Intelligent management system and method for highway comprehensive service

Abstract

The invention discloses an intelligent management system and method for expressway comprehensive services, and belongs to the technical field of intelligent management. A first construction module obtains multi-source heterogeneous data and preprocesses the multi-source heterogeneous data, a feature engineering module carries out feature engineering processing on the processed multi-source heterogeneous data to obtain industry features, and a second construction module carries out feature engineering processing on the industry features; the second construction module inputs the industry characteristics into the trained service model to obtain a service result, the large model module obtains a vertical industry large model based on a large language model, and the engine construction module constructs five service decision engines based on the multi-source heterogeneous data, the industry characteristics, the service result and the vertical industry large model. According to the method, multi-source heterogeneous data can be deeply fused, a unified industry data base can be constructed, post-analysis is changed into pre-prediction and in-process intervention, the perspectiveness and predictability of management are improved, the whole process from analysis to execution can be automatically completed by constructing a decision closed loop, and fine and automatic intelligent management is achieved.

Description

An intelligent management system and method for integrated highway services Technical Field

[0001] This invention belongs to the field of enterprise-level artificial intelligence application technology, specifically relating to an intelligent management system and method for integrated highway services, which is applicable to the operation analysis, strategy generation, resource optimization and risk management of integrated highway service networks. Background Technology

[0002] With the rapid development of Artificial Intelligence (AI) technology, especially big data analytics and machine learning, enterprise-level AI applications are deeply penetrating vertical industries from general-purpose tools. Currently, in the field of business decision support, existing technical solutions mainly exist in the following forms, each with significant limitations: General-purpose AI platforms and analytics tools: These platforms (such as standardized machine learning platforms provided by some cloud service providers) offer general data preprocessing, model training, and deployment capabilities. However, they lack deep integration with the business logic, knowledge systems, and decision-making processes of specific industries (such as transportation and energy services). Enterprise users need to invest significant industry expert resources in secondary development and customization to solve specific business problems, resulting in high technical barriers, long implementation cycles, and high costs. More importantly, these platforms are essentially still "tools," unable to directly output business-specific decision recommendations, let alone achieve automated decision-making across multiple stages.

[0003] Isolated, tool-based AI applications: Some AI solutions on the market target single scenarios, such as standalone sales forecasting systems or customer profiling systems. While these applications can improve efficiency at specific points, they create "data silos" and "algorithm islands." For example, the forecasting system cannot automatically transmit its results to the marketing strategy system, leading to a break in the decision-making chain. They cannot start from the overall business objectives of the enterprise to conduct cross-industry or cross-departmental collaborative optimization, and are ill-equipped to handle comprehensive problems such as "site resource optimization" and "cross-industry traffic redirection" that require multi-source data fusion and complex business logic judgment.

[0004] Traditional Business Intelligence (BI) and Rule Engine Systems: Traditional BI systems excel at multi-dimensional reporting and visualization of historical data, but their analytical capabilities are largely based on preset, static queries and indicators, lacking the ability to predict future trends, deeply attribute anomalies, and uncover unknown patterns. Expert systems based on fixed rules, while capable of encapsulating some business rules, suffer from poor flexibility, cannot autonomously learn and evolve from data, and require frequent manual updates to the rule base when market conditions or business models change, resulting in high maintenance costs and difficulty in handling highly complex non-linear relationships.

[0005] In summary, the core shortcomings of existing technologies are: Lagging analysis: Most systems remain at the level of "telling what happened" and "predicting what will happen," unable to automatically complete decision-making analysis and strategy generation for "what should be done" and "how to do it optimally"; Inability to deeply integrate industry intelligence: General AI technology lacks the "soul" of industry knowledge, while traditional industry information systems lack the "intelligence" of AI, resulting in a failure to deeply integrate the two; Inability to support global optimization decisions: Single-point intelligent applications cannot support the entire chain and global optimization decisions from macro-strategy to micro-execution, from single business models to ecosystem collaboration.

[0006] In the existing integrated transportation service industry (such as gas stations, charging piles, and commercial services in service areas), business decisions face technical challenges such as diverse data sources (such as oil, electricity, retail, and road networks), complex influencing factors (such as weather and holidays), and high real-time requirements.

[0007] Therefore, there is an urgent need to provide an intelligent management system and method for comprehensive highway services to solve the above-mentioned technical problems. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent management system and method for comprehensive highway services, in order to solve the problems of analysis lag, inability to deeply integrate industry intelligence, inability to support global optimization decision-making, diverse data sources, complex influencing factors, and high real-time requirements in the existing technology.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, the present invention provides an intelligent management system for comprehensive highway services, comprising a first construction module, a feature engineering module, a second construction module, a large model module, and an engine construction module; the first construction module is used to acquire multi-source heterogeneous data from the highway energy network, including fuel dispenser transaction data, convenience store POS sales data, charging pile operation status and order data, and highway network gantry traffic flow data; preprocessing the multi-source heterogeneous data to obtain processed multi-source heterogeneous data; and sending the processed multi-source heterogeneous data to the feature engineering module and the large model module; the feature engineering module is used to perform feature engineering processing on the processed multi-source heterogeneous data to obtain industry features, including historical sales sequences of stations, price elasticity coefficients of different customer groups, and holiday moving average sales; and sending the industry features to the second construction module and the large model module; the second construction module is used to acquire historical industry data... According to the data, a pre-built business model is trained based on historical industry data to obtain a trained business model. Industry characteristics are input into the trained business model to obtain business results, which are then sent to a large model module. The large model module is used to integrate multi-source heterogeneous data, industry characteristics, business results, and a pre-built industry knowledge base, using a large language model as the core engine, to obtain a vertical industry large model, which is then sent to an engine construction module. The engine construction module is used to build five major business decision engines based on multi-source heterogeneous data, industry characteristics, business results, and the vertical industry large model. These five business decision engines are used to generate executable decision instructions. Specifically, the five major business decision engines include a business format collaboration and network construction engine for responding to competition and assessing potential, a strategic goal management and sand table simulation engine for decomposing strategic goals, an early warning and root cause management engine for identifying operational anomalies, an operational strategy management engine for generating operational strategies, and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

[0010] In one possible design, a closed-loop processing module is also included. This module is used to obtain executable decision instructions from the five major business decision engines. These executable decision instructions include sales warnings, promotion generation strategies, and / or customer churn recovery strategies. The executable decision instructions are then sent to the corresponding front-line business systems for execution via API interfaces. These front-line business systems include POS systems, CRM systems, and marketing platforms. The closed-loop processing module is also used to obtain the effect data after executing the decision instructions and feed the effect data back to the first construction module.

[0011] In one possible design, the first construction module is further configured to construct a data model based on multi-source heterogeneous data and processed multi-source heterogeneous data. The data model includes a raw data layer, a computational processing layer, and an analysis result layer. The raw data layer stores the multi-source heterogeneous data. The computational processing layer generates industry-specific indicators based on industry business logic. These industry-specific indicators include oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate. It also constructs industry dimension tables and fact tables based on the industry business logic, and builds an industry data warehouse based on the industry-specific indicators, industry dimension tables, and fact tables. The analysis result layer stores the processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines. The first construction module is also configured to obtain effect data fed back from the closed-loop processing module and update the data model in real time based on the effect data.

[0012] Secondly, this invention provides an intelligent management method for integrated highway services, comprising: acquiring multi-source heterogeneous data from the highway energy network, including fuel dispenser transaction data, convenience store POS sales data, charging pile operation status and order data, and highway network gantry traffic flow data; preprocessing the multi-source heterogeneous data to obtain processed multi-source heterogeneous data; performing feature engineering on the processed multi-source heterogeneous data to obtain industry features, including historical sales sequences of stations, price elasticity coefficients of different customer groups, and holiday moving average sales; acquiring historical industry data; training a pre-built business model based on the historical industry data to obtain a trained business model; and inputting the industry features into the trained business model. The business model yields business results; using a large language model as the core engine, it integrates multi-source heterogeneous data, industry characteristics, business results, and a pre-built industry knowledge base to obtain a vertical industry large model; based on multi-source heterogeneous data, industry characteristics, business results, and the vertical industry large model, it constructs five major business decision engines. These five business decision engines are used to generate executable decision instructions. Specifically, the five major business decision engines include a business format collaboration and network construction engine for responding to competition and assessing potential, a strategic goal management and sand table simulation engine for decomposing strategic goals, an early warning and root cause management engine for identifying operational anomalies, an operational strategy management engine for generating operational strategies, and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

[0013] In one possible design, multi-source heterogeneous data is preprocessed to obtain processed multi-source heterogeneous data, including: data cleaning of the multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data; format conversion of the cleaned multi-source heterogeneous data to obtain converted multi-source heterogeneous data; entity data identification of the converted multi-source heterogeneous data to obtain multiple record data of the same entity, and association of the multiple record data of the same entity to obtain associated data; subject domain integration of the converted multi-source heterogeneous data based on a pre-built subject data model to obtain integrated data; time-series alignment of the converted multi-source heterogeneous data to obtain aligned data; and fusion of associated data, integrated data, and aligned data to obtain processed multi-source heterogeneous data.

[0014] In one possible design, after obtaining the processed multi-source heterogeneous data, the process further includes: constructing a data model based on the multi-source heterogeneous data and the processed multi-source heterogeneous data; wherein, the data model includes a raw data layer, a computational processing layer, and an analysis result layer; the raw data layer is used to store the multi-source heterogeneous data; the computational processing layer is used to generate industry-specific indicators based on industry business logic, the industry-specific indicators including oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate, and constructs industry dimension tables and fact tables based on industry business logic, and constructs an industry data warehouse based on industry-specific indicators, industry dimension tables, and fact tables; the analysis result layer is used to store the processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines.

[0015] In one possible design, a pre-built business model is trained based on historical industry data to obtain the trained business model. This includes: performing feature engineering on the historical industry data to obtain historical industry features; obtaining historical business results corresponding to the historical industry data; labeling the historical industry features based on the historical business results to obtain labeled training data; and inputting the labeled training data into the pre-built business model for training to obtain the trained business model.

[0016] In one possible design, the vertical industry big model includes one or more combinations of predictive models, diagnostic models, optimization decision-making models, and insight segmentation models.

[0017] In one possible design, the prediction model is built on a long short-term memory network, the diagnostic model is built on an isolated forest, the optimization decision model is built on a reinforcement learning algorithm, and the insight segmentation model is built on a clustering algorithm.

[0018] The beneficial effects of this invention are as follows: This invention discloses an intelligent management system and method for comprehensive highway services, comprising a first construction module, a feature engineering module, a second construction module, a large model module, and an engine construction module. The engine construction module constructs five major business decision engines based on multi-source heterogeneous data, industry characteristics, business results, and a vertical industry large model. These engines generate executable decision instructions for responding to competition, assessing potential, decomposing strategic goals, identifying operational anomalies, generating operational strategies, evaluating stations, and optimizing scheduling and ordering. The five business decision engines generate decision-making analyses and strategies on "what should be done" and "how to do it optimally," improving the foresight of management and the scientific nature of decision-making. The large model module deeply integrates a large language model, multi-source heterogeneous data, industry characteristics, business results, and a pre-built industry knowledge base, deeply fusing industry knowledge and AI intelligence. Furthermore, by constructing a decision-making closed loop, this invention enables the system to automatically complete the entire process from analysis to execution, improving operational efficiency and achieving refined and automated intelligent management, facilitating application and promotion. Attached Figure Description

[0019] Figure 1 is a block diagram of the intelligent management system for comprehensive highway services provided in an embodiment of the present invention; Figure 2 is a flowchart of the intelligent management method for comprehensive highway services provided in an embodiment of the present invention. Detailed Implementation

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0021] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0022] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0023] Example: As shown in Figure 1, this example provides an intelligent management system for comprehensive highway services. Specifically, comprehensive highway services in this example include, but are not limited to, business services such as gas stations, service areas, and charging piles. The application scenario targeted in this example is comprehensive highway services. In specific implementation, no limitation is made to the application scenario. This example can also be applied to commercial areas, residential areas, or areas centered around highways near highways. The engine construction module generates executable decision instructions, and intelligent operation is realized based on the executable decision instructions. It can realize automatic analysis of business service items and implement the strategies in the executable decision instructions.

[0024] This embodiment provides an intelligent management system for integrated highway services, comprising a first construction module, a feature engineering module, a second construction module, a large model module, and an engine construction module. The first construction module acquires multi-source heterogeneous data from the highway energy network, including transaction data from gas pumps, convenience store POS sales data, charging pile operation status and order data, and highway network gantry traffic flow data. It preprocesses the multi-source heterogeneous data to obtain processed multi-source heterogeneous data and sends it to the feature engineering module and the large model module. The feature engineering module performs feature engineering on the processed multi-source heterogeneous data to obtain industry features, including historical sales sequences at stations, price elasticity coefficients for different customer groups, and holiday moving average sales. It then sends these industry features to the second construction module and the large model module. The second construction module acquires historical industry data and, based on this historical industry data... A pre-built business model is trained to obtain a trained business model. Industry features are input into the trained business model to obtain business results, which are then sent to the large model module. The large model module is used to integrate multi-source heterogeneous data, industry features, business results, and a pre-built industry knowledge base, using a large language model as the core engine, to obtain a vertical industry large model, which is then sent to the engine construction module. The engine construction module is used to build five major business decision engines based on multi-source heterogeneous data, industry features, business results, and the vertical industry large model. These five business decision engines are used to generate executable decision instructions. Specifically, the five major business decision engines include a business format collaboration and network construction engine for responding to competition and assessing potential, a strategic goal management and sand table simulation engine for decomposing strategic goals, an early warning and root cause management engine for identifying operational anomalies, an operational strategy management engine for generating operational strategies, and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

[0025] It should be noted that multi-source heterogeneous data includes, but is not limited to, gas pump transaction records, convenience store POS sales data, charging pile operation status and order data, highway network gantry traffic data (such as vehicle type, time period, and quantity), online membership data and consumption records, and third-party data (such as weather, holidays, and competitor intelligence), without specific limitations here.

[0026] Specifically, the five business decision engines do not operate in isolation, but rather form a collaborative decision-making closed loop through predefined business logic and event-triggered mechanisms. This allows industry expert experience, business rules, and best practices to be encoded into computable and iterative algorithmic processes, thereby achieving full-process automation from business problem perception to strategy generation and evaluation. The five business decision engines include: a business format collaboration and network construction engine, which models site networks and customer cross-business flow based on graph neural networks and uses spatial clustering and regression algorithms to quantify site location potential. Its aim is to achieve dynamic global resource optimization across oil, electricity, and non-oil businesses through multimodal data fusion, breaking down business format data silos; and a strategic goal management and sandbox simulation engine, which decomposes macro-strategic goals into actionable micro-goals and utilizes... Monte Carlo simulation and counterfactual reasoning technologies enable multi-strategy combinations to be simulated and risk-assessed in a digital twin environment, shifting decision-making from "experience-driven" to "simulation-driven." The early warning and root cause management engine, with its built-in causal inference model, automatically drills down into the root causes of operational anomalies along a hierarchical "macro-meso-micro" business indicator model, pinpointing key anomaly nodes and quantifying the causal effects of various influencing factors, thus achieving precise identification of operational problems. The operational strategy management engine generates dynamic operational strategies based on reinforcement learning and price elasticity models, and works in conjunction with the root cause management engine and the sandbox simulation engine to simulate and optimize the strategies. The site operation management engine utilizes data envelopment analysis and / or regression models to generate specific operational plans to improve site efficiency.

[0027] Among them, the executable decision-making instructions include, but are not limited to, instructions for responding to competition and assessing potential, decomposing strategic objectives, identifying operational anomalies, generating operational strategies, evaluating sites, optimizing marketing activities, and optimizing scheduling and ordering.

[0028] In a preferred embodiment, the system further includes a closed-loop processing module. This module acquires executable decision instructions from five business decision engines, including sales warnings, promotion generation strategies, and / or customer churn recovery strategies. The executable decision instructions are then sent to the corresponding front-line business systems (POS system, CRM system, and marketing platform) via API interfaces for execution. The closed-loop processing module also acquires effect data after executing the decision instructions and feeds this data back to the first construction module.

[0029] Furthermore, this feedback data is also used to update the dynamic knowledge graph built within the vertical industry big model. The knowledge graph records historical strategies, execution results, and scenario features in a structured manner, enabling the system to perform semantic matching and strategy recommendation when new problems arise, thereby achieving continuous accumulation and reuse of decision-making wisdom.

[0030] In a preferred embodiment, the first construction module is further configured to construct a data model based on multi-source heterogeneous data and processed multi-source heterogeneous data. The data model includes a raw data layer, a computational processing layer, and an analysis result layer. The raw data layer stores the multi-source heterogeneous data. The computational processing layer generates industry-specific indicators based on industry business logic. These industry-specific indicators include oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate. It also constructs industry dimension tables and fact tables based on the industry business logic, and builds an industry data warehouse based on the industry-specific indicators, industry dimension tables, and fact tables. The analysis result layer stores the processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines. The first construction module is also configured to obtain effect data fed back by the closed-loop processing module and update the data model in real time based on the effect data.

[0031] Based on the aforementioned publicly available information, this system focuses on applying AI technology to industry scenarios (such as highways) to address industry needs. This system differs significantly from existing AI tools, which directly provide answers based on user questions. In contrast, this system delves into various processes within an enterprise, understanding and analyzing problems to provide decision support and achieve intelligent management.

[0032] This embodiment provides an intelligent management system for comprehensive highway services. Through the synergy and triggering mechanism of five major engines, it achieves full-process automation from problem perception to strategy generation, and realizes the leap from data analysis to decision-making. This embodiment can accurately locate the root cause of operational fluctuations through "macro-micro" full-link causal attribution, and achieve precise and targeted governance. At the same time, by constructing a dynamic knowledge graph and reusing it, it has the ability to continuously learn and evolve. Through Monte Carlo simulation, it conducts pre-effect prediction and risk assessment of executable decisions, which greatly reduces the cost of trial and error and improves the scientific nature of executable decisions.

[0033] As shown in Figure 2, the second aspect of this embodiment provides an intelligent management method for integrated highway services. This method can be executed, but is not limited to, by a computer device or virtual machine with certain computing resources, such as a personal computer or smartphone, or by a virtual machine. The intelligent management method includes, but is not limited to, the following steps: S1. Acquiring multi-source heterogeneous data from the highway energy network, including fuel dispenser transaction records, convenience store POS sales data, charging pile operation status and order data, and highway network gantry traffic flow data. Preprocessing the multi-source heterogeneous data yields processed multi-source heterogeneous data. Specifically, in step S1, preprocessing the multi-source heterogeneous data yields processed multi-source heterogeneous data. The process includes: S101. Cleaning the multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data; S102. Converting the format of the cleaned multi-source heterogeneous data to obtain converted multi-source heterogeneous data; S103. Identifying entity data in the converted multi-source heterogeneous data to obtain multiple records of the same entity, and associating the multiple records of the same entity to obtain associated data; S104. Integrating the converted multi-source heterogeneous data into subject domains based on a pre-built subject data model to obtain integrated data; S105. Performing time-series alignment processing on the converted multi-source heterogeneous data to obtain aligned data; S106. Merging the associated data, integrated data, and aligned data to obtain processed multi-source heterogeneous data.

[0034] Furthermore, after obtaining the processed multi-source heterogeneous data, the process further includes: S107. Constructing a data model based on the multi-source heterogeneous data and the processed multi-source heterogeneous data; wherein, the data model includes a raw data layer, a computational processing layer, and an analysis result layer; the raw data layer is used to store the multi-source heterogeneous data; the computational processing layer is used to generate industry-specific indicators based on industry business logic, the industry-specific indicators including oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate, and constructs industry dimension tables and fact tables based on industry business logic, and constructs an industry data warehouse based on industry-specific indicators, industry dimension tables, and fact tables; the analysis result layer is used to store the processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines.

[0035] Specifically, the industry's business logic includes existing business logic, and its analytical dimensions differ from those in existing technologies, such as site dimensions, customer dimensions, and product dimensions. Site dimensions include analytical attributes such as name, number, site type (e.g., hub type, freight type, scenic area type), number of lanes, whether charging piles are installed, and the number of surrounding competitors; customer dimensions include ID, name, customer tags (e.g., commuters, freight drivers, self-driving tourists), value level (e.g., high, medium, low), fuel preference (e.g., diesel, gasoline), and consumption frequency; product dimensions include fuel type (e.g., 92#, 95#, diesel), convenience store goods (e.g., food, beverages, lubricants), and charging services (e.g., fast charging, slow charging). At the same time, a dedicated fact sheet was designed to record "what was sold" and "how it was operated", including a transaction fact sheet, a customer flow fact sheet, an operational fact sheet, and an event fact sheet. The transaction fact sheet records data for each refueling, charging, and retail transaction. The customer flow fact sheet records the number of vehicles entering the station, the number of people entering the store, and their stay duration through cameras or sensors. The operational fact sheet records the manpower hours of the shift, the operating status of equipment, and changes in inventory levels. The event fact sheet records events that affect operations, such as marketing activities, competitor price reductions, holidays, and extreme weather.

[0036] Furthermore, the generation process of specific indicators includes obtaining the calculation formulas, dimensions, and data sources of preset business indicators, such as: sales per square meter = total operating revenue / station operating area, with data sourced from the amount field in the "Transaction Fact Table"; refueling rate = (number of transactions where single refueling amount ≥ vehicle fuel tank capacity * coefficient α) / total number of transactions; charging turnover rate = number of daily charging orders / number of charging piles; cross-business consumption penetration rate = (number of customers with both refueling and convenience store purchase records on the same day) / total number of refueling customers on the same day); generating preset calculation tasks based on the processed multi-source heterogeneous data to obtain indicator results; storing the indicator results in the analysis results layer and providing a unified indicator data query interface for upper-layer applications.

[0037] S2. Perform feature engineering on the processed multi-source heterogeneous data to obtain industry features. The industry features include the historical sales sequence of the site, the price elasticity coefficient of different customer groups, and the moving average sales during holidays. It should be noted that the industry features include, but are not limited to, the historical sales sequence of the site, the moving average sales during holidays, trend items, the price elasticity coefficient of different customer groups, site performance indicators, and keywords extracted from business reports and customer complaints using NLP (Natural Language Processing) algorithms.

[0038] For example, industry characteristics also include spatiotemporal characteristics, cross-features, industry sequence characteristics, and event characteristics. Among them, spatiotemporal characteristics include, but are not limited to, traffic density around highway stations, distribution of vehicle origins, and distance vectors to competing highway stations within a preset time period; cross-features include, but are not limited to, the interaction between oil prices and sales of specific commodities, and the ratio of truck traffic to diesel sales; behavioral sequence characteristics include, but are not limited to, customers' historical consumption category sequences, refueling frequency change rates, and the correlation of cross-industry consumption; event characteristics include, but are not limited to, holiday type codes, severe weather levels, and competitor promotional activity intensity indices. This embodiment does not impose specific limitations on these characteristics.

[0039] Furthermore, in step S2, the feature extraction methods include feature construction based on business rules, feature generation based on statistics and aggregation, and feature learning based on models. The principle of feature construction based on business rules is to construct new features from the original data through programming logic based on the experience of business experts or industry best practices. The principle of feature generation based on statistics and aggregation is to perform rolling, sliding window, or grouping aggregation operations on the data and calculate statistics to capture historical behavior and trends. The principle of feature learning based on models is to automatically learn and generate features from the data using unsupervised or lightly supervised models. Unsupervised or lightly supervised models include, but are not limited to, K-means algorithm, principal component analysis, variational autoencoder, or gradient boosting tree. The above algorithms are all existing algorithms and will not be described in detail here.

[0040] S3. Obtain historical industry data, train a pre-built business model based on the historical industry data to obtain the trained business model, and input industry features into the trained business model to obtain business results; specifically, step S3, training a pre-built business model based on historical industry data to obtain the trained business model, includes: S301. Performing feature engineering on the historical industry data to obtain historical industry features; S302. Obtaining the historical business results corresponding to the historical industry data, and labeling the historical industry features based on the historical business results to obtain labeled training data; S303. Inputting the labeled training data into the pre-built business model for training to obtain the trained business model.

[0041] It should be noted that the business model includes one or more combinations of predictive models, diagnostic models, optimization decision-making models, and insight segmentation models. Specifically, the predictive model is built based on Long Short-Term Memory networks, the diagnostic model is built based on Isolation Forest, the optimization decision-making model is built based on reinforcement learning algorithms, and the insight segmentation model is built based on clustering algorithms.

[0042] Specifically, predictive models can forecast future business conditions, and can be built based on LSTM (Long Short-Term Memory) models, Transformer models, or Prophet algorithms (time series forecasting algorithms) to create sales forecasting models, traffic flow forecasting models, and electricity demand forecasting models; diagnostic models can gain insights into the current situation and trace the causes, and can use isolated forest or autoencoder models to create sales anomaly detection models and equipment failure early warning models, and root cause analysis models based on causal forest or Shapley value attribution to quantify the contribution of various factors such as weather, holiday competition activities, etc., to fluctuations in business indicators; optimization decision-making models are used to generate optimal action strategies, and can use dynamic pricing models and marketing budget allocation models based on reinforcement learning, and human resource scheduling optimization models and inventory replenishment optimization models based on operations research algorithms; insight segmentation models can use customer grouping models and site classification models based on clustering algorithms, and product cross-selling recommendation models based on association rules.

[0043] In this embodiment, a large language model is used as the core engine. By integrating multi-source heterogeneous data, industry features, business results, and a pre-built industry knowledge base, a business model is obtained. The business model is used as a model factory, which can perform unified full lifecycle management of the above models, such as model version control, automated training and deployment pipeline, performance monitoring and decay warning, and strategy effect comparison based on A / B testing. This ensures that the model can continuously and stably serve the business and can be continuously iterated and optimized.

[0044] S4. Using a large language model as the core engine, integrate multi-source heterogeneous data, industry characteristics, business results, and a pre-built industry knowledge base to obtain a vertical industry large model. In this embodiment, the vertical industry large model can provide functions including but not limited to intelligent interactive question answering, intelligent automatic report generation, summarization and explanation of complex strategies and results, and creation of intelligent agents. Specifically, it obtains questions raised by users in natural language, automatically identifies the intent of the questions, calls the corresponding data query interface or AI model for calculation, and generates and displays the results in a comprehensive form such as text, tables, and visualization charts; based on a preset generation period or based on event triggering, it automatically aggregates multi-source heterogeneous data to generate a structured business analysis report; for the results generated by the complex model in step S3 that are difficult to understand intuitively, it transforms them into decision suggestions that are easy for business personnel to understand, improving the transparency and trustworthiness of decision-making; it constructs an automated business intelligent agent dedicated to the large model, such as an operational early warning intelligent agent, which automatically inspects the entire site for abnormal situations, automatically initiates root cause analysis after discovering anomalies, generates corresponding handling suggestions, and can also automatically create task work orders to be dispatched to the relevant responsible persons, initially realizing autonomous decision-making and action.

[0045] S5. Based on multi-source heterogeneous data, industry characteristics, business results, and vertical industry big data models, five major business decision engines are constructed. These five business decision engines are used to generate executable decision instructions. Specifically, the five major business decision engines include a business collaboration and network construction engine for responding to competition and assessing potential, a strategic goal management and sand table simulation engine for decomposing strategic goals, an early warning and root cause management engine for identifying operational anomalies, an operational strategy management engine for generating business strategies, and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

[0046] Specifically, the five business decision engines work together to form a closed-loop decision-making process and encode industry knowledge into computable algorithmic processes; the site operation management engine can also be used to optimize marketing campaigns.

[0047] It should be noted that the business collaboration and network construction engine is used to analyze customer cross-site flow patterns based on graph neural networks (GNN), formulate traffic diversion strategies, and evaluate site location potential based on spatial regression models; the strategic goal management and sandbox simulation engine is used to intelligently decompose company-level strategic goals into executable sub-site goals based on historical data and potential models of each site, and can simulate goal achievement under different market strategies; the early warning and root cause management engine is used to receive abnormal signals from the model factory, automatically trigger the root cause analysis process, locate the source of the problem, and automatically match the historical governance experience library to recommend response strategies; the business strategy management engine is used to automatically generate or optimize business strategies based on forecast and early warning information, and output executable instructions; the site operation management engine is used to provide standardized operational functions such as site performance evaluation, scheduling optimization, and intelligent ordering.

[0048] Furthermore, this embodiment encapsulates the outputs of the five major business decision engines into multiple specific intelligent application scenarios for use by business personnel. The generated decision instructions are sent to the front-line business system for execution via API interface, and the effect data after execution is collected again and stored in the data model, thereby forming a complete business closed loop.

[0049] In a preferred embodiment, the working method of the early warning and root cause management engine includes: real-time monitoring of abnormal fluctuations in top-level KPIs (Key Performance Indicators); automatically drilling down along a pre-built hierarchical business indicator model to locate key abnormal areas or sites causing the fluctuations; invoking a causal inference model to quantify and analyze the causal effect values ​​of multiple potential influencing factors on the indicator fluctuations; and outputting an attribution report that clearly identifies key nodes, core influencing factors, and their contributions. In this embodiment, the hierarchical business indicator model is a tree-like or pyramid-shaped indicator system and data association model built based on business logic. Its purpose is to decompose high-level and generalized business objectives into lower-level quantifiable, attributable, and operable specific business indicators, thereby establishing a penetrating path from macro to micro at the data level.

[0050] In a preferred embodiment, the working method of the strategic goal management and sand table simulation engine includes: decomposing macro-strategic goals into actionable micro-goals based on a prediction model and a multi-objective optimization algorithm; constructing a digital twin simulation environment for the business system; inputting candidate strategies and external variables into the simulation environment and performing large-scale parallel simulations using Monte Carlo simulation; evaluating the expected returns and risk distribution of each strategy based on the simulation results, and outputting strategy ranking and risk warnings. In this embodiment, the prediction model is a set of models, which are constructed by selecting or integrating different core algorithms for different prediction goals and data characteristics. Specifically, the prediction models in this embodiment include, but are not limited to, sales / demand prediction models, traffic / passenger flow prediction models, and electricity / charging demand prediction models. Among them, the sales / demand prediction model is constructed based on long short-term memory networks, Transformer time series models, and the Prophet algorithm. Since sales data is a typical multivariate time series, it is affected by various factors such as trends, seasonality, holidays, weather, or oil prices. Therefore, long short-term memory... Network and Transformer time series models can effectively capture long-term dependencies and complex patterns. The Prophet algorithm has built-in support for seasonality and holiday effects and is highly interpretable. Combining these algorithms improves robustness. The traffic / passenger flow prediction model is built on spatiotemporal graph neural networks and ARIMA. Since highway traffic flow has strong spatial and temporal correlations, ST-GNN can model spatiotemporal dependencies simultaneously. ARIMA is suitable for stationary or differentially stationary time series prediction. The electricity / charging demand prediction model is built on XGBoost, LightGBM, and LSTM. Since charging demand is related to multiple features such as traffic flow, time period, electricity price, and surrounding services, the gradient boosting tree model can efficiently process structured features and give feature importance. LSTM is good at processing time series information. The prediction results obtained by the above models are the core inputs for strategic goal decomposition, resource pre-arrangement, risk warning, and strategy simulation. For example, the sand table simulation engine needs to simulate the incremental effects of different promotion strategies based on sales forecasts.

[0051] In a preferred embodiment, the process of building a dynamic knowledge graph within a large vertical industry model includes: constructing a dynamic knowledge graph architecture with business problems, scenario features, strategy solutions, and execution results as node types; continuously importing historical strategy cases and governance reports; and updating the dynamic knowledge graph using natural language processing technology.

[0052] In a preferred embodiment, the process of reusing dynamic knowledge graphs within a large vertical industry model includes: when a new business problem is identified, semantically matching the current scenario features corresponding to the new business problem with historical cases in the dynamic knowledge graph; retrieving and recommending successful strategies adopted by similar historical cases as candidate solutions.

[0053] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent management system for integrated highway services, characterized in that, The system comprises a first construction module, a feature engineering module, a second construction module, a large model module, and an engine construction module. The first construction module acquires multi-source heterogeneous data from the highway energy network, including transaction data from fuel dispensers, convenience store POS sales data, charging pile operation status and order data, and highway network gantry traffic flow data. It preprocesses the multi-source heterogeneous data to obtain processed multi-source heterogeneous data, which is then sent to the feature engineering module and the large model module. The feature engineering module performs feature engineering on the processed multi-source heterogeneous data to obtain industry features, including historical sales sequences at stations, price elasticity coefficients for different customer groups, and holiday moving average sales. These industry features are then sent to the second construction module and the large model module. The second construction module acquires historical industry data, trains a pre-built business model based on this data, obtains a trained business model, inputs the industry features into the trained business model, and obtains business results, which are then sent to the large model module. The large model module is used to integrate multi-source heterogeneous data, industry characteristics, business results, and pre-built industry knowledge bases with a large language model as the core engine to obtain a vertical industry large model, which is then sent to the engine construction module. The engine construction module is used to build five major business decision engines based on multi-source heterogeneous data, industry characteristics, business results, and the vertical industry large model. The five major business decision engines are used to generate executable decision instructions. These five major business decision engines include a business collaboration and network construction engine for responding to competition and assessing potential, a strategic goal management and sand table simulation engine for decomposing strategic goals, an early warning and root cause management engine for identifying operational anomalies, an operational strategy management engine for generating operational strategies, and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

2. The intelligent management system for integrated highway services according to claim 1, characterized in that, It also includes a closed-loop processing module, which is used to obtain executable decision instructions from the five major business decision engines. These executable decision instructions include sales warnings, promotion generation strategies, and / or customer churn recovery strategies. The executable decision instructions are then sent to the corresponding front-line business systems for execution via API interfaces. These front-line business systems include POS systems, CRM systems, and marketing platforms. The closed-loop processing module is also used to obtain the effect data after executing the decision instructions and feed the effect data back to the first construction module.

3. The intelligent management system for integrated highway services according to claim 2, characterized in that, The first construction module is also used to construct a data model based on multi-source heterogeneous data and processed multi-source heterogeneous data. The data model includes a raw data layer, a computational processing layer, and an analysis result layer. The raw data layer is used to store multi-source heterogeneous data. The computational processing layer is used to generate industry-specific indicators based on industry business logic. The industry-specific indicators include oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate. It also constructs industry dimension tables and fact tables based on industry business logic and builds an industry data warehouse based on industry-specific indicators, industry dimension tables, and fact tables. The analysis result layer is used to store processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines. The first construction module is also used to obtain effect data fed back by the closed-loop processing module and update the data model in real time based on the effect data.

4. An intelligent management method for integrated highway services, applied to the intelligent management system according to any one of claims 1 to 3, characterized in that, include: This process involves acquiring multi-source heterogeneous data from the highway energy network, including transaction records from fuel dispensers, sales data from convenience store POS terminals, operational status and order data from charging piles, and traffic flow data from highway gantry systems. The data is preprocessed to obtain processed multi-source heterogeneous data. Feature engineering is then applied to this processed data to obtain industry features, including historical sales sequences at service stations, price elasticity coefficients for different customer groups, and moving average sales during holidays. Historical industry data is acquired, and a pre-built business model is trained based on this data to obtain a trained business model. The industry features are then input into the trained business model to obtain business results. Finally, the process is described in the context of a large-scale language... The system uses a language model as its core engine, integrating multi-source heterogeneous data, industry characteristics, business results, and a pre-built industry knowledge base to obtain a large-scale vertical industry model. Based on this model, five business decision engines are constructed to generate executable decision instructions. These engines include: a business collaboration and network construction engine for responding to competition and assessing potential; a strategic goal management and simulation engine for decomposing strategic objectives; an early warning and root cause management engine for identifying operational anomalies; an operational strategy management engine for generating business strategies; and a site operation management engine for evaluating sites, optimizing scheduling, and ordering.

5. The intelligent management method for integrated highway services according to claim 4, characterized in that, The process involves preprocessing multi-source heterogeneous data to obtain processed multi-source heterogeneous data, including: data cleaning of the multi-source heterogeneous data to obtain cleaned multi-source heterogeneous data; format conversion of the cleaned multi-source heterogeneous data to obtain converted multi-source heterogeneous data; entity data identification of the converted multi-source heterogeneous data to obtain multiple records of the same entity, and association of the multiple records of the same entity to obtain associated data; subject domain integration of the converted multi-source heterogeneous data based on a pre-built subject data model to obtain integrated data; time-series alignment of the converted multi-source heterogeneous data to obtain aligned data; and fusion of associated data, integrated data, and aligned data to obtain processed multi-source heterogeneous data.

6. The intelligent management method for integrated highway services according to claim 4, characterized in that, After obtaining the processed multi-source heterogeneous data, the process further includes: constructing a data model based on the multi-source heterogeneous data and the processed multi-source heterogeneous data; wherein, the data model includes a raw data layer, a computational processing layer, and an analysis result layer; the raw data layer is used to store the multi-source heterogeneous data; the computational processing layer is used to generate industry-specific indicators based on industry business logic, the industry-specific indicators including oil sales volume, sales per square meter, sales per employee, average order value, fill-up rate, and customer repurchase rate, and constructs industry dimension tables and fact tables based on industry business logic, and constructs an industry data warehouse based on industry-specific indicators, industry dimension tables, and fact tables; the analysis result layer is used to store the processed multi-source heterogeneous data and executable decision instructions generated by the five major business decision engines.

7. The intelligent management method for integrated highway services according to claim 4, characterized in that, The process of training a pre-built business model based on historical industry data to obtain the trained business model includes: performing feature engineering on the historical industry data to obtain historical industry features; obtaining historical business results corresponding to the historical industry data; labeling the historical industry features based on the historical business results to obtain labeled training data; and inputting the labeled training data into the pre-built business model for training to obtain the trained business model.

8. The intelligent management method for integrated highway services according to claim 4, characterized in that, The vertical industry big model includes one or more combinations of predictive models, diagnostic models, optimization decision-making models, and insight segmentation models.

9. The intelligent management method for integrated highway services according to claim 8, characterized in that, The prediction model is based on a long short-term memory network, the diagnostic model is based on an isolated forest, the optimization decision model is based on a reinforcement learning algorithm, and the insight segmentation model is based on a clustering algorithm.

Images