Intelligent reservation and multi-party real-time confirmation cooperation method and system in business travel scene

Abstract

The invention provides an intelligent reservation and multi-party real-time confirmation cooperation method and system in a business travel scene, and relates to the technical field of intellectualization, and the method comprises the steps: receiving a business travel reservation request, extracting travel time, personnel, budget and business travel policy information, and generating a reservation task; a deep learning model is adopted to analyze reservation task information, and traffic, accommodation and conference site resources are matched to generate an initial resource matching scheme; receiving schedule feedback returned by the resource provider, calculating a time sequence correlation degree, a geographic position correlation degree and a cost matching degree among the resources by using a graph neural network model, and generating an optimal resource combination scheme; and broadcasting the optimal scheme to a resource provider node through a block chain network to execute an intelligent contract to perform resource locking, generating reservation confirmation information and electronic certificate information, and writing the reservation confirmation information and the electronic certificate information into a block chain distributed account book. According to the invention, intelligent matching and real-time confirmation of business travel resources are realized, the booking efficiency is improved, and the transaction reliability is ensured.

Description

Technical Field

[0001] The present invention relates to intelligent technology, and in particular to an intelligent reservation and multi-party real-time confirmation collaboration method and system in a business travel scenario. Background Art

[0002] Business travel booking is a complex process that involves booking and coordinating multiple resources such as air tickets, hotels, trains, and conference venues. Traditional business travel booking methods mainly rely on manual operations, such as communicating and confirming with various resource providers through telephone, email, or online platforms. This method is inefficient and prone to problems such as information asymmetry, high communication costs, and cumbersome booking processes. In recent years, with the rapid development of the Internet and mobile Internet technologies, some online business travel booking platforms have emerged that can provide more convenient booking services. However, these platforms still have some limitations.

[0003] Low resource matching efficiency: Traditional business travel booking platforms mostly use keyword searches or simple screening methods to match resources. They are unable to make intelligent recommendations based on users’ personalized needs and travel policies, resulting in low resource matching efficiency. Users need to spend a lot of time screening and comparing.

[0004] Lack of real-time collaboration: Existing booking platforms usually require communication and confirmation with each resource provider separately. The lack of real-time collaboration mechanism makes it easy for resource conflicts or information lags to occur, affecting booking efficiency and user experience.

[0005] Opaque booking process: Traditional booking methods lack transparency, making it difficult for users to understand the real-time price and availability of resources. This can easily lead to price opacity and information asymmetry, which can affect user trust and satisfaction. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for intelligent reservation and multi-party real-time confirmation collaboration in a travel scenario, which can solve the problems in the prior art.

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

[0008] Provide intelligent booking and multi-party real-time confirmation collaboration methods in travel scenarios, including:

[0009] Receive a business travel reservation request and extract travel time information, personnel information, budget information and travel policy information to generate a reservation task; use a deep learning model to intelligently analyze various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan and send the initial resource matching plan to the corresponding resource provider;

[0010] Receive resource schedule feedback returned by each resource provider for the initial resource matching plan, the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model, input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output an optimal resource combination plan that meets each correlation, wherein the optimal resource combination plan associates the specific schedules of each resource in time sequence;

[0011] The optimal resource combination plan is broadcast to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; based on the resource locking status, the reservation confirmation information and the electronic certificate information are generated, and the reservation confirmation information and the electronic certificate information are written into the blockchain distributed ledger; the status of the reservation confirmation information is monitored, and when the status of the reservation confirmation information changes, the corresponding resource unlocking or re-locking operation is triggered through the smart contract.

[0012] Using a deep learning model to intelligently analyze various information in the reservation task, matching transportation resources based on the travel time information and the travel policy information, matching accommodation resources based on the personnel information and the budget information, matching conference venue resources based on the travel time information and the personnel information, generating an initial resource matching plan and sending the initial resource matching plan to the corresponding resource provider includes:

[0013] Convert the travel time information into a timestamp sequence, convert the personnel information into a multidimensional feature vector including rank characteristics, department characteristics, and historical booking preference characteristics, normalize the budget information into a budget vector with a uniform numerical range, convert the travel policy information into a rule vector through natural language processing, and construct a deep learning model to extract features from the timestamp sequence, the multidimensional feature vector, the budget vector, and the rule vector;

[0014] Input the timestamp sequence and the rule vector into the first bidirectional long short-term memory network of the deep learning model, the first bidirectional long short-term memory network encodes the timestamp sequence through the time series encoding layer, calculates the weights of different rules in the rule vector through the policy rule attention layer, matches the transportation resources that meet the travel policy, and generates a transportation resource matching plan including transportation type, departure and arrival time, and seat grade; input the multidimensional feature vector and the budget vector into the second bidirectional long short-term memory network of the deep learning model, the second bidirectional long short-term memory network determines the accommodation standard based on the job grade characteristics in the multidimensional feature vector, screens the accommodation resources that meet the budget based on the budget vector, and generates an accommodation resource matching plan; input the timestamp sequence and the multidimensional feature vector into the third bidirectional long short-term memory network of the deep learning model, the third bidirectional long short-term memory network determines the meeting time interval based on the timestamp sequence, predicts the number of participants based on the multidimensional feature vector and matches the meeting venue resources, and generates a meeting venue matching plan;

[0015] The transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme are input into a multi-task learning framework, and the multi-task learning framework calculates the association weights of different resource types through an attention mechanism, and establishes a spatiotemporal constraint relationship among the transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme; the spatiotemporal constraint relationship is input into a global optimization layer, and the global optimization layer generates an initial resource matching scheme by minimizing an overall cost function; and the initial resource matching scheme is sent to corresponding transportation resource providers, accommodation resource providers, and conference venue resource providers for resource confirmation.

[0016] Receiving resource schedule feedback returned by each resource provider for the initial resource matching solution, wherein the resource schedule feedback identifies the available time period and real-time price of each resource, and constructing a graph neural network model to input the resource schedule feedback into the graph neural network model includes:

[0017] Receive resource schedule feedback returned by each resource provider for the initial resource matching solution, extract time period information identifying the available time period of each resource and price information identifying the real-time price of each resource from the resource schedule feedback; convert the time period information into a binary sequence representing the available state of the resource, combine the price information with historical price fluctuation data to generate a price trend sequence, and combine the binary sequence with the price trend sequence to form a resource feature vector;

[0018] Construct a graph neural network model, input the resource feature vector into the graph neural network model, the graph neural network model constructs different types of resources as nodes of a heterogeneous resource graph, constructs the spatiotemporal association relationship between resources as the edge of the heterogeneous resource graph, and uses an attention mechanism to calculate the association strength between different resource nodes; input the resource feature vector into the gated recurrent unit of the graph neural network model, the gated recurrent unit performs time series encoding on the resource available time period, and calculates the importance of different time slots through the time attention layer;

[0019] The encoded temporal features are input into the graph structure aggregation layer of the graph neural network model. The graph structure aggregation layer extracts the interaction pattern between nodes from multiple feature subspaces based on the multi-head attention mechanism, and transmits the node features on the edge of the heterogeneous resource graph through the message passing algorithm; the node-level features are aggregated into a graph-level representation by using the graph pooling operation, and the original feature information is retained by the residual connection;

[0020] A time-sensitive dynamic update module is constructed based on graph-level representation. The time-sensitive dynamic update module continuously receives new resource schedule feedback through a sliding time window, calculates the impact of real-time price changes on resource nodes based on a price propagation network, identifies new resource associations through an edge prediction model, and updates the topological structure of the heterogeneous resource graph; the updated heterogeneous resource graph is input into a temporal graph convolutional network.

[0021] Based on the graph neural network model, the temporal order correlation, geographical location correlation and cost matching between various resources are calculated, and the optimal resource combination scheme that meets the various correlations is output. The optimal resource combination scheme associates the specific schedules of various resources in chronological order, including:

[0022] Receive the resource node feature vector output by the graph neural network model, extract the time slot information, geographical location information, and price information of each resource node according to the resource node feature vector; input the time slot information into a dynamic time planner, and the dynamic time planner analyzes the schedule overlap, connection interval, and timing dependency between resource nodes to generate a timing dependency matrix;

[0023] The temporal dependency matrix and the geographic location information are input into a spatial attention network, the spatial attention network calculates spatiotemporal accessibility based on the geographic distance between the resource nodes, uses a distance decay function to evaluate the degree of traffic convenience, and outputs a geographic location association matrix; the geographic location association matrix and the price information are input into a cost matching calculation module, the cost matching calculation module converts the price range into a standardized score based on a piecewise linear mapping function, and generates a cost matching matrix in combination with a budget constraint;

[0024] The time dependency matrix, the geographic location correlation matrix, and the cost matching matrix are input into a multidimensional feature fusion network, which extracts multidimensional correlation features through tensor decomposition, uses an adaptive weight mechanism to balance the importance of different dimensions, and outputs a comprehensive correlation score matrix between resource nodes; the comprehensive correlation score matrix is ​​input into a resource combination optimization module, which searches for the maximum weight match based on a hierarchical Hungarian algorithm and generates a candidate resource combination plan;

[0025] The candidate resource combination scheme is input into the constraint verification module, and the constraint verification module uses a graph coloring algorithm to detect and eliminate resource schedule conflicts, determines the optimal path that meets the timing constraints through a dynamic programming algorithm, and outputs the optimal resource combination scheme; a reinforcement learning environment is constructed based on the execution feedback of the optimal resource combination scheme, and the resource combination strategy is alternately optimized and updated through the policy network and the value network. The Monte Carlo tree search is used to expand the search space, and the final resource combination scheme after online optimization is output, and the final resource combination scheme includes the specific schedule arrangements for each resource.

[0026] The optimal resource combination plan is broadcasted to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status including:

[0027] The optimal resource combination plan is converted into a data structure, and the resource schedule information, price information, and reservation conditions are packaged into a transaction data packet using a Merkle tree structure; the transaction data packet is broadcast to each resource provider node through the sharding mechanism of the blockchain network, and the resource provider node verifies the validity of the transaction data packet based on a threshold signature scheme;

[0028] The resource provider node receives the verified transaction data packet and triggers the smart contract to execute the resource locking operation. The smart contract first performs a resource status pre-check to verify the current resource status, price validity, and schedule availability, and generates a resource status check report; based on the resource status check report, the cross-chain atomic exchange protocol is executed to encapsulate the locking requests of multiple resources into one atomic transaction, and a hash time lock contract is used to set a unified lock time window;

[0029] During the execution of atomic transactions by the smart contract, concurrent lock requests are processed through an optimistic concurrency control protocol based on timestamps, and a processing order is assigned to lock requests of different priorities; a deadlock detection algorithm is used to monitor the resource locking status, and a rollback compensation mechanism is triggered when a deadlock is found to automatically release the locked resources;

[0030] The execution result of the atomic transaction is input into the state synchronization module, which uses a vector clock to track the timing relationship of resource state changes and propagates state update information in the blockchain network through the gossip protocol; the state synchronization module stores the resource lock state in multiple nodes in a sharding manner and ensures the security of the state data through a threshold secret sharing scheme;

[0031] A node monitoring network is built based on the heartbeat mechanism to detect the operating status and network connectivity of the resource provider nodes in real time. When a node failure is detected, service recovery is performed through the state machine copy technology. The node monitoring network submits the node status information to the blockchain network, and adopts the Byzantine fault-tolerant consensus mechanism to maintain the stable operation of the network, and finally outputs the resource locking status and returns it to the requester.

[0032] Generating reservation confirmation information and electronic credential information based on the resource locking state, and writing the reservation confirmation information and the electronic credential information into the blockchain distributed ledger; monitoring the state of the reservation confirmation information, and when the state of the reservation confirmation information changes, triggering corresponding resource unlocking or re-locking operations through the smart contract includes:

[0033] Construct reservation confirmation information including reservation identification, resource locking status hash, timestamp, reservation party information, resource provider information, and transaction amount according to the resource locking status, encrypt sensitive data in the reservation confirmation information using homomorphic encryption technology, and generate encrypted reservation confirmation information; based on the encrypted reservation confirmation information, call a smart contract to generate electronic credential information, wherein the electronic credential information includes an anti-counterfeiting verification field generated based on zero-knowledge proof;

[0034] The encrypted reservation confirmation information and the electronic credential information are stored in different nodes of the blockchain network by using sharding technology, and the reservation confirmation information and the electronic credential information are multi-signed by a threshold signature scheme, and the signed information is written into a blockchain distributed ledger; the blockchain distributed ledger uses a hierarchical Merkle tree structure to establish a data index to achieve fast query of status information;

[0035] An event-driven state listener is constructed to monitor the state of the reservation confirmation information in real time. When the state listener captures a state change event, the state change is verified for legitimacy based on the rule engine, and a trigger request containing state transition conditions is generated; the state listener asynchronously transmits the trigger request to the smart contract through a distributed message queue;

[0036] The smart contract that receives the trigger request parses the state transition type based on the state machine model. When the state transition type is to cancel the reservation, it triggers the resource unlocking operation to restore the resource state to an available state; when the state transition type is to change the reservation, it triggers the resource re-locking operation and adopts the timestamp-based optimistic concurrency control protocol to handle resource conflicts.

[0037] According to a second aspect of the embodiments of the present invention,

[0038] Provide intelligent reservation and multi-party real-time confirmation collaboration system in travel scenarios, including:

[0039] The first unit is used to receive a business travel reservation request and extract travel time information, personnel information, budget information and travel policy information to generate a reservation task; use a deep learning model to intelligently analyze various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan and send the initial resource matching plan to the corresponding resource provider;

[0040] The second unit is used to receive resource schedule feedback returned by each resource provider for the initial resource matching plan, the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model to input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output an optimal resource combination plan that meets each correlation, wherein the optimal resource combination plan associates the specific schedules of each resource in time sequence;

[0041] The third unit is used to broadcast the optimal resource combination plan to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; generates reservation confirmation information and electronic certificate information based on the resource locking status, and writes the reservation confirmation information and the electronic certificate information into the blockchain distributed ledger; monitors the status of the reservation confirmation information, and when the status of the reservation confirmation information changes, triggers the corresponding resource unlocking or re-locking operation through the smart contract.

[0042] According to a third aspect of the embodiments of the present invention,

[0043] An electronic device is provided, comprising:

[0044] processor;

[0045] a memory for storing processor-executable instructions;

[0046] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0047] A fourth aspect of the embodiments of the present invention is:

[0048] 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.

[0049] The beneficial effects of this application are as follows:

[0050] 1. Improve booking efficiency: Through the deep learning model, intelligent analysis of travel information, travel policies, budgets, etc., automatically match transportation, accommodation and conference venue resources, quickly generate initial plans, and reduce manual intervention and time costs. Subsequently, the graph neural network model is used to quickly calculate the correlation between resources based on the feedback from the resource party, and output the optimal combination plan, further improving the efficiency of resource matching and plan adjustment.

[0051] 2. Optimize resource allocation: Based on the graph neural network model, the time, geographical location and cost matching between resources are considered to ensure reasonable itinerary arrangements, optimal resource utilization, avoid resource conflicts and waste, and dynamically adjust according to real-time price information to achieve cost control.

[0052] 3. Ensure safe and reliable transactions: Use blockchain technology to broadcast solutions, execute smart contracts, lock resources and record reservation information, ensuring the transparency and security of transactions and preventing information tampering and refusal to perform. Status changes automatically trigger smart contracts to adjust resources, improving the automation and reliability of transaction execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of a collaborative method for intelligent reservation and multi-party real-time confirmation in a business travel scenario according to an embodiment of the present invention;

[0054] Figure 2 It is a structural diagram of the intelligent reservation and multi-party real-time confirmation collaborative system in the business travel scenario according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] 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.

[0056] 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.

[0057] Figure 1 FIG. 1 is a flow chart of a method for coordinating intelligent reservation and multi-party real-time confirmation in a business travel scenario according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] S11. Receive a business travel reservation request and extract travel time information, personnel information, budget information and travel policy information to generate a reservation task; use a deep learning model to intelligently analyze the various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan and send the initial resource matching plan to the corresponding resource provider;

[0059] S12. Receive resource schedule feedback returned by each resource provider for the initial resource matching solution, wherein the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model, input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output the optimal resource combination solution that meets each correlation, wherein the optimal resource combination solution associates the specific schedules of each resource in time sequence;

[0060] S13. Broadcast the optimal resource combination plan to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; generates reservation confirmation information and electronic certificate information based on the resource locking status, and writes the reservation confirmation information and the electronic certificate information into the blockchain distributed ledger; monitors the status of the reservation confirmation information, and when the status of the reservation confirmation information changes, triggers the corresponding resource unlocking or re-locking operation through the smart contract.

[0061] In an optional implementation, a deep learning model is used to intelligently analyze various information in the reservation task, and transportation resources are matched based on the travel time information and the travel policy information, accommodation resources are matched based on the personnel information and the budget information, and conference venue resources are matched based on the travel time information and the personnel information. An initial resource matching plan is generated and sent to the corresponding resource provider, including:

[0062] Convert the travel time information into a timestamp sequence, convert the personnel information into a multidimensional feature vector including rank characteristics, department characteristics, and historical booking preference characteristics, normalize the budget information into a budget vector with a uniform numerical range, convert the travel policy information into a rule vector through natural language processing, and construct a deep learning model to extract features from the timestamp sequence, the multidimensional feature vector, the budget vector, and the rule vector;

[0063] Input the timestamp sequence and the rule vector into the first bidirectional long short-term memory network of the deep learning model, the first bidirectional long short-term memory network encodes the timestamp sequence through the time series encoding layer, calculates the weights of different rules in the rule vector through the policy rule attention layer, matches the transportation resources that meet the travel policy, and generates a transportation resource matching plan including transportation type, departure and arrival time, and seat grade; input the multidimensional feature vector and the budget vector into the second bidirectional long short-term memory network of the deep learning model, the second bidirectional long short-term memory network determines the accommodation standard based on the job grade characteristics in the multidimensional feature vector, screens the accommodation resources that meet the budget based on the budget vector, and generates an accommodation resource matching plan; input the timestamp sequence and the multidimensional feature vector into the third bidirectional long short-term memory network of the deep learning model, the third bidirectional long short-term memory network determines the meeting time interval based on the timestamp sequence, predicts the number of participants based on the multidimensional feature vector and matches the meeting venue resources, and generates a meeting venue matching plan;

[0064] The transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme are input into a multi-task learning framework, and the multi-task learning framework calculates the association weights of different resource types through an attention mechanism, and establishes a spatiotemporal constraint relationship among the transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme; the spatiotemporal constraint relationship is input into a global optimization layer, and the global optimization layer generates an initial resource matching scheme by minimizing an overall cost function; and the initial resource matching scheme is sent to corresponding transportation resource providers, accommodation resource providers, and conference venue resource providers for resource confirmation.

[0065] A deep learning-driven intelligent matching method and system for travel booking resources, which is used to intelligently match transportation, accommodation and conference resources according to travel needs.

[0066] First, the information in the booking task is preprocessed. Travel time information, such as "departure at 8:00 on March 1, 2024, return at 12:00 on March 3, 2024", will be converted into a series of timestamps, indicating key time points such as departure time and return time. Personnel information, such as "Zhang San, senior manager, marketing department", will be converted into a multidimensional feature vector. This vector contains job level features (for example, different levels are represented by numerical values, senior manager is "3"), department features (for example, different departments are represented by one-hot encoding, "marketing department" is "[0,0,1,0]"), and historical booking preference features (for example, "preferred hotel star: five-star", "preferred airline: China Southern Airlines"). Budget information, such as "transportation budget: 2,000 yuan, accommodation budget: 3,000 yuan", will be normalized into a budget vector with a uniform numerical range, such as "[0.2,0.3]", for easy model processing. Travel policy information, such as "economy class is required when flying, and accommodation standards do not exceed four stars", will be converted into rule vectors through natural language processing technology. For example, the flight class rule can be expressed as "[1,0,0]" (representing economy class, business class, and first class, respectively), and the accommodation star rule can be expressed as "[1,1,1,0,0]" (representing one star to five stars, respectively).

[0067] Next, a deep learning model is built for feature extraction and resource matching. The model contains three bidirectional long short-term memory networks (BiLSTMs). The first BiLSTM receives a timestamp sequence and a rule vector as input. The timestamp sequence is encoded by the time series encoding layer, and the rule vector calculates the weights of different rules through the policy rule attention layer. Then, the model matches qualified transportation resources according to the time and travel policy, and generates a transportation resource matching solution including transportation type, departure and arrival time, and seat class. For example, according to the time and "economy class" rule, it matches "economy class air tickets from A to B at 8:00 on March 1, 2024". The second BiLSTM receives a multidimensional feature vector and a budget vector as input. The model determines the accommodation standard based on the job level characteristics, for example, a senior manager corresponds to a four-star hotel. Then, according to the budget vector, the accommodation resources that meet the budget are screened and an accommodation resource matching solution is generated. For example, a four-star hotel with a price of less than 3,000 yuan in B is selected. The third BiLSTM receives a timestamp sequence and a multidimensional feature vector as input. The model determines the meeting time interval based on the timestamp sequence, for example, from the afternoon of March 1, 2024 to the morning of March 3. The model predicts the number of participants based on the multidimensional feature vector, for example, the number of participants is predicted to be 10 based on the department and rank. Then, it matches the meeting venue resources that meet the time and number requirements to generate a meeting venue matching solution. For example, book a meeting room at location B that can accommodate 10 people for the entire day of March 2, 2024.

[0068] Subsequently, the generated transportation, accommodation, and conference resource matching solutions are input into the multi-task learning framework. The framework uses the attention mechanism to calculate the association weights of different resource types and establish spatiotemporal constraints between them. For example, the accommodation location needs to be close to the conference location, and the transportation vehicle arrival time needs to be before the conference start time. These spatiotemporal constraints are input into the global optimization layer. The global optimization layer generates the initial resource matching solution by minimizing the overall cost function, such as total cost and time cost.

[0069] Finally, the initial resource matching plan is sent to the corresponding resource providers, such as airlines, hotels, and conference centers, for resource confirmation.

[0070] The solution of this application can:

[0071] Improve efficiency: Automate the resource matching process, reduce manual operations, and improve booking efficiency. Reduce costs: Reduce the overall cost of travel, including time and economic costs, through global optimization. Personalized service: Provide personalized resource matching solutions based on personnel information and historical preferences to improve user satisfaction.

[0072] In an optional implementation, receiving resource schedule feedback returned by each resource provider for the initial resource matching solution, the resource schedule feedback identifying the available time period and real-time price of each resource, and constructing a graph neural network model to input the resource schedule feedback into the graph neural network model includes:

[0073] Receive resource schedule feedback returned by each resource provider for the initial resource matching solution, extract time period information identifying the available time period of each resource and price information identifying the real-time price of each resource from the resource schedule feedback; convert the time period information into a binary sequence representing the available state of the resource, combine the price information with historical price fluctuation data to generate a price trend sequence, and combine the binary sequence with the price trend sequence to form a resource feature vector;

[0074] Construct a graph neural network model, input the resource feature vector into the graph neural network model, the graph neural network model constructs different types of resources as nodes of a heterogeneous resource graph, constructs the spatiotemporal association relationship between resources as the edge of the heterogeneous resource graph, and uses an attention mechanism to calculate the association strength between different resource nodes; input the resource feature vector into the gated recurrent unit of the graph neural network model, the gated recurrent unit performs time series encoding on the resource available time period, and calculates the importance of different time slots through the time attention layer;

[0075] The encoded temporal features are input into the graph structure aggregation layer of the graph neural network model. The graph structure aggregation layer extracts the interaction pattern between nodes from multiple feature subspaces based on the multi-head attention mechanism, and transmits the node features on the edge of the heterogeneous resource graph through the message passing algorithm; the node-level features are aggregated into a graph-level representation by using the graph pooling operation, and the original feature information is retained by the residual connection;

[0076] A time-sensitive dynamic update module is constructed based on graph-level representation. The time-sensitive dynamic update module continuously receives new resource schedule feedback through a sliding time window, calculates the impact of real-time price changes on resource nodes based on a price propagation network, identifies new resource associations through an edge prediction model, and updates the topological structure of the heterogeneous resource graph; the updated heterogeneous resource graph is input into a temporal graph convolutional network.

[0077] The resource scheduling optimization method aims to efficiently allocate and schedule various resources to meet the needs of complex projects. The core of this method is to use the graph neural network model to dynamically optimize resource allocation by combining resource availability, price fluctuations, and the temporal and spatial correlation between resources.

[0078] First, feedback information is received from various resource providers, which describes in detail the available time period and real-time price of each resource. From these feedback information, key information is extracted: the available time period and real-time price of the resource.

[0079] Next, the available time period information of the resource is converted into a binary sequence. For example, a day is divided into 24 hour periods. If the resource is available in a certain hour period, the binary value of the corresponding position is 1, otherwise it is 0. At the same time, the historical price data and current price information are combined to generate a price trend sequence. For example, the price volatility or growth rate over a period of time can be calculated to form a sequence that reflects the price change trend. The generated binary sequence and price trend sequence are combined to form a feature vector of the resource, which comprehensively characterizes the availability and price dynamics of the resource. For example, a feature vector of a resource can contain 24 binary bits representing the available time period and 7 values ​​representing the price trend over the past week.

[0080] Then, a graph neural network model is constructed. In this model, different types of resources are represented as nodes in a heterogeneous graph, such as human resources, equipment resources, and site resources. The spatiotemporal relationship between resources is represented as edges in the graph. For example, if some resources need to be used in the same place or in the same time period, there is an association relationship between them. The attention mechanism is used to calculate the association strength between different resource nodes. For example, if two resources are often used at the same time, the association strength between them is high.

[0081] The resource feature vector is input into the gated recurrent unit of the graph neural network model. The gated recurrent unit performs temporal encoding of the resource availability period to capture the temporal dependency of resource availability. The importance of different time slots is calculated through the temporal attention layer. For example, if the demand for resources in certain time slots is large, then these time slots are more important. For example, if the availability of a resource in the morning is more important than its availability in the evening, then the morning time slot will be given a higher weight.

[0082] The encoded temporal features are input into the graph structure aggregation layer of the graph neural network model. The graph structure aggregation layer extracts the interaction patterns between nodes from multiple feature subspaces based on the multi-head attention mechanism, and transmits node features on the edge of the graph through the message passing algorithm to realize the propagation of resource information on the graph. The graph pooling operation is used to aggregate the node-level features into a graph-level representation, and the original feature information is retained through residual connections to prevent information loss.

[0083] Build a time-sensitive dynamic update module. This module continuously receives new resource schedule feedback through a sliding time window and updates the resource feature vector and graph structure based on the new information. Calculate the impact of real-time price changes on resource nodes based on the price propagation network. For example, if the price of a resource rises, the demand for other resources related to it may decrease. Identify newly added resource associations through the edge prediction model and update the topology of the graph. For example, if two resources start to be used together frequently, add an edge between them. Input the updated graph into the temporal graph convolutional network to further capture the dynamic changes of resources.

[0084] Case: Assume that you need to schedule three resources A, B, and C to complete a project. Resource A is available in the morning and costs 10; resource B is available in the afternoon and costs 15; resource C is available all day and costs 20. Resources A and B need to be used in the same location and have a spatiotemporal association. Build a graph neural network model and represent resources A, B, and C as nodes. There is an edge between A and B. Convert the available time and price information of resources into feature vectors and input them into the model. Based on resource characteristics and associations, the model calculates the optimal resource scheduling plan: use resource A in the morning, resource B in the afternoon, and resource C at the same time.

[0085] The solution of this application can:

[0086] Improve resource utilization: By considering the available time, price, and relationship between resources, this method can allocate resources more effectively, avoid resource idleness or conflict, and thus improve resource utilization. Reduce costs: By analyzing and predicting price information, this method can help select resources with more appropriate prices, thereby reducing project costs. Enhance scheduling flexibility: This method can dynamically update resource information and graph structure to adapt to changing project requirements and resource conditions, thereby enhancing scheduling flexibility.

[0087] In an optional implementation, the time sequence correlation, geographical location correlation and cost matching degree between various resources are calculated based on the graph neural network model, and an optimal resource combination scheme that satisfies various correlations is output. The optimal resource combination scheme associates the specific schedules of various resources in time sequence, including:

[0088] Receive the resource node feature vector output by the graph neural network model, extract the time slot information, geographical location information, and price information of each resource node according to the resource node feature vector; input the time slot information into a dynamic time planner, and the dynamic time planner analyzes the schedule overlap, connection interval, and timing dependency between resource nodes to generate a timing dependency matrix;

[0089] The temporal dependency matrix and the geographic location information are input into a spatial attention network, the spatial attention network calculates spatiotemporal accessibility based on the geographic distance between the resource nodes, uses a distance decay function to evaluate the degree of traffic convenience, and outputs a geographic location association matrix; the geographic location association matrix and the price information are input into a cost matching calculation module, the cost matching calculation module converts the price range into a standardized score based on a piecewise linear mapping function, and generates a cost matching matrix in combination with a budget constraint;

[0090] The time dependency matrix, the geographic location correlation matrix, and the cost matching matrix are input into a multidimensional feature fusion network, which extracts multidimensional correlation features through tensor decomposition, uses an adaptive weight mechanism to balance the importance of different dimensions, and outputs a comprehensive correlation score matrix between resource nodes; the comprehensive correlation score matrix is ​​input into a resource combination optimization module, which searches for the maximum weight match based on a hierarchical Hungarian algorithm and generates a candidate resource combination plan;

[0091] The candidate resource combination scheme is input into the constraint verification module, and the constraint verification module uses a graph coloring algorithm to detect and eliminate resource schedule conflicts, determines the optimal path that meets the timing constraints through a dynamic programming algorithm, and outputs the optimal resource combination scheme; a reinforcement learning environment is constructed based on the execution feedback of the optimal resource combination scheme, and the resource combination strategy is alternately optimized and updated through the policy network and the value network. The Monte Carlo tree search is used to expand the search space, and the final resource combination scheme after online optimization is output, and the final resource combination scheme includes the specific schedule arrangements for each resource.

[0092] A resource combination optimization method based on graph neural network is used to generate the optimal resource combination plan that meets the time sequence, geographical location and cost matching, and includes the specific schedule arrangement of each resource.

[0093] First, construct a resource graph. Each resource is represented as a node in the graph. The characteristics of the node include time slot information (such as available date and time period), geographical location information (such as longitude and latitude coordinates), price information, etc. The time dependency relationship and geographical proximity relationship between resources can be represented as edges in the graph.

[0094] Next, a graph neural network model is used to learn the feature vectors of resource nodes. The model can capture the complex relationships between resource nodes and encode these relationships into the feature vectors of the nodes. For example, a graph convolutional network (GCN) or a graph attention network (GAT) can be used to learn node features. Suppose there are two resource nodes A and B. A is available from 9 am to 10 am, and B is available from 10 am to 11 am. The geographical distance between A and B is relatively close. The graph neural network can learn that A and B have a strong correlation in both time and space, and reflect this correlation in their feature vectors.

[0095] Then, time slot information, geographic location information and price information are extracted from the resource node feature vector.

[0096] The extracted time slot information is input into the dynamic time planner. The dynamic time planner analyzes the schedule overlap, connection interval, and timing dependency between resource nodes to generate a timing dependency matrix. For example, if the end time of resource A is connected to the start time of resource B, the element values ​​corresponding to A and B in the timing dependency matrix are higher. Assuming that the time slot of resource A is [9,10], the time slot of resource B is [10,11], and the time slot of resource C is [11,12], there is a strong timing dependency between A and B, and between B and C.

[0097] The temporal dependency matrix and geographic location information are input into the spatial attention network. The spatial attention network calculates the spatiotemporal accessibility based on the geographic distance between resource nodes, uses the distance decay function to evaluate the degree of transportation convenience, and outputs the geographic location correlation matrix. For example, if the geographical distance between two resource nodes is close, the geographic location correlation between them is high. Assuming that the location coordinates of resource A are (10,20), the location coordinates of resource B are (11,21), and the location coordinates of resource C are (20,30), the geographic location correlation between A and B is higher than that between A and C.

[0098] The geographic location association matrix and price information are input into the cost matching calculation module. The cost matching calculation module converts the price range into a standardized score based on the piecewise linear mapping function, and generates a cost matching matrix in combination with the budget constraint. For example, if the prices of two resources are within the budget range and the prices are close, the cost matching between them is high. Assuming that the price of resource A is 100 yuan, the price of resource B is 110 yuan, and the budget is 200 yuan, the cost matching between A and B is high.

[0099] The time dependency matrix, geographic location correlation matrix, and cost matching matrix are input into the multidimensional feature fusion network. The multidimensional feature fusion network extracts multidimensional correlation features through tensor decomposition, uses an adaptive weight mechanism to balance the importance of different dimensions, and outputs a comprehensive correlation score matrix between resource nodes.

[0100] The comprehensive correlation score matrix is ​​input into the resource combination optimization module. The resource combination optimization module finds the maximum weight match based on the hierarchical Hungarian algorithm and generates candidate resource combination solutions.

[0101] The candidate resource combination scheme is input into the constraint verification module. The constraint verification module uses the graph coloring algorithm to detect and eliminate resource schedule conflicts, determines the optimal path that meets the timing constraints through the dynamic programming algorithm, and outputs the optimal resource combination scheme, including the specific schedule arrangements for each resource.

[0102] Finally, a reinforcement learning environment is constructed based on the execution feedback of the optimal resource combination plan. The resource combination strategy is optimized and updated alternately through the policy network and the value network. The Monte Carlo tree search is used to expand the search space, and the final resource combination plan that has been optimized online is output. The plan includes the specific schedule arrangements for each resource.

[0103] The solution of this application can:

[0104] Improve resource utilization efficiency: By optimizing resource combination schemes, minimize resource idleness and waste, and improve resource utilization efficiency. Reduce costs: By considering cost matching, select a more cost-effective resource combination to reduce overall costs. Improve user experience: By considering the correlation of time sequence and geographical location, generate a more reasonable resource combination scheme to improve user experience.

[0105] In an optional implementation, the optimal resource combination plan is broadcasted to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status, including:

[0106] The optimal resource combination plan is converted into a data structure, and the resource schedule information, price information, and reservation conditions are packaged into a transaction data packet using a Merkle tree structure; the transaction data packet is broadcast to each resource provider node through the sharding mechanism of the blockchain network, and the resource provider node verifies the validity of the transaction data packet based on a threshold signature scheme;

[0107] The resource provider node receives the verified transaction data packet and triggers the smart contract to execute the resource locking operation. The smart contract first performs a resource status pre-check to verify the current resource status, price validity, and schedule availability, and generates a resource status check report; based on the resource status check report, the cross-chain atomic exchange protocol is executed to encapsulate the locking requests of multiple resources into one atomic transaction, and a hash time lock contract is used to set a unified lock time window;

[0108] During the execution of atomic transactions by the smart contract, concurrent lock requests are processed through an optimistic concurrency control protocol based on timestamps, and a processing order is assigned to lock requests of different priorities; a deadlock detection algorithm is used to monitor the resource locking status, and a rollback compensation mechanism is triggered when a deadlock is found to automatically release the locked resources;

[0109] The execution result of the atomic transaction is input into the state synchronization module, which uses a vector clock to track the timing relationship of resource state changes and propagates state update information in the blockchain network through the gossip protocol; the state synchronization module stores the resource lock state in multiple nodes in a sharding manner and ensures the security of the state data through a threshold secret sharing scheme;

[0110] A node monitoring network is built based on the heartbeat mechanism to detect the operating status and network connectivity of the resource provider nodes in real time. When a node failure is detected, service recovery is performed through the state machine copy technology. The node monitoring network submits the node status information to the blockchain network, and adopts the Byzantine fault-tolerant consensus mechanism to maintain the stable operation of the network, and finally outputs the resource locking status and returns it to the requester.

[0111] A resource combination and locking method based on blockchain is used to optimize resource allocation, improve resource utilization, and ensure transaction security and reliability. This method packages information such as resource schedule, price, and reservation conditions into a transaction data packet, broadcasts it to each resource provider through the blockchain network, and realizes automatic resource locking.

[0112] First, the data structure of the optimal resource combination plan is converted. The resource schedule information (for example, July 1 to July 10, 2024), price information (for example, 1,000 yuan per day), booking conditions (for example, reservations must be made three days in advance) and other information are organized in the form of key-value pairs, such as {"date":"2024-07-01","price":"1000","conditions":"book three days in advance"}. Then, these key-value pairs are combined into a list to represent the complete information of a resource. The information of multiple resources forms a larger list. Finally, using the Merkle tree structure, this list containing all resource information is hashed to generate a unique Merkle root as the identifier of the transaction data packet. For example, the information of resource A and resource B are {"date":"2024-07-01","price":"1000"} and {"date":"2024-07-05","price":"1500"} respectively. After hash calculation, the Merkle root HASH (HASH(A), HASH(B)) is generated.

[0113] Next, the transaction data packet is broadcast to each resource provider node through the sharding mechanism of the blockchain network. The blockchain network is divided into multiple shards, each of which is responsible for processing a portion of the transaction data. The transaction data packet is assigned to different shards according to certain rules, for example, sharding is performed based on the geographical location of the resource provider. The resource provider node only needs to monitor the transaction data of the shard to which it belongs, reducing the network load.

[0114] After receiving the transaction data packet, each resource provider node verifies its validity based on the threshold signature scheme. Assuming the threshold value is 3, at least 3 nodes' signatures are required to verify the validity of the transaction data packet. Each node signs the Merkle root with its own private key and broadcasts the signature to other nodes. When a node collects a sufficient number of valid signatures, it can verify the validity of the transaction data packet. For example, node 1, node 2, and node 3 sign the Merkle root respectively. After node 4 collects these three signatures, it can verify the validity of the transaction data packet.

[0115] After verification, the smart contract is triggered to perform resource locking operations. The smart contract first performs a preliminary check of the resource status. For example, check whether the resource is currently available, whether the price is consistent with the transaction data package, whether the schedule is available, etc. Generate a resource status check report, for example, {"Resource A":"Available","Resource B":"Available","Price":"Match","Schedule":"Available"}.

[0116] Based on the resource status check report, execute the cross-chain atomic exchange protocol. Encapsulate the lock requests of multiple resources into one atomic transaction. For example, encapsulate the lock requests of resource A and resource B into one atomic transaction. Use the hash time lock contract to set a unified lock time window, for example, set the lock time to 10 minutes.

[0117] During the execution of atomic transactions by smart contracts, concurrent lock requests are processed through an optimistic concurrency control protocol based on timestamps. For example, the processing order is assigned to lock requests of different priorities according to the timestamps of the requests arriving. The requests that arrive first are processed first. A deadlock detection algorithm is used to monitor the resource locking status. For example, if resource A and resource B wait for each other to release, a deadlock is formed. When a deadlock is found, the rollback compensation mechanism is triggered to automatically release the locked resources.

[0118] The execution result of the atomic transaction is input into the state synchronization module. The state synchronization module uses vector clocks to track the timing relationship of resource state changes. For example, the vector clock is used to record the latest update time of each resource. The state update information is propagated in the blockchain network through the gossip protocol. The state synchronization module stores the resource lock state in multiple nodes in a sharding manner. For example, the lock state of resource A is stored in shard 1, and the lock state of resource B is stored in shard 2. The security of state data is guaranteed by the threshold secret sharing scheme.

[0119] A node monitoring network is built based on the heartbeat mechanism to detect the operating status and network connectivity of the resource provider nodes in real time. For example, the node sends a heartbeat message to other nodes at regular intervals. When a node failure is detected, for example, if the heartbeat message is not received three times in a row, the service is restored through the state machine replica technology. The node monitoring network submits the node status information to the blockchain network and uses the Byzantine fault-tolerant consensus mechanism to maintain the stable operation of the network. Finally, the resource locking status is output and returned to the requester, for example, {"resource A":"locked","resource B":"locked"}.

[0120] The solution of this application can:

[0121] Improve resource utilization: Through the optimal resource combination plan, integrate the resources of different resource providers to avoid idle resources and maximize resource utilization. Ensure transaction security and reliability: Use blockchain technology and smart contracts to ensure that the transaction process is transparent, safe and reliable, and prevent malicious tampering and fraud. Improve transaction efficiency: Automate the resource locking process, reduce manual intervention, improve transaction efficiency, and reduce transaction costs.

[0122] In an optional implementation, generating reservation confirmation information and electronic credential information based on the resource locking state, writing the reservation confirmation information and the electronic credential information into a blockchain distributed ledger; monitoring the state of the reservation confirmation information, and when the state of the reservation confirmation information changes, triggering a corresponding resource unlocking or relocking operation through the smart contract includes:

[0123] Construct reservation confirmation information including reservation identification, resource locking status hash, timestamp, reservation party information, resource provider information, and transaction amount according to the resource locking status, encrypt sensitive data in the reservation confirmation information using homomorphic encryption technology, and generate encrypted reservation confirmation information; based on the encrypted reservation confirmation information, call a smart contract to generate electronic credential information, wherein the electronic credential information includes an anti-counterfeiting verification field generated based on zero-knowledge proof;

[0124] The encrypted reservation confirmation information and the electronic credential information are stored in different nodes of the blockchain network by using sharding technology, and the reservation confirmation information and the electronic credential information are multi-signed by a threshold signature scheme, and the signed information is written into a blockchain distributed ledger; the blockchain distributed ledger uses a hierarchical Merkle tree structure to establish a data index to achieve fast query of status information;

[0125] An event-driven state listener is constructed to monitor the state of the reservation confirmation information in real time. When the state listener captures a state change event, the state change is verified for legitimacy based on the rule engine, and a trigger request containing state transition conditions is generated; the state listener asynchronously transmits the trigger request to the smart contract through a distributed message queue;

[0126] The smart contract that receives the trigger request parses the state transition type based on the state machine model. When the state transition type is to cancel the reservation, it triggers the resource unlocking operation to restore the resource state to an available state; when the state transition type is to change the reservation, it triggers the resource re-locking operation and adopts the timestamp-based optimistic concurrency control protocol to handle resource conflicts.

[0127] A resource booking and management method based on blockchain, the specific implementation method is as follows:

[0128] First, the resource lock and booking confirmation information is generated. When a user initiates a resource booking request, the system will first check the lock status of the target resource. If the resource is available, the system will lock the resource and generate a booking identifier based on the resource lock status. The booking identifier can be a globally unique UUID. At the same time, the system calculates the hash value of the resource lock status for subsequent verification and auditing. For example, the resource lock status can be represented as a JSON structure containing information such as resource ID and lock time, and then the hash value is calculated using the SHA256 algorithm. Next, the system records the current timestamp, obtains the information of the booking party and resource provider, and data such as the transaction amount. In order to protect sensitive data, the system uses homomorphic encryption technology to encrypt the booking party information, resource provider information, and transaction amount. For example, the transaction amount can be encrypted using the Paillier homomorphic encryption algorithm, so that the encrypted amount can still be added without decryption. Finally, the system assembles the booking identifier, resource lock status hash, timestamp, encrypted booking party information, encrypted resource provider information, and encrypted transaction amount into booking confirmation information.

[0129] Subsequently, an electronic certificate is generated and written to the blockchain. Based on the encrypted booking confirmation information, the system calls the smart contract pre-deployed on the blockchain to generate electronic certificate information. The electronic certificate information contains an anti-counterfeiting verification field generated based on zero-knowledge proof. For example, zk-SNARKs technology can be used to generate a proof that the data in the booking confirmation information is valid and has not been tampered with without revealing the specific booking information. In order to improve the efficiency and security of data storage, the system uses sharding technology to store the encrypted booking confirmation information and electronic certificate information in different nodes of the blockchain network. At the same time, the system uses a threshold signature scheme to perform multi-party signatures on the booking confirmation information and electronic certificate information to ensure the integrity and non-repudiation of the data. For example, the BLS threshold signature scheme can be used, and a valid signature can be generated when the preset signature threshold is reached. Finally, the signed booking confirmation information and electronic certificate information are written to the blockchain distributed ledger. The blockchain distributed ledger uses a hierarchical Merkle tree structure to establish a data index, which is convenient for subsequent rapid query of status information. For example, the transaction information of each block can be stored in the leaf node of the Merkle tree, and then the hash value is calculated layer by layer to finally obtain a Merkle root and store it in the block header.

[0130] Next, the reservation status is monitored in real time and the corresponding actions are triggered. The system builds an event-driven state listener to monitor the status of the reservation confirmation information in real time. When the reservation status changes, such as when the user cancels or changes the reservation, the state listener will capture the corresponding event. Once the state change event is captured, the system will verify the legitimacy of the state change based on the pre-defined rule engine. For example, verify whether the user has the authority to cancel or change the reservation. After the verification is passed, the system generates a trigger request containing the state transition condition. The state listener asynchronously passes the trigger request to the smart contract through a distributed message queue.

[0131] Finally, the smart contract processes the state change request. After receiving the trigger request, the smart contract parses the state transition type based on the state machine model. If the state transition type is to cancel the reservation, the smart contract triggers the resource unlocking operation to restore the resource state to the available state. If the state transition type is to change the reservation, the smart contract triggers the resource relocking operation and uses the timestamp-based optimistic concurrency control protocol to handle potential resource conflicts. For example, if the same resource receives multiple change reservation requests in a short period of time, the smart contract will compare the timestamps of these requests, prioritize the request with the earliest timestamp, and reject other requests.

[0132] The solution of this application can:

[0133] Enhance data security and privacy protection: Protect users' sensitive information and prevent information leakage and tampering through homomorphic encryption and zero-knowledge proof technology. Improve booking process efficiency and transparency: Use blockchain technology and smart contracts to automate the booking process, reduce manual intervention, and improve efficiency and transparency. Optimize resource management and conflict resolution: Monitor resource status in real time through state listeners and optimistic concurrency control protocols, effectively resolve resource conflicts, and improve resource utilization.

[0134] Figure 2 FIG. 1 is a schematic diagram of the structure of the intelligent reservation and multi-party real-time confirmation collaborative system in the business travel scenario according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0135] The first unit is used to receive a business travel reservation request and extract travel time information, personnel information, budget information and travel policy information to generate a reservation task; use a deep learning model to intelligently analyze various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan and send the initial resource matching plan to the corresponding resource provider;

[0136] The second unit is used to receive resource schedule feedback returned by each resource provider for the initial resource matching plan, the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model to input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output an optimal resource combination plan that meets each correlation, wherein the optimal resource combination plan associates the specific schedules of each resource in time sequence;

[0137] The third unit is used to broadcast the optimal resource combination plan to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; generates reservation confirmation information and electronic certificate information based on the resource locking status, and writes the reservation confirmation information and the electronic certificate information into the blockchain distributed ledger; monitors the status of the reservation confirmation information, and when the status of the reservation confirmation information changes, triggers the corresponding resource unlocking or re-locking operation through the smart contract.

[0138] According to a third aspect of the embodiments of the present invention,

[0139] An electronic device is provided, comprising:

[0140] processor;

[0141] a memory for storing processor-executable instructions;

[0142] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0143] A fourth aspect of the embodiments of the present invention is:

[0144] 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.

[0145] 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.

[0146] 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. Intelligent booking and multi-party real-time confirmation collaboration method in business travel scenarios, characterized by: include: Receive business travel booking requests and extract travel time information, personnel information, budget information, and travel policy information to generate booking tasks; A deep learning model is used to intelligently analyze various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan, and send the initial resource matching plan to the corresponding resource provider; Receive resource schedule feedback returned by each resource provider for the initial resource matching plan, the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model, input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output an optimal resource combination plan that meets each correlation, wherein the optimal resource combination plan associates the specific schedules of each resource in time sequence; The optimal resource combination plan is broadcast to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; based on the resource locking status, the reservation confirmation information and the electronic certificate information are generated, and the reservation confirmation information and the electronic certificate information are written into the blockchain distributed ledger; the status of the reservation confirmation information is monitored, and when the status of the reservation confirmation information changes, the corresponding resource unlocking or re-locking operation is triggered through the smart contract.

2. The method according to claim 1, characterized in that Using a deep learning model to intelligently analyze various information in the reservation task, matching transportation resources based on the travel time information and the travel policy information, matching accommodation resources based on the personnel information and the budget information, matching conference venue resources based on the travel time information and the personnel information, generating an initial resource matching plan and sending the initial resource matching plan to the corresponding resource provider includes: Convert the travel time information into a timestamp sequence, convert the personnel information into a multidimensional feature vector including rank characteristics, department characteristics, and historical booking preference characteristics, normalize the budget information into a budget vector with a uniform numerical range, convert the travel policy information into a rule vector through natural language processing, and construct a deep learning model to extract features from the timestamp sequence, the multidimensional feature vector, the budget vector, and the rule vector; Input the timestamp sequence and the rule vector into the first bidirectional long short-term memory network of the deep learning model, the first bidirectional long short-term memory network encodes the timestamp sequence through the time series encoding layer, calculates the weights of different rules in the rule vector through the policy rule attention layer, matches the transportation resources that meet the travel policy, and generates a transportation resource matching plan including transportation type, departure and arrival time, and seat grade; input the multidimensional feature vector and the budget vector into the second bidirectional long short-term memory network of the deep learning model, the second bidirectional long short-term memory network determines the accommodation standard based on the job grade characteristics in the multidimensional feature vector, screens the accommodation resources that meet the budget based on the budget vector, and generates an accommodation resource matching plan; input the timestamp sequence and the multidimensional feature vector into the third bidirectional long short-term memory network of the deep learning model, the third bidirectional long short-term memory network determines the meeting time interval based on the timestamp sequence, predicts the number of participants based on the multidimensional feature vector and matches the meeting venue resources, and generates a meeting venue matching plan; The transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme are input into a multi-task learning framework, and the multi-task learning framework calculates the association weights of different resource types through an attention mechanism, and establishes a spatiotemporal constraint relationship among the transportation resource matching scheme, the accommodation resource matching scheme, and the conference venue matching scheme; the spatiotemporal constraint relationship is input into a global optimization layer, and the global optimization layer generates an initial resource matching scheme by minimizing an overall cost function; and the initial resource matching scheme is sent to corresponding transportation resource providers, accommodation resource providers, and conference venue resource providers for resource confirmation.

3. The method according to claim 1, characterized in that Receiving resource schedule feedback returned by each resource provider for the initial resource matching solution, wherein the resource schedule feedback identifies the available time period and real-time price of each resource, and constructing a graph neural network model to input the resource schedule feedback into the graph neural network model includes: Receive resource schedule feedback returned by each resource provider for the initial resource matching solution, extract time period information identifying the available time period of each resource and price information identifying the real-time price of each resource from the resource schedule feedback; convert the time period information into a binary sequence representing the available state of the resource, combine the price information with historical price fluctuation data to generate a price trend sequence, and combine the binary sequence with the price trend sequence to form a resource feature vector; Construct a graph neural network model, input the resource feature vector into the graph neural network model, the graph neural network model constructs different types of resources as nodes of a heterogeneous resource graph, constructs the spatiotemporal association relationship between resources as the edge of the heterogeneous resource graph, and uses an attention mechanism to calculate the association strength between different resource nodes; input the resource feature vector into the gated recurrent unit of the graph neural network model, the gated recurrent unit performs time series encoding on the resource available time period, and calculates the importance of different time slots through the time attention layer; The encoded temporal features are input into the graph structure aggregation layer of the graph neural network model. The graph structure aggregation layer extracts the interaction pattern between nodes from multiple feature subspaces based on the multi-head attention mechanism, and transmits the node features on the edge of the heterogeneous resource graph through the message passing algorithm; the node-level features are aggregated into a graph-level representation by using the graph pooling operation, and the original feature information is retained by the residual connection; A time-sensitive dynamic update module is constructed based on graph-level representation. The time-sensitive dynamic update module continuously receives new resource schedule feedback through a sliding time window, calculates the impact of real-time price changes on resource nodes based on a price propagation network, identifies new resource associations through an edge prediction model, and updates the topological structure of the heterogeneous resource graph; the updated heterogeneous resource graph is input into a temporal graph convolutional network.

4. The method according to claim 3, characterized in that Based on the graph neural network model, the temporal order correlation, geographical location correlation and cost matching between various resources are calculated, and the optimal resource combination scheme that meets the various correlations is output. The optimal resource combination scheme associates the specific schedules of various resources in chronological order, including: Receive the resource node feature vector output by the graph neural network model, extract the time slot information, geographical location information, and price information of each resource node according to the resource node feature vector; input the time slot information into a dynamic time planner, and the dynamic time planner analyzes the schedule overlap, connection interval, and timing dependency between resource nodes to generate a timing dependency matrix; The temporal dependency matrix and the geographic location information are input into a spatial attention network, the spatial attention network calculates spatiotemporal accessibility based on the geographic distance between the resource nodes, uses a distance decay function to evaluate the degree of traffic convenience, and outputs a geographic location association matrix; the geographic location association matrix and the price information are input into a cost matching calculation module, the cost matching calculation module converts the price range into a standardized score based on a piecewise linear mapping function, and generates a cost matching matrix in combination with a budget constraint; The time dependency matrix, the geographic location correlation matrix, and the cost matching matrix are input into a multidimensional feature fusion network, which extracts multidimensional correlation features through tensor decomposition, uses an adaptive weight mechanism to balance the importance of different dimensions, and outputs a comprehensive correlation score matrix between resource nodes; the comprehensive correlation score matrix is ​​input into a resource combination optimization module, which searches for the maximum weight match based on a hierarchical Hungarian algorithm and generates a candidate resource combination plan; The candidate resource combination scheme is input into the constraint verification module, and the constraint verification module uses a graph coloring algorithm to detect and eliminate resource schedule conflicts, determines the optimal path that meets the timing constraints through a dynamic programming algorithm, and outputs the optimal resource combination scheme; a reinforcement learning environment is constructed based on the execution feedback of the optimal resource combination scheme, and the resource combination strategy is alternately optimized and updated through the policy network and the value network. The Monte Carlo tree search is used to expand the search space, and the final resource combination scheme after online optimization is output, and the final resource combination scheme includes the specific schedule arrangements for each resource.

5. The method according to claim 4, characterized in that The optimal resource combination plan is broadcasted to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status including: The optimal resource combination plan is converted into a data structure, and the resource schedule information, price information, and reservation conditions are packaged into a transaction data packet using a Merkle tree structure; the transaction data packet is broadcast to each resource provider node through the sharding mechanism of the blockchain network, and the resource provider node verifies the validity of the transaction data packet based on a threshold signature scheme; The resource provider node receives the verified transaction data packet and triggers the smart contract to execute the resource locking operation. The smart contract first performs a resource status pre-check to verify the current resource status, price validity, and schedule availability, and generates a resource status check report; based on the resource status check report, the cross-chain atomic exchange protocol is executed to encapsulate the locking requests of multiple resources into one atomic transaction, and a hash time lock contract is used to set a unified lock time window; During the execution of atomic transactions by the smart contract, concurrent lock requests are processed through an optimistic concurrency control protocol based on timestamps, and a processing order is assigned to lock requests of different priorities; a deadlock detection algorithm is used to monitor the resource locking status, and a rollback compensation mechanism is triggered when a deadlock is found to automatically release the locked resources; The execution result of the atomic transaction is input into the state synchronization module, which uses a vector clock to track the timing relationship of resource state changes and propagates state update information in the blockchain network through the gossip protocol; the state synchronization module stores the resource lock state in multiple nodes in a sharding manner and ensures the security of the state data through a threshold secret sharing scheme; A node monitoring network is built based on the heartbeat mechanism to detect the operating status and network connectivity of the resource provider nodes in real time. When a node failure is detected, service recovery is performed through the state machine copy technology. The node monitoring network submits the node status information to the blockchain network, and adopts the Byzantine fault-tolerant consensus mechanism to maintain the stable operation of the network, and finally outputs the resource locking status and returns it to the requester.

6. The method according to claim 5, characterized in that Generating reservation confirmation information and electronic credential information based on the resource locking state, and writing the reservation confirmation information and the electronic credential information into the blockchain distributed ledger; monitoring the state of the reservation confirmation information, and when the state of the reservation confirmation information changes, triggering corresponding resource unlocking or re-locking operations through the smart contract includes: Construct reservation confirmation information including reservation identification, resource locking status hash, timestamp, reservation party information, resource provider information, and transaction amount according to the resource locking status, encrypt sensitive data in the reservation confirmation information using homomorphic encryption technology, and generate encrypted reservation confirmation information; based on the encrypted reservation confirmation information, call a smart contract to generate electronic credential information, wherein the electronic credential information includes an anti-counterfeiting verification field generated based on zero-knowledge proof; The encrypted reservation confirmation information and the electronic credential information are stored in different nodes of the blockchain network by using sharding technology, and the reservation confirmation information and the electronic credential information are multi-signed by a threshold signature scheme, and the signed information is written into a blockchain distributed ledger; the blockchain distributed ledger uses a hierarchical Merkle tree structure to establish a data index to achieve fast query of status information; An event-driven state listener is constructed to monitor the state of the reservation confirmation information in real time. When the state listener captures a state change event, the state change is verified for legitimacy based on the rule engine, and a trigger request containing state transition conditions is generated; the state listener asynchronously transmits the trigger request to the smart contract through a distributed message queue; The smart contract that receives the trigger request parses the state transition type based on the state machine model. When the state transition type is to cancel the reservation, it triggers the resource unlocking operation to restore the resource state to an available state; when the state transition type is to change the reservation, it triggers the resource re-locking operation and adopts the timestamp-based optimistic concurrency control protocol to handle resource conflicts.

7. An intelligent reservation and multi-party real-time confirmation collaborative system in a business travel scenario, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to receive a business travel booking request and extract travel time information, personnel information, budget information and travel policy information to generate a booking task; A deep learning model is used to intelligently analyze various information in the reservation task, match transportation resources based on the travel time information and the travel policy information, match accommodation resources based on the personnel information and the budget information, match conference venue resources based on the travel time information and the personnel information, generate an initial resource matching plan, and send the initial resource matching plan to the corresponding resource provider; The second unit is used to receive resource schedule feedback returned by each resource provider for the initial resource matching plan, the resource schedule feedback identifies the available time period and real-time price of each resource, construct a graph neural network model to input the resource schedule feedback into the graph neural network model, calculate the time sequence correlation, geographical location correlation and cost matching between each resource based on the graph neural network model, and output an optimal resource combination plan that meets each correlation, wherein the optimal resource combination plan associates the specific schedules of each resource in time sequence; The third unit is used to broadcast the optimal resource combination plan to the nodes of each resource provider through the blockchain network. After receiving the optimal resource combination plan, each resource provider node executes the smart contract to lock the resources and returns the resource locking status; generates reservation confirmation information and electronic certificate information based on the resource locking status, and writes the reservation confirmation information and the electronic certificate information into the blockchain distributed ledger; monitors the status of the reservation confirmation information, and when the status of the reservation confirmation information changes, triggers the corresponding resource unlocking or re-locking operation through the smart contract.

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