Intelligent service combination method and system for tourism hotel

Abstract

The invention provides an intelligent service combination method and system for a tourist hotel, and the method comprises the steps: generating a multi-dimensional state vector containing the potential demands of a user through multi-modal data fusion and behavior prediction; constructing a contextualized service network by using an evolved knowledge graph with a static mode layer and a dynamic instance layer; the service combination is formalized into a multi-agent collaborative decision-making problem for multi-objective optimization solution; the scheme is converted into an interpretable service script for flexible delivery and local re-planning; the invention relates to the technical field of smart hotels and service computing, in particular to a smart hotel and service computing system based on a closed-loop data driving system continuous evolution, which realizes the jump from passive response to active prediction, from static rules to dynamic learning and from centralized control to distributed coordination, solves the problems of service splitting, recommendation rigidity and frequent conflicts, and improves the service quality. And the user experience and the operation efficiency are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of smart hotel and service computing technology, and in particular to a method and system for intelligent service combination in tourist hotels. Background Technology

[0002] With consumption upgrades and digital transformation, the hotel industry is shifting from providing standardized accommodations to offering personalized experiences. Currently, mainstream technological solutions suffer from the following significant drawbacks: Service silos and fragmented experiences: Hotel property management systems, room control systems, and catering and entertainment systems often operate independently, lacking information flow and resulting in fragmented service processes. Guests need to communicate repeatedly with multiple departments or terminals, leading to a poor experience.

[0003] Static Recommendations and Lack of Context Awareness: Most existing intelligent recommendation systems rely on historical order data for collaborative filtering or content recommendation, failing to integrate users' dynamic behavior, physiological state, and external environment during their stay in real time. As a result, the recommendation results are lagging, simplistic, and lack contextual relevance.

[0004] Rigid rules and frequent conflicts: Simple IF-THEN rule engines struggle to handle complex resource competition, timing conflicts, and logical contradictions between services. Conflicts often require manual resolution after the fact, which is inefficient.

[0005] System closure and lack of evolutionary capability: Existing system functions are rigid and cannot learn new user preferences, service association patterns or better scheduling strategies from continuous service interactions, resulting in stagnant intelligence.

[0006] Centralized decision-making and poor fault tolerance: A centralized decision-making system will cause a complete service outage if it fails or experiences network latency. Furthermore, the decision-making process is opaque, making it difficult to trace and adjust solutions when problems arise.

[0007] With consumption upgrades and digital transformation, the hotel industry is shifting from providing standardized accommodations to offering personalized experiences. Current mainstream technological solutions suffer from significant drawbacks such as service silos, static recommendations, rigid rules, closed systems, and centralized decision-making, leading to fragmented user experiences, low operational efficiency, and high conflict resolution costs.

[0008] Therefore, there is an urgent need for an intelligent service portfolio solution that can perceive the context in real time, intelligently resolve conflicts, and automatically coordinate the execution of hotel resources. Summary of the Invention

[0009] To overcome existing problems, embodiments of this application provide a method and system for intelligent service combination in tourism hotels, aiming to achieve a fundamental transformation from responsive to predictive, from isolated to collaborative, from rigid to flexible, and from static to self-evolving, and to improve the interpretability of the system.

[0010] The technical solution adopted by the embodiments of this application to solve its technical problem is: A method and system for intelligent service combination in tourism hotels includes the following steps: S1: Multimodal fusion and intent prediction: Real-time collection of user explicit interaction data, behavioral trajectory data, physiological state data, and environmental data is performed for spatiotemporal alignment and fusion. Based on the fused time-series data stream, a pre-trained behavior prediction model is used to infer the user's immediate intent, potential derivative needs, and corresponding urgency of needs. A structured multi-dimensional user state vector is output, which includes at least intent labels, demand sets, sentiment values, and time sensitivity. Among them, through multimodal behavior prediction, the system can predict the needs of users that have not yet been explicitly expressed, realizing a leap from passive response to proactive prediction, from static rules to dynamic learning, and from centralized control to distributed collaboration. This solves the problems of service fragmentation, rigid recommendations, and frequent conflicts, and significantly improves user experience and operational efficiency. S2: Querying and Context Construction of Evolved Service Knowledge Graphs Using the user state vector as the initial trigger signal, an evolving hotel service knowledge graph is queried. The evolving hotel service knowledge graph consists of a static schema layer and a dynamic instance layer. The query process includes: first, semantic matching and rule reasoning are performed in the static schema layer to generate a preliminary candidate service schema subgraph that is semantically related to the user state; then, the schema subgraph is bound and instantiated with the current resource availability and real-time status of the service execution queue in the dynamic instance layer to form a contextualized and executable candidate service network. S3: Cooperative combinatorial decision-making based on multi-objective constraint optimization: The candidate service network is formalized as a multi-agent collaborative decision-making problem. Each candidate service is considered as a proposal from an agent with specific capabilities, resource requirements, and costs, while the hotel's various physical and human resources are considered as a shared environment. A multi-objective optimization function is established, including user personalized utility, total service execution time, hotel operating costs, and resource balance. By running a collaborative decision-making algorithm based on game theory constraint optimization, an optimal set of collaborative service combinations is found under the premise of satisfying all hard constraints, clearly defining the execution subject, timing, and parameters of each service. S4: Flexible Delivery and Execution Evolution of the Solution: The collaborative service combination scheme generated by the decision is transformed into a set of interpretable service scripts with execution priorities, dependencies and fault tolerance instructions, and distributed to the relevant execution terminals. During the execution of the scripts, the execution progress and abnormal events of each node are collected in real time through a distributed state bus. If the deviation from the expectation exceeds the threshold or the user actively intervenes, a lightweight local replanning is triggered, which only makes rapid adjustments to the affected service chain segments, and the adjustment results and experience are fed back to the dynamic instance layer of the evolutionary knowledge graph in an incremental form. Among them, the distributed multi-agent architecture avoids the risk of single point of failure; the service script and local replanning mechanism enable the system to gracefully handle execution anomalies and requirement changes, ensuring the resilience and flexibility of the service process, and enabling its understanding of user preferences, discovery of service associations, and optimization of decision-making strategies to deepen over time, thus building a long-term competitive barrier. S5: The Continuous Evolution of Knowledge Graphs After each service interaction loop, the system analyzes the data across the entire chain from intent prediction to end-user feedback. Through an online learning mechanism, it updates the weights of the relationships between entities in the evolving knowledge graph, discovers new potential association rules, and optimizes the parameters of the behavior prediction model and the multi-objective optimization function, thereby achieving continuous autonomous evolution of the system's overall decision-making capabilities.

[0011] Preferably, the behavior prediction model adopts a Transformer-based sequence model, whose input is an encoded multimodal time-series data segment, and whose output is the probability distribution of user intent and the corresponding confidence level over a future period.

[0012] Preferably, the static schema layer of the evolutionary knowledge graph is constructed through a hotel domain ontology, defining service, resource, context, and strategy entities and their semantic relationships and logical rules; the dynamic instance layer is updated through real-time data streams, recording the current state, attributes, and relationships of all entity instances in the form of an attribute graph.

[0013] Preferably, the candidate service network is formalized as a multi-agent collaborative decision-making problem, which specifically includes: modeling the service's use of resources, including time, space, materials, and manpower, as constraints, and modeling user historical preferences, the preceding and succeeding relationships between services, and business strategies as soft constraints or part of the optimization objective.

[0014] Preferably, the collaborative decision-making algorithm employs a multi-agent reinforcement learning algorithm based on a partially observable Markov decision process, or a distributed constraint optimization algorithm.

[0015] Preferably, the interpretable service script is defined using a structured description language based on a directed acyclic graph, where nodes represent atomic service actions, edges represent execution dependencies or data flows, and nodes and edges are attached with execution parameters, success criteria, and exception handling logic.

[0016] Preferably, the lightweight local replanning specifically involves: when an anomaly is detected, locking the completed and unaffected parts of the current service script, extracting the subgraph containing the anomaly node and its subsequent dependent nodes, and treating it as a new, scaled-down multi-agent collaborative decision-making problem, and re-executing steps S2 to S3 for rapid solution and replacement.

[0017] include: The multimodal perception and prediction module is configured to access user terminal data, IoT sensor data and hotel business system logs through multiple heterogeneous data interfaces, and embeds a behavior prediction model to execute step S1 and output a multi-dimensional user state vector. An evolutionary knowledge graph management module, configured to store and maintain an evolutionary hotel service knowledge graph, further includes: The graph storage and query engine unit is used to efficiently store the static pattern layer and the dynamic instance layer, and to respond to graph query requests. Instantiation and binder units are used to bind static schemas with real-time dynamic data to generate a contextualized candidate service network; A multi-agent collaborative decision engine module is configured to receive a candidate service network and formalize and solve it. The multi-agent collaborative decision engine module further includes: The problem formalization unit is used to automatically transform the service network into a multi-agent decision problem model based on a preset constraint template and objective function library; A distributed solver cluster unit consists of multiple solver instances working in parallel, used to run collaborative decision-making algorithms and generate the optimal solution set; The solution evaluation and selector unit is used to select the final collaborative service combination solution from the solution set based on the current strategy; The multi-agent collaborative decision-making engine module adopts a hybrid architecture, including: a centralized solution arbitrator and global optimizer, and multiple distributed domain-specialized agents; wherein, the domain-specialized agents correspond to different domains such as guest room service, catering service, and entertainment facilities, and are responsible for generating service proposals within their respective domains and performing local optimization; the solution arbitrator and global optimizer is responsible for receiving proposals from each domain, resolving conflicts, and coordinating and making final decisions under the global objective; A flexible execution coordination and monitoring module, configured to deliver and manage the lifecycle of service composition schemes, further includes: A service script compiler unit is used to compile cooperative service composition schemes into interpretable service scripts specific to the execution terminal. The service script compiler unit supports compiling the same service script into instruction formats recognizable by multiple heterogeneous execution terminals. Distributed execution agent units, deployed on each service execution terminal, are used to receive, interpret, and execute service scripts; The status monitoring and replanning trigger unit monitors execution deviations by subscribing to status events on the distributed message bus and triggers lightweight local replanning when conditions are met. An online evolutionary learning module, configured to collect end-to-end data from perception to feedback, further includes: The experience replay buffer unit is used to store structured interactive experience data; The model updater unit is used to periodically update the behavior prediction model, knowledge graph association weights, and decision model parameters using data from the experience replay buffer unit. The distributed message bus module connects all the above modules and uses a publish-deployment model for asynchronous and decoupled data and event communication.

[0018] The advantages of the embodiments of this application are: Through multimodal behavior prediction, the system can anticipate user needs that have not yet been explicitly expressed, achieving a leap from passive response to proactive prediction, from static rules to dynamic learning, and from centralized control to distributed collaboration. This solves the problems of service fragmentation, rigid recommendations, and frequent conflicts, significantly improving user experience and operational efficiency.

[0019] Modeling a problem as a multi-agent collaborative optimization not only avoids conflicts but also balances user experience, operating costs, and resource utilization from a global perspective, finding a Pareto optimal solution that benefits all parties and maximizing business value.

[0020] The distributed multi-agent architecture avoids the risk of single point of failure. Service scripts and local replanning mechanisms enable the system to gracefully handle execution anomalies and requirement changes, ensuring the resilience and flexibility of the service process. This allows the system to continuously deepen its understanding of user preferences, discover service relationships, and optimize decision-making strategies over time, thus building a long-term competitive barrier. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. In addition, for the sake of convenience, the terms "upper," "lower," "left," and "right" are equivalent to the upper, lower, left, and right directions of the accompanying drawings themselves, and the terms "first," "second," etc., are used for descriptive purposes and have no other special meaning.

[0023] This application provides a smart service combination method and system for tourism hotels, which solves the problems in the prior art. Through multimodal behavior prediction, the system can predict the needs of users that have not yet been explicitly expressed, realizing a leap from passive response to active prediction, from static rules to dynamic learning, and from centralized control to distributed collaboration. It solves the problems of service fragmentation, rigid recommendations, and frequent conflicts, and significantly improves user experience and operational efficiency.

[0024] Modeling a problem as a multi-agent collaborative optimization not only avoids conflicts but also balances user experience, operating costs, and resource utilization from a global perspective, finding a Pareto optimal solution that benefits all parties and maximizing business value.

[0025] The distributed multi-agent architecture avoids the risk of single point of failure. Service scripts and local replanning mechanisms enable the system to gracefully handle execution anomalies and requirement changes, ensuring the resilience and flexibility of the service process. This allows the system to continuously deepen its understanding of user preferences, discover service relationships, and optimize decision-making strategies over time, thus building a long-term competitive barrier.

[0026] The technical solution in this application is to solve the above problems, and the overall approach is as follows: Example

[0027] This embodiment provides a method for intelligent service combination in tourism hotels, including the following steps: Multimodal fusion and intent prediction: The system collects explicit user interaction data, behavioral trajectory data, physiological state data, and environmental data in real time, and performs spatiotemporal alignment and fusion. Based on the fused time-series data stream, it uses a pre-trained behavior prediction model to infer the user's immediate intent, potential derived needs, and corresponding urgency of needs. It outputs a structured multi-dimensional user state vector, which includes at least intent tags, need sets, sentiment values, and time sensitivity. Through multimodal behavior prediction, the system can predict needs that users have not yet explicitly expressed. This represents a leap from passive response to proactive prediction, from static rules to dynamic learning, and from centralized control to distributed collaboration. It solves the problems of service fragmentation, rigid recommendations, and frequent conflicts, and significantly improves user experience and operational efficiency. Among them, the behavior prediction model adopts a Transformer-based sequence model. Its input is an encoded multimodal time series data segment, and its output is the probability distribution of user intent and the corresponding confidence level over a future period of time. Querying and context construction of evolutionary service knowledge graphs: Using the user state vector as the initial trigger signal, an evolving hotel service knowledge graph is queried. The evolving hotel service knowledge graph consists of a static schema layer and a dynamic instance layer. The query process includes: first, semantic matching and rule reasoning are performed in the static schema layer to generate a preliminary candidate service schema subgraph that is semantically related to the user state; then, the schema subgraph is bound and instantiated with the current resource availability and real-time status of the service execution queue in the dynamic instance layer to form a contextualized and executable candidate service network. Among them, the static schema layer of the evolutionary knowledge graph is constructed through the hotel domain ontology, defining service, resource, context, and strategy entities and their semantic relationships and logical rules; the dynamic instance layer is updated through real-time data streams, recording the current state, attributes, and relationships of all entity instances in the form of an attribute graph. Cooperative combinatorial decision-making based on multi-objective constrained optimization: The candidate service network is formalized as a multi-agent collaborative decision-making problem. Each candidate service is considered as a proposal from an agent with specific capabilities, resource requirements, and costs, while the hotel's various physical and human resources are considered as a shared environment. A multi-objective optimization function is established, including user personalized utility, total service execution time, hotel operating costs, and resource balance. By running a collaborative decision-making algorithm based on game theory constraint optimization, an optimal set of collaborative service combinations is found under the premise of satisfying all hard constraints, clearly defining the execution subject, timing, and parameters of each service. Specifically, the candidate service network is formalized as a multi-agent collaborative decision-making problem, which includes: modeling the resource consumption of services, including time, space, materials, and manpower, as constraints; modeling user historical preferences, the pre- and post-service relationships, and business strategies as soft constraints or part of the optimization objective; and the distributed multi-agent architecture avoids the risk of single point of failure. Service scripts and local replanning mechanisms enable the system to gracefully handle execution anomalies and demand changes, ensuring the resilience and flexibility of the service process. This allows the system to continuously deepen its understanding of user preferences, the discovery of service relationships, and the optimization of decision-making strategies over time, thus building a long-term competitive barrier. Evolution of Flexible Delivery and Execution of the Solution: The collaborative service combination scheme generated by the decision is transformed into a set of interpretable service scripts with execution priorities, dependencies and fault tolerance instructions, and distributed to the relevant execution terminals. During the execution of the scripts, the execution progress and abnormal events of each node are collected in real time through a distributed state bus. If the deviation from the expectation exceeds the threshold or the user actively intervenes, a lightweight local replanning is triggered, which only makes rapid adjustments to the affected service chain segments, and the adjustment results and experience are fed back to the dynamic instance layer of the evolutionary knowledge graph in an incremental form. The continuous evolution of knowledge graphs: After each service interaction loop, the system analyzes the data across the entire chain from intent prediction to end-user feedback. Through an online learning mechanism, it updates the weights of the relationships between entities in the evolving knowledge graph, discovers new potential association rules, and optimizes the parameters of the behavior prediction model and the multi-objective optimization function, thereby achieving continuous autonomous evolution of the system's overall decision-making capabilities.

[0028] The collaborative decision-making algorithm employs a multi-agent reinforcement learning algorithm based on a partially observable Markov decision process, or a distributed constraint optimization algorithm.

[0029] The interpretable service script is defined using a structured description language based on a directed acyclic graph. Nodes represent atomic service actions, edges represent execution dependencies or data flows, and nodes and edges are attached with execution parameters, success criteria, and exception handling logic.

[0030] The lightweight local replanning is as follows: when an anomaly is detected, the completed and unaffected parts of the current service script are locked, the subgraph containing the anomaly node and its subsequent dependent nodes is extracted, and the subgraph is treated as a new, scaled-down multi-agent collaborative decision-making problem. Steps S2 to S3 are then re-executed for rapid solution and replacement. Example

[0031] This embodiment provides an intelligent service combination system for tourist hotels, including: The multimodal perception and prediction module is configured to access user terminal data, IoT sensor data and hotel business system logs through multiple heterogeneous data interfaces, and embeds a behavior prediction model to execute step S1 and output a multi-dimensional user state vector. The evolutionary knowledge graph management module is configured to store and maintain an evolutionary hotel service knowledge graph. The evolutionary knowledge graph management module further includes: The graph storage and query engine unit is used to efficiently store the static pattern layer and the dynamic instance layer, and to respond to graph query requests. Instantiation and binder units are used to bind static schemas with real-time dynamic data to generate a contextualized candidate service network; The multi-agent collaborative decision engine module is configured to receive candidate service networks, formalize and solve them. The multi-agent collaborative decision engine module further includes: The problem formalization unit is used to automatically transform the service network into a multi-agent decision problem model based on a preset constraint template and objective function library; A distributed solver cluster unit consists of multiple solver instances working in parallel, used to run collaborative decision-making algorithms and generate the optimal solution set; The solution evaluation and selector unit is used to select the final collaborative service combination solution from the solution set based on the current strategy; The multi-agent collaborative decision-making engine module adopts a hybrid architecture, including: a centralized solution arbitrator and global optimizer, and multiple distributed domain-specialized agents; the domain-specialized agents correspond to different domains such as guest room service, catering service, and entertainment facilities, and are responsible for generating service proposals within their respective domains and performing local optimization; the solution arbitrator and global optimizer is responsible for receiving proposals from various domains, resolving conflicts, and coordinating and making final decisions under the global objective; The flexible execution coordination and monitoring module is configured to deliver and manage the lifecycle of service composition solutions. The flexible execution coordination and monitoring module further includes: The service script compiler unit is used to compile cooperative service composition schemes into interpretable service scripts specific to the execution terminal. The service script compiler unit supports compiling the same service script into instruction formats recognizable by multiple heterogeneous execution terminals. Distributed execution agent units, deployed on each service execution terminal, are used to receive, interpret, and execute service scripts; The status monitoring and replanning trigger unit monitors execution deviations by subscribing to status events on the distributed message bus and triggers lightweight local replanning when conditions are met. The online evolutionary learning module is configured to collect end-to-end data from perception to feedback. The online evolutionary learning module further includes: The experience replay buffer unit is used to store structured interactive experience data; The model updater unit is used to periodically update the behavior prediction model, knowledge graph association weights, and decision model parameters using data from the experience replay buffer unit. The distributed message bus module connects all the above modules and uses a publish-deployment model for asynchronous and decoupled data and event communication.

[0032] Real-time collection of user-related multimodal data, including: user explicit request voice, APP orders, behavioral data, indoor positioning, device usage logs, status data, physiological sensor data, schedules, and environmental data such as room temperature and humidity, time, and weather.

[0033] Multimodal data is fused and cleaned, and a pre-trained natural language processing model for the hotel industry is used to parse the deep semantics of users’ explicit requests. Combined with behavioral and state data, users’ immediate intentions and potential needs are inferred to generate structured user intent vectors.

[0034] Based on a pre-built hotel service knowledge graph, nodes include service items, hardware resources, user profiles, and service rules, while edges represent semantic, logical, or resource relationships between them.

[0035] Using the user intent vector as input, multi-hop queries are performed in the graph to retrieve the core service nodes and their directly related service nodes, forming an initial service subgraph. Based on real-time environmental data such as time and weather, relevant contextualized rule nodes in the graph are activated to dynamically expand and weight the initial service subgraph.

[0036] The expanded service subgraph is transformed into a constraint satisfaction problem, where the variables are the candidate services, the domain is the service parameters such as time and specifications, and the constraints include: timing constraints, resource mutual exclusion constraints, logical preference constraints, and business rule constraints.

[0037] The CSP solver is run to find the service combination scheme with the optimal comprehensive utility function, while satisfying all constraints. If an unsolvable conflict is detected, the negotiation engine is triggered to provide the user with several alternative optimization schemes.

[0038] The finalized service combination plan is encapsulated into executable atomic service instruction packages, which are then distributed to execution terminals such as hotel property management systems, IoT central control systems, and third-party service provider systems through standardized API interfaces.

[0039] During service execution, continuously monitor the execution status and user feedback. If execution fails or user requirements change, return to step S1 and initiate the dynamic replanning process for service composition.

[0040] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for intelligent service combination in tourism hotels, characterized in that, Includes the following steps: S1: Multimodal fusion and intent prediction: Real-time collection of user explicit interaction data, behavioral trajectory data, physiological state data, and environmental data; spatiotemporal alignment and fusion. S2: Querying and Context Construction of Evolved Service Knowledge Graphs: Using the user state vector as the initial trigger signal, a query is performed on an evolving hotel service knowledge graph, which includes a static schema layer and a dynamic instance layer. S3: Cooperative combinatorial decision-making based on multi-objective constraint optimization: The candidate service network is formalized as a multi-agent collaborative decision-making problem; where each candidate service is regarded as an agent proposal with specific capabilities, resource requirements and costs, and the hotel's various physical and human resources are regarded as a shared environment. S4: Flexible Delivery and Execution Evolution of the Solution: The collaborative service combination scheme generated by the decision is transformed into a set of interpretable service scripts with execution priorities, dependencies and fault tolerance instructions, and distributed to the relevant execution terminals; S5: The Continuous Evolution of Knowledge Graphs After each service interaction loop, the system analyzes the data across the entire chain from intent prediction to end-user feedback. Through an online learning mechanism, it updates the weights of the relationships between entities in the evolving knowledge graph, discovers new potential association rules, and optimizes the parameters of the behavior prediction model and the multi-objective optimization function, thereby achieving continuous autonomous evolution of the system's overall decision-making capabilities.

2. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The behavior prediction model adopts a Transformer-based sequence model. Its input is an encoded multimodal time series data segment, and its output is the probability distribution of user intent and the corresponding confidence level over a future period.

3. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The static schema layer of the evolutionary knowledge graph is constructed through a hotel domain ontology, defining service, resource, context, and strategy entities and their semantic relationships and logical rules; the dynamic instance layer is updated through real-time data streams, recording the current state, attributes, and relationships of all entity instances in the form of an attribute graph.

4. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The candidate service network is formalized as a multi-agent collaborative decision-making problem, which specifically includes: modeling the service's resource occupancy, including time, space, materials, and manpower, as constraints; and modeling user historical preferences, the preceding and succeeding relationships between services, and business strategies as soft constraints or part of the optimization objective.

5. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The collaborative decision-making algorithm employs a multi-agent reinforcement learning algorithm based on partially observable Markov decision processes, or a distributed constraint optimization algorithm.

6. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The interpretable service script is defined using a structured description language based on a directed acyclic graph. Nodes represent atomic service actions, edges represent execution dependencies or data flows, and nodes and edges are attached with execution parameters, success criteria, and exception handling logic.

7. The intelligent service combination method for tourist hotels according to claim 1, characterized in that, The lightweight local replanning specifically involves: when an anomaly is detected, locking the completed and unaffected parts of the current service script, extracting the subgraph containing the anomaly node and its subsequent dependent nodes, and treating it as a new, scaled-down multi-agent collaborative decision-making problem, and re-executing steps S2 to S3 for rapid solution and replacement.

8. A smart service combination system for a tourist hotel for implementing the method of any one of claims 1-7, characterized in that, include: The multimodal perception and prediction module is configured to access user terminal data, IoT sensor data and hotel business system logs through multiple heterogeneous data interfaces, and embeds a behavior prediction model to execute step S1 and output a multi-dimensional user state vector. An evolutionary knowledge graph management module, configured to store and maintain an evolutionary hotel service knowledge graph, further includes: The graph storage and query engine unit is used to efficiently store static mode layers and dynamic instance layers, and respond to graph query requests. Instantiation and binder units are used to bind static schemas with real-time dynamic data to generate a contextualized candidate service network; A multi-agent collaborative decision engine module is configured to receive a candidate service network and formalize and solve it. The multi-agent collaborative decision engine module further includes: The problem formalization unit is used to automatically transform the service network into a multi-agent decision problem model based on a preset constraint template and objective function library; A distributed solver cluster unit consists of multiple solver instances working in parallel, used to run collaborative decision-making algorithms and generate the optimal solution set; The solution evaluation and selector unit is used to select the final collaborative service combination solution from the solution set based on the current strategy; A flexible execution coordination and monitoring module, configured to deliver and manage the lifecycle of service composition schemes, further includes: The service playbook compiler unit is used to compile cooperative service composition schemes into interpretable service playbooks specific to the execution terminal; Distributed execution agent units, deployed on each service execution terminal, are used to receive, interpret, and execute service scripts; The status monitoring and replanning trigger unit monitors execution deviations by subscribing to status events on the distributed message bus and triggers lightweight local replanning when conditions are met. An online evolutionary learning module, configured to collect end-to-end data from perception to feedback, further includes: The experience replay buffer unit is used to store structured interactive experience data; The model updater unit is used to periodically update the behavior prediction model, knowledge graph association weights, and decision model parameters using data from the experience replay buffer unit. The distributed message bus module connects all the above modules and uses a publish-publish model for asynchronous and decoupled data and event communication.

9. The intelligent service combination system for tourist hotels according to claim 8, characterized in that, The multi-agent collaborative decision-making engine module adopts a hybrid architecture, including: a centralized solution arbitrator and global optimizer, and multiple distributed domain-specialized agents; wherein, the domain-specialized agents correspond to different domains such as guest room service, catering service, and entertainment facilities, and are responsible for generating service proposals within their respective domains and performing local optimization; the solution arbitrator and global optimizer is responsible for receiving proposals from each domain, resolving conflicts, and coordinating and making final decisions under the global objective.

10. The intelligent service combination system for tourist hotels according to claim 8, characterized in that, The service script compiler unit supports compiling the same service script into instruction formats recognizable by multiple heterogeneous execution terminals.

Images