Automatic data management system and method based on AI engine

Abstract

The invention discloses an automatic data management system and method based on an AI engine, and relates to the technical field of data management. The method comprises the following steps: a hospital guide robot starts an AI service engine; the hospital guide robot guides the user to operate on the hospital guide robot through voice; the hospital guide robot records the use operation of the user or inquires the illness state of the user to generate feature words; calculating a user portrait feature value for a corresponding user in each feature operation record, and locking the feature user; and obtaining a destination intended by the user, and recommending an optimal travel route of the destination to each feature user by an AI service engine. The AI service engine is started through the hospital guide robot, inquiry operation between the user and the hospital guide robot is achieved through the AI service engine, the feature operation, the feature words and the user portrait feature values are obtained to judge the destination of the user, path navigation information is directly generated, the accuracy and efficiency of the hospital guide system are improved, and the user experience is improved. The illness state of the user and the consulting room registration can be accurately identified.

Description

Technical Field

[0001] The present invention belongs to the technical field of data management, and particularly relates to an automated data management system and method based on an AI engine. Background Art

[0002] The AI service engine is mainly used to provide diversified scenario services, making the device more intelligent and enhancing the user's experience of using the machine. AI refers to artificial intelligence, which is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence. Artificial intelligence is a branch of computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. The research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems, etc.

[0003] The basic environment and hospital orientation sign system of most modern hospitals are in a state of continuous optimization and development. Among them, the navigation system can help first-time hospital visitors quickly understand the overall layout of the hospital, quickly find out which department they should register for, quickly locate the corresponding department, and receive doctor's diagnosis and treatment in a timely manner; however, the intelligent navigation function of the current navigation system is not perfect enough to accurately identify the user's condition and needs. It is necessary to fully consider the hospital layout and the situation of different patients, conduct careful research, continuously improve and modify, and finally determine the plan. Summary of the Invention

[0004] The purpose of the present invention is to provide an automated data management system and method based on an AI engine. By starting the AI service engine through a navigation robot, using the AI service engine to realize the inquiry operation between the user and the navigation robot, obtaining user information to judge the user's destination and directly generating path navigation information, the problem that the existing hospital navigation system has imperfect functions and cannot accurately identify the user's condition and needs is solved.

[0005] To solve the above technical problems, the present invention is realized through the following technical solutions:

[0006] The present invention is an automated data management method based on an AI engine, including the following steps:

[0007] Step S1: Create an AI service engine based on hospital basic information, medical staff information text, and medical literature knowledge;

[0008] Step S2: When the operation time of the user on the navigation robot is greater than the time threshold or the navigation robot finds that there is a user wandering, the navigation robot starts the AI service engine;

[0009] Step S3: The triage robot has an intelligent conversation with the user and provides voice guidance for the user to operate on the triage robot;

[0010] Step S4: The triage robot records the user's operation or asks the user about their condition, records the keywords of the user's answer, and generates feature words;

[0011] Step S5: Based on the feature words, extract the user's operation sequence for feature analysis, calculate the user portrait feature values for the user corresponding to each feature operation record, and lock in on the feature user;

[0012] Step S6: Obtain the destination of the user's intention. The AI service engine recommends the best travel route to the destination for each feature user and generates a route QR code for display;

[0013] Step S7: The user scans the QR code with a mobile intelligent terminal to view the travel route in real time.

[0014] As a preferred technical solution, in step S1, a hospital triage pre-training model is constructed using the hospital's basic information, medical staff information text, and medical literature knowledge. Based on the pre-training model, an AI engine based on medical knowledge is constructed by combining the marked corpus in the medical literature.

[0015] As a preferred technical solution, in step S2, the triage robot moves within a designated area of the hospital building, continuously collects and observes the surrounding user images through a high-definition camera, and evaluates the user's movement trajectory. The specific evaluation process of the user's movement trajectory is as follows:

[0016] Step S21: Preprocess and extract features from the user trajectory data collected by the high-definition camera;

[0017] Step S22: Extract and classify the motion features of the user trajectory. The motion features focus on determining whether it is a wandering;

[0018] Step S23: Extract and classify the morphological features of the user trajectory. The morphological features focus on determining the category of the trajectory;

[0019] Step S24: Train a trajectory classification model through an LSTM neural network;

[0020] Step S25: Use the trained trajectory classification model to classify and judge the user's movement trajectory.

[0021] As a preferred technical solution, in step S12, when extracting the motion features, the user is estimated as a sequence of points, tgj = <P 1 , P 2 ,..., P i,...,P n >, P i is the i-th point, denoted as (x i , y i , t i ), where x i is the abscissa of the i-th point, y i is the ordinate of the i-th point, and t i is the timestamp of the i-th point; then the user speed v i , acceleration a i and curvature s i are calculated by the following formulas:

[0022]

[0023] In the formula, dist(P i , P i+1 ) is the Euclidean distance between P i and P i+1 , Δt i is the event interval between the i-th moment and the (i + 1)-th moment, dierct is the direction, turnAng is the turning angle. The direction is the moving direction between consecutive sampling points, which is represented by the angle between the direction and the basic direction (such as the due north direction). The turning angle can be obtained by calculating the difference between consecutive directions; l i+1 -l i represents the moving distance of the user between the i-th moment and the (i + 1)-th moment, L i+1 -L i represents the straight-line distance between the user's position at the i-th moment and the position at the (i + 1)-th moment. The curvature s i is used to judge the degree of curvature of the walking path.

[0024] As a preferred technical solution, in step S23, when extracting morphological features, trajectory scanning grid analysis is adopted, that is, the target trajectory is segmented by a grid of size b×b. The grid is represented in coordinate form, and each grid is marked with coordinates (e, d). All points (x, y) on the trajectory are mapped into the grid conversion formula; the grid conversion formula is:

[0025]

[0026] Since the behavior of a wandering user is usually relatively chaotic, in order to improve the accuracy of judging whether a user is wandering, a trajectory scanning grid analysis method is introduced, that is, the target trajectory is segmented by a grid of size b×b. The morphological features after gridification include: n e,d represents the number of trajectory points falling into the grid (e, d), reflecting the degree of overlap of the trajectory; n naclid represents the number of grids where n e,d > 0, that is, the number of effective grids; ne,d >i represents the number of grids in the effective grid with a count greater than i, where i = 1, 2, 3. Emphasize the trajectory cases with a higher degree of overlap to determine the target suspicious wandering area; s is the approximate area of the trajectory, and this feature can estimate the overall proportion size; n center represents the number of non-effective grids in the 3×3 area near the center point, and this feature is used to judge the trajectory shape.

[0027] As a preferred technical solution, in step S3, the guiding robot calls the speech recognition API through the AI engine for speech recognition. After processing the recognized text, corresponding information is extracted from the database according to the set rules and fed back in the form of characters or speech. According to the Q&A dialogue with the user, the timing relationship and law between the doctor and the user are obtained. The specific processing flow is as follows:

[0028] Step S31: Record the interaction information between the user and the guiding robot according to their conversation, specifically including: event type and timestamp;

[0029] Step S32: Clean the data to remove incomplete or incorrect data;

[0030] Step S33: Encode the event type;

[0031] Step S34: Build a timing model. Define a time window according to the user's purpose, build a state transition matrix, and build a time series model using the collected and encoded data; Let the state set be S = {s 1 , s 2 ,..., s m}, then the state transition matrix M is an m×m matrix; then the matrix element M i,j transfers from state s i to state s j The probability calculation formula is as follows:

[0032]

[0033] Step S35: Obtain the frequent sequence patterns in the interaction events between the user and the doctor, and analyze the trend between events; Let the time series data be {y t}, and use the linear regression model y t = β 0 + β 1 t + α t for fitting. In the formula, β 0 is the intercept, β 1 is the slope, and α t is the error term; then:

[0034]

[0035] As a preferred technical solution, in step S5, during the interaction between the user and the medical guidance robot, the age range division, gender category, regional division, visit frequency, department preference, appointment time preference, registration method, disease type distribution, disease severity trend, occupation visit time, and medical insurance type are obtained; then, for the user corresponding to each feature operation record, the calculation formula for the user's facial feature value is as follows: Assume that the user portrait has n features, and the feature value is FV i (i = 1, 2,..., n), then the corresponding weight is Then the total user portrait feature value FV total The calculation formula is:

[0036]

[0037] As a preferred technical solution, in step S6, the AI service engine digitally processes the hospital's building layout in advance, and marks the floor plan, department distribution, and passage connection relationship, abstracting the hospital's building structure into a graph data structure, where departments, elevators, stairs, and corridor intersections are used as nodes, and the passages connecting the nodes are used as edges, and the shortest path from the user's location to the target department is found in the graph.

[0038] The present invention is an automated data management system based on an AI engine, including a hospital layout generation module, a user path recognition module, an intelligent question - answering module, a feature operation record extraction module, a feature word extraction module, a user portrait feature value calculation module, and a path navigation module, characterized in that:

[0039] The hospital layout generation module is used to digitally process the hospital's building layout and mark the floor plan, department distribution, and passage connection relationship; the user path recognition module is used to judge the user's movement path; the intelligent question - answering module is used for interaction between the medical guidance robot and the user; the feature operation record extraction module is used to extract records of the user's operations on the medical guidance robot; the feature word extraction module is used to extract keywords from the records of the user's disease inquiries; the user portrait feature value calculation module is used to extract feature values according to the user's age, gender, disease type distribution, and severity trend; the path navigation module is used to generate a navigation path according to the user's destination.

[0040] The present invention has the following beneficial effects:

[0041] (1) By starting the AI service engine through the medical guidance robot, the present invention uses the AI service engine to implement the inquiry operation between the user and the medical guidance robot, obtains feature operations, feature words, and user portrait feature values to judge the user's destination and directly generate path navigation information, improving the accuracy and efficiency of the hospital's medical guidance system, and accurately identifying the user's condition and the registration consulting room.

[0042] (2) The present invention enables the guiding robot to identify the walking paths of surrounding users, determine whether the users are wandering around, promptly detect users in difficulty and approach them to inquire, thereby enhancing the functionality and intelligence of the guiding robot.

[0043] (3) After determining the destination that the user needs to go to, the present invention combines the generated hospital building layout map to generate the shortest path from the user's location to the target department, which is convenient for the user to view. A QR code is also generated, enabling the user to directly view the 3D path navigation by scanning the QR code with a mobile terminal, thus improving the user experience.

[0044] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for describing the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 It is a flowchart of an automated data management method based on an AI engine according to the present invention;

[0047] Figure 2 It is a schematic structural diagram of an automated data management system based on an AI engine according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0049] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] To make the objectives, technical solutions, and advantages of the present application clearer, the following will Figure 1 further describe the embodiments of the present application in detail.

[0051] Before introducing the embodiments of the present application, relevant descriptions of the AI engine technology are first made.

[0052] An AI engine is a software system for implementing artificial intelligence functions, covering multiple aspects such as data collection, processing, analysis, model training, and inference. Its aim is to help developers quickly develop and deploy AI applications, and improve the performance, scalability, and maintainability of AI applications. The following is some common information about AI engines:

[0053] Main functions:

[0054] (1) Data processing and analysis: The AI engine can collect and preprocess large amounts of data, including operations such as data cleaning, feature extraction, and data normalization, in order to provide high-quality data for subsequent model training. At the same time, it can also conduct in-depth analysis of the data to discover potential patterns and regularities in the data.

[0055] (2) Model training: It supports various machine learning and deep learning algorithms, such as neural networks, decision trees, support vector machines, etc. Through learning and training on large amounts of data, the model can automatically identify and understand the patterns and regularities in the data, thereby achieving prediction and classification of unknown data.

[0056] (3) Inference and prediction: After the model training is completed, the AI engine can use the trained model to perform inference and prediction on new data, such as tasks like image recognition, speech recognition, and natural language processing. It can quickly generate corresponding output results based on the input data and has a certain degree of accuracy and reliability.

[0057] Application areas:

[0058] (1) Search engines: Search engines such as Baidu and Google use AI engines to understand users' search intentions and provide more accurate and personalized search results. AI search engines can not only analyze the keywords entered by users but also understand the context and semantics of the questions, thus better meeting users' search needs.

[0059] (2) Natural language processing: This includes machine translation, text generation, sentiment analysis, question-and-answer systems, etc. The AI engine can help computers understand and generate human language, enabling natural interaction between humans and machines. For example, intelligent voice assistants such as Siri and Xiaoai Tongxue are natural language processing applications based on AI engines.

[0060] (3) Image recognition and computer vision: Used for tasks such as image classification, object detection, face recognition, and image generation. The AI engine can, through learning from large amounts of image data, identify objects, scenes, and people in images and conduct corresponding analysis and processing. For example, it has extensive applications in fields such as security monitoring, autonomous driving, and medical image diagnosis.

[0061] (4) Intelligent recommendation system: Based on the user's behavior, interests, and preferences, the AI engine can provide personalized recommended content for the user, such as product recommendations, news recommendations, video recommendations, etc. By analyzing and mining user data, the intelligent recommendation system can predict the user's needs, improve user satisfaction, and increase the conversion rate of the platform.

[0062] In order to make the objectives, technical solutions, and advantages of this application clearer and more understandable, the following further details this application in conjunction with the Figure 1-2 accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0063] Embodiment 1

[0064] Please refer to Figure 1 as shown. The present invention is an automated data management method based on an AI engine, including the following steps:

[0065] Step S1: Create an AI service engine based on hospital basic information, medical staff information text, and medical literature knowledge;

[0066] Step S2: When the operation time of the user on the navigation robot is greater than the time threshold or the navigation robot detects that there is a user wandering, the navigation robot activates the AI service engine;

[0067] Step S3: The navigation robot has an intelligent conversation with the user and provides voice guidance for the user to operate on the navigation robot;

[0068] Step S4: The navigation robot records the user's operation or asks the user about their condition, records the keywords of the user's answer, and generates feature words;

[0069] Step S5: Based on the feature words, extract the user's feature analysis operation sequence, calculate the user portrait feature value for the user corresponding to each feature operation record, and lock it with the feature user;

[0070] Step S6: Obtain the destination of the user's intention, and the AI service engine recommends the best travel route to the destination for each feature user and generates a route QR code for display;

[0071] Step S7: The user scans the QR code through a mobile intelligent terminal to view the travel route in real time.

[0072] In step S1, a hospital navigation pre-training model is constructed based on hospital basic information, medical staff information text, and medical literature knowledge. On the basis of the pre-training model, an AI engine based on medical knowledge is constructed by combining the marked corpus in medical literature.

[0073] The pre-trained model mainly consists of four levels: the input layer (Input), the embedding layer (Embedding), the network layer (TransformerEncoder), and the target layer (Target).

[0074] Among them, the input layer: mainly corresponds to the construction work of the pre-trained corpus, giving full play to the advantages of the original data of the medical literature information center. The original Chinese model of BERT, BERT-base-Chinese, uses the Chinese Wikipedia corpus. For the pre-trained model of hospital triage in this paper, the pre-trained data corpus collected and sorted includes: ① Chinese medical literature abstracts; ② full texts of Chinese medical papers; ③ network information texts in specific medical fields; ④ relevant book texts in specific medical fields; ⑤ relevant term lists and knowledge bases in specific medical fields. On this basis, a pre-trained corpus covering all fields and pre-trained corpora in the fields of biomedicine, physics, and chemistry are constructed.

[0075] The embedding layer: is mainly used for the feature description of the original corpus. Based on BERT, more external knowledge features are fused by adding a domain vocabulary and POS tags.

[0076] The network layer: consists of a complex Transformer neural network to achieve automatic learning. By improving the Mask matrix, the attention learning mechanism in Transformer is improved to enhance the learning effect.

[0077] The target layer: completes the model training by designing the training objectives of the language model. For different corpus characteristics, multiple training objectives are designed to complete multi-task learning.

[0078] The rich medical knowledge in medical data can be easily converted into labeled corpus for machine learning. By constructing the corresponding relationships such as data - department, data - registration, data - hospital, data - classification number, data - keyword, abstract - move, sentence - definition, etc., the training corpus for deep learning can be formed. Starting from the pre-trained language model, a classification layer is added to the model to form the corresponding deep learning model. By adopting architectures such as microservices, the rapid invocation of the model can be realized. Thus, for any scientific and technological materials, real-time recommendation interfaces for hospitals, departments, doctors, etc. can be provided, and a medical data knowledge engine for practical applications can be built.

[0079] In step S2, the triage robot moves within the designated area of the hospital building, continuously collects and observes the surrounding user images through a high-definition camera, and evaluates the user's movement trajectory. The specific evaluation process of the user movement trajectory is as follows:

[0080] Step S21: Preprocess and extract features from the user trajectory data collected by the high-definition camera;

[0081] Step S22: Extract and classify the motion features of the user trajectory. The motion features focus on determining whether it is a wandering state.

[0082] Step S23: Extract and classify the morphological features of the user trajectory. The morphological features focus on determining the type of the trajectory.

[0083] Step S24: Train a trajectory classification model through an LSTM neural network.

[0084] Step S25: Use the trained trajectory classification model to classify and judge the user's movement trajectory.

[0085] In step S22, when extracting the motion features, the user is estimated as a sequence of points, tgj = <P 1 , P 2 ,..., P i ,..., P n . P i is the i-th point, expressed as (x i , y i , t i ), where x i is the abscissa of the i-th point, y i is the ordinate of the i-th point, and t i is the timestamp of the i-th point. Then the calculation formulas for the user speed v i , acceleration a i , and curvature s i are as follows:

[0086]

[0087] In the formula, dist(P i , P i+1 ) is the Euclidean distance between P i and P i+1 . Δt i is the time interval between the i-th moment and the (i + 1)-th moment. Dierct is the direction, and turnAng is the turning angle. The direction is the moving direction between consecutive sampling points, which is represented by the angle between the direction and the basic direction (such as the due north direction). The turning angle can be obtained by calculating the difference between consecutive directions; l i+1 -l i represents the moving distance of the user between the i-th moment and the (i + 1)-th moment, and L i+1 -L i represents the straight-line distance between the user's position at the i-th moment and the position at the (i + 1)-th moment. The curvature s i is used to judge the degree of curvature of the walking path.

[0088] In step S23, when extracting the morphological features, trajectory scanning grid analysis is used, that is, the target trajectory is divided by a grid of size b×b, the grid is represented in the form of coordinates, each grid is marked with coordinates (e, d), and all points (x, y) on the trajectory are mapped to the grid conversion formula; the grid conversion formula is:

[0089]

[0090] Since wandering user behavior is usually more changeable, in order to improve the accuracy of determining whether they are wandering, a trajectory scanning grid analysis method is introduced, that is, the target trajectory is segmented using a grid of size b×b. The morphological features after gridding include: e,d Indicates the number of trajectory points that fall into the grid (e, d), reflecting the degree of trajectory overlap; n naclid Indicates n e,d The number of grids > 0 is the effective number of grids; n e,d >i represents the number of grids with counts greater than i in the valid grid, where i = 1, 2, 3, emphasizing the situation of tracks with a high degree of overlap and determining the target's suspected wandering area; s is the approximate area of ​​the track, and this feature can estimate the overall proportion; n center Indicates the number of invalid grids in the 3×3 area near the center point. This feature is used to determine the trajectory shape.

[0091] In step S24, the wandering behavior LSTM model designed in this paper has the dimension of the input layer, which is the velocity, acceleration, curvature, direction and angle feature vector obtained within 150 seconds. The temporal features are extracted through several LSTM layers, and the classification layer outputs the classification result (wandering / non-wandering). When classifying trajectories, the trajectory obtained by the trajectory screening model is used as input to perform fine-grained classification of the trajectory, and the trajectory morphology is considered more. The wandering behavior trajectory of the elderly with dementia has different morphologies, but combined with the existing wandering behavior of the elderly with dementia in nursing homes and the architectural design requirements, this paper mainly studies four wandering patterns, including elliptical trajectory, round-trip trajectory, elliptical-round-trip trajectory, and elliptical-round-trip-elliptical trajectory. The trajectory classification model uses trajectory morphological features for training to explore the morphological differences between different wandering trajectories.

[0092] In step S3, the guidance robot calls the speech recognition API through the AI ​​engine to perform speech recognition. After processing the recognized text, it extracts the corresponding information from the database according to the set rules and gives feedback in the form of characters or voice. It obtains the temporal relationship and rules between the doctor and the user based on the question-and-answer dialogue with the user. The specific processing flow is as follows:

[0093] Step S31: Record the interaction information between the user and the triage robot according to their conversation, specifically including: event type (such as diagnosis, prescribing medicine, surgery, etc.) and timestamp accurate to the date, hour, or even minute, depending on the requirements. For example, record that user A had a first diagnosis at 9 am on January 1, 2024, and doctor B made a preliminary diagnosis of a cold; at 3 pm on January 3, user A had a follow-up visit, and doctor B adjusted the treatment plan, etc.

[0094] Step S32: Clean the data to remove incomplete or incorrect data. For example, some records may have incorrect timestamps or ambiguous event descriptions (such as only writing "treatment" without specifying the treatment method), and these data need to be corrected or deleted.

[0095] Step S33: Encode the event types. For example, the diagnosis event is encoded as 1, the prescribing medicine event is encoded as 2, the surgery event is encoded as 3, etc.

[0096] Step S34: Build a time series model. Define the time window according to the user's purpose. If it is to study the short-term treatment effect, it may be in days; if it is to study the long-term management of chronic diseases, it may be in months or years. For example, for the study of acute disease treatment, set a 7-day time window, observe the doctor's treatment behavior and the user's response within this time period, build a state transition matrix, assuming there are a finite number of states (such as stable condition, improved condition, deteriorated condition, etc.), and build a state transition matrix based on the collected data. For example, within the time window, the probability of transitioning from the stable condition state to the improved condition state can be determined by statistically analyzing the proportion of users in the stable condition who improved after a certain treatment. This matrix can help analyze the temporal relationship between the doctor's treatment measures and the change of the user's disease state, and build a time series model using the collected and encoded data. For example, for a user's treatment process, an event sequence such as [1 (first diagnosis), 2 (prescribing medicine), 1 (follow-up visit), 3 (adjusting medicine)] may be formed. By analyzing the event sequences of a large number of users, some common patterns can be found. For example, it is found that for a certain disease, most users' event sequences are first a first diagnosis, then prescribing medicine, and then a follow-up visit to adjust the medicine. Let the state set be S = {s 1 , s 2 ,..., s m}. Therefore, we set the state transition matrix M to be an m×m matrix; then the matrix element M i,j from state s i transitioning to state s j has the following probability calculation formula:

[0097]

[0098] Step S35: Obtain the frequent sequence patterns in the user-doctor interaction events and analyze the trends between the events; Given the time series data {y t}, use the linear regression model y t = β 0 + β 1 t + α t for fitting, where β 0 is the intercept, β 1 is the slope, and α t is the error term; then:

[0099]

[0100] In Step S5, obtain the age range division, gender category, regional division, visit frequency, department preference, appointment time preference, registration method, disease type distribution, disease severity trend, occupation visit time, and medical insurance type of the user during the interaction with the triage robot; then the calculation formula for the user's facial feature value corresponding to each feature operation record is as follows: Given that the user portrait has n features with feature values FV i (i = 1, 2,..., n), then the corresponding weights are Then the total user portrait feature value FV total The calculation formula is:

[0101] In Step S6, the AI service engine digitally processes the hospital's building layout in advance and marks the floor plan, department distribution, and passage connection relationships. The hospital's building structure is abstracted into a graph data structure, where departments, elevators, stairs, and corridor intersections are used as nodes, and the passages connecting the nodes are used as edges. Find the shortest path from the user's location to the target department in the graph.

[0102] After the AI service engine calculates the path planning, it also combines other situations, specifically as follows:

[0103] (1) People flow and congestion situation:

[0104] Set up people flow monitoring devices (such as cameras combined with intelligent analysis software, infrared sensors, etc.) in the hospital to collect real-time people flow information in various passages, elevators, department waiting areas, etc.

[0105] A congestion coefficient can be introduced for path calculation. For example, when the people flow in a certain corridor exceeds a certain threshold, the congestion coefficient of this path will increase. In the path search algorithm, in addition to considering the distance factor, the impact of the congestion coefficient on the path also needs to be considered. For example, a path with a slightly longer distance but less people flow may be better than a path with a shorter distance but congestion.

[0106] (2) Elevator operation situation:

[0107] Connect to the data interface of the hospital elevator system to obtain information such as the current position, running direction, and passenger capacity of the elevator.

[0108] If the elevator is full or moving away from the current floor, then in path planning, it may consider guiding the user to take the stairs or wait for the next elevator, and at the same time calculate the time cost of waiting for the elevator and add it to the total path time estimation.

[0109] After the path is generated, the system obtains the user's current position (which can be determined through the hospital's internal positioning system, such as Bluetooth positioning, Wi-Fi positioning, etc., or let the user enter their location at the consultation terminal, such as "hospital lobby") and the location of the target department (based on the department that the user needs to go to as judged before). Based on the graph model and path search algorithm, combined with real-time factors and personalized factors, calculate the optimal path from the starting point to the ending point. For example, use Dijkstra's algorithm. Start from the starting point, gradually expand the search range, calculate the shortest path passing through each node until the node corresponding to the target department is found. Present the generated path to the user in an intuitive way. It can be to display a roadmap on the hospital's electronic consultation screen, using arrows and text to guide the direction (such as "Go straight 50 meters, turn left into the corridor, take the No. 2 elevator to the 3rd floor"); it can also send voice navigation or navigation information with a map through a mobile application.

[0110] Embodiment 2

[0111] Refer to Figure 2 As shown, the present invention is an automated data management system based on an AI engine, which can be used to execute the method content of Embodiment 1 of the present invention, including: a hospital layout generation module, a user path recognition module, an intelligent question-and-answer module, a feature operation record extraction module, a feature word extraction module, a user portrait feature value calculation module, and a path navigation module;

[0112] The hospital layout generation module is used to digitally process the hospital's architectural layout and mark the floor plan, department distribution, and passage connection relationship; the user path recognition module is used to judge the user's movement path; the intelligent question-and-answer module is used for interaction between the consultation robot and the user; the feature operation record extraction module is used to extract records of the user's operations on the consultation robot; the feature word extraction module is used to extract keywords from the records of the user's disease inquiries; the user portrait feature value calculation module is used to extract feature values according to the user's age, gender, disease type distribution, and severity trend; the path navigation module is used to generate a navigation path according to the user's destination.

[0113] It should be noted that in the above system embodiments, the included units are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0114] In addition, those of ordinary skill in the art can understand that all or part of the steps in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0115] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An automated data management method based on an AI engine, characterized in that: The steps include: Step S1: Create an AI service engine based on basic hospital information, medical information text, and medical literature knowledge; Step S2: When the user operates the guidance robot for a time longer than the time threshold or the guidance robot finds that the user is wandering, the guidance robot starts the AI ​​service engine; Step S3: The medical guidance robot conducts an intelligent dialogue with the user and voice-guides the user to operate on the medical guidance robot; Step S4: The medical guidance robot records the user's operation or inquires about the user's condition, records the user's answer keywords, and generates feature words; Step S5: extracting users based on feature words to perform feature analysis using operation sequences, calculating user portrait feature values ​​for users corresponding to each feature operation record, and locking with feature users; Step S6: Obtain the destination intended by the user, and the AI ​​service engine recommends the best route to the destination for each characteristic user, and generates a route QR code for display; Step S7: The user scans the QR code through a mobile smart terminal to view the route in real time.

2. The AI ​​engine-based automated data management method according to claim 1, characterized in that: In step S1, a hospital guidance pre-training model is constructed based on hospital basic information, medical information text and medical literature knowledge, and an AI engine based on medical knowledge is constructed based on the pre-training model and combined with the marked corpus in the medical literature.

3. The AI ​​engine-based automated data management method according to claim 1, characterized in that: In step S2, the medical guide robot moves in the designated area of ​​the hospital building, continuously collects and observes images of surrounding users through a high-definition camera, and evaluates the user's movement trajectory. The specific evaluation process of the user's movement trajectory is as follows: Step S21: preprocessing and feature extraction of user trajectory data collected by the high-definition camera; Step S22: extracting and classifying the motion features of the user trajectory; Step S23: extracting and classifying the morphological features of the user trajectory; Step S24: training a trajectory classification model through an LSTM neural network; Step S25: Use the trained trajectory classification model to classify and judge the user's movement trajectory.

4. The AI ​​engine-based automated data management method according to claim 2, characterized in that: In step S22, when extracting motion features, the user estimate is represented as a point sequence, tgj = P1, P2, ..., P i ,...,P n , P i is the i-th point, expressed as (x i ,y i ,t i ), where x i is the horizontal coordinate of the i-th point, y i is the ordinate of the i-th point, t i is the timestamp of the i-th point; then calculate the user speed v i , acceleration a i and the curvature s i The calculation formula is: In the formula, dist(P i ,P i+1 ) is P i , P i+1 The Euclidean distance between i is the event interval between time i and time i+1, dierct is the direction, turnAng is the turning angle, l i+1 -l i represents the distance the user moves between time i and time i+1, L i+1 -L i Indicates the straight-line distance between the user's location at time i and the user's location at time i+1.

5. The AI ​​engine-based automated data management method according to claim 2, characterized in that: In step S23, when extracting features based on morphological features, trajectory scanning grid analysis is used, that is, the target trajectory is segmented using a grid of size b×b, the grid is represented in the form of coordinates, each grid is marked with coordinates (e, d), and all points (x, y) on the trajectory are mapped to the grid conversion formula; the grid conversion formula is:

6. The AI ​​engine-based automated data management method according to claim 1, characterized in that: In step S3, the medical guidance robot calls the speech recognition API through the AI ​​engine to perform speech recognition. After processing the recognized text, it extracts the corresponding information from the database according to the set rules and gives feedback in the form of characters or voice. It obtains the temporal relationship and rules between the doctor and the user based on the question-and-answer dialogue with the user. The specific processing flow is as follows: Step S31: According to the dialogue between the user and the medical guidance robot, the interaction information between them is recorded, including: event type and timestamp; Step S32: Clean the data to remove incomplete or erroneous data; Step S33: Encode the event type; Step S34: construct a time series model, define a time window according to the user's purpose, construct a state transition matrix, and use the collected and encoded data to construct a time series model; let the state set be S = {s1, s2, ..., s m }, then the state transfer matrix M is an m×m matrix; then the matrix element M i,j From the state i Transfer to state s j The probability calculation formula is as follows: Step S35: Obtain frequent sequence patterns in the user-doctor interaction events and analyze the trends between events; suppose the time series data {y t }, using the linear regression model y t =β0+β1t+α t Fitting, where β0 is the intercept, β1 is the slope, and α t is the error term; then:

7. The AI ​​engine-based automated data management method according to claim 1, characterized in that: In step S5, the user's age range, gender category, region, frequency of visits, department preference, appointment time preference, registration method, disease type distribution, disease severity trend, occupational visit time, and medical insurance type are obtained during the user's interaction with the medical guidance robot; the user's facial feature value calculation formula corresponding to each feature operation record is as follows: Assume that the user portrait has n features, and the feature value is FV i (i=1,2,...,n), the corresponding weight is Then the total user portrait feature value FV total The calculation formula is:

8. The AI ​​engine-based automated data management system and method according to claim 1, characterized in that: In step S6, the AI ​​service engine digitizes the hospital's architectural layout in advance and marks the floor plan, department distribution, and channel connection relationships, abstracting the hospital's architectural structure into a graph data structure, in which departments, elevators, stairs, and corridor intersections are used as nodes, and channels connecting nodes are used as edges, and the shortest path from the user's location to the target department is found in the graph.

9. An automated data management system based on an AI engine, comprising a hospital layout generation module, a user path recognition module, an intelligent question-answering module, a feature operation record extraction module, a feature word extraction module, a user portrait feature value calculation module and a path navigation module, characterized in that: The hospital layout generation module is used to digitize the hospital's architectural layout and mark it with floor plans, department distribution, and channel connection relationships; the user path identification module is used to determine the user's movement path; the intelligent question-answering module is used for interaction between the medical guidance robot and the user; the feature operation record extraction module is used to record and extract the user's operations on the medical guidance robot; the feature word extraction module is used to extract keywords from the user's condition inquiry records; The user portrait feature value calculation module is used to extract feature values ​​according to the user's age, gender, disease type distribution and severity trend; The path navigation module is used to generate a navigation path according to a user's destination.

Images