Method and device for cost optimization prediction through multi-dimensional travel behavior analysis

Abstract

The invention provides a method and device for carrying out cost optimization prediction through multi-dimensional business travel behavior analysis, and belongs to the technical field of enterprise business travel management.The method comprises the steps that business travel data are collected, cleaning, duplicate removal and entity alignment operation are carried out, and a business travel behavior data set is constructed; executing a preset query script on the business travel behavior data set, extracting characteristic indexes from preset dimensions, executing descriptive analysis, diagnostic analysis and clustering analysis, and generating a business travel behavior analysis report; and inputting the business travel behavior data set and the business travel behavior analysis report into a cost optimization prediction model to carry out business travel demand prediction, business travel price prediction and business travel travel optimization, generating an optimal travel travel scheme set, pushing the optimal travel travel scheme set to the employee user, collecting a feedback result of the employee user on the optimal travel travel scheme set, updating the business travel behavior data set, and obtaining a business travel result. And incremental training and iterative optimization are carried out on the cost optimization prediction model. Through multi-dimensional analysis and prediction, the travel cost is optimized, and the management efficiency is improved.

Description

Technical Field

[0001] This application belongs to the field of corporate travel management technology, specifically involving a method and system for cost optimization and prediction through multi-dimensional travel behavior analysis. Background Technology

[0002] With the development of enterprise digital transformation, travel management has become an important part of enterprise management. Traditional travel management methods mostly rely on manual statistics, Excel spreadsheets, or simple reimbursement systems, which can only achieve post-trip expense summary and budget comparison, lacking in-depth analysis of travel behavior, let alone predicting cost optimization.

[0003] Currently, companies typically control travel costs through centralized procurement, negotiated pricing, and advance booking. However, the following problems persist: First, data analysis is limited in scope, failing to provide systematic analysis across multiple dimensions such as personnel, time, itinerary, cost, compliance, and suppliers. Second, there is a lack of predictive and optimization capabilities, making it impossible to predict future travel demand and price trends based on historical data, let alone recommend optimal travel options. Third, data silos exist, with travel data scattered across multiple systems such as finance, human resources, and office automation (OA), making integration and analysis difficult. Finally, balancing employee satisfaction and compliance is challenging, lacking personalized recommendations and making it difficult to improve employee satisfaction while maintaining compliance.

[0004] Therefore, there is an urgent need for a travel expense analysis and cost optimization method that can integrate multi-source data, perform multi-dimensional analysis and intelligent prediction, so as to support enterprises in achieving digital transformation from passive reimbursement to proactive management. Summary of the Invention

[0005] In a first aspect, embodiments of this application provide a method for cost optimization prediction through multi-dimensional travel time analysis, comprising the following steps: S1. Extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. Clean, deduplicate and entity-align the collected historical and real-time travel data to construct a travel behavior dataset. S2. Execute a preset query script on the business travel dataset to extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance, and suppliers. Perform descriptive analysis, diagnostic analysis, and cluster analysis on the extracted feature indicators to generate a business travel analysis report. S3. Use the travel data set and travel analysis report to train the constructed cost optimization prediction model, and use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries to generate the optimal itinerary set; S4. Push the optimal travel plan set to employee users, collect feedback from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model.

[0006] Furthermore, the internal systems of the enterprise in step S1 include ERP system, human resources HR system, and office automation (OA) approval system; External systems for enterprises include airline ticket price query interfaces, hotel group room price interfaces, and travel service management systems; Historical travel data of ERP expense reimbursement type is extracted from the database interface of the ERP system. The ERP expense reimbursement data includes expense order number, employee ID, expense type, amount, and reimbursement status. Historical travel data of HR organizational data type is extracted from the database interface of the human resources HR system. The HR organizational data includes employee ID, department, job level and position. Historical travel data of OA approval data type is extracted from the database interface of the office OA approval system. The OA approval data includes approval form ID, employee ID, trip purpose, departure point, destination and planned departure time. Real-time travel data of TMC order type is extracted from the travel service management system. The TMC order data includes order ID, employee ID, flight number, hotel name, reservation time, check-in time, check-out time, and order amount. The entity alignment operation specifically involves using employee ID and department code within the enterprise as the primary key to associate ERP expense reimbursement data, HR organization data, OA approval data, and TMC order data with the same travel behavior fact record, thereby completing the construction of the travel behavior dataset.

[0007] Furthermore, the specific steps of step S2 are as follows: S21. Using business travel as the dataset as input, execute the pre-defined query script to extract the following business travel features: Extract employee job level, department, and travel frequency features from a personnel perspective; The frequency of business trips is calculated by counting the number of records in the OA approval data. Extract booking advance time period and travel date attribute features from the time dimension; The advance booking period is calculated by determining the time difference between the booking time in the TMC order data and the planned departure time in the OA approval data. Extract features of departure point, destination, and mode of transportation from the perspective of travel itinerary; Extract the characteristics of per capita cost and cost composition ratio from the cost dimension; Extract characteristics of travel restriction violation types, violation amounts, and approval time from a compliance perspective; Extract supplier usage concentration and agreement price usage characteristics from the supplier dimension; S22. Perform analysis using the extracted travel routes as features: First, descriptive analysis is used to generate historical trend and percentage reports, outputting the monthly total travel cost trend and the ranking of expense percentages for each department; Then, using drill-down and association rule methods in diagnostic analysis, the root causes of cost anomalies or travel restriction violations are analyzed, and a cost overrun root cause analysis report is output. Finally, cluster analysis was used to divide employees into different travel behavior groups using travel frequency, average advance booking days, average discount rate, and hotel usage rate as feature vectors, and output employee travel behavior classification labels. S23. Integrate the results of descriptive analysis, diagnostic analysis, and cluster analysis to generate a travel experience analysis report, which includes at least: List of high-frequency travelers, ranking of departments with cost overruns, statistics on violations of travel restrictions, supplier performance evaluations, and classification results of employee travel behavior groups.

[0008] Furthermore, in step S22, the K-Means algorithm is used for cluster analysis to output classification labels for employee travel behavior; The objective function of the K-Means algorithm for cluster analysis is to minimize the sum of squares within each cluster.

[0009] in, The preset number of clusters, For clustering schemes, It is the i-th cluster. It is the feature vector of employee business travel behavior. It is a cluster The mean center; It should be noted that, It is composed of characteristics such as travel frequency, average advance booking days, average discount rate, and hotel occupancy rate.

[0010] Furthermore, the specific steps of step S3 are as follows: S31. Construct a unified training data source based on the travel data set and the travel analysis report: Historical business trips were extracted from OA approval data in the business trip dataset to construct a time series of business trips. Historical airfare price sequences were extracted from TMC order data and ERP expense reimbursement data, which were used as a dataset for business travel. Extract cost overrun tags, travel restriction violation tags, and employee travel behavior group classification tags from the business travel behavior analysis report; S32. Perform demand forecasting, price forecasting, and route optimization using a cost optimization forecasting model: Pre-trained cost optimization prediction models include demand prediction models, price prediction models, and route optimization models; Input the time series of business trips into a time series-type demand forecasting model to predict the total business travel demand within a future set time period. By inputting historical airfare price sequences, advance booking periods, and current fuel prices and holiday indicators obtained from external systems into a machine learning regression-type price prediction model, the airfare for future routes can be predicted. The total predicted travel demand within a specified future time period, the predicted airfare prices for specified future routes, employee travel behavior classification labels and travel restriction compliance requirements obtained from the travel behavior analysis report, and the contract hotel prices obtained from TMC order data and external hotel interface data in the travel behavior dataset are input into the trip optimization model using the optimization constraint algorithm. The optimization calculation is performed with the goal of minimizing total cost and maximizing travel efficiency, generating a candidate trip plan set containing several candidate options. S33. Sort and format the candidate itinerary options set to generate the optimal itinerary option set, which includes the total cost of the option, the total travel time, the combination of transportation and accommodation, and the description of compliance with travel restrictions.

[0011] Furthermore, the specific training steps for the cost optimization prediction model in step S32 are as follows: S321. Extract historical time series data from the OA approval data of the business trip dataset, including the number of business trips and the dates of business trips, to form the first training sample; S322. Use the LSTM model to build a demand prediction model, and then use the first training sample to train the LSTM model to obtain a trained demand prediction model. S323. Extract historical feature data from the TMC order data and ERP reimbursement data of the travel data set, including historical airfare prices, booking advance period, and routes, and integrate current oil prices and holiday markers obtained from external systems as feature vectors, using historical airfare prices as labels to construct a second training sample; S324. Use the second training sample to train the gradient boosting tree regression model to obtain a trained price prediction model; S325. Extract historical excellent travel plans from the business travel dataset as positive samples. The positive samples must meet the following requirements: low total cost, high travel efficiency, and compliance with business travel policies. S326. Using the departure point, destination, time, employee travel behavior classification labels, and predicted future flight ticket prices of the analysis report as features, a recommendation model based on constraint optimization is trained using positive samples to obtain a trained itinerary optimization model.

[0012] Furthermore, in step S326, the objective function of the constraint-based recommendation model is to minimize the total travel cost and travel time:

[0013] in, It's the cost of the airfare. It's the hotel's cost. It is the cost of ground transportation. It is the total travel time. , These are preset weighting coefficients. ; The constraints of the recommendation model based on constraint optimization include: Flight departure time should not be earlier than the earliest departure time of employees, hotel star rating should not be higher than the employee's job level standard, and total cost should not exceed the budget threshold; We optimized the weighting of cost and duration using the travel behavior dataset and the travel restriction compliance requirements and employee travel behavior classification results from the travel behavior analysis report.

[0014] Furthermore, the specific steps of step S4 are as follows: S41. Push the generated optimal itinerary plan set to the target employee's user terminal through the application programming interface of the enterprise's internal travel management system; The optimal travel plan set includes at least two plans: The lowest total cost pair marked with the cost-optimal solution; The route with the shortest total travel time, marked as the most efficient route; Each option provides details on the total cost, total travel time, transportation and accommodation combination, and compliance with travel restrictions. S42. Collect employee users' selection results of push notification options in real time through the application programming interface of the enterprise's internal travel management system; S43. After associating the employee user's selection result with the corresponding travel application record, write it back as a new data record and update the travel behavior dataset; S44. Using the updated travel data as the new training data, incrementally train the cost optimization prediction model to achieve continuous iterative optimization of the cost optimization prediction model.

[0015] Furthermore, a travel constraint simulation and optimization step is included between step S3 and step S4: S3A. Using business travel as the dataset and business travel as the analysis report as the basic data, and business travel restriction parameters as variable inputs; Using the Monte Carlo simulation method, we simulated travel behavior and costs under different travel constraint parameters in batches. By comparing simulation results, a report is generated recommending adjustments to travel restrictions, including expected cost savings, expected changes in compliance rates, and potential impacts on employee satisfaction.

[0016] Secondly, embodiments of this application also provide an apparatus for cost optimization prediction through multi-dimensional difference travel time analysis, comprising: The travel data acquisition module is used to extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. It also performs cleaning, deduplication, and entity alignment operations on the collected historical and real-time travel data to construct a travel behavior dataset. The travel feature extraction module is used to execute preset query scripts on the travel behavior dataset, extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance and supplier, and perform descriptive analysis, diagnostic analysis and cluster analysis on the extracted feature indicators to generate a travel behavior analysis report. The cost optimization prediction module is used to train the constructed cost optimization prediction model using the travel data set and travel analysis report, and to use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries, generating the optimal itinerary set. The recommendation and feedback execution module is used to push the optimal travel plan set to employee users, collect feedback results from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model.

[0017] As can be seen from the above technical solutions, this application has the following advantages: The method and system for cost optimization and prediction through multi-dimensional travel behavior analysis provided in this application extracts feature indicators from multiple dimensions, including personnel, time, itinerary, cost, compliance, and suppliers, and uses various analytical methods to comprehensively identify potential problems and patterns in travel data. By cleaning, deduplicating, and aligning the collected data, the accuracy, completeness, and consistency of the data are ensured, providing a data foundation for analysis. By integrating multi-source data, data silos are broken down, approval processes are simplified, and employees are freed from tedious travel arrangements and reimbursement tasks. Through multi-dimensional analysis and monitoring of travel behavior, violations can be detected and corrected in a timely manner, reducing the risk of violations and avoiding additional costs and risks arising from them. Through predictive analysis and optimization recommendations, enterprises can accurately grasp travel price trends, select more cost-effective travel solutions, reduce unnecessary travel expenses, and lower costs. Attached Figure Description

[0018] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the method for cost optimization and prediction based on multi-dimensional travel behavior analysis according to the present invention.

[0020] Figure 2 This is a schematic diagram of the system for cost optimization prediction based on multi-dimensional travel behavior analysis according to the present invention. Detailed Implementation

[0021] Various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the method for cost optimization prediction through multidimensional difference travel behavior analysis. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0022] For example, under the current digital transformation of enterprises, traditional travel management methods have exposed many drawbacks. On the one hand, relying on manual statistics and simple tools, data analysis is superficial and cannot comprehensively analyze travel behavior from multiple dimensions such as personnel, time, itinerary, cost, compliance, and suppliers. On the other hand, data is scattered across financial, human resources, and OA systems, making it difficult to aggregate and analyze. Furthermore, it lacks the ability to predict future travel demand and price trends, as well as optimization functions to recommend optimal travel plans. Moreover, while ensuring compliance, it is difficult to simultaneously address employees' personalized needs and improve satisfaction. Therefore, enterprises urgently need a new travel management solution that can integrate multi-source data, achieve multi-dimensional in-depth analysis and intelligent prediction, enabling enterprises to transform travel management from passive recording to proactive optimization, achieving a balance between cost control and employee experience.

[0023] To address the aforementioned issues, this embodiment provides a method for cost optimization and prediction through multi-dimensional travel behavior analysis, realizing a shift from passive reimbursement to proactive management. By optimizing travel costs through multi-dimensional data analysis and intelligent prediction, the efficiency and effectiveness of travel management are improved.

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 The diagram shows a flowchart of a method for cost optimization prediction through multi-dimensional travel behavior analysis in a specific embodiment. The method includes the following steps: S1. Extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. Clean, deduplicate and entity-align the collected historical and real-time travel data to construct a travel behavior dataset. It should be noted that by integrating data from internal and external systems and performing data cleaning, deduplication, and entity alignment operations, a travel data dataset is constructed to provide a data foundation for subsequent analysis and prediction. S2. Execute a preset query script on the business travel dataset to extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance, and suppliers. Perform descriptive analysis, diagnostic analysis, and cluster analysis on the extracted feature indicators to generate a business travel analysis report. It should be noted that feature indicators are extracted from multiple dimensions such as personnel, time, itinerary, cost, compliance, and suppliers. Descriptive analysis, diagnostic analysis, and cluster analysis are used to analyze business travel behavior, identify potential problems and patterns, and generate detailed business travel behavior analysis reports to provide a basis for corporate business travel decisions. S3. Use the travel data set and travel analysis report to train the constructed cost optimization prediction model, and use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries to generate the optimal itinerary set; It should be noted that the cost optimization prediction model is trained based on the travel data set and analysis report to predict and optimize travel demand, price trends and itinerary plans, generate the optimal itinerary plan set, help companies plan travel arrangements in advance, optimize resource allocation, and enhance the company's initiative in travel management. S4. Push the optimal travel plan set to employee users, collect feedback from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model. It should be noted that by pushing the optimal itinerary plan set to employee user terminals and collecting employee feedback in real time, the travel behavior dataset is updated in a timely manner. This allows for incremental training and iterative optimization of the cost optimization prediction model, enabling continuous improvement and updating of the model. This ensures the accuracy of the prediction results and the relevance of the recommended plans, thereby enhancing the effectiveness and quality of travel management.

[0026] This embodiment enables comprehensive analysis of business travel behavior and intelligent prediction and optimization of costs, allowing enterprises to shift from post-processing to pre-planning and in-process control, thereby improving the initiative and rationality of business travel management.

[0027] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process in this embodiment, another method for cost optimization prediction through multi-dimensional travel behavior analysis is provided. This method includes the following steps: S1. Extract historical travel data from the database interface of the enterprise's internal systems and collect real-time travel data from the data interface of the enterprise's external systems. Clean, deduplicate, and align the collected historical and real-time travel data to construct a travel behavior dataset. The enterprise's internal systems in step S1 include the ERP system, the human resources HR system, and the office automation (OA) approval system. External systems for enterprises include airline ticket price query interfaces, hotel group room price interfaces, and travel service management systems; Historical travel data of ERP expense reimbursement type is extracted from the database interface of the ERP system. The ERP expense reimbursement data includes expense order number, employee ID, expense type, amount, and reimbursement status. Historical travel data of HR organizational data type is extracted from the database interface of the human resources HR system. The HR organizational data includes employee ID, department, job level and position. Historical travel data of OA approval data type is extracted from the database interface of the office OA approval system. The OA approval data includes approval form ID, employee ID, trip purpose, departure point, destination and planned departure time. Real-time travel data of TMC order type is extracted from the travel service management system. The TMC order data includes order ID, employee ID, flight number, hotel name, reservation time, check-in time, check-out time, and order amount. The entity alignment operation specifically involves using employee ID and department code within the enterprise as the primary key to associate ERP expense reimbursement data, HR organization data, OA approval data, and TMC order data with the same travel behavior fact record, thereby completing the construction of the travel behavior dataset. For example, consider an internet company (department coding rules: technology department is "01", marketing department is "02"): ERP expense reimbursement data: Expense number "BX202405001", employee ID "E001", expense type "airfare + hotel", amount "3200 yuan", reimbursement status "approved"; HR organizational data: Employee ID "E001", Department "02 (Marketing Department)", Job Level "Supervisor", Position "Marketing Promotion"; OA Approval Data: Approval Form ID "OA202405001", Employee ID "E001", Trip Purpose "Client Visit", Departure Point "Beijing", Destination "Shanghai", Planned Departure Time "2024-05-10 09:00"; TMC Order Data: Order ID "TMC202405001", Employee ID "E001", Flight Number "xxxx", Hotel Name "Shanghai xxxx Hotel", Booking Time "2024-05-05 14:30", Check-in Time "2024-05-10", Check-out Time "2024-05-12", Order Amount "3200 RMB"; When aligning entities, the above four types of data are associated as a single travel record using “Employee ID=E001” and “Department Code=02” as the primary keys, forming a complete entry in the travel entity dataset. S2. Execute a pre-defined query script on the business travel dataset to extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance, and supplier. Then, perform descriptive analysis, diagnostic analysis, and cluster analysis on the extracted feature indicators to generate a business travel analysis report. The specific steps of step S2 are as follows: S21. Using business travel as the dataset as input, execute the pre-defined query script to extract the following business travel features: Extract employee job level, department, and travel frequency features from a personnel perspective; The frequency of business trips is calculated by counting the number of records in the OA approval data. Extract booking advance time and travel date attributes (such as whether it is a holiday or peak season) from the time dimension. The advance booking period is calculated by determining the time difference between the booking time in the TMC order data and the planned departure time in the OA approval data. Extract features of departure point, destination, and mode of transportation from the perspective of travel itinerary; Extract the characteristics of per capita cost and cost composition ratio from the cost dimension; Extract characteristics of travel restriction violation types, violation amounts, and approval time from a compliance perspective; Extract supplier usage concentration and agreement price usage characteristics from the supplier dimension; For example, based on a company's business travel data from January to April 2024, the following features are extracted: Personnel Dimension: E001 (Marketing Department Supervisor), travel frequency "4 times / month" (4 OA approval records); E002 (Technical Department Engineer), travel frequency "1 time / month"; Time dimension: The TMC booking time for E001 is "2024-05-05" and the OA planned departure time is "2024-05-10", with a booking advance period of "5 days"; the travel date "2024-05-10" is "non-holiday (working day)"; Itinerary details: E001 Departure point "Beijing", destination "Shanghai", mode of transportation "Airplane (Economy Class)"; Cost dimension: The average travel cost per person in the marketing department is "3,000 yuan / trip", with the cost breakdown as "airfare 60% (1,800 yuan), hotel 35% (1,050 yuan), and ground transportation 5% (150 yuan)". Compliance Dimension: E003 (Finance Department Specialist) violated travel restrictions by "staying in a hotel exceeding the job level (the job level allows three-star hotels, but actually stayed in a four-star hotel)," with a violation amount of "200 yuan / time" and an approval time of "24 hours" (company regulations stipulate an approval time of ≤12 hours). Supplier Dimension: The company has a total of 5 partner hotels, with Shanghai xxx Hotel having the highest utilization rate at 65% (the highest supplier concentration); the agreement price utilization rate is 80% (80 out of 100 orders use the agreement price). S22. Perform analysis using the extracted travel routes as features: First, descriptive analysis is used to generate historical trend and percentage reports, outputting the monthly total travel cost trend and the ranking of expense percentages for each department; Then, using drill-down and association rule methods in diagnostic analysis, the root causes of cost anomalies or travel restriction violations are analyzed, and a cost overrun root cause analysis report is output. Finally, cluster analysis methods (such as the K-Means algorithm) are used to divide employees into different travel behavior groups (such as "economy type", "comfort type", and "efficiency priority type") using travel frequency, average advance booking days, average discount rate and contracted hotel usage rate as feature vectors, and output employee travel behavior classification labels; in step S22, the K-Means algorithm is used for cluster analysis to output employee travel behavior classification labels. The objective function of the K-Means algorithm for cluster analysis is to minimize the sum of squares within each cluster.

[0028] in, The preset number of clusters, For clustering schemes, It is the i-th cluster. It is the feature vector of employee business travel behavior. It is a cluster The mean center; It should be noted that, It is composed of characteristics such as travel frequency, average advance booking days, average discount rate, and hotel occupancy rate. For example, descriptive analysis: The trend of total monthly travel costs from January to April 2024 is "120,000 yuan in January → 150,000 yuan in February → 130,000 yuan in March → 140,000 yuan in April"; the ranking of expenses by department is "Marketing Department 40% (highest) → Technology Department 25% → Finance Department 15% → Administration Department 20%". Diagnostic analysis: The marketing department's travel costs exceeded the budget in March (budgeted at 120,000 yuan, actual at 150,000 yuan). Drill-down analysis revealed that "the frequency of customer visits in the Shanghai area increased by 30%, and 60% of orders were not booked 7 days in advance (advance booking can get a discount of up to 20%, while last-minute bookings only get a 9.5% discount)," which was the root cause of the cost overrun. Cluster analysis (K-Means algorithm, K=3): Cluster 1 (Economy): Travel frequency "1-2 times / month", average advance booking "7 days", average discount rate "75%", and hotel occupancy rate "90%" (e.g., E004, Administrative Specialist). Cluster 2 (Efficiency Priority): Business travel frequency "4-5 times / month", average advance booking "1-2 days", average discount rate "9.2%", and hotel occupancy rate "60%" (e.g., E001, Marketing Manager). Cluster 3 (Comfort): Travel frequency "2-3 times / month", average advance booking "3-4 days", average discount rate "8.5%", and hotel occupancy rate "70%" (e.g., E005, Technical Department Manager). S23. Integrate the results of descriptive analysis, diagnostic analysis, and cluster analysis to generate a travel experience analysis report, which includes at least: List of high-frequency travelers, ranking of departments with cost overruns, statistics on violations of travel restrictions, supplier performance evaluations, and classification results of employee travel behavior groups; For example, a list of high-frequency travelers: E001 (Marketing Department, 16 times / April), E006 (Sales Department, 14 times / April), E008 (Technical Department, 12 times / April). Departments with cost overruns: 1. Marketing Department (overrun 30,000 yuan), 2. Sales Department (overrun 15,000 yuan), 3. Technical Department (overrun 8,000 yuan). Travel restriction violation statistics: There were 20 violations from January to April 2024, including 12 violations of "accommodation exceeding the job level" (accounting for 60%) and 8 violations of "failure to use the contracted supplier" (accounting for 40%). Supplier performance evaluation: Shanghai xxx Hotel (satisfaction rate 90%, utilization rate 65%, ranked first), Beijing xxx Airlines (discount rate 7.8%, utilization rate 50%, ranked second); Employee travel behavior is categorized as follows: Economy (30 people, 30%), Efficiency-first (25 people, 25%), and Comfort (45 people, 45%). S3. Train the constructed cost optimization prediction model using the travel data set and travel analysis report, and use the trained cost optimization prediction model to predict travel demand, travel prices, and travel itineraries to generate an optimal itinerary set; the specific steps of step S3 are as follows: S31. Construct a unified training data source based on the travel data set and the travel analysis report: Historical business trips were extracted from OA approval data in the business trip dataset to construct a time series of business trips. Historical airfare price sequences were extracted from TMC order data and ERP expense reimbursement data, which were used as a dataset for business travel. Extract cost overrun tags, travel restriction violation tags, and employee travel behavior group classification tags from the business travel behavior analysis report; For example, the time series of business trips (January-April 2024, unit: person-times): January 80 → February 100 → March 90 → April 95; Historical airfare prices for the Beijing-Shanghai route (January-April 2024, unit: RMB): January 1200 → February 1500 (Spring Festival) → March 1300 → April 1400; Tag data: E001 (travel behavior tag "efficiency priority", no cost overrun tag, no violation tag), E003 (travel behavior tag "comfort", cost overrun tag "overrun by 200 yuan", violation tag "over-class accommodation"); S32. Perform demand forecasting, price forecasting, and route optimization using a cost optimization forecasting model: Pre-trained cost optimization prediction models include demand prediction models, price prediction models, and route optimization models; Input the time series of business trips into a time series-type demand forecasting model to predict the total business travel demand within a future set time period. By inputting historical airfare price sequences, advance booking periods, and current fuel prices and holiday indicators obtained from external systems into a machine learning regression-type price prediction model, the airfare for future routes can be predicted. The total predicted travel demand within a specified future time period, the predicted airfare prices for specified future routes, employee travel behavior classification labels and travel restriction compliance requirements obtained from the travel behavior analysis report, and the contract hotel prices obtained from TMC order data and external hotel interface data in the travel behavior dataset are input into the trip optimization model using the optimization constraint algorithm. The optimization calculation is performed with the goal of minimizing total cost and maximizing travel efficiency, generating a candidate trip plan set containing several candidate options. The specific training steps for the cost optimization prediction model in step S32 are as follows: S321. Extract historical time series data from the OA approval data of the business trip dataset, including the number of business trips and the dates of business trips, to form the first training sample; S322. Use the LSTM model to build a demand prediction model, and then use the first training sample to train the LSTM model to obtain a trained demand prediction model. S323. Extract historical feature data from the TMC order data and ERP reimbursement data of the travel data set, including historical airfare prices, booking advance period, and routes, and integrate current oil prices and holiday markers obtained from external systems as feature vectors, using historical airfare prices as labels to construct a second training sample; S324. Use the second training sample to train the gradient boosting tree regression model to obtain a trained price prediction model; S325. Extract historical excellent travel plans from the business travel dataset as positive samples. The positive samples must meet the following requirements: low total cost, high travel efficiency, and compliance with business travel policies. S326. Using the departure point, destination, time, employee travel behavior classification labels, and predicted future flight ticket prices of the analysis report as features, a recommendation model based on constraint optimization is trained using positive samples to obtain a well-trained itinerary optimization model. In step S326, the objective function of the constraint-based recommendation model is to minimize the total travel cost and travel time.

[0029] in, It's the cost of the airfare. It's the hotel's cost. It is the cost of ground transportation. It is the total travel time. , These are preset weighting coefficients. ; The constraints of the recommendation model based on constraint optimization include: Flight departure time should not be earlier than the earliest departure time of employees, hotel star rating should not be higher than the employee's job level standard, and total cost should not exceed the budget threshold; The weights of cost and duration were optimized using the business travel dataset and the business travel restriction compliance requirements and employee travel behavior classification results in the business travel analysis report. Demand Forecast (LSTM model, forecast for May 2024): Based on the time series of passenger trips from January to April, the total business travel demand in May is predicted to be "100 passenger trips", of which the marketing department accounts for 40% (40 passenger trips). Price prediction (gradient boosting tree regression model, predicting the airfare for the Beijing-Shanghai route on May 10, 2024): Input "historical price 1200-1500 yuan, book 5 days in advance, current fuel price 8.5 yuan / liter, non-holiday", the predicted price is "1350 yuan (economy class)"; if book 7 days in advance, the predicted price is "1150 yuan (8.5% discount)"; Itinerary optimization (for E001, May 10, 2024, Beijing-Shanghai, "efficiency-first", job level allows "four-star hotel"): Option 1: Book xxx flight (09:00-11:30, ticket price 1350 yuan) 1 day in advance, stay at xxx four-star hotel in Shanghai (2 nights, 1200 yuan), and take a taxi for ground transportation (200 yuan); total cost "2750 yuan", total duration "2.5 hours (flight) + 1 hour (transportation) = 3.5 hours"; Option 2: Book xxx flight 7 days in advance (08:00-10:30, ticket price 1150 yuan), stay at Shanghai XX four-star hotel (2 nights, 1200 yuan), ground transportation subway (50 yuan); total cost "2400 yuan", total duration "2.5 hours (flight) + 1.5 hours (transportation) = 4 hours"; Option 3: Book xxx flight (10:00-12:30, ticket price 1250 yuan) 3 days in advance, stay at Shanghai YY non-contract four-star hotel (2 nights, 1500 yuan), and take a taxi for ground transportation (200 yuan); total cost "2950 yuan", total duration "2.5 hours (flight) + 1 hour (transportation) = 3.5 hours"; S33. Sort and format the candidate itinerary options set to generate the optimal itinerary option set, which includes the total cost of the option, the total travel time, the combination of transportation and accommodation, and the description of compliance with travel restrictions; For example, the optimal itinerary plan set is shown in Table 1 below: Table 1

[0030] S4. Push the optimal travel plan set to employee users, collect feedback from employee users on the optimal travel plan set, update the business travel data using the feedback results, and then incrementally train and iteratively optimize the cost optimization prediction model; the specific steps of step S4 are as follows: S41. Push the generated optimal itinerary plan set to the target employee's user terminal through the application programming interface of the enterprise's internal travel management system; The optimal travel plan set includes at least two plans: The lowest total cost pair marked with the cost-optimal solution; The route with the shortest total travel time, marked as the most efficient route; Each option provides details on the total cost, total travel time, transportation and accommodation combination, and compliance with travel restrictions. S42. Collect employee users' selection results of push notification options in real time through the application programming interface of the enterprise's internal travel management system; S43. After associating the employee user's selection result with the corresponding travel application record, write it back as a new data record and update the travel behavior dataset; S44. Using the updated business travel dataset as new training data, incrementally train the cost optimization prediction model (especially the price prediction model and the trip optimization model) to achieve continuous iterative optimization of the cost optimization prediction model; For example, the solution push (E001, Marketing Department Manager, "Efficiency Priority"): The above set of optimal itinerary solutions is pushed through the corporate travel management APP, marked "Cost-optimal solution (saves 350 yuan)" and "Efficiency-optimal solution (saves 0.5 hours)"; Feedback collection: E001 selected the most efficient solution and reported in the APP that due to the need to hold a meeting with the client that afternoon, they prioritized shortening travel time and were willing to bear slightly higher costs; Dataset update: For E001, 2024-05-10 Beijing-Shanghai, select the most efficient solution and prioritize the time option, link it to the original travel application record, write it back to the travel data set, and add a user preference label that prioritizes time. Model iteration: The trip optimization model is incrementally trained using the updated dataset (containing 20 new data points showing "efficiency-first employees choosing time-first options"), and the objective function weights are adjusted (originally...). =0.6、 =0.4, after iteration targeting "efficiency-first" employees =0.4、 =0.6), and when pushing solutions to this type of employee in the future, the combination of "close to the optimal time and slightly higher cost" will be displayed first.

[0031] In one embodiment of the present invention, unlike the embodiments described above, a travel restriction simulation and optimization step is included between step S3 and step S4: S3A. Using business travel as the dataset and business travel as the analysis report as the basic data, and business travel restriction parameters as variable inputs; The Monte Carlo simulation method is used to simulate business travel behavior and costs under different travel constraints (such as advance booking days and hotel star ratings) in batches. By comparing simulation results, a report is outputting recommendations for adjusting travel restrictions, including expected cost savings, expected changes in compliance rates, and potential impacts on employee satisfaction. For example, given the company's current travel restrictions: "Book at least 3 days in advance" and "Supervisors are allowed to stay in four-star hotels," let's simulate two adjustment scenarios: Option A: Adjust to "Book at least 5 days in advance" and "Supervisor permits 3-star hotels"; Monte Carlo simulation (1000 iterations): Expected annual cost savings of RMB 150,000 (original annual travel costs of RMB 1.2 million, a saving of 12.5%), compliance rate increased from 80% to 95%, and employee satisfaction decreased from 85 points to 70 points (40% dissatisfaction rate among "efficiency-first" employees). Option B: Adjusted to "Book at least 4 days in advance" and "Supervisor's permission to select a four-star hotel + a contracted hotel is mandatory"; Monte Carlo simulation (1000 iterations): Expected annual cost savings of 100,000 yuan (8.3% savings), compliance rate increased from 80% to 92%, and employee satisfaction decreased from 85 points to 80 points (15% dissatisfaction rate of "efficiency-first" employees). Recommended adjustment: Option B is recommended, which balances "cost savings (100,000 yuan)," "compliance rate improvement (12%)," and "employee satisfaction (only a 5-point decrease)," avoiding employee resistance caused by the sharp drop in satisfaction in Option A.

[0032] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0033] like Figure 2 As shown, the following are embodiments of the apparatus for cost optimization prediction through multi-dimensional travel behavior analysis provided in this disclosure. This apparatus and the method for cost optimization prediction through multi-dimensional travel behavior analysis in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the apparatus for cost optimization prediction through multi-dimensional travel behavior analysis, please refer to the embodiments of the method for cost optimization prediction through multi-dimensional travel behavior analysis described above.

[0034] The device includes: The travel data acquisition module is used to extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. It also performs cleaning, deduplication, and entity alignment operations on the collected historical and real-time travel data to construct a travel behavior dataset. The travel feature extraction module is used to execute preset query scripts on the travel behavior dataset, extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance and supplier, and perform descriptive analysis, diagnostic analysis and cluster analysis on the extracted feature indicators to generate a travel behavior analysis report. The cost optimization prediction module is used to train the constructed cost optimization prediction model using the travel data set and travel analysis report, and to use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries, generating the optimal itinerary set. The recommendation and feedback execution module is used to push the optimal travel plan set to employee users, collect feedback results from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model.

[0035] This embodiment integrates internal and external enterprise data through the interactive collaboration of a travel data collection module, a travel feature extraction module, a cost optimization prediction module, and a recommendation and feedback execution module. It analyzes travel behavior from multiple dimensions, accurately predicts cost trends, intelligently optimizes itinerary plans, reduces travel costs, improves management efficiency and employee satisfaction, and enhances compliance.

[0036] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for cost optimization and prediction using multi-dimensional difference travel analysis, characterized in that, Includes the following steps: S1. Extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. Clean, deduplicate and entity-align the collected historical and real-time travel data to construct a travel behavior dataset. S2. Execute a preset query script on the business travel dataset to extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance, and suppliers. Perform descriptive analysis, diagnostic analysis, and cluster analysis on the extracted feature indicators to generate a business travel analysis report. S3. Use the travel data set and travel analysis report to train the constructed cost optimization prediction model, and use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries to generate the optimal itinerary set; S4. Push the optimal travel plan set to employee users, collect feedback from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model.

2. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 1, characterized in that, In step S1, the enterprise's internal systems include the ERP system, the human resources (HR) system, and the office automation (OA) approval system. External systems for enterprises include airline ticket price query interfaces, hotel group room price interfaces, and travel service management systems; Historical travel data of ERP expense reimbursement type is extracted from the database interface of the ERP system. The ERP expense reimbursement data includes expense order number, employee ID, expense type, amount, and reimbursement status. Historical travel data of HR organizational data type is extracted from the database interface of the human resources HR system. The HR organizational data includes employee ID, department, job level and position. Historical travel data of OA approval data type is extracted from the database interface of the office OA approval system. The OA approval data includes approval form ID, employee ID, trip purpose, departure point, destination and planned departure time. Real-time travel data of TMC order type is extracted from the travel service management system. The TMC order data includes order ID, employee ID, flight number, hotel name, reservation time, check-in time, check-out time, and order amount. The entity alignment operation specifically involves using employee ID and department code within the enterprise as the primary key to associate ERP expense reimbursement data, HR organization data, OA approval data, and TMC order data with the same travel behavior fact record, thereby completing the construction of the travel behavior dataset.

3. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 1, characterized in that, The specific steps of step S2 are as follows: S21. Using business travel as the dataset as input, execute the pre-defined query script to extract the following business travel features: Extract employee job level, department, and travel frequency features from a personnel perspective; The frequency of business trips is calculated by counting the number of records in the OA approval data. Extract booking advance time period and travel date attribute features from the time dimension; The advance booking period is calculated by determining the time difference between the booking time in the TMC order data and the planned departure time in the OA approval data. Extract features of departure point, destination, and mode of transportation from the perspective of travel itinerary; Extract the characteristics of per capita cost and cost composition ratio from the cost dimension; Extract characteristics of travel restriction violation types, violation amounts, and approval time from a compliance perspective; Extract supplier usage concentration and agreement price usage characteristics from the supplier dimension; S22. Perform analysis using the extracted travel routes as features: First, descriptive analysis is used to generate historical trend and percentage reports, outputting the monthly total travel cost trend and the ranking of expense percentages for each department; Then, using drill-down and association rule methods in diagnostic analysis, the root causes of cost anomalies or travel restriction violations are analyzed, and a cost overrun root cause analysis report is output. Finally, cluster analysis was used to divide employees into different travel behavior groups using travel frequency, average advance booking days, average discount rate, and hotel usage rate as feature vectors, and output employee travel behavior classification labels. S23. Integrate the results of descriptive analysis, diagnostic analysis, and cluster analysis to generate a travel experience analysis report, which includes at least: List of high-frequency travelers, ranking of departments with cost overruns, statistics on violations of travel restrictions, supplier performance evaluations, and classification results of employee travel behavior groups.

4. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 3, characterized in that, In step S22, the K-Means algorithm is used for cluster analysis to output classification labels for employee travel behavior; The objective function of the K-Means algorithm for cluster analysis is to minimize the sum of squares within each cluster. in, The preset number of clusters, For clustering schemes, It is the i-th cluster. It is the feature vector of employee business travel behavior. It is a cluster The mean center; It should be noted that, It is composed of characteristics such as travel frequency, average advance booking days, average discount rate, and hotel occupancy rate.

5. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 3, characterized in that, The specific steps of step S3 are as follows: S31. Construct a unified training data source based on the travel data set and the travel analysis report: Historical business trips were extracted from OA approval data in the business trip dataset to construct a time series of business trips. Historical airfare price sequences were extracted from TMC order data and ERP expense reimbursement data, which were used as a dataset for business travel. Extract cost overrun tags, travel restriction violation tags, and employee travel behavior group classification tags from the business travel behavior analysis report; S32. Perform demand forecasting, price forecasting, and route optimization using a cost optimization forecasting model: Pre-trained cost optimization prediction models include demand prediction models, price prediction models, and route optimization models; Input the time series of business trips into a time series-type demand forecasting model to predict the total business travel demand within a future set time period. By inputting historical airfare price sequences, advance booking periods, and current fuel prices and holiday indicators obtained from external systems into a machine learning regression-type price prediction model, the airfare for future routes can be predicted. The total predicted travel demand within a specified future time period, the predicted airfare prices for specified future routes, employee travel behavior classification labels and travel restriction compliance requirements obtained from the travel behavior analysis report, and the contract hotel prices obtained from TMC order data and external hotel interface data in the travel behavior dataset are input into the trip optimization model using the optimization constraint algorithm. The optimization calculation is performed with the goal of minimizing total cost and maximizing travel efficiency, generating a candidate trip plan set containing several candidate options. S33. Sort and format the candidate itinerary options set to generate the optimal itinerary option set, which includes the total cost of the option, the total travel time, the combination of transportation and accommodation, and the description of compliance with travel restrictions.

6. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 5, characterized in that, The specific training steps for the cost optimization prediction model in step S32 are as follows: S321. Extract historical time series data from the OA approval data of the business trip dataset, including the number of business trips and the dates of business trips, to form the first training sample; S322. Use the LSTM model to build a demand prediction model, and then use the first training sample to train the LSTM model to obtain a trained demand prediction model. S323. Extract historical feature data from the TMC order data and ERP reimbursement data of the travel data set, including historical airfare prices, booking advance period, and routes, and integrate current oil prices and holiday markers obtained from external systems as feature vectors, using historical airfare prices as labels to construct a second training sample; S324. Use the second training sample to train the gradient boosting tree regression model to obtain a trained price prediction model; S325. Extract historical excellent travel plans from the business travel dataset as positive samples. The positive samples must meet the following requirements: low total cost, high travel efficiency, and compliance with business travel policies. S326. Using the departure point, destination, time, employee travel behavior classification labels, and predicted future flight ticket prices of the analysis report as features, a recommendation model based on constraint optimization is trained using positive samples to obtain a trained itinerary optimization model.

7. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 6, characterized in that, In step S326, the objective function of the constraint-based recommendation model is to minimize the total travel cost and travel time. in, It's the cost of the airfare. It's the hotel's cost. It is the cost of ground transportation. It is the total travel time. , These are preset weighting coefficients. ; The constraints of the recommendation model based on constraint optimization include: Flight departure time should not be earlier than the earliest departure time of employees, hotel star rating should not be higher than the employee's job level standard, and total cost should not exceed the budget threshold; We optimized the weighting of cost and duration using the travel behavior dataset and the travel restriction compliance requirements and employee travel behavior classification results from the travel behavior analysis report.

8. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. Push the generated optimal itinerary plan set to the target employee's user terminal through the application programming interface of the enterprise's internal travel management system; The optimal travel plan set includes at least two plans: The lowest total cost pair marked with the cost-optimal solution; The route with the shortest total travel time, marked as the most efficient route; Each option provides details on the total cost, total travel time, transportation and accommodation combination, and compliance with travel restrictions. S42. Collect employee users' selection results of push notification options in real time through the application programming interface of the enterprise's internal travel management system; S43. After associating the employee user's selection result with the corresponding travel application record, write it back as a new data record and update the travel behavior dataset; S44. Using the updated travel data as the new training data, incrementally train the cost optimization prediction model to achieve continuous iterative optimization of the cost optimization prediction model.

9. The method for cost optimization and prediction through multi-dimensional travel behavior analysis according to claim 1, characterized in that, Between step S3 and step S4, there is also a travel restriction simulation and optimization step: S3A. Using business travel as the dataset and business travel as the analysis report as the basic data, and business travel restriction parameters as variable inputs; Using the Monte Carlo simulation method, we simulated travel behavior and costs under different travel constraint parameters in batches. By comparing simulation results, a report is generated recommending adjustments to travel restrictions, including expected cost savings, expected changes in compliance rates, and potential impacts on employee satisfaction.

10. An apparatus for cost optimization prediction through multi-dimensional difference travel analysis, characterized in that, include: The travel data acquisition module is used to extract historical travel data from the database interface of the enterprise's internal system and collect real-time travel data from the data interface of the enterprise's external system. It also performs cleaning, deduplication, and entity alignment operations on the collected historical and real-time travel data to construct a travel behavior dataset. The travel feature extraction module is used to execute preset query scripts on the travel behavior dataset, extract feature indicators from the dimensions of personnel, time, itinerary, cost, compliance and supplier, and perform descriptive analysis, diagnostic analysis and cluster analysis on the extracted feature indicators to generate a travel behavior analysis report. The cost optimization prediction module is used to train the constructed cost optimization prediction model using the travel data set and travel analysis report, and to use the trained cost optimization prediction model to predict travel demand, travel prices and optimize travel itineraries, generating the optimal itinerary set. The recommendation and feedback execution module is used to push the optimal travel plan set to employee users, collect feedback results from employee users on the optimal travel plan set, update the travel data set with the feedback results, and then perform incremental training and iterative optimization on the cost optimization prediction model.

Images