes). The interaction network of service providers is iteratively updated through a multi-agent reinforcement learning (MARL) algorithm (such as MADDPG), so that the service force portrait of each agent can reflect its current performance ability in real time (for example, when the service load of car wash B exceeds the limit due to equipment failure, the "response speed" score in the portrait automatically drops by 15%).

[0095] Furthermore, in the above steps, based on the user portrait and service force portrait, combined with the user's current location, time, and historical preferences, a personalized service list for the user is generated, including:

[0096] Based on the user portrait, combined with the user's current location, time, and historical preferences, predict the user's real-time service needs, and generate a personalized service candidate set based on the prediction results; based on the service force portrait, calibrate the performance status score, service quality score, response speed, and load status of each local service provider in the personalized service candidate set as a service status set, screen the target service providers that meet the service provider ability evaluation conditions from the service status set, and construct the personalized service list with the target service providers.

[0097] Exemplarily, the preference embedding vector output by the user portrait (such as "high-frequency consumption of breakfast + within 500 meters of the current location + working hours at 7:30") is feature-aligned with the service provider status of the service capacity portrait (such as the sufficient inventory of Steamed Bun Shop A and a delivery time of 10 minutes) through the attention mechanism. First, a set of 10-20 candidate services is generated based on the real-time demand prediction of the user (such as the probability of the demand for "convenient breakfast" during the morning rush hour reaching 85% through the LSTM model). Then, the fulfillment status scores of each candidate service provider are calibrated through the service capacity portrait (combining the comprehensive service quality, response speed, and load situation, with the threshold set to 0.7). The target service providers with scores ≥ 0.7 are selected (such as retaining Steamed Bun Shop A and excluding Steamed Bun Shop B with insufficient inventory). After sorting by distance priority (prioritizing within 500 meters) and user preference weights (such as preferentially displaying whole wheat bread options for users with a preference for "sugar-free foods"), a personalized service list is formed (such as recommending "Steamed Bun Shop A - Fresh Meat Buns - 90 remaining in stock - 3 minutes' walk" at the top).

[0098] Next, in the above steps, a service matching strategy for the user is generated by combining the user's historical consumption data, the current required business attributes, the inventory of surrounding service points, and the service time window.

[0099] Exemplarily, by integrating the user's historical consumption data (such as an average consumption of 30 yuan), the current business attributes (such as "urgent refueling" needs to be processed first), the inventory of surrounding service points (such as the inventory of 95# gasoline at Gas Station A is 5000 liters), and the service time window (the user hopes to complete the service within 10 minutes), the optimal matching solution is obtained through a mixed-integer programming model. For example, for the composite demand of "car wash + convenience store shopping", the algorithm preferentially matches the gas station where the car wash bay is idle and the convenience store has sufficient inventory, while considering the time window constraint (40 minutes for car wash + 10 minutes for shopping, total time consumption ≤ 60 minutes). The optimal path is calculated through the Dijkstra algorithm (such as recommending Gas Station C, 1.2 kilometers away from the current location, with an estimated arrival time of 8 minutes), and an integrated strategy including service reservation, inventory reservation, and route navigation is generated.

[0100] Finally, in the above steps, based on the personalized service list and the service matching strategy, corresponding personalized push information is displayed in the interaction interface of the local life service platform to interact with the user and provide the user with corresponding local life service resources.

[0101] The above mechanism realizes the dynamic and precise matching of "user demand-service supply". The delay of user portrait reconstruction is controlled within 200ms, and the construction time of new user cold start portrait is shortened from 24 hours of traditional methods to 8 minutes, and the fit between recommended content and user real-time preferences is improved by 40%; the service force portrait reflects the status changes of service providers in real time, which increases the resource scheduling response speed in sudden load scenarios by 60% and reduces the service quality non-compliance rate by 25%; the personalized service list increases the recommendation click rate by 35% through situational awareness (location + time + preference), and shortens the average user decision time from 3 minutes to 1.5 minutes. The service matching strategy combines inventory and time window optimization to improve the completion efficiency of composite services and reduce user waiting time. Finally, the overall service conversion rate of the platform is improved, the user repurchase rate is increased, and a data-driven intelligent recommendation closed loop is formed, which effectively improves the personalization and convenience of local life services.

[0102] As an optional embodiment, in step S105, based on the personalized service list and the service matching strategy, the corresponding personalized push information is displayed in the interactive interface of the local life service platform, and after interacting with the user, the user's switching behavior between the mobile terminal and its in-vehicle system can also be detected. If the switching behavior is detected, the interaction data integration analysis is started synchronously, the interactive interface adapted to the switched device is re-layouted, and the interaction information before the switch is synchronously displayed in the switched device through the re-layout interactive interface, so as to realize the integrated interaction between the mobile terminal and the in-vehicle system.

[0103] In step S105, the interactive integration across devices is achieved by detecting the switching behavior of the user between the mobile terminal and the vehicle-mounted system.

[0104] Specifically, when the user enters the vehicle with a mobile device and triggers Bluetooth connection, NFC sensing, or geographical location fencing, the system captures the switching event in real time through device fingerprint recognition technology (such as binding IMEI with the in-vehicle system ID). Subsequently, the interactive data integration and analysis module is launched to associate the interactive data before switching (such as the car wash service details page browsed on the mobile device and the convenience store products added to favorites) based on the timestamp and user ID. The data is lightly processed by the edge computing node (such as compressing the picture size and streamlining the text information) to adapt to the display resolution and interactive logic of the in-vehicle system (such as giving priority to voice control and optimizing button operations). The interactive interface is dynamically reconstructed according to the screen size and operation method of the in-vehicle device (such as steering wheel buttons and voice commands). For example, the text and picture mixed layout on the mobile device is adjusted to a large-font, high-contrast card view, and a voice interaction entry (such as "Click here to place an order by voice") is embedded in the appropriate position. At the same time, the interactive state before switching (such as the scrolling position and the selected service specifications) is synchronized to the in-vehicle system through a WebSocket long connection, so that the in-vehicle interface automatically displays the operations not completed by the user on the mobile device (such as the unsubmitted car wash reservation order), avoiding repeated input. If the switch occurs during the service (such as the user has placed an order but has not arrived at the store), the in-vehicle interface will give priority to displaying the navigation function and service countdown, generate the optimal route in combination with real-time traffic data, and announce the order details through in-vehicle voice broadcast.

[0105] Through device linkage perception, real-time data synchronization, and interface adaptive rendering, a seamless interactive link between mobile pre-browsing and in-vehicle fast operation is constructed, reducing the cross-device operation latency and improving the user task completion efficiency. At the same time, the interface reconstruction based on scenario perception (such as automatically dimming the brightness in the night in-vehicle mode) further improves the operation safety in the driving scenario, reduces the interactive errors caused by distraction, and reduces the interactive accidental touch rate, forming an intelligent service closed-loop for multi-device collaboration.

[0106] Further optionally, collect the energy consumption data of gas station equipment, and perform intelligent scheduling on equipment such as lighting systems, air conditioning systems, fuel dispensers, and charging piles in the gas station in combination with the prediction results.

[0107] Specifically, intelligent sensors (such as power metering modules, temperature sensors, and intelligent charging pile controllers) deployed on various devices at gas stations are used to collect real-time device energy consumption data, including the real-time power of the lighting system, the operating frequency of the air conditioning system, the standby energy consumption of fuel dispensers, and the charging load of charging piles. After the data is cleaned and filtered by the edge computing node, it is transmitted to the platform data center through the MQTT protocol. Combining the user traffic prediction (such as the vehicle flow during peak hours) output in step S103, weather data (such as high-temperature warnings), and the historical device operation data, a time series prediction model (such as LSTM-Attention) is used to construct the energy consumption demand prediction curves of each device. For example, it is predicted that the charging pile load will reach 85% of the peak value and the energy consumption of the air conditioning system will increase by 20% from 2 to 4 pm on weekends. Based on the prediction results, the intelligent scheduling module optimizes the device operation in the following ways: for the lighting system, a dynamic dimming strategy is adopted, and the brightness is adjusted to 30% during off-peak hours (such as from 1 to 5 am), and automatic lights-off in unoccupied areas is achieved through infrared sensors; the air conditioning system combines the indoor and outdoor temperature and humidity and the personnel density, automatically adjusts the set temperature through a fuzzy control algorithm (such as maintaining 26°C during summer peak hours and raising it to 28°C during off-peak hours), and optimizes the start-stop frequency of the compressor to reduce standby energy consumption; the fuel dispenser and the charging pile implement off-peak scheduling, automatically perform device maintenance operations during the grid valley electricity period (such as at night), and avoid overloading of a single device through a load balancing algorithm during peak hours (such as when the load of a certain charging pile exceeds 70%, automatically guiding vehicles to adjacent idle piles).

[0108] In addition, a reinforcement learning algorithm (such as DQN) is further introduced to construct a device collaborative scheduling model. The energy consumption of each device, the user service experience (such as the refueling waiting time), the scheduling cost (such as the valley electricity price), etc. are used as the state space, and operations such as device start-stop and power adjustment are used as the action space. The scheduling strategy is optimized through a long-term reward function (such as the comprehensive energy consumption reduction rate + service satisfaction). For example, when it is predicted that the number of inbound vehicles will decrease due to heavy rain, the model automatically reduces the lighting power in non-business areas to 10%, switches the air conditioning to the fresh air mode, and at the same time starts the regular inspection program of the fuel dispenser, reducing the overall energy consumption while ensuring basic services. The entire scheduling process dynamically displays the device status and energy consumption data through a real-time monitoring large screen, supports manual intervention and automatic mode switching, forms a closed loop of "data collection - prediction analysis - intelligent scheduling - effect feedback", and finally realizes the reduction of the energy consumption of gas station devices, the reduction of operation and maintenance costs, and at the same time ensures service continuity and user experience.

[0109] In the embodiments of this application, the efficient integration of multiple heterogeneous local life service providers is achieved through the local life service integration method, which has strong dynamic resource allocation and flexible supply network management capabilities.

[0110] Please refer to Figure 2 , Figure 2A local life service integration device 200 provided by an embodiment of the present application. The local life service integration device includes: a collection module, which is used to integrate multiple heterogeneous local life service providers into a gas station service platform to obtain a local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and to collect the real-time service data of each local life service provider in real time;

[0111] A configuration module, which is used to predict the non-oil product sales volume fluctuation of the gas station based on the real-time service data, and dynamically adjust the order quantity and delivery cycle of upstream and downstream suppliers according to the prediction result; construct a multi-dimensional data analysis model for the gas station scenario, integrate the historical user traffic data of the gas station, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predict the user traffic and the demand quantity of various services at different time periods every day, and configure the corresponding local life service time window based on the prediction result to dynamically optimize the local life service resource configuration of the gas station at different time periods; dynamically update the local life service resources that can be called by the interactive interface in the local life service platform based on the order quantity and delivery cycle of upstream and downstream suppliers and the local life service resource configuration at different time periods, and dynamically allocate the human resources of the gas station;

[0112] An interaction module, which is used for users in the local life service platform to generate a personalized service list for the user by using the collaborative filtering algorithm and the reinforcement learning algorithm, and interact with the user in the interactive interface based on the personalized service list to provide the corresponding local life service resources for the user.

[0113] In some embodiments, the local life service integration device can be applied to a terminal device.

[0114] It should be noted that for the convenience and conciseness of description, the specific working process of the above-described local life service integration device can refer to the corresponding process in the foregoing embodiment of the local life service integration method, and will not be elaborated herein.

[0115] Please refer to Figure 3 , Figure 3 A schematic block diagram of the structure of a terminal device provided by an embodiment of the present application.

[0116] As Figure 3As shown, the terminal device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a bus 303, which is, for example, an I2C bus. Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 may be a central processing unit, or it may also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc. Specifically, the memory 302 may be a Flash chip, a read-only memory disk, an optical disk, a USB flash drive, or a mobile hard disk, etc.

[0117] Those skilled in the art can understand that Figure 3 the structure shown in is only a block diagram of some structures related to the solution of the embodiment of the present application, and does not constitute a limitation on the terminal device to which the solution of the embodiment of the present application is applied. A specific server may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. Among them, the processor is used to run the computer program stored in the memory and implement any one of the local life service integration methods provided by the embodiment of the present application when executing the computer program. It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described terminal device can refer to the foregoing embodiment of the local life service integration method and will not be elaborated herein.

[0118] The embodiment of the present application also provides a storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the local life service integration methods provided in the specification of the embodiment of the present application.

Claims

1. A local life service integration method, characterized in that, The method includes: Integrating multiple heterogeneous local life service providers into the gas station service platform to obtain a local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and collecting real-time service data of each local life service provider in real time; Predicting the non-oil product sales volume fluctuations of the gas station based on the real-time service data, and dynamically adjusting the order quantities and delivery cycles of upstream and downstream suppliers according to the prediction results; Constructing a multi-dimensional data analysis model for the gas station scenario, integrating the historical user traffic data of the gas station, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predicting the user traffic and various service demand quantities at different time periods of each day, and configuring corresponding local life service time windows based on the prediction results to dynamically optimize the local life service resource allocation of the gas station at different time periods; Dynamically updating the local life service resources that can be called by the interactive interface in the local life service platform based on the order quantities and delivery cycles of upstream and downstream suppliers and the local life service resource allocation at different time periods, and dynamically allocating the human resources of the gas station; For the users in the local life service platform, using collaborative filtering algorithms and reinforcement learning algorithms to generate personalized service lists for the users, and interacting with the users in the interactive interface based on the personalized service lists to provide corresponding local life service resources for the users.

2. The method according to claim 1, wherein The step of, for the users in the local life service platform, using collaborative filtering algorithms and reinforcement learning algorithms to generate personalized service lists for the users, and interacting with the users in the interactive interface based on the personalized service lists to provide corresponding local life service resources for the users includes: For the users in the local life service platform, using collaborative filtering algorithms to dynamically reconstruct user portraits in real time, using reinforcement learning algorithms to dynamically reconstruct service force portraits in real time, and generating personalized service lists for the users based on the user portraits and service force portraits, combined with the user's current location, time, and historical preferences; Generating a service matching strategy for the user by combining the user's historical consumption data, current required business attributes, inventory of surrounding service points, and service time windows; Based on the personalized service lists and the service matching strategy, displaying corresponding personalized push information in the interactive interface of the local life service platform, interacting with the users, and providing corresponding local life service resources for the users.

3. The method according to claim 2, wherein After the step of, based on the personalized service lists and the service matching strategy, displaying corresponding personalized push information in the interactive interface of the local life service platform and interacting with the users, it further includes: Detecting the switching behavior between the user's mobile device and the in-vehicle system; If a switching behavior is detected, synchronously starting the integrated analysis of interactive data, re-layouting the interactive interface adapted to the switched device, and synchronously displaying the interactive information before switching in the switched device through the re-layouted interactive interface to achieve the integrated interaction between the mobile device and the in-vehicle system.

4. The method according to claim 2, wherein The step of using collaborative filtering algorithms to dynamically reconstruct user portraits in real time includes: Semantic extraction is performed based on user registration data to obtain user basic attribute features; and the first browsing behavior in the user's historical behavior is analyzed to obtain initial browsing behavior features; meta-learning is used to transfer the initial migration vector from a similar user group, and the user basic attribute features and the initial browsing behavior features are fused to obtain the initial user portrait of the new user; Extract the user's historical consumption data, map the product ID, amount, and category in the user's historical consumption data into low-dimensional vectors through the TransE algorithm, and perform normalization processing according to the consumption timestamp to generate a time series vector containing a time decay factor, which is used as the user's consumption time series vector; Use the graph convolutional network GCN to construct a user-product interaction graph, with users and products as nodes, and the user's real-time behavior and historical behavior respectively forming connection edges. The attention weight of the key product nodes in the user's behavior path is calculated through the graph attention mechanism GAT to generate the user's behavior feature vector; The Flink stream processing engine is used to capture user behavior events in real time, and the neural collaborative filtering NCF model is used to process the high-dimensional sparse features in the user behavior events. A multi-modal input feature including the consumption time series vector, the behavior feature vector, and the high-dimensional sparse feature is constructed, and the non-linear interaction feature of the multi-modal input feature is learned through a multi-layer perceptron to output the user preference embedding vector; Update the user preference embedding vector to the initial user portrait or the user portrait reconstructed and updated last time to realize the real-time reconstruction of the user portrait.

5. The method according to claim 2, characterized in that, The real-time dynamic reconstruction of the service force portrait using the reinforcement learning algorithm includes: Define local life service providers as independent agents and the local life service platform as a coordinating agent; Collect multi-dimensional state data of local life service providers, and use reinforcement learning to construct the service provider state space corresponding to each agent to obtain the initial service force portrait; the multi-dimensional state data includes: service quality, response speed, user evaluation, service load; Adopt a graph attention network to establish a spatial association relationship between different agents based on the geographical location of local life service providers, enable different agents to perceive the states of surrounding agents through the environmental state transition model, and obtain the collaborative strategy vector between different agents by extracting graph structure features; Based on the historical performance of local life service providers, dynamically adjust the service evaluation weight of the coordinating agent through a credit assignment mechanism; Adopt the adjusted collaborative strategy vector and service evaluation weight, and iterate and update the service provider interaction network formed by the independent agent and the coordinating agent according to the elastic evaluation period associated with the activity hotspot to reconstruct the service force portrait corresponding to each independent agent.

6. The method according to claim 2, characterized in that, The generation of the user's personalized service list based on the user portrait and the service force portrait, combined with the user's current location, time, and historical preferences, includes: Based on the user portrait, combined with the user's current location, time, and historical preferences, predict the user's real-time service needs, and generate a personalized service candidate set based on the prediction results; Based on the service force portrait, calibrate the fulfillment status score, service quality score, response speed, and load condition of each local life service provider in the personalized service candidate set as the service status set, screen for target service providers whose service status set meets the service provider ability evaluation conditions, and construct the personalized service list with the target service providers.

7. The method according to claim 1, characterized in that, The method for predicting the non-oil product sales volume fluctuation of a gas station based on the real-time service data and dynamically adjusting the order quantity and delivery cycle of upstream and downstream suppliers includes: Using historical non-oil product sales data and associated geographical locations, surrounding business ecosystems, emergencies, seasonal data, and weather data, determine the optimal combination weights through Bayesian optimization, and construct a three-dimensional dynamic decision-making model including sales volume prediction, inventory status, and delivery capacity; Establish a supply chain virtual mirror corresponding to the real-time service data; Using the three-dimensional dynamic decision-making model, adaptively construct multiple ordering strategies matching the supply chain virtual mirror, dynamically simulate each ordering strategy, and obtain the dynamic simulation results under each ordering strategy; where the dynamic simulation results include: predicted sales volume fluctuation value, predicted inventory fluctuation value, and predicted delivery load value; Introduce the Copula function as a connection function, identify the tail correlation between the sales volume fluctuation and supply delay phenomena in the supply chain virtual mirror by constructing a non-parametric joint distribution model, and obtain the tail risk characteristics under each ordering strategy; Based on the tail risk characteristics, screen for the target ordering strategy with the inventory fluctuation reaching the safety inventory level, and execute the target ordering strategy to balance the order quantity and delivery cycle of upstream and downstream suppliers.

8. The method according to claim 1, characterized in that, After dynamically updating the local life service resources that can be called by the interactive interface in the local life service platform and dynamically allocating the human resources of the gas station, it further includes: Detect whether the local life service providers in the local life service platform reach the set service resource quantity. If they do not reach the set service resource quantity, screen for alternative local life service providers from the candidate service provider pool according to the preset rules, and reconnect the alternative local life service providers in the local life service platform; Detect the change trend of user traffic. If a sudden increase in user traffic is detected, analyze the historical user data to find the matching reasons for the sudden increase in traffic, and predict the service demand growth area based on the real-time user traffic; Based on the reasons for the sudden increase in traffic and the service demand growth area, dynamically expand the matching local life service providers and the types of local life services that can be provided in the local life service platform.

9. An integrated device for local life services, characterized in that, The device includes the following modules: An acquisition module, which is used to integrate multiple heterogeneous local life service providers into the gas station service platform to obtain the local life service platform by formulating a standardized interface protocol, unifying the data exchange format, and establishing a resource update strategy, and to collect the real-time service data of each local life service provider in real time; A configuration module, which is used to predict the fluctuation of non-oil product sales volume of a gas station based on the real-time service data, and dynamically adjust the order quantity and delivery cycle of upstream and downstream suppliers according to the prediction results; construct a multi-dimensional data analysis model for the gas station scenario, integrate the historical user traffic data of the gas station, the historical demand data of different local life services, the surrounding traffic flow, weather change data, and community activities, predict the user traffic and the demand for various services at different time periods of each day, and configure the corresponding local life service time window based on the prediction results to dynamically optimize the allocation of local life service resources in the gas station at different time periods; based on the order quantity and delivery cycle of upstream and downstream suppliers, and the allocation of local life service resources at different time periods, dynamically update the local life service resources that can be called by the interactive interface in the local life service platform, and dynamically allocate the human resources of the gas station. An interaction module, which is used for users in the local life service platform, generates a personalized service list for users by using collaborative filtering algorithms and reinforcement learning algorithms, and interacts with users in the interactive interface based on the personalized service list to provide corresponding local life service resources for users.

10. A terminal device, characterized in that, The terminal device includes a processor and a memory; The memory is used for storing a computer program; The processor is used for executing the computer program and implementing the local life service integration method according to any one of claims 1 to 8 when executing the computer program.

Images