Business business travel itinerary recommendation and cost optimization method and system based on big data

Abstract

The invention provides a business travel itinerary recommendation and cost optimization method and system based on big data, and relates to the technical field of big data, and the method comprises the steps: obtaining the real-time traffic information and hotel information of a destination city; a machine learning model is utilized to analyze user travel characteristics, traffic information is intelligently screened, and a traffic scheme conforming to enterprise travel standards is generated. And determining an accommodation time interval according to the traffic scheme and the business activity time, and screening candidate hotels. And performing time sequence analysis on the air ticket price and the house fee, predicting a price fluctuation trend, determining the optimal reservation time, and generating a business travel itinerary initial scheme. And finally, pushing a travel confirmation request to a related contact person, dynamically optimizing the initial scheme according to feedback information, and outputting a final travel travel scheme with the lowest total cost. According to the invention, big data analysis, machine learning and time sequence analysis technologies are combined to intelligently recommend the business travel scheme which accords with enterprise business travel standards and has the lowest cost, so that the business travel efficiency is improved, and the business travel cost is reduced.

Description

Technical Field

[0001] The present invention relates to big data technology, and in particular to a business travel itinerary recommendation and cost optimization method and system based on big data. Background Art

[0002] Business travel is an important part of corporate operations. Efficient travel management can effectively control costs, improve employee satisfaction and increase work efficiency. Traditional travel booking methods usually rely on manual operations, such as booking air tickets and hotels through travel agents or searching by yourself. This method is time-consuming and labor-intensive, and is easily affected by human factors, making it difficult to achieve cost optimization. With the rapid development of the Internet and mobile Internet technologies, online travel booking platforms have gradually emerged, providing companies and employees with more convenient travel booking services. However, existing online travel booking platforms still have some defects and shortcomings.

[0003] First, most platforms only provide simple search and price comparison functions, lacking personalized recommendations and intelligent optimization capabilities. Information such as users' historical travel preferences and corporate travel standards is often not fully utilized, resulting in a deviation between the recommendation results and actual needs, making it difficult to meet companies' requirements for travel cost control and efficiency improvement.

[0004] Secondly, existing travel platforms are not capable of integrating real-time information. Real-time information such as traffic delays and hotel price fluctuations is crucial for travel arrangements, but many platforms fail to obtain and effectively utilize this information in a timely manner, resulting in a lack of flexibility in travel arrangements and being easily affected by emergencies.

[0005] Finally, the existing travel platform lacks effective integration with the internal systems of enterprises. Corporate travel management usually involves approval processes, expense reimbursement and other links. If the travel platform cannot be connected with the internal systems of enterprises, it will increase the management burden and reduce efficiency. Summary of the invention

[0006] The embodiments of the present invention provide a business travel itinerary recommendation and cost optimization method and system based on big data, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] Receive travel demand information input by a user, wherein the travel demand information includes a destination city and a business activity time; based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and simultaneously obtain real-time traffic information and hotel information of the destination city from a third-party data interface; input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model;

[0009] Analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently filter the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; determine the accommodation time interval based on the arrival time of the traffic plan and the business activity time; and filter out a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards;

[0010] Perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predict the price fluctuation trend in the next seven days, and determine the optimal booking time; generate an initial business trip itinerary plan based on the optimal booking time; push an itinerary confirmation request to the user's business activity-related contacts; receive feedback information on the itinerary confirmation request, dynamically optimize the initial business trip itinerary plan based on the feedback information, and output a final business trip itinerary plan with the lowest total cost.

[0011] Analyzing the user travel characteristics in the historical travel records by the machine learning model, intelligently screening the real-time traffic information according to the travel characteristics, and generating a traffic plan that meets the corporate travel standards includes:

[0012] Extracting time features, geographic features, traffic preference features, and price sensitivity features from the historical travel records, mapping the time features, the geographic features, the traffic preference features, and the price sensitivity features to a continuous feature space through a feature embedding vector, and generating a user feature vector;

[0013] A multi-head self-attention mechanism is used to process the user feature vector, wherein attention weights are calculated through a query matrix, a key matrix, and a value matrix, multiple attention heads are generated based on the attention weights, and the multiple attention heads are concatenated and multiplied with a parameter matrix to obtain a user difference travel behavior representation;

[0014] A two-layer optimization model is constructed based on the user's business trip as a representation, wherein the upper optimization goal of the two-layer optimization model is to minimize the total travel cost, and the lower constraint conditions include the travel budget constraint, the arrival time window constraint and the comfort constraint; and real-time traffic data is input into the two-layer optimization model;

[0015] Using a long short-term memory network in combination with an attention mechanism to predict transportation fares, calculating attention weights according to hidden states of the long short-term memory network, and weighting the hidden states based on the attention weights to obtain predicted prices;

[0016] Calculate the cost-effectiveness scores of different transportation solutions, where the cost-effectiveness scores are obtained by weighted calculation of the cost score, the comfort score, and the time score; sort the transportation solutions according to the cost-effectiveness scores and select the optimal transportation solution.

[0017] A two-layer optimization model is constructed based on the user's business trip as a representation. The upper optimization goal of the two-layer optimization model is to minimize the total travel cost. The lower constraints include travel budget constraints, arrival time window constraints and comfort constraints, including:

[0018] Constructing an upper optimization objective function of a two-layer optimization model, wherein the upper optimization objective function includes a traffic scheme cost item and a time cost item, wherein the traffic scheme cost item is the sum of the products of the costs of each traffic scheme and the corresponding 0-1 decision variables, and the time cost item is the sum of the products of the time costs of each traffic scheme and the corresponding 0-1 decision variables, and introducing a time cost weight coefficient;

[0019] Construct lower-level constraints based on the corporate travel budget, which include a budget constraint function, a time window constraint function, and a comfort constraint function. The budget constraint function requires that the sum of the product of the actual expenditure of each plan and the corresponding 0-1 decision variable must be less than the budget upper limit, and the time window constraint function requires that the sum of the product of the arrival time of each plan and the corresponding 0-1 decision variable must be within a preset time window;

[0020] Construct a multidimensional comfort evaluation function, and multiply the evaluation values ​​of the three dimensions of transportation grade score, transfer number penalty item, and seat comfort by the corresponding weights and sum them up. The transportation grade score is the sum of the products of each transportation scheme grade score and the corresponding 0-1 decision variable, the transfer number penalty item is the exponential function transformation value of the transfer number, and the seat comfort is the sum of the products of each transportation scheme seat comfort score and the corresponding 0-1 decision variable;

[0021] Introducing slack variables to construct a Lagrangian optimization objective function, the Lagrangian optimization objective function includes an original objective function term and a constraint penalty term, wherein the constraint penalty term is the sum of the product of each constraint violation degree and the corresponding Lagrangian multiplier; updating the Lagrangian multiplier by iteration, the update formula of the Lagrangian multiplier includes the original multiplier value, the step size parameter and the constraint violation degree;

[0022] The primal-dual iterative algorithm is used to solve the Lagrangian optimization objective function. In each iteration, the optimal value of the primal variable is first solved, and then the dual variable is updated based on the optimal value of the primal variable. The optimal solution is determined when the difference between two iterations of the primal variable is less than a first threshold and the constraint violation degree is less than a second threshold.

[0023] Performing time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predicting the price fluctuation trend in the next seven days, and determining the optimal booking time; generating an initial travel itinerary plan based on the optimal booking time includes:

[0024] Decomposing the original price sequence into a trend component, a seasonal component and a random fluctuation component; constructing a price feature vector, wherein the price feature vector includes a historical price sequence, date-related features, historical statistical features and external factor features, wherein the historical statistical features include a mean, a variance, a maximum value and a minimum value;

[0025] A multi-scale time series prediction model is constructed based on the price feature vector, short-term price fluctuations are modeled by multiple basic learners, and a short-term price prediction value is obtained by weighted combination of the basic learner integrated weights; a long short-term memory network is used in combination with an attention mechanism to model medium- and long-term price trends, and the short-term price prediction value is merged with the medium- and long-term price trend prediction value to obtain a final price prediction value;

[0026] Calculate the standard deviation of historical price returns to obtain volatility, and calculate the risk value based on the volatility, expected price and confidence level; input the risk value, volatility and the time interval from the departure date into the utility function, wherein the input parameters of the utility function also include the historical highest price;

[0027] Under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, determining the optimal booking time by maximizing the utility function value, and the dynamic decision rule also considers the budget constraint condition;

[0028] Calculate the average absolute percentage error between the final price forecast value and the actual price to evaluate the forecast accuracy; calculate the saving ratio of the price corresponding to the optimal booking time relative to the average booking price to evaluate the profit level;

[0029] A loss function is constructed based on the mean absolute percentage error and the savings ratio, and the parameters of the prediction model are adaptively updated by the gradient descent method, wherein the parameters of the prediction model include the basic learner ensemble weights, long short-term memory network parameters, attention mechanism parameters and utility function weight coefficients, and the optimized travel expense prediction model and optimal booking time are output.

[0030] Under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the utility function value includes:

[0031] Acquire historical price data, extract the historical highest price from the historical price data, and calculate the time interval between the current time and the departure date based on the historical price data;

[0032] Constructing a utility function, wherein the utility function includes a price benefit term, a volatility risk term, and a time interval term, wherein the price benefit term is the difference between the historical highest price and the predicted price multiplied by a first weight coefficient, the volatility risk term is the price volatility multiplied by a second weight coefficient, and the time interval term is the time interval from the departure date multiplied by a third weight coefficient;

[0033] Setting a time window constraint, wherein the time window constraint includes the current time and the latest booking time, wherein the latest booking time is the departure time minus the minimum advance booking time, and the minimum advance booking time is determined based on historical ticket booking data;

[0034] Constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the output value of the utility function under the premise of satisfying the time window constraint and the budget constraint, wherein the budget constraint is determined based on the corporate travel standard;

[0035] The weight coefficient of the utility function is dynamically adjusted by using a sliding time window method, and the weight coefficient combination with the highest benefit is selected as the optimal weight coefficient by calculating the booking benefits under different weight coefficient combinations;

[0036] Substitute the optimal weight coefficient into the utility function, recalculate the optimal booking time based on the time window constraint and the budget constraint, and output the final travel booking time recommendation.

[0037] Receiving feedback information of the itinerary confirmation request, dynamically optimizing the initial itinerary plan for the business trip according to the feedback information, and outputting a final itinerary plan with the lowest total cost includes:

[0038] The feedback information is divided into time preference feedback information, comfort feedback information and price sensitivity feedback information; the feedback information is quantified, and a feedback quantization value is obtained based on the weight coefficient and score of each feedback dimension;

[0039] The time window constraint is adjusted according to the feedback quantization value of the time preference feedback information, and the start and end time of the original time window is dynamically adjusted; the comfort threshold is updated based on the feedback quantization value of the comfort feedback information, and the product of the original comfort threshold and the comfort adjustment coefficient is used as the updated comfort threshold; the budget constraint is adjusted according to the feedback quantization value of the price sensitivity feedback information, and the product of the original budget and the budget adjustment coefficient is used as the updated budget constraint;

[0040] Constructing a multi-objective optimization model, wherein the objective function of the multi-objective optimization model includes a total cost function term, a time loss function term, and a comfort function term, and introducing corresponding weight coefficients; dynamically adjusting the weight coefficients based on the feedback quantization value, and calculating the updated weight coefficients through the weight adjustment rate;

[0041] The incremental optimization strategy is used to optimize the difference travel itinerary plan. The solution with the smallest increment of the objective function is searched in the neighborhood solution space of the current plan as the local adjustment plan. The simulated annealing algorithm is combined for global optimization, and the acceptance probability of the plan is calculated according to the increment of the objective function.

[0042] Construct a comprehensive scoring function, and perform a weighted summation of the product of the evaluation function value of each evaluation dimension and the corresponding feedback quantization value to obtain the solution score; when the difference between the solution scores of two adjacent iterations is less than the preset convergence threshold, select the solution with the higher score as the final optimization solution;

[0043] Generate a travel timeline, cost composition analysis and comfort evaluation report based on the final optimization plan; perform a sensitivity analysis on the final optimization plan to evaluate the impact of price fluctuations, time window changes and comfort demand changes on the plan, and output a business trip itinerary plan with the best comprehensive score.

[0044] A second aspect of an embodiment of the present invention provides a business travel itinerary recommendation and cost optimization system based on big data, including:

[0045] The first unit is used to receive travel demand information input by a user, wherein the travel demand information is a destination city and a business activity time; based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and simultaneously obtain the real-time traffic information and hotel information of the destination city from a third-party data interface; and input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model;

[0046] The second unit is used to analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently screen the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; determine the accommodation time interval based on the arrival time of the traffic plan and the business activity time; and screen a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards;

[0047] The third unit is used to perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predict the price fluctuation trend in the next seven days, and determine the optimal booking time; generate an initial business trip itinerary plan according to the optimal booking time; push an itinerary confirmation request to the user's business activity-related contacts; receive feedback information of the itinerary confirmation request, dynamically optimize the initial business trip itinerary plan according to the feedback information, and output a final business trip itinerary plan with the lowest total cost.

[0048] A third aspect of the embodiments of the present invention

[0049] An electronic device is provided, comprising:

[0050] processor;

[0051] a memory for storing processor-executable instructions;

[0052] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0053] A fourth aspect of the embodiments of the present invention is:

[0054] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0055] The beneficial effects of this application are as follows:

[0056] 1. Personalized recommendations: Based on the user's historical travel characteristics and corporate travel standards, intelligently screen transportation and accommodation plans to provide travel arrangements that are more in line with user preferences.

[0057] 2. Cost optimization: Use time series analysis to predict the price fluctuation trend of air tickets and hotels, determine the best booking time, and combine it with dynamic itinerary optimization to minimize the total travel cost.

[0058] 3. Improve efficiency: Automatically generate travel itinerary plans and push confirmations, simplify the booking process, and save users time and energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of a business travel itinerary recommendation and cost optimization method based on big data according to an embodiment of the present invention;

[0060] Figure 2 It is a structural diagram of a business travel itinerary recommendation and cost optimization system based on big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0063] Figure 1 FIG. 1 is a flow chart of a business travel itinerary recommendation and cost optimization method based on big data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0064] S101. Receive travel demand information input by a user, wherein the travel demand information includes a destination city and a business activity time; based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and simultaneously obtain real-time traffic information and hotel information of the destination city from a third-party data interface; input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model;

[0065] S102. Analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently filter the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; determine the accommodation time interval based on the arrival time of the traffic plan and the business activity time; and filter out a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards;

[0066] S103. Perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predict the price fluctuation trend in the next seven days, and determine the optimal booking time; generate an initial business trip itinerary plan based on the optimal booking time; push an itinerary confirmation request to the user's business activity-related contacts; receive feedback information on the itinerary confirmation request, dynamically optimize the initial business trip itinerary plan based on the feedback information, and output a final business trip itinerary plan with the lowest total cost.

[0067] In an optional implementation, analyzing the user travel characteristics in the historical travel records by the machine learning model, intelligently screening the real-time traffic information according to the travel characteristics, and generating a traffic plan that meets the corporate travel standards includes:

[0068] Extracting time features, geographic features, traffic preference features, and price sensitivity features from the historical travel records, mapping the time features, the geographic features, the traffic preference features, and the price sensitivity features to a continuous feature space through a feature embedding vector, and generating a user feature vector;

[0069] A multi-head self-attention mechanism is used to process the user feature vector, wherein attention weights are calculated through a query matrix, a key matrix, and a value matrix, multiple attention heads are generated based on the attention weights, and the multiple attention heads are concatenated and multiplied with a parameter matrix to obtain a user difference travel behavior representation;

[0070] A two-layer optimization model is constructed based on the user's business trip as a representation, wherein the upper optimization goal of the two-layer optimization model is to minimize the total travel cost, and the lower constraint conditions include the travel budget constraint, the arrival time window constraint and the comfort constraint; and real-time traffic data is input into the two-layer optimization model;

[0071] Using a long short-term memory network in combination with an attention mechanism to predict transportation fares, calculating attention weights according to hidden states of the long short-term memory network, and weighting the hidden states based on the attention weights to obtain predicted prices;

[0072] Calculate the cost-effectiveness scores of different transportation solutions, where the cost-effectiveness scores are obtained by weighted calculation of the cost score, the comfort score, and the time score; sort the transportation solutions according to the cost-effectiveness scores and select the optimal transportation solution.

[0073] A method for generating a transportation plan that meets corporate travel standards based on historical travel records and real-time traffic information is specifically implemented as follows:

[0074] First, extract the user's travel characteristics from historical travel records. For example, by analyzing the user's historical travel records, extract the user's travel time pattern, such as when they usually depart and how long the trip lasts; extract the user's travel location information, such as which cities they often travel between and where they stay; extract the user's transportation preference, such as whether they prefer to take airplanes, high-speed trains, or trains; and extract the user's price sensitivity, such as whether they prefer to choose economy class or business class.

[0075] For example, if a user's historical business trip record shows that he often departs from Beijing to Shanghai at 8 am on Monday and returns to Beijing at 5 pm on Friday, and usually chooses a second-class high-speed rail seat. Then, the user's time feature is "departure at 8 am on Monday and return at 5 pm on Friday", the geographical feature is "Beijing-Shanghai", the transportation preference feature is "high-speed rail", and the price sensitivity feature is "medium".

[0076] Then, the extracted time features, geographic features, traffic preference features, and price sensitivity features are converted into numerical representations that can be understood by computers. This can be achieved through feature embedding technology, which maps each feature to a point in a high-dimensional vector space, thereby converting discrete features into continuous numerical vectors, which is convenient for subsequent machine learning model processing. For example, "Beijing" can be represented as [0.2, 0.5, 0.8, ...], and "high-speed rail" can be represented as [0.1, 0.7, 0.2, ...]. Assume that the feature vectors of the above users are time feature vector T, geographic feature vector G, traffic preference feature vector P, and price sensitivity feature vector S.

[0077] Next, the user feature vector is processed using a multi-head self-attention mechanism. The user's feature vectors T, G, P, and S are concatenated into a large feature vector. The multi-head self-attention mechanism calculates the correlation between different parts of the feature vector and generates multiple "attention heads". Each attention head focuses on different aspects of the feature vector. For example, one attention head may focus more on the correlation between time and place, and another attention head may focus more on the correlation between transportation and price. The results of multiple attention heads are combined to obtain the final user travel behavior representation. For example, the user's travel behavior representation may indicate that he tends to choose a cost-effective travel method.

[0078] Based on the user's business trip, a two-level optimization model is constructed. The upper level goal of the model is to minimize the total travel cost, and the lower level constraints include travel budget, arrival time window, and comfort requirements. Real-time transportation data (such as flight, high-speed rail, train, etc.) are input into the model. For example, the model considers that the user's budget limit is 1,000 yuan, requires arriving in Shanghai before 6 pm on Friday, and tends to choose a more comfortable means of transportation.

[0079] At the same time, the long short-term memory network combined with the attention mechanism is used to predict transportation fares. The long short-term memory network can learn the fluctuation pattern of historical ticket prices and predict future ticket price trends. The attention mechanism can help the model focus on historical information related to the current prediction. For example, the model may pay more attention to the ticket price data in the same time period in the past few weeks. Suppose the prediction results show that the second-class ticket price of the high-speed rail from Beijing to Shanghai is 500 yuan, and the economy class ticket price of the plane is 800 yuan.

[0080] Finally, based on the predicted ticket price and the user's travel standards, the cost-effectiveness scores of different transportation options are calculated. The cost-effectiveness score takes into account factors such as cost, comfort, and time. For example, the cost score of the high-speed rail option is 5 points (out of 10 points), the comfort score is 4 points, and the time score is 3 points; the cost score of the airplane option is 3 points, the comfort score is 5 points, and the time score is 5 points. According to the preset weights, the final score of each option is calculated, and the transportation options are sorted according to the score, and the best option is finally recommended to the user. For example, assuming that the final score of the high-speed rail option is 4.5 points and the score of the airplane option is 4.2 points, the high-speed rail option is recommended to the user.

[0081] The beneficial effects of this method can be summarized in the following three aspects:

[0082] Personalized recommendations provide personalized transportation plan recommendations based on the user's historical travel habits and preferences to meet the travel needs of different users.

[0083] Cost saving: Through the two-layer optimization model and fare prediction mechanism, it helps users choose the most cost-effective travel plan and save travel costs.

[0084] Improve efficiency: Automatically generate transportation plans that meet corporate travel standards, reduce manual intervention, and improve travel arrangement efficiency.

[0085] In an optional implementation, a two-layer optimization model is constructed based on the user's business trip as a representation, the upper optimization goal of the two-layer optimization model is to minimize the total travel cost, and the lower constraint conditions include travel budget constraints, arrival time window constraints and comfort constraints, including:

[0086] Constructing an upper optimization objective function of a two-layer optimization model, wherein the upper optimization objective function includes a traffic scheme cost item and a time cost item, wherein the traffic scheme cost item is the sum of the products of the costs of each traffic scheme and the corresponding 0-1 decision variables, and the time cost item is the sum of the products of the time costs of each traffic scheme and the corresponding 0-1 decision variables, and introducing a time cost weight coefficient;

[0087] Construct lower-level constraints based on the corporate travel budget, which include a budget constraint function, a time window constraint function, and a comfort constraint function. The budget constraint function requires that the sum of the product of the actual expenditure of each plan and the corresponding 0-1 decision variable must be less than the budget upper limit, and the time window constraint function requires that the sum of the product of the arrival time of each plan and the corresponding 0-1 decision variable must be within a preset time window;

[0088] Construct a multidimensional comfort evaluation function, and multiply the evaluation values ​​of the three dimensions of transportation grade score, transfer number penalty item, and seat comfort by the corresponding weights and sum them up. The transportation grade score is the sum of the products of each transportation scheme grade score and the corresponding 0-1 decision variable, the transfer number penalty item is the exponential function transformation value of the transfer number, and the seat comfort is the sum of the products of each transportation scheme seat comfort score and the corresponding 0-1 decision variable;

[0089] Introducing slack variables to construct a Lagrangian optimization objective function, the Lagrangian optimization objective function includes an original objective function term and a constraint penalty term, wherein the constraint penalty term is the sum of the product of each constraint violation degree and the corresponding Lagrangian multiplier; updating the Lagrangian multiplier by iteration, the update formula of the Lagrangian multiplier includes the original multiplier value, the step size parameter and the constraint violation degree;

[0090] The primal-dual iterative algorithm is used to solve the Lagrangian optimization objective function. In each iteration, the optimal value of the primal variable is first solved, and then the dual variable is updated based on the optimal value of the primal variable. The optimal solution is determined when the difference between two iterations of the primal variable is less than a first threshold and the constraint violation degree is less than a second threshold.

[0091] A two-level optimization model based on user differential travel is constructed to minimize the total travel cost while satisfying constraints such as budget, time, and comfort. The implementation method and steps are described in detail below.

[0092] First, collect and preprocess travel data. This includes the user's historical travel records, preferences, and other related information, such as transportation type, departure point, destination, departure time, arrival time, cost, seat class, etc. Clean and organize the collected data, such as removing outliers and filling missing values, to ensure the accuracy and completeness of the data. For example, collect travel data of 100 users, including their travel records in the past year, and clean the data to remove invalid records and outliers.

[0093] Next, build a travel plan set. Generate a set of feasible travel plans based on the user's travel needs and preprocessed data. Each plan contains information such as specific transportation combinations, itinerary arrangements, estimated costs and time. For example, three plans are generated for a user traveling from Beijing to Shanghai: Plan 1, take the high-speed rail directly; Plan 2, take a plane directly; Plan 3, take a train to Nanjing first, and then transfer to the high-speed rail to Shanghai.

[0094] Then, a two-level optimization model is constructed. The upper-level goal is to minimize the total travel cost, including the transportation plan cost and the time cost. The transportation plan cost refers to the actual cost of each transportation plan, and the time cost refers to the time spent by the user on the road. The time cost is quantified by introducing the time cost weight coefficient, which reflects the user's emphasis on the time cost. The lower-level constraints include budget constraints, time window constraints, and comfort constraints. The budget constraint means that the user's total expenditure cannot exceed the preset budget limit. The time window constraint means that the user's arrival time must be within the preset time range. The comfort constraint is measured by a multidimensional comfort evaluation function.

[0095] The multidimensional comfort evaluation function comprehensively considers three dimensions: transportation grade, number of transfers, and seat comfort. Transportation grade is scored according to the type of transportation and service level. For example, the grade of high-speed rail is higher than that of ordinary trains. The transfer number penalty term transforms the number of transfers into an exponential function to reflect the user's aversion to frequent transfers. Seat comfort is scored according to the type and size of the seat. For example, the seat comfort score of business class is higher than that of economy class. The evaluation value of each dimension is multiplied by the corresponding weight coefficient, and then the sum is obtained to obtain the final comfort score. For example, suppose the weight of transportation grade is 0.4, the weight of transfer number is 0.3, and the weight of seat comfort is 0.3. Option 1: high-speed rail direct access, grade score is 9, number of transfers is 0, and seat comfort is 8; Option 2: airplane direct access, grade score is 10, number of transfers is 0, and seat comfort is 7; Option 3: train to high-speed rail, grade score is 7, number of transfers is 1, and seat comfort is 6.

[0096] Slack variables and Lagrange multipliers are introduced to integrate constraints into the objective function and construct the Lagrange optimization objective function. This function contains the original objective function term and the constraint penalty term, which is obtained by multiplying the constraint violation degree and the corresponding Lagrange multiplier.

[0097] The primal-dual iterative algorithm is used to solve the Lagrangian optimization objective function. In each iteration, the Lagrangian multiplier is first fixed to solve the optimal value of the original variable (i.e., the 0-1 decision variable of each solution); then, the Lagrangian multiplier is updated based on the optimal value of the original variable. The original variable and the dual variable (i.e., the Lagrangian multiplier) are continuously updated iteratively until the preset stopping criteria are met, such as the difference between the two iterations of the original variable is less than the first threshold and the constraint violation is less than the second threshold.

[0098] Finally, the optimal travel plan is selected based on the solution results. This plan can minimize the user's total travel cost while meeting constraints such as budget, time, and comfort.

[0099] Beneficial effects:

[0100] 1. Reduce travel costs: By optimizing the model, users’ travel costs, including transportation costs and time costs, can be effectively reduced. The model can select the most economical transportation plan while meeting user needs.

[0101] 2. Improve travel efficiency: The model can help users quickly find travel plans that meet their requirements, reduce the time and energy users spend on planning itineraries, and improve travel efficiency.

[0102] 3. Improve travel experience: The model takes into account the user's comfort needs and can provide users with more comfortable travel plans and improve their travel experience.

[0103] In an optional implementation, a time series analysis is performed on the air ticket prices in the transportation plan and the room rates in the candidate hotel list to predict the price fluctuation trend in the next seven days and determine the optimal booking time; generating an initial travel itinerary plan according to the optimal booking time includes:

[0104] Decomposing the original price sequence into a trend component, a seasonal component and a random fluctuation component; constructing a price feature vector, wherein the price feature vector includes a historical price sequence, date-related features, historical statistical features and external factor features, wherein the historical statistical features include a mean, a variance, a maximum value and a minimum value;

[0105] A multi-scale time series prediction model is constructed based on the price feature vector, short-term price fluctuations are modeled by multiple basic learners, and a short-term price prediction value is obtained by weighted combination of the basic learner integrated weights; a long short-term memory network is used in combination with an attention mechanism to model medium- and long-term price trends, and the short-term price prediction value is merged with the medium- and long-term price trend prediction value to obtain a final price prediction value;

[0106] Calculate the standard deviation of historical price returns to obtain volatility, and calculate the risk value based on the volatility, expected price and confidence level; input the risk value, volatility and the time interval from the departure date into the utility function, wherein the input parameters of the utility function also include the historical highest price;

[0107] Under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, determining the optimal booking time by maximizing the utility function value, and the dynamic decision rule also considers the budget constraint condition;

[0108] Calculate the average absolute percentage error between the final price forecast value and the actual price to evaluate the forecast accuracy; calculate the saving ratio of the price corresponding to the optimal booking time relative to the average booking price to evaluate the profit level;

[0109] A loss function is constructed based on the mean absolute percentage error and the savings ratio, and the parameters of the prediction model are adaptively updated by the gradient descent method, wherein the parameters of the prediction model include the basic learner ensemble weights, long short-term memory network parameters, attention mechanism parameters and utility function weight coefficients, and the optimized travel expense prediction model and optimal booking time are output.

[0110] The method for predicting transportation travel expenses and determining the optimal booking time is specifically implemented as follows:

[0111] First, obtain the historical price data of air tickets and hotels. For example, collect the daily air ticket price and hotel price data for the past year, and clean and preprocess the data, such as handling missing values ​​and outliers.

[0112] Then, the price time series is decomposed. The original price series is decomposed into trend component, seasonal component and random fluctuation component. For example, the STL decomposition method is used to decompose the time series into these three components. Assume that the prices of a flight from January 1 to 7 are: 800, 850, 900, 880, 920, 1000, 950 yuan. The trend is identified by the moving average method, and it is found that the price is generally on an upward trend; the seasonality is identified by periodic analysis, and it is found that the price is generally higher on weekends; the remaining fluctuations are regarded as random fluctuations.

[0113] Next, construct a price feature vector. This vector contains historical price series, date-related features, historical statistical features, and external factor features. Historical price series include prices for the past 7 days; date-related features include day of the week, month, and whether it is a holiday; historical statistical features include the mean, variance, maximum, and minimum of prices for the past 7 days; and external factor features include weather, large events, etc. Taking the above flight as an example, the feature vector for January 7 can include: prices for the previous 7 days (800, 850, 900, 880, 920, 1000, 950), date (Sunday, January), historical statistical features (mean 885.7, variance 5142.9, maximum 1000, minimum 800), and the assumption that there are no external factors.

[0114] After that, a multi-scale time series prediction model is constructed. Multiple basic learners are used to model short-term price fluctuations. For example, models such as linear regression, support vector regression, and decision tree regression are used to predict the prices for the next 1-3 days. Then, the prediction results of each model are combined to obtain the short-term price forecast value through ensemble learning methods, such as weighted average. Assuming that the linear regression predicts that the price on January 8 is 960 yuan, the support vector regression predicts 970 yuan, and the decision tree regression predicts 980 yuan, and the weights are assigned to them as 0.3, 0.4, and 0.3 respectively, the integrated short-term forecast price is 960*0.3+970*0.4+980*0.3=969 yuan.

[0115] At the same time, the long short-term memory network combined with the attention mechanism is used to model the medium- and long-term price trends, such as predicting the prices in the next 4-7 days. The attention mechanism can focus on the time period in the historical price that has a greater impact on the future price. The short-term price prediction value is merged with the medium- and long-term price trend prediction value to obtain the final price prediction value. For example, assuming that the long short-term memory network predicts that the price on January 8 is 975 yuan, the final predicted price can be the weighted average of the short-term predicted price and the long-term predicted price, for example (969+975) / 2=972 yuan.

[0116] Then, the standard deviation of the historical price return is calculated to get the volatility. The risk value is calculated based on the volatility, expected price and confidence level. For example, assuming that the standard deviation of the historical price return is 0.05, the expected price is 972 yuan, and the confidence level is 95%, the risk value can be calculated based on the normal distribution.

[0117] Next, the risk value, volatility, time interval to departure date, and historical highest price are input into the utility function. The design of the utility function needs to consider the user's risk preference and price sensitivity. For example, a simple utility function can be: Utility = - Risk Value - Volatility * Risk Aversion Coefficient + (Historical Highest Price - Expected Price) * Price Sensitivity Coefficient - Time Interval to Departure Date * Time Discount Coefficient.

[0118] Under the condition of satisfying the minimum advance booking time constraint, a dynamic decision rule is constructed based on the utility function to determine the optimal booking time by maximizing the utility function value. The dynamic decision rule also takes the budget constraint into consideration. For example, the utility value is calculated every day, and when the utility value reaches the maximum and satisfies the budget constraint and the minimum advance booking time constraint, it is determined as the optimal booking time.

[0119] Finally, evaluate the performance of the prediction model. Calculate the average absolute percentage error between the final price prediction and the actual price to evaluate the prediction accuracy. Calculate the savings ratio of the price corresponding to the optimal booking time relative to the average booking price to evaluate the profit level. Construct a loss function based on the average absolute percentage error and the savings ratio, and adaptively update the parameters of the prediction model through the gradient descent method to optimize the prediction model.

[0120] Beneficial effects:

[0121] 1. Improve prediction accuracy: This method can more accurately predict future air ticket and hotel prices through multi-scale time series prediction models and feature engineering.

[0122] 2. Optimize booking strategies: Through dynamic decision rules and utility functions, it can help users determine the optimal booking time, thereby saving travel expenses.

[0123] 3. Adaptive learning and optimization: Adaptive updating of model parameters through gradient descent method can continuously improve the performance and profit level of the prediction model.

[0124] In an optional implementation, under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the utility function value includes:

[0125] Acquire historical price data, extract the historical highest price from the historical price data, and calculate the time interval between the current time and the departure date based on the historical price data;

[0126] Constructing a utility function, wherein the utility function includes a price benefit term, a volatility risk term, and a time interval term, wherein the price benefit term is the difference between the historical highest price and the predicted price multiplied by a first weight coefficient, the volatility risk term is the price volatility multiplied by a second weight coefficient, and the time interval term is the time interval from the departure date multiplied by a third weight coefficient;

[0127] Setting a time window constraint, wherein the time window constraint includes the current time and the latest booking time, wherein the latest booking time is the departure time minus the minimum advance booking time, and the minimum advance booking time is determined based on historical ticket booking data;

[0128] Constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the output value of the utility function under the premise of satisfying the time window constraint and the budget constraint, wherein the budget constraint is determined based on the corporate travel standard;

[0129] The weight coefficient of the utility function is dynamically adjusted by using a sliding time window method, and the weight coefficient combination with the highest benefit is selected as the optimal weight coefficient by calculating the booking benefits under different weight coefficient combinations;

[0130] Substitute the optimal weight coefficient into the utility function, recalculate the optimal booking time based on the time window constraint and the budget constraint, and output the final travel booking time recommendation.

[0131] Get historical price data and preprocess it:

[0132] First, obtain historical price data for a specified route and a period of time before and after a specified date from the air ticket booking platform or other data sources. This data may contain information such as date, time, price, and cabin class.

[0133] For example, we obtained economy class airfare price data from Beijing to Shanghai between March 1, 2024 and March 31, 2024, recording prices at multiple times every day.

[0134] Next, the acquired historical price data is preprocessed, such as removing outliers and processing missing values. Missing values ​​can be processed using methods such as mean value filling and linear interpolation. Outliers can be identified and removed by setting thresholds or using statistical methods, such as treating data outside the range of three standard deviations as outliers.

[0135] Extract the historical highest price and calculate the time interval:

[0136] Extract the highest historical price from the preprocessed historical price data. For example, in the above example, we found that the highest price of economy class from Beijing to Shanghai between March 1, 2024 and March 31, 2024 is 1,200 yuan.

[0137] Calculate the time interval between the current time and the departure date. For example, if the current date is February 1, 2024, and the departure date is March 15, 2024, the time interval is 44 days.

[0138] Construct the utility function:

[0139] Construct a utility function that is used to evaluate the value of different booking times. The utility function contains price return terms, volatility risk terms, and time interval terms.

[0140] Price benefit item: The difference between the highest price in history and the predicted price multiplied by the first weight coefficient. The predicted price can be predicted by time series analysis and other methods. For example, if the ticket price on March 15, 2024 is predicted to be 1,000 yuan and the highest price in history is 1,200 yuan, then the price benefit item is (1,200-1,000)*the first weight coefficient.

[0141] Volatility risk item: price volatility multiplied by the second weight coefficient. Price volatility can be obtained by calculating the standard deviation of historical price data or other volatility indicators. For example, if the standard deviation of the Beijing to Shanghai airfare from March 1 to 31, 2024 is calculated to be 100 yuan, then the volatility risk item is 100*the second weight coefficient.

[0142] Time interval item: the time interval to the departure date multiplied by the third weight coefficient. For example, if there are 44 days to the departure date, the time interval item is 44*the third weight coefficient.

[0143] Set time window constraints and budget constraints:

[0144] Set time window constraints, including the current time and the latest booking time. The latest booking time is the departure time minus the minimum advance booking time. The minimum advance booking time can be determined based on historical booking data. For example, if statistical analysis shows that 90% of users will book tickets 7 days before departure, the minimum advance booking time is set to 7 days.

[0145] Set budget constraints, for example, the company's travel standards stipulate that the price of economy class tickets cannot exceed 1,100 yuan.

[0146] Determine the best time to book:

[0147] Under the premise of satisfying the time window constraint and the budget constraint, the optimal booking time is determined by maximizing the output value of the utility function. For example, by traversing each day in the time window, calculating the utility function value of each day, and selecting the day with the largest utility function value as the optimal booking time.

[0148] Dynamically adjust weight coefficients:

[0149] The sliding time window method is used to dynamically adjust the weight coefficient of the utility function. For example, taking 7 days as a time window, the booking revenue under different weight coefficient combinations in the past 7 days is calculated, and the weight coefficient combination with the highest revenue is selected as the optimal weight coefficient.

[0150] Recalculate the best booking time:

[0151] Substitute the optimal weight coefficient into the utility function, recalculate the optimal booking time based on the time window constraints and budget constraints, and output the final travel booking time recommendation.

[0152] Beneficial effects:

[0153] 1. Improve economic benefits: Through intelligent air ticket booking strategies, you can effectively reduce air ticket costs while meeting travel needs, thereby improving the economic benefits of the company.

[0154] 2. Optimize travel management: Automated booking processes can reduce manual intervention, improve travel management efficiency, save labor costs, and reduce the risk of human errors.

[0155] 3. Improve user experience: Provide users with a more convenient and intelligent travel booking experience and improve user satisfaction.

[0156] In an optional implementation, receiving feedback information of the trip confirmation request, dynamically optimizing the initial business trip itinerary plan according to the feedback information, and outputting a final business trip itinerary plan with the lowest total cost includes:

[0157] The feedback information is divided into time preference feedback information, comfort feedback information and price sensitivity feedback information; the feedback information is quantified, and a feedback quantization value is obtained based on the weight coefficient and score of each feedback dimension;

[0158] The time window constraint is adjusted according to the feedback quantization value of the time preference feedback information, and the start and end time of the original time window is dynamically adjusted; the comfort threshold is updated based on the feedback quantization value of the comfort feedback information, and the product of the original comfort threshold and the comfort adjustment coefficient is used as the updated comfort threshold; the budget constraint is adjusted according to the feedback quantization value of the price sensitivity feedback information, and the product of the original budget and the budget adjustment coefficient is used as the updated budget constraint;

[0159] Constructing a multi-objective optimization model, wherein the objective function of the multi-objective optimization model includes a total cost function term, a time loss function term, and a comfort function term, and introducing corresponding weight coefficients; dynamically adjusting the weight coefficients based on the feedback quantization value, and calculating the updated weight coefficients through the weight adjustment rate;

[0160] The incremental optimization strategy is used to optimize the difference travel itinerary plan. The solution with the smallest increment of the objective function is searched in the neighborhood solution space of the current plan as the local adjustment plan. The simulated annealing algorithm is combined for global optimization, and the acceptance probability of the plan is calculated according to the increment of the objective function.

[0161] Construct a comprehensive scoring function, and perform a weighted summation of the product of the evaluation function value of each evaluation dimension and the corresponding feedback quantization value to obtain the solution score; when the difference between the solution scores of two adjacent iterations is less than the preset convergence threshold, select the solution with the higher score as the final optimization solution;

[0162] Generate a travel timeline, cost composition analysis and comfort evaluation report based on the final optimization plan; perform a sensitivity analysis on the final optimization plan to evaluate the impact of price fluctuations, time window changes and comfort demand changes on the plan, and output a business trip itinerary plan with the best comprehensive score.

[0163] Receive the user's travel confirmation request and corresponding feedback information, and based on this, dynamically optimize the initial travel plan and finally output the plan with the lowest total cost. The plan includes the itinerary timeline, cost composition analysis, and comfort evaluation report.

[0164] First, the collected user feedback information is classified into three dimensions: time preference, comfort, and price sensitivity. Then, the feedback information of each dimension is quantified. For example, time preference is divided into five levels: "very flexible", "relatively flexible", "average", "not very flexible", and "very inflexible", and each is assigned a score of 1 to 5. The other two dimensions are also quantified in a similar way. Next, according to the pre-set weight coefficient, the quantitative value of the feedback of each dimension is calculated, and finally a comprehensive feedback quantitative value is obtained. For example, assuming that the weight coefficients of time preference, comfort, and price sensitivity are 0.4, 0.3, and 0.3 respectively, and the user's scores for the three dimensions are 4, 3, and 2 respectively, then the comprehensive feedback quantitative value is 0.4*4+0.3*3+0.3*2=3.1.

[0165] According to the quantified feedback information, the constraints of the initial travel plan are dynamically adjusted. For example, if the user's time preference feedback quantification value is high, indicating that the time is more flexible, the original time window can be expanded; otherwise, the time window can be reduced. Assuming that the original departure time is from 8 to 10 a.m., and the user feedback is very flexible, the departure time can be adjusted to 7 to 12 a.m. The comfort and budget constraints are also adjusted similarly. Assuming that the initial comfort threshold is a 3-star hotel, and the user has a high comfort requirement, the comfort threshold can be increased to a 4-star hotel; if the user is more sensitive to price, the budget can be reduced. Assuming that the original budget is 1,000 yuan, and the user feedback is highly price sensitive, the budget can be adjusted to 800 yuan.

[0166] Next, a multi-objective optimization model is constructed. The objective function of this model includes three aspects: total cost, time loss, and comfort. Each objective is assigned a corresponding weight coefficient. These weight coefficients are dynamically adjusted according to the quantitative value of user feedback. For example, if the user is very sensitive to price, the weight coefficient of total cost is increased; if the user has strict time requirements, the weight coefficient of time loss is increased. Assuming that the initial weight coefficients of total cost, time loss, and comfort are 0.5, 0.3, and 0.2 respectively, and the user feedback shows that the price sensitivity is very high, the weight coefficient of total cost can be adjusted to 0.7, and the weight coefficients of time loss and comfort can be adjusted to 0.2 and 0.1 accordingly.

[0167] The incremental optimization strategy is combined with the simulated annealing algorithm to search for the optimal solution. First, the solution with the smallest increment of the objective function is searched in the neighborhood solution space of the current solution as the local adjustment solution. Then, the simulated annealing algorithm is used for global optimization, and the acceptance probability of the solution is calculated according to the increment of the objective function.

[0168] In order to evaluate the quality of the optimization scheme, a comprehensive scoring function is constructed. This function multiplies the evaluation function value of each evaluation dimension with the corresponding feedback quantization value, and then weighted sums them to obtain the scheme score. When the difference in scheme scores between two adjacent iterations is less than the preset convergence threshold, the scheme with the higher score is selected as the final optimization scheme. Assuming that the scheme scores of the two iterations are 8.5 and 8.6 respectively, and the convergence threshold is 0.1, the scheme with a score of 8.6 is selected as the final optimization scheme.

[0169] Finally, based on the final optimization plan, a travel timeline, cost composition analysis and comfort evaluation report are generated, and a sensitivity analysis is performed to evaluate the impact of price fluctuations, time window changes and comfort demand changes on the plan.

[0170] The beneficial effects of this program are reflected in three aspects:

[0171] First, personalized customization: dynamically adjust optimization goals and constraints based on user feedback to achieve true personalized customization and improve user satisfaction.

[0172] Second, cost control: Through multi-objective optimization models and incremental optimization strategies, the total travel cost can be reduced as much as possible while meeting user needs.

[0173] Third, solution stability: Through sensitivity analysis, the impact of various factors on the final solution is evaluated to improve the stability and reliability of the solution.

[0174] Figure 2 FIG. 1 is a schematic diagram of the structure of a business travel itinerary recommendation and cost optimization system based on big data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0175] The first unit is used to receive travel demand information input by a user, wherein the travel demand information is a destination city and a business activity time; based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and simultaneously obtain the real-time traffic information and hotel information of the destination city from a third-party data interface; and input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model;

[0176] The second unit is used to analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently screen the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; determine the accommodation time interval based on the arrival time of the traffic plan and the business activity time; and screen a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards;

[0177] The third unit is used to perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predict the price fluctuation trend in the next seven days, and determine the optimal booking time; generate an initial business trip itinerary plan according to the optimal booking time; push an itinerary confirmation request to the user's business activity-related contacts; receive feedback information of the itinerary confirmation request, dynamically optimize the initial business trip itinerary plan according to the feedback information, and output a final business trip itinerary plan with the lowest total cost.

[0178] According to a third aspect of the embodiments of the present invention,

[0179] An electronic device is provided, comprising:

[0180] processor;

[0181] a memory for storing processor-executable instructions;

[0182] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0183] A fourth aspect of the embodiments of the present invention is:

[0184] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0185] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A business travel itinerary recommendation and cost optimization method based on big data, characterized in that: include: Receive travel demand information input by a user, wherein the travel demand information includes a destination city and a business activity time; Based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and obtain real-time traffic information and hotel information of the destination city from a third-party data interface; input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model; Analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently filter the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; determine the accommodation time range based on the arrival time of the traffic plan and the business activity time; Filtering a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards; Perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list to predict the price fluctuation trend in the next seven days and determine the optimal booking time; Generate an initial travel itinerary plan according to the optimal booking time; Pushing a travel confirmation request to contacts related to the user's business activities; receiving feedback information of the travel confirmation request, dynamically optimizing the initial travel itinerary plan according to the feedback information, and outputting a final travel itinerary plan with the lowest total cost.

2. The method according to claim 1, characterized in that Analyzing the user travel characteristics in the historical travel records by the machine learning model, intelligently screening the real-time traffic information according to the travel characteristics, and generating a traffic plan that meets the corporate travel standards includes: Extracting time features, geographic features, traffic preference features, and price sensitivity features from the historical travel records, mapping the time features, the geographic features, the traffic preference features, and the price sensitivity features to a continuous feature space through a feature embedding vector, and generating a user feature vector; A multi-head self-attention mechanism is used to process the user feature vector, wherein attention weights are calculated through a query matrix, a key matrix, and a value matrix, multiple attention heads are generated based on the attention weights, and the multiple attention heads are concatenated and multiplied with a parameter matrix to obtain a user difference travel behavior representation; A two-layer optimization model is constructed based on the user's business trip as a representation, wherein the upper optimization goal of the two-layer optimization model is to minimize the total travel cost, and the lower constraint conditions include the travel budget constraint, the arrival time window constraint and the comfort constraint; and real-time traffic data is input into the two-layer optimization model; Using a long short-term memory network in combination with an attention mechanism to predict transportation fares, calculating attention weights according to hidden states of the long short-term memory network, and weighting the hidden states based on the attention weights to obtain predicted prices; Calculate the cost-effectiveness scores of different transportation solutions, where the cost-effectiveness scores are obtained by weighted calculation of the cost score, the comfort score, and the time score; sort the transportation solutions according to the cost-effectiveness scores and select the optimal transportation solution.

3. The method according to claim 2, characterized in that A two-layer optimization model is constructed based on the user's business trip as a representation. The upper optimization goal of the two-layer optimization model is to minimize the total travel cost. The lower constraints include travel budget constraints, arrival time window constraints and comfort constraints, including: Constructing an upper optimization objective function of a two-layer optimization model, wherein the upper optimization objective function includes a traffic scheme cost item and a time cost item, wherein the traffic scheme cost item is the sum of the products of the costs of each traffic scheme and the corresponding 0-1 decision variables, and the time cost item is the sum of the products of the time costs of each traffic scheme and the corresponding 0-1 decision variables, and introducing a time cost weight coefficient; Construct lower-level constraints based on the corporate travel budget, which include a budget constraint function, a time window constraint function, and a comfort constraint function. The budget constraint function requires that the sum of the product of the actual expenditure of each plan and the corresponding 0-1 decision variable must be less than the budget upper limit, and the time window constraint function requires that the sum of the product of the arrival time of each plan and the corresponding 0-1 decision variable must be within a preset time window; Construct a multidimensional comfort evaluation function, and multiply the evaluation values ​​of the three dimensions of transportation grade score, transfer number penalty item, and seat comfort by the corresponding weights and sum them up. The transportation grade score is the sum of the products of each transportation scheme grade score and the corresponding 0-1 decision variable, the transfer number penalty item is the exponential function transformation value of the transfer number, and the seat comfort is the sum of the products of each transportation scheme seat comfort score and the corresponding 0-1 decision variable; Introducing slack variables to construct a Lagrangian optimization objective function, the Lagrangian optimization objective function includes an original objective function term and a constraint penalty term, wherein the constraint penalty term is the sum of the product of each constraint violation degree and the corresponding Lagrangian multiplier; updating the Lagrangian multiplier by iteration, the update formula of the Lagrangian multiplier includes the original multiplier value, the step size parameter and the constraint violation degree; The primal-dual iterative algorithm is used to solve the Lagrangian optimization objective function. In each iteration, the optimal value of the primal variable is first solved, and then the dual variable is updated based on the optimal value of the primal variable. The optimal solution is determined when the difference between two iterations of the primal variable is less than a first threshold and the constraint violation degree is less than a second threshold.

4. The method according to claim 1, characterized in that: Perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list to predict the price fluctuation trend in the next seven days and determine the optimal booking time; Generating an initial travel itinerary plan according to the optimal booking time includes: Decomposing the original price sequence into a trend component, a seasonal component and a random fluctuation component; constructing a price feature vector, wherein the price feature vector includes a historical price sequence, date-related features, historical statistical features and external factor features, wherein the historical statistical features include a mean, a variance, a maximum value and a minimum value; A multi-scale time series prediction model is constructed based on the price feature vector, short-term price fluctuations are modeled by multiple basic learners, and a short-term price prediction value is obtained by weighted combination of the basic learner integrated weights; a long short-term memory network is used in combination with an attention mechanism to model medium- and long-term price trends, and the short-term price prediction value is merged with the medium- and long-term price trend prediction value to obtain a final price prediction value; Calculate the standard deviation of historical price returns to obtain volatility, and calculate the risk value based on the volatility, expected price and confidence level; input the risk value, volatility and the time interval from the departure date into the utility function, wherein the input parameters of the utility function also include the historical highest price; Under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, determining the optimal booking time by maximizing the utility function value, and the dynamic decision rule also considers the budget constraint condition; Calculate the average absolute percentage error between the final price forecast value and the actual price to evaluate the forecast accuracy; calculate the saving ratio of the price corresponding to the optimal booking time relative to the average booking price to evaluate the profit level; A loss function is constructed based on the mean absolute percentage error and the savings ratio, and the parameters of the prediction model are adaptively updated by the gradient descent method, wherein the parameters of the prediction model include the basic learner ensemble weights, long short-term memory network parameters, attention mechanism parameters and utility function weight coefficients, and the optimized travel expense prediction model and optimal booking time are output.

5. The method according to claim 4, characterized in that Under the condition of satisfying the minimum advance booking time constraint, constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the utility function value includes: Acquire historical price data, extract the historical highest price from the historical price data, and calculate the time interval between the current time and the departure date based on the historical price data; Constructing a utility function, wherein the utility function includes a price benefit term, a volatility risk term, and a time interval term, wherein the price benefit term is the difference between the historical highest price and the predicted price multiplied by a first weight coefficient, the volatility risk term is the price volatility multiplied by a second weight coefficient, and the time interval term is the time interval from the departure date multiplied by a third weight coefficient; Setting a time window constraint, wherein the time window constraint includes the current time and the latest booking time, wherein the latest booking time is the departure time minus the minimum advance booking time, and the minimum advance booking time is determined based on historical ticket booking data; Constructing a dynamic decision rule based on the utility function, and determining the optimal booking time by maximizing the output value of the utility function under the premise of satisfying the time window constraint and the budget constraint, wherein the budget constraint is determined based on the corporate travel standard; The weight coefficient of the utility function is dynamically adjusted by using a sliding time window method, and the weight coefficient combination with the highest benefit is selected as the optimal weight coefficient by calculating the booking benefits under different weight coefficient combinations; Substitute the optimal weight coefficient into the utility function, recalculate the optimal booking time based on the time window constraint and the budget constraint, and output the final travel booking time recommendation.

6. The method according to claim 1, characterized in that Receiving feedback information of the itinerary confirmation request, dynamically optimizing the initial itinerary plan for the business trip according to the feedback information, and outputting a final itinerary plan with the lowest total cost includes: The feedback information is divided into time preference feedback information, comfort feedback information and price sensitivity feedback information; the feedback information is quantified, and a feedback quantization value is obtained based on the weight coefficient and score of each feedback dimension; The time window constraint is adjusted according to the feedback quantization value of the time preference feedback information, and the start and end time of the original time window is dynamically adjusted; the comfort threshold is updated based on the feedback quantization value of the comfort feedback information, and the product of the original comfort threshold and the comfort adjustment coefficient is used as the updated comfort threshold; the budget constraint is adjusted according to the feedback quantization value of the price sensitivity feedback information, and the product of the original budget and the budget adjustment coefficient is used as the updated budget constraint; Constructing a multi-objective optimization model, wherein the objective function of the multi-objective optimization model includes a total cost function term, a time loss function term, and a comfort function term, and introducing corresponding weight coefficients; dynamically adjusting the weight coefficients based on the feedback quantization value, and calculating the updated weight coefficients through the weight adjustment rate; The incremental optimization strategy is used to optimize the difference travel itinerary plan. The solution with the smallest increment of the objective function is searched in the neighborhood solution space of the current plan as the local adjustment plan. The simulated annealing algorithm is combined for global optimization, and the acceptance probability of the plan is calculated according to the increment of the objective function. Construct a comprehensive scoring function, and perform a weighted summation of the product of the evaluation function value of each evaluation dimension and the corresponding feedback quantization value to obtain the solution score; when the difference between the solution scores of two adjacent iterations is less than the preset convergence threshold, select the solution with the higher score as the final optimization solution; Generate a travel timeline, cost composition analysis and comfort evaluation report based on the final optimization plan; perform a sensitivity analysis on the final optimization plan to evaluate the impact of price fluctuations, time window changes and comfort demand changes on the plan, and output a business trip itinerary plan with the best comprehensive score.

7. A business travel itinerary recommendation and cost optimization system based on big data, used to implement the method as claimed in any one of claims 1 to 6, characterized in that: include: The first unit is used to receive travel demand information input by a user, wherein the travel demand information is a destination city and a business activity time; Based on the travel demand information, extract the user's historical travel records and corporate travel standards from the corporate travel management system, and obtain real-time traffic information and hotel information of the destination city from a third-party data interface; input the historical travel records, corporate travel standards, real-time traffic information and hotel information into a preset machine learning model; The second unit is used to analyze the user travel characteristics in the historical travel records through the machine learning model, intelligently filter the real-time traffic information according to the travel characteristics, and generate a traffic plan that meets the corporate travel standards; based on the arrival time of the traffic plan and the business activity time, determine the accommodation time interval; Filtering a list of candidate hotels from the hotel information according to the accommodation time interval and the corporate travel standards; The third unit is used to perform time series analysis on the air ticket prices in the transportation plan and the room rates in the candidate hotel list, predict the price fluctuation trend in the next seven days, and determine the optimal booking time; Generate an initial travel itinerary plan according to the optimal booking time; Pushing a travel confirmation request to contacts related to the user's business activities; receiving feedback information of the travel confirmation request, dynamically optimizing the initial travel itinerary plan according to the feedback information, and outputting a final travel itinerary plan with the lowest total cost.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images