Multi-protocol compatible intelligent travel terminal AI service self-adaption method

Abstract

The invention relates to the technical field of intelligent travel terminals, and discloses a multi-protocol compatible intelligent travel terminal AI service self-adaption method. The method comprises the steps of firstly collecting travel service request data in a multi-protocol environment, then constructing a multi-protocol adaptation model through dynamic protocol conversion middleware, and unifying heterogeneous protocol requests; and dynamically allocating service resources by adopting a reinforcement learning framework, predicting user travel demands based on a hidden Markov model, and adjusting service recommendation priorities. And constructing a multi-modal interaction adaptation engine to fuse the multi-modal features to generate a unified service response. In addition, the system is also provided with a protocol compatibility verification module, a service priority dynamic scheduling algorithm, a federated learning driven user preference updating model and a fault-tolerant protocol adaptation engine. According to the invention, the AI service adaptation capability of the intelligent travel terminal in a multi-protocol environment is improved, the resource allocation is optimized, the user demand prediction accuracy and the interaction experience are improved, and the reliability and the fault tolerance of the system are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of smart travel terminals, and in particular to a multi-protocol compatible smart travel terminal AI service self-adaptation method. Background Art

[0002] With the rapid development of technology, smart travel devices are playing an increasingly important role in people's travel. They integrate multiple functions to meet users' diverse travel needs, such as hotel reservations, attraction recommendations, and transportation planning. However, current smart travel devices face numerous operational challenges, seriously impacting their service quality and user experience.

[0003] In a multi-protocol environment, the internet hosts a variety of network protocols, including HTTP, HTTPS, MQTT, and CoAP. Different travel service providers employ different protocols. For example, some small travel service platforms may use the lightweight MQTT protocol to reduce communication overhead, while large online travel platforms prefer HTTP / HTTPS to ensure stable and secure data transmission. This forces smart travel terminals to handle multiple heterogeneous protocols when interacting with different service providers. Due to the significant differences in syntax, semantics, and data formats among these protocols, terminals struggle to uniformly handle requests from various protocols, increasing system complexity and development costs. This also increases the risk of compatibility issues, leading to service request failures or data transmission errors.

[0004] From the perspective of service resource allocation, smart travel devices have limited hardware resources, such as CPU processing power, memory capacity, and network bandwidth. Multiple different service requests may occur concurrently at the same time, such as simultaneously inquiring about hotel reservations and loading real-time videos of tourist attractions. Traditional resource allocation methods are often static or based on simple rules, unable to dynamically adjust based on the device's real-time resource usage and the characteristics of service requests. This can easily lead to irrational resource allocation, with some services experiencing slow responses due to insufficient resources, while others occupy excessive idle resources, reducing overall system performance and user experience.

[0005] Predicting user needs is also a major challenge for smart travel terminals. User travel needs are dynamic and diverse, influenced by multiple factors such as time, location, and personal preferences. Existing travel service systems typically only make simple recommendations based on a user's historical behavior and lack the ability to accurately predict a user's future travel needs. For example, they might recommend similar locations based solely on a user's past travel destinations, ignoring factors such as the user's current travel purpose, time of year, and potential changes in interests. This can lead to a mismatch between recommended services and actual user needs, failing to meet users' demands for personalized, precise services.

[0006] In terms of multimodal interaction, while smart travel terminals support multiple interaction methods such as voice, text, and images, current interaction adaptation technology is insufficient. Inefficient feature extraction and fusion of input information from different modalities prevents the system from fully understanding the user's intent. For example, when a user simultaneously enters text containing the destination name and an image of a landmark building, the system struggles to accurately correlate and analyze this information, failing to quickly provide travel planning suggestions that meet the user's expectations. This impacts the efficiency and satisfaction of user interaction with the terminal.

[0007] Furthermore, numerous issues exist in areas such as protocol compatibility verification, service priority scheduling, user preference updates, and fault-tolerance processing. Existing technologies lack effective protocol compatibility verification mechanisms, making it impossible to ensure that service instructions after protocol conversion meet security and liveness properties. Service priority scheduling is inflexible and cannot be adjusted in real time based on the urgency of requests, resource requirements, and service-level agreements. User preference update models struggle to accurately reflect users' latest preference changes while protecting their privacy. When protocol conversion anomalies occur, there is a lack of reliable fault-tolerance processing, leading to system service interruptions. These issues severely restrict the development and application of smart travel terminals, and a comprehensive solution is urgently needed to improve their performance and user experience. Summary of the Invention

[0008] The purpose of the present invention is to provide a multi-protocol compatible smart travel terminal AI service self-adaptation method to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a multi-protocol compatible smart travel terminal AI service self-adaptation method, the method comprising:

[0010] Step T1: Collect travel service request data in a multi-protocol environment, including protocol type, service parameters, user context information, and terminal hardware resource status;

[0011] Step T2: Based on the dynamic protocol conversion middleware, a multi-protocol adaptation model is constructed to unify heterogeneous protocol requests into standardized service instructions through the protocol parsing tree and semantic mapping rules;

[0012] Step T3: Use the reinforcement learning framework to design a dynamic service resource allocation strategy. Define the state space as the real-time resource occupancy rate and service queue length, the action space as the resource allocation ratio, and the reward function as the weighted inverse of service response time and resource consumption.

[0013] Step T4: Predict user travel needs based on the hidden Markov model, calculate the temporal distribution of future service requests through the state transition probability matrix, and dynamically adjust the service recommendation priority;

[0014] Step T5: Build a multimodal interaction adaptation engine, integrate voice, text, and image inputs, use the attention mechanism to fuse multimodal features, and generate a unified service response.

[0015] Preferably, the step T2 further comprises:

[0016] Step T21: define the protocol parsing tree nodes as the syntax structure of the protocol fields, the edges as the dependencies between the fields, and generate the protocol abstract syntax tree through depth-first traversal;

[0017] Step T22: Establish a semantic mapping rule base to map each protocol field to the semantic label of the standardized service instruction. The rule format is:

[0018] Map(F proto ,C context )→S std

[0019] Among them, F proto is the protocol field, C context is the context constraint, S std is the target instruction, Map is the mapping function;

[0020] Step T23: A dynamic programming algorithm is used to optimize the protocol conversion path, and the objective function is a linear combination of conversion delay and semantic consistency.

[0021] Preferably, in step T3, the specific implementation of the reinforcement learning framework includes:

[0022] Step T31: Define the state vector s t =[R t ,Q t ], where R t is the real-time occupancy rate of CPU, memory and network bandwidth, Q t The number of pending requests in each service queue;

[0023] Step T32: The action space is the resource allocation ratio vector a t =[α1,α2,…,α n ],satisfy

[0024] Step T33: The reward function is designed as:

[0025]

[0026] Among them, T resp is the average response time, E res is the total resource consumption, λ1,λ2 are weight coefficients.

[0027] Preferably, in step T4, the construction of the hidden Markov model includes:

[0028] Step T41: define the hidden state as the user's travel intention category and the observed state as the historical service request sequence;

[0029] Step T42: Estimate the state transition matrix A and observation probability matrix B using the Baum-Welch algorithm. The calculation formula is:

[0030]

[0031] Among them, A ij is the state transition probability, which represents the probability of transitioning from hidden state i to state j, ξ t (i, j) is the probability of transitioning from state i to j at time t, γ t (i) is the probability of being in state i at time t, and T is the total time step of the observation sequence;

[0032] Step T43: Use the Viterbi algorithm to decode the most likely state sequence for the next k steps and generate a service recommendation priority list.

[0033] Preferably, in step T5, the specific steps of multimodal feature fusion include:

[0034] Step T51: Use a pre-trained convolutional neural network to extract image features, a bidirectional long short-term memory network to extract text features, and a mel-spectrogram convolutional network to extract speech features;

[0035] Step T52: Calculate the cross-modal attention weight matrix through the multi-head attention mechanism and dynamically weight the image, text and speech features;

[0036] Step T53: concatenate the weighted multimodal feature vectors and input them into the fully connected layer to generate a unified service response instruction.

[0037] Preferably, the method further comprises:

[0038] Step T6: Design a protocol compatibility verification module, use the temporal logic formulas in the formal method to describe the protocol interaction constraints, and use the model checking tool to verify whether the converted service instructions meet the security and liveness properties.

[0039] Preferably, the method further comprises:

[0040] Step T7: Implement the service priority dynamic scheduling algorithm, define the service level agreement weight matrix, calculate the priority score in real time according to the deadline, resource requirements and service level agreement weight of each request, and generate a processing queue by arranging the scores in descending order.

[0041] Preferably, the method further comprises:

[0042] Step T8: Build a federated learning-driven user preference update model. By training the preference prediction network locally on each terminal, the aggregation server fuses the local model parameters and adds differential privacy noise to update the global model parameters.

[0043] Preferably, the method further comprises:

[0044] Step T9: Design a fault-tolerant protocol adaptation engine. When a protocol conversion anomaly is detected, a similarity graph is constructed based on historical successful conversion records, and the K-nearest neighbor algorithm is used to match the optimal alternative conversion path.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] In terms of multi-protocol compatibility and processing, a multi-protocol adaptation model is constructed through dynamic protocol conversion middleware, and heterogeneous protocol requests are unified into standardized service instructions using protocol parsing trees and semantic mapping rules. This enables smart travel terminals to easily handle service requests from a variety of different protocols, whether HTTP, MQTT, or other protocols, and accurately parse and process them. Compared to traditional methods, this greatly reduces the complexity of system protocol processing, improves protocol compatibility, and improves data transmission accuracy. For example, when booking a hotel, different hotel booking platforms may use different protocols. This invention can quickly convert various protocol requests into unified instructions, reducing the probability of errors and failures and improving service reliability.

[0047] The dynamic service resource allocation strategy utilizes a reinforcement learning framework, defining the state space and action space and designing a reasonable reward function. This enables the system to dynamically adjust resource allocation based on real-time resource utilization and service queue lengths. For example, if the system detects network bandwidth constraints and a long hotel reservation queue, it automatically increases bandwidth allocation to the hotel reservation service, prioritizing urgent and resource-intensive tasks. This effectively improves resource utilization and reduces service response time.

[0048] Predicting user travel needs based on a hidden Markov model can more accurately grasp user intent. The state transition probability matrix is used to calculate the temporal distribution of future service requests and dynamically adjust the priority of service recommendations. For example, if a user plans a trip to a coastal city, the model can predict the user's likely needs for booking a beach hotel, inquiring about water activities, and purchasing sunscreen, and recommend corresponding services ranked by likelihood and importance. The multimodal interaction adaptation engine integrates voice, text, and image inputs, using an attention mechanism to fuse multimodal features and generate unified service responses. This enables the terminal to more comprehensively and accurately understand user intent. For example, when a user simultaneously expresses "I want to go somewhere with mountains and water" and uploads a picture of the scenery, the system can combine the voice and image information to quickly recommend destinations that meet the requirements, such as Guilin and Huangshan. This improves interaction efficiency and service quality, providing users with a more convenient and intelligent interactive experience.

[0049] The protocol compatibility verification module uses temporal logic formulas from formal methods to describe protocol interaction constraints and verifies the converted service instructions using model checking tools. This measure ensures that service instructions meet safety and liveness properties, prevents the system from entering an illegal state, and ensures the stability and reliability of system operation. The service priority dynamic scheduling algorithm calculates and ranks priority scores in real time based on each request's deadline, resource requirements, and service level agreement weight, ensuring that important and urgent requests are processed first, further improving the efficiency and quality of service processing.

[0050] The federated learning-driven user preference update model, through local training on each terminal and server aggregation, ensures timely and accurate updates of user preferences while protecting user privacy. Even without leaking private user data, the system can adjust preference predictions based on the user's latest behavior, making recommendation services more tailored to their actual needs. When a protocol conversion anomaly is detected, the fault-tolerant protocol adaptation engine constructs a similarity graph based on historical successful conversion records and uses the K-nearest neighbor algorithm to match the optimal alternative conversion path. This ensures the system can continue to provide services normally under abnormal circumstances, enhancing the system's fault tolerance and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a working principle diagram of the multi-protocol compatible smart travel terminal AI service self-adaptation method described in the present invention;

[0052] Figure 2 Flowchart for building a multi-protocol adaptation model;

[0053] Figure 3 Flowchart of the dynamic allocation strategy of service resources under the reinforcement learning framework;

[0054] Figure 4Flowchart for building a federated learning-driven user preference update model. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] See also Figure 1-4 The present invention provides a multi-protocol compatible smart travel terminal AI service self-adaptation method, and its overall implementation scheme is as follows:

[0057] Step T1: Collect travel service request data in a multi-protocol environment. This data includes protocol type, service parameters, user context, and terminal hardware resource status. For example, when a user requests a hotel reservation through a smart travel terminal, the collected protocol type might be HTTP / HTTPS, MQTT, etc. Service parameters include check-in date, check-out date, and hotel star rating requirements. User context includes the user's current location, travel purpose, and past travel preferences. Terminal hardware resource status includes information such as CPU usage, remaining memory space, and network bandwidth.

[0058] Step T2: Build a multi-protocol adaptation model based on dynamic protocol conversion middleware. Through specific protocol parsing trees and semantic mapping rules, heterogeneous protocol requests are unified into standardized service instructions, allowing requests in different protocol formats to be effectively processed by the system.

[0059] Step T3: Design a dynamic service resource allocation strategy using a reinforcement learning framework. Define the state space as real-time resource utilization and service queue length, and the action space as resource allocation ratio. Use a reward function to balance service response time and resource consumption, achieving optimal resource allocation.

[0060] Step T4: Predict user travel needs based on the Hidden Markov Model. The state transition probability matrix is used to calculate the temporal distribution of future service requests, and the service recommendation priority is dynamically adjusted to provide users with service recommendations that better meet their needs.

[0061] Step T5: Build a multimodal interaction adaptation engine that integrates voice, text, and image input. Use an attention mechanism to fuse multimodal features and generate unified service responses, enhancing the user interaction experience and meeting diverse user input needs.

[0062] The implementation of the present invention will be further described below with reference to Examples 1 to 5.

[0063] Example 1:

[0064] This embodiment mainly describes the specific details of the construction of the multi-protocol adaptation model in step T2, as well as the design of the protocol compatibility verification module, to ensure that heterogeneous protocol requests can be accurately converted into standardized service instructions and verify that the converted instructions meet the security and liveness properties.

[0065] In step T2, the multi-protocol adaptation model is constructed as follows:

[0066] Step T21: Define the protocol parsing tree nodes as the grammatical structure of the protocol fields, and the edges as the dependencies between fields. Taking the HTTP protocol request as an example, the protocol fields may include the request method (such as GET, POST), URL, request header (such as Content-Type, User-Agent), etc. Generate the protocol abstract syntax tree by traversing these fields in a depth-first manner. During the traversal process, starting from the root node, the child nodes are visited first, and the tree structure is constructed according to the hierarchical relationship and dependency relationship between the fields, so as to clearly present the grammatical hierarchy of the protocol.

[0067] Step T22: Establish a semantic mapping rule base to map each protocol field to the semantic label of the standardized service instruction. The rule format is:

[0068] Map(F proto ,C context )→S std

[0069] Among them, F proto It is a protocol field, such as the URL field in the HTTP protocol; C context Context constraints, such as the user's current location, current time, etc.; S std The target instruction is the converted standardized service instruction; Map is the mapping function. For example, if the protocol field is a reservation request URL for a specific hotel booking platform, combined with the user's current location (context constraint), the mapping function converts it into a standardized service instruction such as "Query for hotels near a specified area."

[0070] Step T23: A dynamic programming algorithm is used to optimize the protocol conversion path. The objective function is a linear combination of conversion latency and semantic consistency. In practical applications, different protocol conversion paths may result in different conversion latency and semantic accuracy. The dynamic programming algorithm calculates the cost of each possible conversion step and selects the optimal conversion path to minimize conversion latency and ensure semantic consistency.

[0071] For the protocol compatibility verification module:

[0072] A protocol compatibility verification module was designed, using temporal logic formulas from formal methods to describe protocol interaction constraints. Temporal logic formulas can precisely describe the protocol's behavioral constraints at different time points. For example, a requirement that a response instruction must be received within a certain timeframe after a reservation request instruction is sent. Model checking tools, such as SPIN, are used to verify that the converted service instructions meet safety and liveness properties. Safety properties ensure that the system does not enter an illegal state, while liveness properties ensure that the system can continue to provide the expected service. If the converted instructions are found to not meet the relevant properties, the protocol conversion process is adjusted and optimized.

[0073] Example 2:

[0074] In step T3, the specific implementation of the reinforcement learning framework is as follows:

[0075] Step T31: Define the state vector s t =[R t ,Q t ], where R t The real-time usage of CPU, memory and network bandwidth. Use system monitoring tools to obtain the real-time usage of these resources. For example, collect CPU usage, memory usage and network bandwidth usage at regular intervals (such as 1 second) and calculate their usage. t The system maintains a queue for each service type to record the number of requests waiting to be processed.

[0076] Step T32: The action space is the resource allocation ratio vector a t =[α1,α2,…,α n ],satisfy For example, in a system that includes services such as hotel reservations, attraction recommendations, and transportation reservations, it might indicate the proportion of CPU resources allocated to the hotel reservation service, the proportion of memory resources allocated to the attraction recommendation service, etc. These resource allocation ratios are dynamically adjusted based on the current state and needs of the system.

[0077] Step T33: The reward function is designed as:

[0078]

[0079] Among them, T resp The average response time is obtained by recording the time from the issuance of each service request to the receipt of the response and calculating the average value. resλ is the total resource consumption, which counts the total usage of CPU, memory, network, and other resources during the service request processing. λ1 and λ2 are weight coefficients, set based on the system's performance goals and resource constraints. For example, if the system prioritizes response time, increase the weight of λ1; if it prioritizes resource consumption, increase the weight of λ2.

[0080] The service priority dynamic scheduling algorithm is implemented as follows:

[0081] Implement a dynamic scheduling algorithm for service priority and define a service level agreement weight matrix. This matrix determines the weight of each service request based on factors such as the importance of different services and the user's payment level. For example, a higher weight is given to a hotel reservation request from a premium member, while a lower weight is given to a general attraction query request from an ordinary user. The priority score is calculated in real time based on the deadline, resource requirements, and service level agreement weight of each request. Assume that the deadline for the request is T deadline , the resource requirement is R require , the service level agreement weight is W SLA , then the priority score calculation formula can be designed as:

[0082]

[0083] Among them, T current The current time. A processing queue is generated in descending order of scores to ensure that important and urgent requests are processed first.

[0084] Example 3:

[0085] This embodiment introduces in detail the construction process of the hidden Markov model, which is used to accurately predict user travel needs and provide a basis for dynamically adjusting the service recommendation priority.

[0086] In step T4, the hidden Markov model is constructed as follows:

[0087] Step T41: Define the hidden state as the user's travel intention category, such as sightseeing, business travel, leisure vacation, etc. The observed state is the historical service request sequence, which records the service requests issued by the user in the past, such as "book a hotel in Beijing" and "check Shanghai attraction tickets".

[0088] Step T42: Estimate the state transition matrix A and observation probability matrix B using the Baum-Welch algorithm. The calculation formula is:

[0089]

[0090] Among them, A ijis the state transition probability, which represents the probability of transitioning from hidden state i to state j. For example, the probability of transitioning from the "tourism" intention to the "leisure vacation" intention. t (i, j) is the probability of transitioning from state i to j at time t, γ t (i) is the probability of being in state i at time t, and T is the total time step of the observation sequence. By analyzing and calculating a large amount of historical service request data, we obtain the state transition matrix and observation probability matrix, which describe the changing patterns of user travel intentions and the probability of generating specific service requests under different intentions.

[0091] Step T43: Use the Viterbi algorithm to decode the most likely state sequence for the next k steps and generate a priority list of service recommendations. Assuming k = 3, based on the state transition matrix and the observation probability matrix, the Viterbi algorithm can calculate the most likely state sequence of the user's travel intention for the next three steps. For example, it can predict that the user may have needs such as "booking transportation," "booking a local restaurant," or "purchasing attraction tickets." A priority list of services is generated based on the likelihood and importance of these needs, prioritizing services that meet these needs.

[0092] Example 4:

[0093] In step T5, the specific steps of multimodal feature fusion are as follows:

[0094] Step T51: Use a pre-trained convolutional neural network to extract image features. For example, a ResNet model pre-trained on a large-scale image dataset (such as ImageNet) can be used to input the input image data into the network. Through the network's convolutional layer, pooling layer and other operations, the feature vector of the image is extracted. These feature vectors contain information such as the color, texture, and shape of the image. Text features are extracted through a bidirectional long short-term memory network. The bidirectional long short-term memory network can simultaneously consider the forward and backward information of the text to better capture the semantics and contextual relationships of the text. The input text data is encoded by word or character, input into the bidirectional long short-term memory network, and the text feature vector is output. Speech features are extracted based on the Mel-spectrogram convolutional network. First, the speech signal is converted into a Mel-spectrogram. The Mel-spectrogram can better simulate human auditory perception. The Mel-spectrogram is then input into the convolutional network to extract the feature vectors of the speech. These feature vectors reflect the frequency, rhythm and other characteristics of the speech.

[0095] Step T52: Calculate the cross-modal attention weight matrix through the multi-head attention mechanism and dynamically weighted fuse the image, text, and speech features. The multi-head attention mechanism uses multiple different attention heads to focus on the features of different modalities from different perspectives. Calculate the attention weights for each attention head. These weights represent the importance of different modal features in generating a unified service response. For example, when processing a user request containing an image and text, one attention head may pay more attention to the key objects in the image, while another attention head may pay more attention to the keywords in the text. Through dynamic weighted fusion, the features of different modalities are combined according to their importance.

[0096] Step T53: The weighted multimodal feature vectors are concatenated and fed into a fully connected layer to generate a unified service response. The image, text, and speech feature vectors, fused through the multi-head attention mechanism, are concatenated to form a comprehensive feature vector. This comprehensive feature vector is then fed into a fully connected layer, which processes the feature vectors through a series of weight matrices and activation functions, ultimately generating a unified service response, such as "recommend travel routes that meet the user's needs."

[0097] Example 5:

[0098] This embodiment describes the construction of a federated learning-driven user preference update model and the design of a fault-tolerant protocol adaptation engine. It aims to solve the problem of updating the user preference model while protecting user data privacy and ensuring the normal operation of the system when anomalies occur in protocol conversion.

[0099] Build a federated learning-driven user preference update model:

[0100] A preference prediction network is trained locally on each smart travel terminal device. The terminal continuously collects user travel behavior data on the device, such as user preferences for different travel destinations, hotels, and transportation methods, their evaluations of various travel services, and the amount of time spent browsing specific travel information pages. A preference prediction network is constructed based on a neural network, using this collected user behavior data as input. By continuously adjusting the network parameters, the network learns the relationship between user behavior and potential travel preferences, thereby predicting the user's future travel preferences. For example, if a user frequently books hotels in a coastal city and browses a lot of information about water activities, the network may predict that the user will be more inclined to choose seaside vacation travel services.

[0101] The aggregation server is responsible for fusing the local model parameters uploaded by each terminal. Once a terminal completes local model training, it uploads the trained model parameters to the aggregation server. The server uses specific fusion strategies, such as weighted averaging, to assign different weights to each terminal's model parameters based on factors such as the quality and quantity of each terminal's data, and then performs fusion calculations. To protect user data privacy, differential privacy noise is added to the fusion process. Differential privacy noise is a carefully designed random perturbation that, while ensuring the usability of the fused model, increases the difficulty for attackers to infer user privacy data from the model parameters. After fusion and noise addition, the global model parameters are updated and distributed to each terminal. The terminal uses the updated global model parameters to optimize its local preference prediction network, thereby improving the accuracy of its predictions of user travel preferences.

[0102] Design a fault-tolerant protocol adaptation engine:

[0103] During system operation, the fault-tolerant protocol adaptation engine monitors the protocol conversion process in real time. Once a protocol conversion anomaly is detected, a processing flow based on historical successful conversion records is immediately initiated. The system pre-stores a record of each successful protocol conversion, including the protocol type before and after the conversion, related parameters, and key information from the conversion process. Based on these historical records, the similarity between the current abnormal request and the historical successful requests is calculated to construct a similarity graph. The nodes in the graph represent different historical successful requests, and the weights of the edges between the nodes reflect the degree of similarity between the corresponding requests. When calculating similarity, algorithms such as cosine similarity and Euclidean distance can be used to compare key elements of the request, such as protocol fields and service parameters.

[0104] The K-nearest neighbor algorithm is used to match the optimal alternative conversion path in the similarity graph. The K-nearest neighbor algorithm will search the graph for the K historical successful requests that are most similar to the current abnormal request. From the conversion paths corresponding to these K requests, the optimal alternative path is selected based on certain evaluation criteria. The evaluation criteria may include the complexity of the conversion path, the historical conversion success rate, the degree of adaptation to the current request, etc. For example, if a conversion path has successfully converted similar requests many times in history and has low complexity, then it is more likely to be selected as the optimal alternative path. After selecting the optimal alternative conversion path, the system uses this path to process the current abnormal request, ensuring that the service can continue when an abnormality occurs in the protocol conversion, thereby improving the reliability and stability of the system.

[0105] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0106] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-protocol compatible smart travel terminal AI service self-adaptation method, characterized in that: The steps include: Step T1: Collect travel service request data in a multi-protocol environment, including protocol type, service parameters, user context information, and terminal hardware resource status; Step T2: Based on the dynamic protocol conversion middleware, a multi-protocol adaptation model is constructed to unify heterogeneous protocol requests into standardized service instructions through the protocol parsing tree and semantic mapping rules; Step T3: Use the reinforcement learning framework to design a dynamic service resource allocation strategy. Define the state space as the real-time resource occupancy rate and service queue length, the action space as the resource allocation ratio, and the reward function as the weighted inverse of service response time and resource consumption. Step T4: Predict user travel needs based on the hidden Markov model, calculate the temporal distribution of future service requests through the state transition probability matrix, and dynamically adjust the service recommendation priority; Step T5: Build a multimodal interaction adaptation engine, integrate voice, text, and image inputs, use the attention mechanism to fuse multimodal features, and generate a unified service response.

2. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: The step T2 further comprises: Step T21: define the protocol parsing tree nodes as the syntax structure of the protocol fields, the edges as the dependencies between the fields, and generate the protocol abstract syntax tree through depth-first traversal; Step T22: Establish a semantic mapping rule base to map each protocol field to the semantic label of the standardized service instruction. The rule format is: Map(F proto ,C context )→S std Among them, F proto is the protocol field, C context is the context constraint, S std is the target instruction, Map is the mapping function; Step T23: A dynamic programming algorithm is used to optimize the protocol conversion path, and the objective function is a linear combination of conversion delay and semantic consistency.

3. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: In step T3, the specific implementation of the reinforcement learning framework includes: Step T31: Define the state vector s t =[R t ,Q t ], where R t is the real-time occupancy rate of CPU, memory and network bandwidth, Q t The number of pending requests in each service queue; Step T32: The action space is the resource allocation ratio vector a t =[α1,α2,…,α n ],satisfy Step T33: The reward function is designed as: Among them, T resp is the average response time, E res is the total resource consumption, λ1,λ2 are weight coefficients.

4. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: In step T4, the construction of the hidden Markov model includes: Step T41: define the hidden state as the user's travel intention category and the observed state as the historical service request sequence; Step T42: Estimate the state transition matrix A and observation probability matrix B using the Baum-Welch algorithm. The calculation formula is: Among them, A ij is the state transition probability, which represents the probability of transitioning from hidden state i to state j, ξ t (i, j) is the probability of transitioning from state i to j at time t, γ t (i) is the probability of being in state i at time t, and T is the total time step of the observation sequence; Step T43: Use the Viterbi algorithm to decode the most likely state sequence for the next k steps and generate a service recommendation priority list.

5. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: In step T5, the specific steps of multimodal feature fusion include: Step T51: Use a pre-trained convolutional neural network to extract image features, a bidirectional long short-term memory network to extract text features, and a mel-spectrogram convolutional network to extract speech features; Step T52: Calculate the cross-modal attention weight matrix through the multi-head attention mechanism and dynamically weight the image, text and speech features; Step T53: concatenate the weighted multimodal feature vectors and input them into the fully connected layer to generate a unified service response instruction.

6. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: Also includes: Step T6: Design a protocol compatibility verification module, use the temporal logic formulas in the formal method to describe the protocol interaction constraints, and use the model checking tool to verify whether the converted service instructions meet the security and liveness properties.

7. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: Also includes: Step T7: Implement the service priority dynamic scheduling algorithm, define the service level agreement weight matrix, calculate the priority score in real time according to the deadline, resource requirements and service level agreement weight of each request, and generate a processing queue by arranging the scores in descending order.

8. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: Also includes: Step T8: Build a federated learning-driven user preference update model. By training the preference prediction network locally on each terminal, the aggregation server fuses the local model parameters and adds differential privacy noise to update the global model parameters.

9. The multi-protocol compatible smart travel terminal AI service self-adaptation method according to claim 1, characterized in that: Also includes: Step T9: Design a fault-tolerant protocol adaptation engine. When a protocol conversion anomaly is detected, a similarity graph is constructed based on historical successful conversion records, and the K-nearest neighbor algorithm is used to match the optimal alternative conversion path.

Images