Search

CN-122025182-A - Health management system and method based on multi-mode agent

CN122025182ACN 122025182 ACN122025182 ACN 122025182ACN-122025182-A

Abstract

The invention discloses a health management system and method based on a multi-modal intelligent agent, and relates to the technical field of artificial intelligence, wherein the health time-space diagram is input into a time-space diagram neural network model, and fusion and nonlinear conversion are carried out by combining a space dependency relationship and a time evolution mode between nodes to generate a high-risk prediction event and a key node abnormal mode; according to the high risk prediction event and the key node abnormal mode, a causal chain is generated by performing cross-modal causal structure learning and pathological mechanism reasoning through a causal discovery algorithm, a causal action scheme is generated by combining the causal chain and the high risk prediction event and performing counter-facts reasoning, and a multi-modal intelligent body monitors real-time physiological data and behavior data of a user according to the intervention action scheme and performs fusion analysis with the causal chain to generate a health assessment report. The invention realizes multi-mode data fusion and space-time dependence deep learning, provides high-precision characteristic representation, and improves the coverage and prediction accuracy of risk assessment dimensions.

Inventors

  • WANG YONG

Assignees

  • 北京新瑞时科技有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (10)

  1. 1. A health management method based on multi-mode agent is characterized by comprising the following steps, Collecting multi-mode health data and preprocessing, and constructing a health time-space diagram by taking the preprocessed multi-mode health data as a time-space diagram node; inputting a health space-time diagram into a space-time diagram neural network model, and combining a spatial dependency relationship and a time evolution mode among nodes to perform fusion and nonlinear conversion so as to generate a high-risk prediction event and a key node abnormal mode; according to the high risk prediction event and the key node abnormal mode, performing cross-modal causal structure learning and pathological mechanism reasoning through a causal discovery algorithm to generate a causal chain; combining a causal chain with a high risk prediction event, and performing inverse fact reasoning to generate an intervention action scheme; The multi-mode intelligent agent monitors real-time physiological data and behavior data of the user according to the intervention action scheme, and performs fusion analysis with a causal chain to generate a health assessment report.
  2. 2. The method for health management based on multi-modal intelligent agent as set forth in claim 1, wherein said multi-modal health data comprises physiological signal time series, behavioral data, environmental parameters, and subjective information; the preprocessing comprises data cleaning, format conversion, deduplication, normalization and outlier processing.
  3. 3. The health management method based on multi-modal intelligent agent as set forth in claim 2, wherein the steps of constructing a health time-space diagram using the preprocessed multi-modal health data as a time-space diagram node, Taking the index type of the preprocessed multi-mode health data as a space-time diagram node, and establishing a directed edge and a non-directed edge to generate a topological network structure; And embedding the preprocessed multi-mode health data into space-time diagram nodes in a topological network structure according to time sequence to generate a health space-time diagram.
  4. 4. The method for health management based on multi-modal agents as claimed in claim 3, wherein the step of inputting the health time space diagram into the time space diagram neural network model is as follows, Constructing a space-time diagram neural network model based on the diagram convolution layer, the time convolution layer, the full connection layer and the interlayer connection relation; and inputting the health space-time diagram into a space-time diagram neural network model diagram convolution layer, and generating a space characterization vector by aggregating characteristic information among space-time diagram nodes and learning space dependency relations among the nodes.
  5. 5. The method for health management based on multi-modal intelligent agent as set forth in claim 4, wherein the combining of spatial dependency and time evolution patterns between nodes, the fusing and nonlinear conversion, the generation of high risk prediction events and critical node anomaly patterns, is performed as follows, Inputting the new node characterization vector into a time convolution layer, and capturing a dynamic mode of the evolution of each health index along with time by applying a one-dimensional convolution kernel to generate a space-time fusion feature vector; Mapping the high-dimensional features to the low-dimensional space through the full-connection layer based on the space-time fusion feature vectors and performing weighted combination to generate risk discrimination feature representation; and marking high-risk prediction events through a softmax classifier according to the risk discriminant feature representation, and identifying the abnormal modes of the key nodes through a weight analysis mechanism.
  6. 6. The method for health management based on multi-modal agents as set forth in claim 5, wherein the step of generating a causal chain by performing cross-modal causal structure learning and pathological mechanism reasoning through a causal discovery algorithm based on high risk prediction events and key node anomaly patterns, Extracting the specific type of the high-risk prediction event and the attribute of the abnormal mode of the key node through a causal discovery algorithm, and generating a targeted causal analysis query instruction; executing a targeted causal analysis query instruction, retrieving similar historical case data from a historical health database, and generating a training data set; based on a training data set, learning a causal network structure among space-time diagram node variables by a continuous optimization method and combining priori medical knowledge as constraint conditions; combining the causal network structure with a medical knowledge base, carrying out reasoning verification, and identifying a complete causal path from a critical node abnormal mode to a high risk prediction event to generate a causal chain.
  7. 7. The method for multimodal intelligent agent-based health management as defined in claim 6 wherein said combining causal links with high risk prediction events, performing counterfacts reasoning, generating intervention action scenarios, steps are as follows, Converting node variables, directed edge relations and conditional probability distribution in the causal chain into a structural causal model diagram, and identifying key nodes with the greatest influence on high-risk prediction events in the causal chain; Simulating the change of conditional probability distribution of the dry prognosis of the key nodes through the do-operator operation, deducing the theoretical change value of the probability of the dry prognosis high-risk prediction event, and generating a counterfactual reasoning result; Based on the counterfactual reasoning result, searching an effective intervention measure set from a historical health database aiming at the key node, and generating a preliminary intervention measure candidate list; The preliminary intervention candidate list is classified into immediate execution measures, short-term adjustment plans, and long-term improvement schemes according to the degree and type of implementation urgency, and is integrated into an intervention action scheme.
  8. 8. The method for health management based on a multi-modal agent as set forth in claim 7, wherein the multi-modal agent monitoring the user's real-time physiological data and behavioral data based on the intervention action plan is based on the intervention action plan, the multi-modal agent initiating a high frequency data collection mode, monitoring the user's real-time physiological data and behavioral data, and generating a multi-modal time series dataset for performing a intervention.
  9. 9. The method for multimodal intelligent agent-based health management as defined in claim 8 wherein the fusion analysis with a causal chain generates a health assessment report by the steps of, Based on the multi-mode time sequence data set of the dry and dry state execution and the multi-mode health data before intervention, the multi-mode time sequence data set is combined with a causal chain to be compared, the improvement degree and the change trend of health indexes are obtained, and the results are integrated into causal path influence assessment results; and according to the causal path influence evaluation result, combining the health evaluation standard in the historical health database, carrying out multidimensional quantitative scoring on the current health state of the user, and integrating the current health state into a health evaluation report.
  10. 10. A health management system based on a multi-modal intelligent agent is characterized by comprising the health management method based on the multi-modal intelligent agent according to any one of claims 1-9, The space-time diagram construction module is used for acquiring the multi-modal health data and preprocessing the multi-modal health data, and constructing a health space-time diagram by taking the preprocessed multi-modal health data as space-time diagram nodes; the risk prediction module is used for inputting a health space-time diagram into a space-time diagram neural network model, combining a space-dependent relationship and a time evolution mode among nodes, and performing fusion and nonlinear conversion to generate a high-risk prediction event and a key node abnormal mode; The causal reasoning module is used for carrying out cross-modal causal structure learning and pathological mechanism reasoning through a causal discovery algorithm according to the high risk prediction event and the key node abnormal mode to generate a causal chain; the anti-facts reasoning module is used for combining the causal chain with the high risk prediction event to conduct anti-facts reasoning and generate an intervention action scheme; and the report generation module is used for monitoring the real-time physiological data and the behavior data of the user by the multi-mode agent according to the intervention action scheme, and carrying out fusion analysis with the causal chain to generate a health evaluation report.

Description

Health management system and method based on multi-mode agent Technical Field The invention relates to the technical field of artificial intelligence, in particular to a health management system and method based on a multi-mode intelligent agent. Background The explosive development of artificial intelligence technology provides a revolutionary technical path for health management, wherein the rapid iteration of the multi-mode artificial intelligence technology enables the system to seamlessly integrate multiple data types such as text, images, audio and physiological signals, and the comprehensiveness and accuracy of health assessment are improved. In recent years, a deep learning-based health management model advances in the fields of health risk prediction and early disease intervention, and the intelligent and data-driven process of health management is effectively promoted. The breakthrough development of the intelligent agent technology, especially the large-scale application of the large multi-mode intelligent agent (LMAs), further realizes the full-closed loop management from multi-source data perception and intelligent analysis to dynamic decision execution, and provides key support for the crossing of health management from experience driving to scientific decision. The current health management technology mainly faces two major core challenges, namely a health data processing mechanism is limited to single-mode analysis, multi-source heterogeneous data such as physiological signal time sequences, behavior data, environmental parameters and subjective information are difficult to effectively fuse, dimension coverage of health risk assessment is insufficient and prediction accuracy is limited, and the existing system lacks deep causal reasoning capability on health risks, cannot realize accurate intervention based on a pathological mechanism, only can provide surface layer risk early warning and is difficult to form closed loop management. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a health management method based on multi-mode agents, which solves the problems of multi-mode health data fusion and deep causal reasoning capability deficiency. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the present invention provides a multi-modal agent-based health management method, comprising, Collecting multi-mode health data and preprocessing, and constructing a health time-space diagram by taking the preprocessed multi-mode health data as a time-space diagram node; inputting a health space-time diagram into a space-time diagram neural network model, and combining a spatial dependency relationship and a time evolution mode among nodes to perform fusion and nonlinear conversion so as to generate a high-risk prediction event and a key node abnormal mode; according to the high risk prediction event and the key node abnormal mode, performing cross-modal causal structure learning and pathological mechanism reasoning through a causal discovery algorithm to generate a causal chain; combining a causal chain with a high risk prediction event, and performing inverse fact reasoning to generate an intervention action scheme; The multi-mode intelligent agent monitors real-time physiological data and behavior data of the user according to the intervention action scheme, and performs fusion analysis with a causal chain to generate a health assessment report. As a preferable scheme of the health management method based on the multi-mode intelligent agent, the multi-mode health data comprises a physiological signal time sequence, behavior data, environmental parameters and subjective information; the preprocessing comprises data cleaning, format conversion, deduplication, normalization and outlier processing. The method for health management based on multi-modal intelligent agent of the invention is an optimal scheme, wherein the preprocessed multi-modal health data is used as a space-time diagram node to construct a health space-time diagram, the steps are as follows, Taking the index type of the preprocessed multi-mode health data as a space-time diagram node, and establishing a directed edge and a non-directed edge to generate a topological network structure; And embedding the preprocessed multi-mode health data into space-time diagram nodes in a topological network structure according to time sequence to generate a health space-time diagram. As an optimal scheme of the health management method based on the multi-modal intelligent agent, the method inputs the health space-time diagram into the space-time diagram neural network model, comprises the following steps, Constructing a space-time diagram neural network model based on the diagram convolution layer, the time convolution layer, the full connection layer and the interlayer connectio