A method for identifying and conditioning improvement of traditional Chinese medicine constitution

Abstract

The application discloses a traditional Chinese medicine constitution identification conditioning improvement method and relates to the technical field of traditional Chinese medicine intelligent identification, and comprises the following steps: inputting a constitution state vector and an individual physiological prior vector into a pre-constructed constitution-physiology parameter-syndrome feature association graph, and outputting constitution identification data and constitution bias degree data corresponding to a subject; acquiring a core monitoring dimension according to the constitution identification data and the constitution bias degree data, and performing constraint clipping and standardization processing on the core monitoring dimension according to a preset safety boundary to obtain a standardized conditioning evaluation vector set matched with the constitution state of the subject; and inputting the standardized conditioning evaluation vector set and the constitution state vector into a continuum constitution evolution calculation unit together, and outputting a constitution evolution prediction trajectory set corresponding to each conditioning suggestion. The application realizes dynamic tracking of a constitution improvement process, constitution state mode classification updating and deviation feedback correction.

Description

Technical Field

[0001] This invention relates to the field of TCM intelligent identification technology, and in particular to a TCM constitution identification, conditioning and improvement method. Background Technology

[0002] With the development of TCM's concept of preventive medicine, wearable sensing technology, pattern classification technology, and intelligent health management technology, TCM constitution identification is gradually evolving from traditional questionnaires and manual judgment to multimodal data fusion, constitution characteristic pattern classification, and individualized analysis. Existing technologies have incorporated pulse signals, heart rate variability signals, skin electrical signals, tongue images, facial images, and medical records into constitution characteristic identification. Pattern classification methods are used to differentiate between different constitution states and imbalances, thereby improving the objectivity of TCM constitution identification and providing a data foundation for generating treatment recommendations.

[0003] However, existing TCM constitution identification and conditioning improvement techniques still have shortcomings: on the one hand, multimodal data mostly adopt simple splicing or independent discrimination methods, lacking correlation reasoning between constitution status, physiological parameters, syndrome characteristics and individual physiological baseline, making it difficult to accurately reflect individual constitution bias differences; on the other hand, conditioning suggestions are mostly based on fixed matching of constitution type, lacking a mechanism to dynamically correct the constitution evolution process by combining real retest data, resulting in insufficient individual adaptability and continuous feedback ability in the conditioning improvement process. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for identifying, regulating, and improving constitution in Traditional Chinese Medicine (TCM) to solve the problems of lack of individualized correlation reasoning in constitution identification and lack of dynamic feedback correction in the regulation and improvement process in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for TCM constitution identification, regulation, and improvement, comprising: collecting multimodal physiological signals and medical history text from subjects and performing feature alignment and dimensionality reduction mapping to obtain a constitution state vector and an individual physiological prior vector; inputting the constitution state vector and the individual physiological prior vector into a pre-constructed constitution-physiological parameter-syndrome feature association graph to output the constitution identification data and constitution bias degree data corresponding to the subject; obtaining core monitoring dimensions based on the constitution identification data and constitution bias degree data, and performing constraint pruning and standardization processing on the core monitoring dimensions according to a preset safety boundary to obtain a result corresponding to the subject's constitution. A standardized conditioning assessment vector set matching the physical condition is generated. The standardized conditioning assessment vector set and the physical condition vector are input into the continuous physical condition evolution calculation unit, and the set of physical condition evolution prediction trajectories corresponding to each conditioning suggestion is output. The expected improvement baseline curve is determined based on the set of physical condition evolution prediction trajectory, and the actual physical condition sequence is reconstructed based on continuously collected real retest data. The dynamic deviation residual between the actual physical condition sequence and the expected improvement baseline curve is calculated. The edge weight reset reliability of the physical condition-physiological parameter-syndrome feature correlation diagram and the parameters of the continuous physical condition evolution calculation unit are corrected based on the dynamic deviation residual.

[0007] As a preferred embodiment of the TCM constitution identification, conditioning and improvement method described in this invention, the multimodal physiological signals include pulse signals, heart rate variability signals, skin electrical signals, tongue images and facial images.

[0008] As a preferred embodiment of the TCM constitution identification, regulation, and improvement method described in this invention, the specific steps for obtaining the constitution state vector and the individual physiological prior vector are as follows: Multimodal physiological signals and medical records of subjects were collected and synchronously labeled and standardized according to subject identification and collection time to obtain a standardized multimodal dataset. Physiological signal features, image representation features, and medical history text features are extracted from a standardized multimodal dataset and mapped to a unified physical evidence space for alignment, resulting in an aligned set of physical evidence features. The dimensionality reduction of the aligned physical evidence feature set is fused to obtain the physical state vector, and the individual physiological prior vector is generated by combining the resting stable physiological features.

[0009] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method described in this invention, the specific steps for outputting the constitution identification data and constitution imbalance data corresponding to the subject are as follows: The physical state vector is used as the current physical evidence anchor point, and the individual physiological prior vector is used as the individual baseline anchor point. The current physical evidence anchor point and the individual baseline anchor point are respectively connected to the physical node, physiological parameter node and syndrome feature node in the physical constitution-physiological parameter-syndrome feature association graph to form an individualized physical constitution relationship graph. Based on the individualized constitution relationship diagram, the evidence matching strength between the current constitution evidence anchor point and each physiological parameter node is calculated, and the target physiological parameter node and target syndrome feature node related to the subject's current state are activated according to the evidence matching strength. Based on the target physiological parameter node and the target syndrome feature node, read the node relationship and edge weight between the target physiological parameter node, the target syndrome feature node and each candidate constitution node, and convert the offset between the individual baseline anchor point and the target physiological parameter node into the individual offset evidence edge. Individual offset evidence edges are used to individually correct node relationships and edge weights, and combined with the constitution bias discrimination rule to suppress conflicting propagation paths, forming a set of constitution relationship diffusion paths for subjects. The candidate physical nodes are subjected to multi-hop relation-aware reasoning using the set of physical relation diffusion paths, and the relation diffusion confidence of each candidate physical node is calculated. The candidate physical nodes are then sorted from high to low relation diffusion confidence. When at least two candidate physical nodes have the same relation diffusion confidence, they are sorted from high to low number of effective propagation paths. When the number of effective propagation paths is still the same, they are sorted from high to low average edge weight, resulting in a sorted sequence of candidate physical nodes. Based on the candidate constitution node ranking sequence, the dominant constitution node and the secondary constitution node are determined to form the constitution identification data corresponding to the subject. Then, according to the relationship diffusion confidence, individual deviation evidence edge strength and constitution bias discrimination rules corresponding to the dominant constitution node and the secondary constitution node, the bias degree value and bias level corresponding to each constitution node are determined to form the constitution bias degree data corresponding to the subject.

[0010] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method described in this invention, the specific steps for obtaining the core monitoring dimensions are as follows: The dominant constitution node, the secondary constitution node, and the relationship diffusion confidence are read from the constitution identification data. The bias degree value and bias level corresponding to each constitution node are read from the constitution bias degree data. The dominant constitution node, secondary constitution node, relationship diffusion confidence, bias degree value, and bias level are sorted and organized to form a constitution bias contribution data table. Using the constitution bias contribution data table, the dominant constitution nodes and the secondary constitution nodes are matched to the preset conditioning dimension library, and the bias contribution matching value of each conditioning dimension is calculated to form a set of candidate intervention dimensions. Based on the set of candidate intervention dimensions, the bias contribution matching value and bias level corresponding to each candidate intervention dimension are read respectively, and the candidate intervention dimensions are sorted according to the rule that the higher the bias contribution matching value and the more severe the bias level, the higher the priority. The candidate intervention dimensions that are ranked first and meet the preset selection requirements are determined as the core monitoring dimensions.

[0011] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method described in this invention, the specific steps for obtaining a standardized conditioning assessment vector set that matches the subject's constitution state are as follows: Extract the bias contribution matching value, bias level and priority from the core monitoring dimensions, and call the corresponding suggestion intensity range, suggestion frequency range and suggestion duration range from the adjustment dimension library to form the core monitoring dimension parameter table; Based on the core monitoring dimension parameter table, conditioning suggestion parameters are generated according to the bias contribution matching value and bias level, and the conditioning suggestion parameters are mapped to the safety boundary for constraint trimming. After the constraint trimming conditioning suggestion parameters are coded with unified dimensions, the priority of each core monitoring dimension is determined according to the bias contribution matching value from high to low and the bias level from severe to mild. Based on the priority, a standardized conditioning evaluation vector set that can characterize the conditioning direction, conditioning intensity and conditioning cycle is formed.

[0012] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method described in this invention, the specific steps for outputting the set of constitution evolution prediction trajectories corresponding to each conditioning suggestion are as follows: The intensity, frequency, and duration of each treatment suggestion are read from the standardized treatment assessment vector set and converted into a continuous treatment driving sequence according to the time axis of the simulation. By using a continuous conditioning driving sequence, each conditioning suggestion is concatenated with the physical condition vector, and combined with the observation period, continuous inference input data corresponding to each conditioning suggestion is generated. Based on continuous inference input data, the continuous physical evolution calculation unit continuously infers the changes in the physical state of the subjects on the inference time axis, forming a physical evolution state sequence. The trajectory is then organized according to the conditioning suggestion identifier and the inference time order to obtain the set of physical evolution prediction trajectories corresponding to each conditioning suggestion.

[0013] As a preferred embodiment of the TCM constitution identification, regulation, and improvement method described in this invention, the specific steps for determining the expected improvement baseline curve based on the constitution evolution prediction trajectory set are as follows: Read the physical evolution state sequence corresponding to each conditioning suggestion from the set of physical evolution prediction trajectories, and calculate the decrease in physical bias and trajectory smoothness corresponding to each physical evolution state sequence; Based on the magnitude of the decrease in physical bias and the smoothness of the trajectory corresponding to each physical evolution state sequence, the physical evolution state sequences are sorted and screened, and the physical evolution state sequences that meet the preset baseline determination requirements are determined as the expected improvement baseline curves.

[0014] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method described in this invention, the specific steps for reconstructing the actual constitution state sequence based on continuously collected real retest data and calculating the dynamic deviation residual between the actual constitution state sequence and the expected improvement baseline curve are as follows: Based on the expected improvement of the baseline curve, real retest data are continuously collected according to the preset retest cycle, and the actual physical condition vector corresponding to each retest time point is reconstructed through multimodal feature extraction, cross-modal alignment and physical evidence space dimensionality reduction mapping. Based on the actual physical condition vector, the data is serialized and organized according to the retest time sequence to form the actual physical condition sequence. The expected improvement baseline curve was mapped to the same retest time axis using the actual physical condition sequence, and the dynamic deviation residual between the actual physical condition sequence and the expected improvement baseline curve was calculated.

[0015] As a preferred embodiment of the TCM constitution identification, conditioning, and improvement method of the present invention, the step of correcting the edge weight reset reliability of the constitution-physiological parameter-syndrome feature association graph and the parameters of the continuous constitution evolution calculation unit based on the dynamic deviation residual refers to using the dynamic deviation residual to screen the target constitution state component whose deviation exceeds the preset tolerance threshold, and determining the target physiological parameter node and target syndrome feature node that cause the deviation according to the preset mapping relationship between the target constitution state component and the physiological parameter node and syndrome feature node; correcting the edge weight reset reliability of the relevant connecting edge in the constitution-physiological parameter-syndrome feature association graph according to the deviation direction and deviation magnitude, and using the dynamic deviation residual as a calibration error signal to adaptively calibrate the parameters of the continuous constitution evolution calculation unit.

[0016] The beneficial effects of this invention are as follows: By connecting the constitution state vector and the individual physiological prior vector into the constitution-physiological parameter-syndrome feature association diagram, and combining the constitution feature pattern classification results, the association reasoning between constitution state, physiological parameters, syndrome features and individual physiological baseline is realized, thereby improving the accuracy of constitution identification, the stability of constitution pattern classification and the reliability of constitution bias judgment; by determining the expected improvement baseline curve through the constitution evolution prediction trajectory set, and reconstructing the actual constitution state sequence by combining real retest data, the dynamic tracking of the constitution improvement process, the updating of constitution state pattern classification and the deviation feedback correction are realized, thereby improving the individual adaptability and continuous monitoring stability of the conditioning and improvement process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for methods of identifying, regulating, and improving constitution in Traditional Chinese Medicine.

[0019] Figure 2 A flowchart for generating body constitution identification data and body constitution bias data.

[0020] Figure 3 A flowchart for generating a set of predicted trajectory sets for physical evolution and a baseline curve expected to improve.

[0021] Figure 4 This is a flowchart for correcting dynamic deviation residuals.

[0022] Figure 5 This is a heatmap of parameters for the constitution correlation diagram.

[0023] Figure 6 This is a graph showing the change in the dynamic deviation residual decrease and the feedback correction trigger ratio. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-6 This is one embodiment of the present invention, which provides a method for identifying, regulating, and improving constitution according to traditional Chinese medicine, including the following steps: S1. Collect the multimodal physiological signals and consultation text of the subjects and perform feature alignment and dimensionality reduction mapping to obtain the physical state vector and the individual physiological prior vector.

[0028] Multimodal physiological signals and medical records of subjects were collected and synchronously labeled and standardized according to subject identification and collection time to obtain a standardized multimodal dataset.

[0029] It should be noted that during the same subject testing process, pulse signals were collected using a pulse acquisition device, heart rate variability signals were collected using a heart rate monitoring device, skin conductance signals were collected using a skin conductance acquisition device, tongue and facial images were collected using an image acquisition device, and consultation text was collected using a consultation form.

[0030] The pulse signal, heart rate variability signal, and electrodermal signal were denoised, abnormal fragments were removed, time windows were segmented, and amplitude normalization was performed. The tongue and facial images were processed for sharpness screening, target region cropping, and brightness correction. The consultation text was processed for typo correction, word segmentation, and invalid content removal. The processed pulse signal, heart rate variability signal, electrodermal signal, tongue image, facial image, and consultation text were uniformly stored according to the subject identification and collection time to obtain a standardized multimodal dataset.

[0031] Physiological signal features, image representation features, and medical history text features are extracted from a standardized multimodal dataset and mapped to a unified physical evidence space for alignment, resulting in an aligned set of physical evidence features.

[0032] The specific process includes reading pulse signals, heart rate variability signals, electrodermal signals, tongue images, facial images, and consultation text from a standardized multimodal dataset; extracting waveform amplitude, pulse interval, and waveform variation features from the pulse signals; extracting heart rate fluctuation features from the heart rate variability signals; extracting skin conductance level and skin conductance fluctuation features from the electrodermal signals; extracting color, texture, and regional morphological features from the tongue and facial images; and extracting symptom terms, frequency information, and semantic category information related to physical condition from the consultation text, thus forming physiological signal features, image representation features, and consultation text features.

[0033] It should be noted that the unified constitution evidence space refers to a unified feature expression space that is pre-established using various TCM constitutions and their corresponding syndromes and physiological parameters as evidence dimensions. It is used to convert different modal features such as physiological signals, images and consultation texts into constitution evidence vectors with the same dimension and the same order.

[0034] Based on physiological signal features, image representation features, and medical history text features, features are classified and normalized according to the dimensions of physical constitution evidence in a unified physical constitution evidence space, so that physiological signal features, image representation features, and medical history text features are all converted into comparable physical constitution evidence features. The converted physical constitution evidence features are arranged according to subject identification, collection time, and physical constitution evidence dimensions to eliminate time and dimension differences between different data sources, resulting in an aligned set of physical constitution evidence features.

[0035] The dimensionality reduction of the aligned physical evidence feature set is fused to obtain the physical state vector, and the individual physiological prior vector is generated by combining the resting stable physiological features.

[0036] The specific process includes: based on the aligned physical evidence feature set, summarizing similar evidence from physiological signal features, image representation features, and medical history text features according to the physical evidence dimension; determining the fusion weight of evidence from different sources based on collection quality, time consistency, and feature completeness; performing weighted fusion on the aligned physical evidence feature set with the completed weight configuration; and using principal component dimensionality reduction to compress redundant dimensions, retaining the main evidence components that can characterize the differences in the physical state of the subjects, to obtain the physical state vector.

[0037] Using the body state vector, pulse signal features, heart rate variability signal features, and electrodermal signal features with stable fluctuations in the resting state are read from a standardized multimodal dataset. These features are then organized into resting stable physiological features. After normalization, the resting stable physiological features are fused with the body state vector to generate individual physiological prior vectors that characterize the basic physiological differences of the subjects.

[0038] S2. Input the physical condition vector and the individual physiological prior vector into the pre-constructed physical condition-physiological parameter-syndrome feature association graph, and output the physical condition identification data and physical condition bias data corresponding to the subject.

[0039] The physical state vector is used as the current physical evidence anchor point, and the individual physiological prior vector is used as the individual baseline anchor point. The current physical evidence anchor point and the individual baseline anchor point are respectively connected to the physical node, physiological parameter node and syndrome feature node in the physical constitution-physiological parameter-syndrome feature association graph to form an individualized physical constitution relationship graph.

[0040] The specific process includes: marking the physical condition vector as the current physical condition evidence anchor point and marking the individual physiological prior vector as the individual baseline anchor point; based on the physical condition evidence dimension in the current physical condition evidence anchor point, finding the corresponding physical condition nodes, physiological parameter nodes, and syndrome feature nodes in the physical condition-physiological parameter-syndrome feature association graph, and establishing the evidence connection relationship between the current physical condition evidence anchor point and the corresponding physical condition nodes, physiological parameter nodes, and syndrome feature nodes; based on the resting stable physiological characteristics in the individual baseline anchor point, finding the corresponding physiological parameter nodes in the physical condition-physiological parameter-syndrome feature association graph, and establishing the baseline connection relationship between the individual baseline anchor point and the corresponding physiological parameter nodes; adding the evidence connection relationship and the baseline connection relationship to the physical condition-physiological parameter-syndrome feature association graph, so that the subject's current physical condition evidence and basic physiological differences jointly participate in the association expression between physical condition nodes, physiological parameter nodes, and syndrome feature nodes, forming an individualized physical condition relationship graph.

[0041] It should be noted that the constitution-physiological parameter-syndrome feature association graph is established based on the TCM constitution classification rules to create constitution nodes such as balanced constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-dampness constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution, and special constitution. Physiological parameter nodes are established based on detectable indicators extracted from pulse signals, heart rate variability signals, skin electrodermatology signals, tongue images, and facial images. Syndrome feature nodes are established based on constitution-related manifestations extracted from the consultation text, such as aversion to cold, fatigue, dry mouth, excessive sweating, chest tightness, abnormal sleep, emotional abnormalities, and abnormal eating. After the nodes are established, according to the TCM constitution discrimination rules, constitution annotation data in historical samples, and the correspondence between physiological parameters and syndrome features, connection edges are established between constitution nodes and physiological parameter nodes, constitution nodes and syndrome feature nodes, and physiological parameter nodes and syndrome feature nodes. Each connection edge is assigned an edge weight to represent the support strength, forming a pre-constructed constitution-physiological parameter-syndrome feature association graph that can express the association between constitution type, physiological parameters, and syndrome features.

[0042] Individualized constitution relationship maps are used to embed the subject's current constitution evidence and individual physiological baseline into the correlation between constitution, physiological parameters, and syndrome characteristics, thereby improving the individual fit and judgment accuracy of constitution identification data and constitution bias data.

[0043] Based on the individualized constitution relationship diagram, the evidence matching strength between the current constitution evidence anchor point and each physiological parameter node is calculated, and the target physiological parameter node and target syndrome feature node related to the subject's current state are activated according to the evidence matching strength.

[0044] The specific process includes reading the current physical condition evidence anchor point, the feature vector of each physiological parameter node, the data credibility of each physiological parameter node, and the normalized offset of each physiological parameter node relative to the individual's physiological prior from the individualized physical condition relationship graph one by one; comparing the similarity between the current physical condition evidence anchor point and the feature vector of each physiological parameter node, and calculating the evidence matching strength by combining the data credibility and normalized offset of each physiological parameter node, so that each physiological parameter node forms a corresponding evidence matching strength.

[0045] Physiological parameter nodes whose evidence matching strength reaches the preset activation requirements are identified as target physiological parameter nodes. The edge weights corresponding to the connecting edges between target physiological parameter nodes and syndrome feature nodes in the individualized constitution relationship graph are read, and syndrome feature nodes whose edge weights reach the preset association requirements are identified as target syndrome feature nodes, thereby activating target physiological parameter nodes and target syndrome feature nodes related to the subject's current state.

[0046] It should be noted that the activation requirement is based on the matching distribution between physiological parameter nodes and constitution identification data in historical samples, selecting the range of evidence matching strength that can distinguish between relevant and irrelevant physiological parameter nodes as the activation judgment condition for physiological parameter nodes; the association requirement is based on the co-occurrence relationship between physiological parameter nodes and syndrome feature nodes in historical samples and the TCM constitution discrimination rules, selecting the range of edge weights that can represent stable association relationships as the association judgment condition for syndrome feature nodes.

[0047] The expression for calculating the evidence match strength between the current physical condition evidence anchor and each physiological parameter node is as follows: ; in, This indicates the current anchor point of physical evidence and the first The strength of evidence matching between physiological parameter nodes The index of the physiological parameter node. This represents the Sigmoid function. The moderating coefficient representing the similarity between the physical state vector and the node feature vector. This represents the normalized physical state vector. Represents the normalized i-th Each physiological parameter node feature vector An adjustment coefficient representing the reliability of node data. Indicates the first The reliability of data from each physiological parameter node. The inhibition coefficient represents the individual's physiological prior bias. Indicates the first The normalized offset of each physiological parameter node relative to the individual's physiological prior.

[0048] It should be noted that, It is obtained by fitting and calibrating the similarity distribution between the physical condition vector and the labeled physiological parameter nodes in historical samples. It was determined after statistical evaluation based on the missing rate, noise level and repeated measurement stability of each physiological parameter node in the historical data; It is obtained by calibration during the historical sample verification process based on the degree of influence of individual physiological prior bias on the body constitution identification bias.

[0049] Based on the target physiological parameter node and the target syndrome feature node, the node relationship and edge weight between the target physiological parameter node, the target syndrome feature node and each candidate constitution node are read, and the offset between the individual baseline anchor point and the target physiological parameter node is converted into individual offset evidence edge.

[0050] The specific process includes: based on the target physiological parameter node and the target syndrome feature node, searching the connection relationship between the target physiological parameter node and each candidate constitution node, as well as the connection relationship between the target syndrome feature node and each candidate constitution node in the individualized constitution relationship graph, and reading the node relationship and edge weight corresponding to each connection relationship one by one, and organizing them according to the target physiological parameter node, target syndrome feature node, candidate constitution node, node relationship and edge weight, so that each candidate constitution node forms the association evidence from the target physiological parameter node and the target syndrome feature node; Based on the associated evidence, the resting stable physiological characteristics corresponding to the target physiological parameter node in the individual baseline anchor point are read, and the current physiological evidence corresponding to the target physiological parameter node in the current physical condition evidence anchor point is read. The resting stable physiological characteristics and the current physiological evidence are compared with the same dimension to obtain the offset direction and offset degree of the target physiological parameter node relative to the individual baseline anchor point. The offset direction and offset degree are written into the connection relationship between the individual baseline anchor point and the target physiological parameter node to form the individual offset evidence edge.

[0051] Individual offset evidence edges are used to individually modify node relationships and edge weights, and combined with the constitution bias discrimination rule to suppress conflicting propagation paths, forming a set of constitution relationship diffusion paths for subjects.

[0052] The specific process includes: using individual offset evidence edges to read the offset direction and degree of the target physiological parameter node relative to the individual baseline anchor point; comparing the offset direction and degree with the node relationships and edge weights between the target physiological parameter node, the target syndrome feature node, and each candidate constitution node; when the offset direction is consistent with the constitution bias direction supported by the node relationship, increasing the confidence level of the edge weight of the corresponding connection edge; when the offset direction is inconsistent with the constitution bias direction supported by the node relationship, decreasing the confidence level of the edge weight of the corresponding connection edge; verifying the propagation path formed by each connection edge according to the constitution bias discrimination rule, and suppressing propagation paths that contradict the constitution bias discrimination rule (e.g., when the target physiological parameter node, the target syndrome feature node, and the target physiological parameter ... The elevated skin conductance level corresponding to the physiological parameter node and the aversion to cold corresponding to the target syndrome feature node both point to the candidate constitution node of damp-heat constitution. However, if the aversion to cold does not support the judgment of damp-heat constitution bias in the constitution bias discrimination rule, the propagation path is suppressed. The propagation path that conforms to the constitution bias discrimination rule and has completed individualized correction is retained (for example, when the elevated skin conductance level corresponding to the target physiological parameter node and the easy sweating manifestation corresponding to the target syndrome feature node are activated, and the connection path points to the candidate constitution node of qi deficiency constitution, if the elevated skin conductance level and easy sweating manifestation both support the judgment of qi deficiency constitution bias in the constitution bias discrimination rule, the propagation path is retained), forming a set of constitution relationship diffusion paths for the subject.

[0053] It should be noted that the constitution bias discrimination rule is pre-organized based on the TCM constitution classification standard, constitution annotation data in historical samples, and the correspondence between physiological parameter nodes and syndrome feature nodes. It is used to determine the type and degree of constitution bias supported by different combinations of physiological parameters and syndrome features.

[0054] The set of physical relationship diffusion paths is used to perform multi-hop relationship-aware reasoning on each candidate physical node, and the relationship diffusion confidence of each candidate physical node is calculated. The candidate physical nodes are sorted from high to low according to the relationship diffusion confidence. When at least two candidate physical nodes have the same relationship diffusion confidence, they are sorted from most to least effective propagation paths. When the number of effective propagation paths is still the same, they are sorted from highest to lowest according to the average edge weight, thus obtaining the candidate physical node sorting sequence.

[0055] The specific process includes: reading the set of diffusion paths related to physical constitution for the subjects; identifying the diffusion paths that have undergone individualized correction and are not suppressed by the physical constitution bias discrimination rule as effective diffusion paths; and grouping them according to the candidate physical constitution nodes that the effective diffusion paths ultimately point to, so that each candidate physical constitution node corresponds to a set of effective diffusion paths; for each set of effective diffusion paths, reading the path length, edge weights, and evidence matching strength of the effective diffusion path; where path length is used to characterize the indirectness of the diffusion relationship, the shorter the path length, the more direct the support of the target physiological parameter node and the target syndrome feature node to the candidate physical constitution node; edge weights are used to characterize the support strength of the node relationship, the higher the edge weight, the stronger the association. The more stable the evidence, the stronger the evidence matching strength, which characterizes the degree of matching between the target physiological parameter node and the subject's current state. The higher the evidence matching strength, the better the effective propagation path reflects the subject's current physical condition. The path length, edge weight, and evidence matching strength are used together to calculate the propagation contribution of each effective propagation path, and the propagation contributions pointing to the same candidate physical condition node are summarized to obtain the relation diffusion confidence of each candidate physical condition node. The candidate physical condition nodes are sorted from high to low according to the relation diffusion confidence. When the relation diffusion confidence is the same, they are sorted from high to low according to the number of effective propagation paths. When the number of effective propagation paths is still the same, they are sorted from high to low according to the average edge weight to obtain the candidate physical condition node ranking sequence.

[0056] The expression for calculating the relation diffusion confidence for each candidate physical node is as follows: ; in, Indicates the first The relationship diffusion confidence level corresponding to each candidate physical node. Indicates the index of the candidate physical node. This represents a path for the diffusion of physical relationships to a candidate physical node. Indicates pointing to the first The set of diffusion paths of physical relationships for each candidate physical node. Represents the natural constant. This represents the path length attenuation coefficient. Indicates diffusion path Path length, Indicates diffusion path The average weight of each connecting edge in the middle. Indicates diffusion path The average value of the evidence matching strength at each node. This represents the total number of candidate physical nodes. It represents a very small positive number.

[0057] It should be noted that, The parameters were determined by statistically fitting the changes in the contribution of different path lengths to the accuracy of body constitution identification in historical samples, and selecting parameter values ​​that allow the contribution of long paths to decrease step by step while keeping the evidence of short paths stable.

[0058] Based on the candidate constitution node ranking sequence, the dominant constitution node and the secondary constitution node are determined to form the constitution identification data corresponding to the subject. Then, according to the relationship diffusion confidence, individual deviation evidence edge strength and constitution bias discrimination rules corresponding to the dominant constitution node and the secondary constitution node, the bias degree value and bias level corresponding to each constitution node are determined to form the constitution bias degree data corresponding to the subject.

[0059] The specific process includes: based on the candidate constitution node ranking sequence, identifying the candidate constitution node with the highest ranking and whose relationship diffusion confidence meets the preset dominance judgment requirements as the dominant constitution node; continuing to read candidate constitution nodes located after the dominant constitution node, and identifying a concurrent constitution node when the difference in relationship diffusion confidence between the candidate constitution node and the dominant constitution node is within the preset consolidation judgment range; and organizing the dominant constitution node, concurrent constitution node, relationship diffusion confidence, number of effective propagation paths, and average edge weight to form the constitution identification data corresponding to the subject.

[0060] The relationship diffusion confidence scores of dominant and secondary constitution nodes are extracted from the constitution identification data of the subjects. The degree of deviation of the dominant and secondary constitution nodes from the individual's physiological prior is determined by combining this with the strength of the individual's deviation evidence edge. Based on the constitution bias discrimination rules, constitution nodes with high relationship diffusion confidence scores and weak individual deviation evidence edge strength are identified as mildly biased; those with high relationship diffusion confidence scores and significantly deviated individual deviation evidence edge strength are identified as moderately biased; and those with high relationship diffusion confidence scores and consistently significantly deviated individual deviation evidence edge strength are identified as severely biased. The dominant constitution node, secondary constitution node, relationship diffusion confidence score, individual deviation evidence edge strength, bias degree value, and mild, moderate, or severe bias level are then organized to form the constitution bias degree data corresponding to the subjects.

[0061] It should be noted that the dominance determination requirement is based on the confidence distribution of the relationship diffusion between the confirmed dominant physical nodes in historical samples, and a confidence range that can stably distinguish between the dominant physical nodes and non-dominant physical nodes is pre-set; the coexistence determination range is based on the difference distribution of the relationship diffusion confidence between the confirmed coexistence physical nodes and the dominant physical nodes in historical samples, and a difference range that can characterize the coexistence tendency of coexistence physical nodes is pre-set.

[0062] like Figure 5A heatmap of parameters in the constitution-physiological parameter-syndrome feature correlation diagram illustrates the influence of key parameters on the constitution identification results. Using the evidence matching strength threshold and edge weights as bivariate variables, the diagram observes the changes in the overall constitution identification score. It reveals that when the evidence matching strength threshold is too low, weakly correlated physiological parameter nodes and syndrome feature nodes are easily introduced; conversely, when the evidence matching strength threshold is too high, effective nodes may be missed. Furthermore, the more accurately the edge weights reflect the support relationships between constitution nodes, physiological parameter nodes, and syndrome feature nodes, the more stable the overall constitution identification score becomes.

[0063] S3. Based on the physical condition identification data and the physical condition bias data, obtain the core monitoring dimensions, and perform constraint trimming and standardization processing on the core monitoring dimensions according to the preset safety boundary to obtain a set of standardized conditioning assessment vectors that match the physical condition of the subject.

[0064] The dominant constitution node, the secondary constitution node, and the relationship diffusion confidence are read from the constitution identification data. The bias degree value and bias level corresponding to each constitution node are read from the constitution bias degree data. The dominant constitution node, secondary constitution node, relationship diffusion confidence, bias degree value, and bias level are sorted and organized to form a constitution bias contribution data table.

[0065] The specific process includes: extracting the dominant constitution nodes, secondary constitution nodes, and the corresponding relationship diffusion confidence scores from the constitution identification data; extracting the bias values ​​and bias levels corresponding to the dominant constitution nodes and secondary constitution nodes from the constitution bias data; and, using the dominant constitution nodes and secondary constitution nodes as the sorting objects, verifying the correspondence between the constitution node names, relationship diffusion confidence scores, bias values, and bias levels item by item, and using the relationship diffusion confidence scores as the basis for determining the relationship between the constitution nodes and the subjects. The supporting evidence for the pre-constitutional state is to use the degree of bias as the basis for the strength of the deviation of the constitution node from the individual's physiological prior, and the degree of bias as the grade indicator of the strength of the constitution bias. After completing the corresponding verification, the data are organized in the same table according to the subject identifier, constitution node type, relationship diffusion confidence, degree of bias, and degree of bias. The data is arranged in the order of dominant constitution node first and secondary constitution node second, forming a constitution bias contribution data table to characterize the source of the subject's constitution bias, the strength of the constitution bias, and the contribution relationship of the constitution bias.

[0066] Using the constitution bias contribution data table, the dominant constitution nodes and the secondary constitution nodes are matched to the preset conditioning dimension library, and the bias contribution matching value of each conditioning dimension is calculated to form a set of candidate intervention dimensions.

[0067] The specific process includes: using the constitution bias contribution data table to read the dominant constitution node, the secondary constitution node, the relationship diffusion confidence, the bias degree value, and the bias level; and searching for the conditioning dimensions related to the dominant constitution node and the secondary constitution node based on the correspondence between the dominant constitution node and the secondary constitution node in the conditioning dimension library; the conditioning dimension library pre-records the conditioning correspondence between constitution nodes and dietary conditioning, daily life conditioning, exercise conditioning, and emotional conditioning, and records the conditioning correlation strength between different constitution nodes and different conditioning dimensions, so that the dominant constitution node and the secondary constitution node can be matched to the corresponding conditioning dimension.

[0068] Based on the conditioning dimensions, and combining the relationship diffusion confidence, bias degree value, bias level, and conditioning association strength, the contribution of each conditioning dimension to the subject's current physical bias is calculated, and the bias contribution matching value of each conditioning dimension is obtained. The conditioning dimensions are arranged from high to low according to the bias contribution matching value. The conditioning dimensions whose bias contribution matching value meets the preset candidate conditions are matched with the corresponding dominant physical node, secondary physical node, bias level, and bias contribution matching value to form a set of candidate intervention dimensions.

[0069] It should be noted that the conditioning dimension library is pre-organized based on the TCM constitution conditioning standards, records of constitution biases in historical samples, and the corresponding relationships of conditioning dimensions. It is used to record the strength of conditioning correlation between constitution nodes and dietary conditioning, daily life conditioning, exercise conditioning, and emotional conditioning. The candidate conditions are pre-determined based on the distribution of the bias contribution matching values ​​of effective conditioning dimensions in historical samples. They are used to screen conditioning dimensions that can stably reflect the source of the subject's current constitution bias.

[0070] The expression for calculating the bias contribution matching value of each conditioning dimension is as follows: ; in, Indicates the first The biased contribution matching value of each conditioning dimension Index representing the conditioning dimension, This represents the total number of dominant and secondary constitution nodes. The index representing the dominant or secondary constitution node. Indicates the first Relationship diffusion confidence level for each dominant or concurrent prime node. Indicates the first The quantification value of the bias level corresponding to the dominant constitution node or the concurrent constitution node. Indicates the first The dominant constitution node or concurrent constitution node and the first The strength of the conditioning correlation between each conditioning dimension.

[0071] It should be noted that, It is obtained by performing multi-hop relationship perception reasoning on each candidate physical node through the set of physical relationship diffusion paths, and summarizing the path length, edge weight and evidence matching strength; It is obtained by converting the mild, moderate or severe imbalance levels in the physical imbalance data according to the level quantification rules; It is read from the conditioning dimension library, which is pre-organized based on the correspondence between constitution nodes and conditioning dimensions in traditional Chinese medicine constitution conditioning standards and historical samples.

[0072] Based on the set of candidate intervention dimensions, the bias contribution matching value and bias level corresponding to each candidate intervention dimension are read respectively, and the candidate intervention dimensions are sorted according to the rule that the higher the bias contribution matching value and the more severe the bias level, the higher the priority. The candidate intervention dimensions that are ranked first and meet the preset selection requirements are determined as the core monitoring dimensions.

[0073] The specific process includes: based on the candidate intervention dimension set, reading the bias contribution matching value, bias level, source constitution node, and conditioning dimension name corresponding to each candidate intervention dimension, and initially sorting the candidate intervention dimensions according to the bias contribution matching value from high to low; when the bias contribution matching values ​​of at least two candidate intervention dimensions are the same or close, further adjusting the sorting position according to the bias level from severe bias level, moderate bias level to mild bias level, so that the candidate intervention dimensions with higher bias contribution matching values ​​and more severe bias levels are arranged first; after the sorting is completed, the sorting position, bias contribution matching value, and bias level of each candidate intervention dimension are compared with the optimization requirements, and the candidate intervention dimensions with higher sorting positions, bias contribution matching values ​​that meet the optimization requirements, and can correspond to the dominant constitution node or also contain the bias source of the constitution node are retained, and the retained candidate intervention dimensions are determined as core monitoring dimensions.

[0074] It should be noted that the preferred requirements are predetermined based on the priority distribution of effective conditioning dimensions and records of improvement in physical imbalances in historical samples.

[0075] The bias contribution matching value, bias level and priority are extracted from the core monitoring dimensions, and the corresponding suggestion intensity range, suggestion frequency range and suggestion duration range are called from the adjustment dimension library to form the core monitoring dimension parameter table.

[0076] The specific process includes: reading the conditioning dimension name, source constitution node, bias contribution matching value, bias level, and priority from the core monitoring dimensions; searching for matching conditioning records in the conditioning dimension library according to the conditioning dimension name; retrieving the suggestion intensity range, suggestion frequency range, and suggestion duration range corresponding to the conditioning dimension name from the conditioning records, so that each core monitoring dimension has parameter boundaries for constraint pruning; organizing the core monitoring dimensions, source constitution node, bias contribution matching value, bias level, priority, suggestion intensity range, suggestion frequency range, and suggestion duration range accordingly, and arranging them according to the priority of the core monitoring dimensions to form a core monitoring dimension parameter table.

[0077] Based on the core monitoring dimension parameter table, conditioning suggestion parameters are generated according to the bias contribution matching value and bias level, and the conditioning suggestion parameters are mapped to the safety boundary for constraint trimming. After the constraint trimming conditioning suggestion parameters are coded with unified dimensions, the priority of each core monitoring dimension is determined according to the bias contribution matching value from high to low and the bias level from severe to mild. Based on the priority, a standardized conditioning evaluation vector set that can characterize the conditioning direction, conditioning intensity and conditioning cycle is formed.

[0078] The specific process includes: based on the core monitoring dimension parameter table, reading the core monitoring dimension, bias contribution matching value, bias level, recommended intensity range, recommended frequency range, and recommended duration range item by item; determining the recommended intensity, recommended frequency, and recommended duration range for each core monitoring dimension according to the degree of physical bias contribution reflected by the bias contribution matching value and the strength of physical bias reflected by the bias level, forming conditioning recommendation parameters; comparing the conditioning recommendation parameters with the recommended intensity range, recommended frequency range, and recommended duration range respectively, and performing boundary recovery on recommended intensity, recommended frequency, and recommended duration ranges that exceed the range, so that the conditioning recommendation parameters remain within the safe boundaries.

[0079] Using the conditioning suggestion parameters after constraint pruning, the suggestion intensity, suggestion frequency, and suggestion duration are converted into coded values ​​of the same scale according to their corresponding ranges. The conditioning dimension name, suggestion intensity coded value, suggestion frequency coded value, and suggestion duration coded value are combined to form a standardized conditioning evaluation vector for the corresponding core monitoring dimension. The core monitoring dimensions are arranged in order of bias contribution matching value from high to low and bias level from severe to mild. All standardized conditioning evaluation vectors are then summarized in order to obtain a set of standardized conditioning evaluation vectors that can characterize conditioning direction, conditioning intensity, and conditioning cycle.

[0080] It should be noted that the safety boundary is predetermined based on the TCM constitution conditioning guidelines, safe conditioning records in historical samples, and the recommended intensity range, recommended frequency range, and recommended duration range. It is used to limit the conditioning recommendations to be within a safe and feasible range.

[0081] S4. Input the standardized conditioning assessment vector set and the physical condition state vector into the continuous physical condition evolution calculation unit, and output the set of physical condition evolution prediction trajectories corresponding to each conditioning suggestion.

[0082] The intensity, frequency, and duration of each treatment suggestion are read from the standardized treatment assessment vector set and converted into a continuous treatment driving sequence according to the time axis.

[0083] The specific process includes: reading the suggestion intensity, suggestion frequency, and suggestion duration period corresponding to each treatment suggestion from the standardized treatment assessment vector set according to the priority of the core monitoring dimensions, and establishing a projection time axis based on the preset observation period; determining the scope of the treatment suggestion on the projection time axis by the suggestion duration period, determining the effect interval by the suggestion frequency, determining the effect magnitude by the suggestion intensity, and organizing the effect magnitude of each treatment suggestion at different time points according to the projection time axis order to form a continuous treatment driving sequence.

[0084] It should be noted that the observation period is predetermined based on the suggested duration range in the conditioning dimension library, the retest interval in historical samples, and the observation duration required for changes in physical imbalance, and is used to limit the extrapolation time range of the continuous conditioning-driven sequence.

[0085] The continuous conditioning driving sequence is used to concatenate each conditioning suggestion with the physical condition vector, and combined with the observation period to generate continuous inference input data corresponding to each conditioning suggestion.

[0086] The specific process includes: using a continuous conditioning-driven sequence, reading the conditioning direction content, prediction time point, and conditioning driving amplitude item by item according to the conditioning suggestion identifier; and mapping the conditioning driving amplitude corresponding to each conditioning suggestion to the physical condition vector according to the prediction time axis; at each prediction time point, using the physical condition vector as the initial physical condition expression of the subject, and concatenating the conditioning direction content, conditioning driving amplitude, suggestion frequency, and suggestion duration of the corresponding conditioning suggestion at the corresponding prediction time point with the physical condition vector to form a continuous prediction data segment containing information on the subject's current physical condition and the effect of the conditioning suggestion; limiting the start and end range of the continuous prediction data segment by the observation period, and serializing and organizing the continuous prediction data segments corresponding to the same conditioning suggestion according to the prediction time axis to obtain the continuous prediction input data corresponding to each conditioning suggestion.

[0087] Based on continuous inference input data, the continuous physical evolution calculation unit continuously infers the changes in the physical state of the subjects on the inference time axis, forming a physical evolution state sequence. The trajectory is then organized according to the conditioning suggestion identifier and the inference time order to obtain the set of physical evolution prediction trajectories corresponding to each conditioning suggestion.

[0088] The specific process includes: based on the continuous deduction input data corresponding to each conditioning suggestion, reading the physical state vector, conditioning direction content, conditioning driving amplitude, suggestion frequency, and suggestion duration period corresponding to each deduction time point in sequence according to the deduction time axis; the continuous physical evolution calculation unit takes the physical state vector corresponding to the first deduction time point as the initial state, combines the conditioning driving amplitude and conditioning direction content corresponding to the current deduction time point, updates the change trend of the subject's physical state at the current deduction time point, and continues to use the updated subject's physical state for the state deduction of the next deduction time point, so that the subject's physical state changes continuously along the deduction time axis, forming a physical evolution state sequence.

[0089] The construction process of the continuous physical fitness evolution calculation unit is as follows: Sample data containing the subject's initial physical fitness state vector, individual physiological prior vector, historical conditioning suggestion parameters, physical fitness state vectors corresponding to multiple retest time points, and physical fitness bias values ​​are pre-collected. The initial physical fitness state vector is used as the state variable, the conditioning driving sequence encoded by conditioning direction, suggestion intensity, suggestion frequency, and suggestion duration period is used as the external driving variable, and the individual physiological prior vector is used as the individual difference constraint variable. The relationship of physical fitness state changes is established according to the projection timeline. Between adjacent retest time points, based on the current... The predicted physical condition at the next time point is obtained from the pre-physical condition, conditioning drive sequence, and individual physiological prior vector, so that the physical condition is continuously updated along the time axis. Then, with the goal of minimizing the difference between the predicted physical condition and the actual retested physical condition, the conditioning response coefficient, individual difference correction coefficient, and time decay coefficient in the state change relationship are trained or calibrated. At the same time, the decreasing direction of the physical bias value, the conditioning safety boundary, and the smoothness of the state change are used as constraints to obtain a continuous physical evolution calculation unit that can infer the future physical change trend based on the subject's current physical condition and different conditioning suggestions.

[0090] Based on the sequence of physical constitution evolution states, the content of state changes is collected according to the conditioning suggestions, and the content of physical constitution evolution state, conditioning driving amplitude and conditioning direction is organized in the order of the projection time to obtain the set of physical constitution evolution prediction trajectories corresponding to each conditioning suggestion.

[0091] S5. Determine the expected improvement baseline curve based on the set of predicted physical constitution evolution trajectories, and reconstruct the actual physical constitution state sequence based on continuously collected real retest data. Calculate the dynamic deviation residual between the actual physical constitution state sequence and the expected improvement baseline curve. Adjust the edge weight reset reliability of the physical constitution-physiological parameters-syndrome characteristic correlation diagram and the parameters of the continuous physical constitution evolution calculation unit based on the dynamic deviation residual.

[0092] Read the constitution evolution state sequence corresponding to each conditioning suggestion from the constitution evolution prediction trajectory set, and calculate the decrease in constitution bias and trajectory smoothness corresponding to each constitution evolution state sequence. The expression is:

[0093] in, Indicates the first The degree of decrease in physical bias corresponding to the sequence of physical evolution states. An index representing a sequence of physical evolutionary states. Indicates the first The degree of physical bias in the sequence of physical evolution states at the initial time point. Indicates the first The degree of physical bias at the termination time point of a sequence of physical evolution states. This indicates the termination point of the sequence of physical evolution states. Indicates the first The trajectory smoothness corresponding to the evolutionary state sequence of the strip body. Indicates the first The sequence of physical evolutionary states in the first stage A vector of physical condition at each time point. Index representing a point in time. Indicates the first The sequence of physical evolutionary states in the first... A vector of physical state at each time point.

[0094] The specific process includes: reading the predicted physical evolution trajectory for each conditioning suggestion from the set of predicted physical evolution trajectories, item by item according to the conditioning suggestion identifier; extracting the physical evolution state sequence, conditioning driving amplitude, and conditioning direction according to the time sequence of the prediction; in each physical evolution state sequence, reading the physical imbalance degree value corresponding to the starting prediction time point and the physical imbalance degree value corresponding to the ending prediction time point; comparing the difference between the physical imbalance degree value corresponding to the starting prediction time point and the physical imbalance degree value corresponding to the ending prediction time point; and combining the difference between the starting prediction time point and the ending prediction time point. The degree of physical imbalance is relativized to obtain the reduction in physical imbalance under the corresponding conditioning recommendations. At the same time, the physical evolution state vectors corresponding to two adjacent time points are read in the order of the simulation time. The degree of change between adjacent physical evolution state vectors is calculated, and all adjacent degree of change is summarized and evaluated. If the degree of adjacent change is small overall and the change process is continuous, the trajectory smoothness is high. If the degree of adjacent change fluctuates greatly and the change process is unstable, the trajectory smoothness is low. The reduction in physical imbalance and trajectory smoothness corresponding to each physical evolution state sequence are obtained.

[0095] Based on the magnitude of the decrease in physical bias and the smoothness of the trajectory corresponding to each physical evolution state sequence, the physical evolution state sequences are sorted and screened, and the physical evolution state sequences that meet the preset baseline determination requirements are determined as the expected improvement baseline curves.

[0096] The specific process includes reading the conditioning suggestion identifier, the magnitude of the decrease in physical imbalance, and the trajectory smoothness corresponding to each physical constitution evolution state sequence one by one. First, sort them from high to low according to the magnitude of the decrease in physical imbalance. When the magnitude of the decrease in physical imbalance is the same or close, sort them from high to low according to the trajectory smoothness. Eliminate physical constitution evolution state sequences with insufficient decrease, low trajectory smoothness, or discontinuous state fluctuations, and retain physical constitution evolution state sequences with stable improvement trends.

[0097] Based on the preserved physical evolution state sequence, the magnitude of the decrease in physical imbalance, the smoothness of the trajectory, the magnitude of the conditioning drive, and the direction of conditioning are read and compared with the baseline determination requirements. When the magnitude of the decrease in physical imbalance reaches the expected improvement level, the smoothness of the trajectory meets the requirements for stable change, and the magnitude of the conditioning drive and the direction of conditioning can correspond to the source of the subject's current physical imbalance, the corresponding physical evolution state sequence is determined as the expected improvement baseline curve.

[0098] It should be noted that the baseline determination requirement is to pre-determine statistically based on the records of improvement in physical fitness bias in historical samples, retest intervals, distribution of the magnitude of decline in physical fitness bias, and distribution of trajectory smoothness.

[0099] Based on the expected improvement of the baseline curve, real retest data are continuously collected according to the preset retest cycle, and the actual physical condition vector corresponding to each retest time point is reconstructed through multimodal feature extraction, cross-modal alignment and physical evidence space dimensionality reduction mapping.

[0100] The specific process includes determining the expected physical evolution status at each projection time point based on the expected improvement baseline curve, and collecting pulse signals, heart rate variability signals, skin conductance signals, tongue images, facial images, and medical records according to the retest cycle; denoising, anomaly removal, image correction, and text processing of the collected data to form real retest data.

[0101] Physiological signal features, image representation features, and medical history text features are extracted from the real retest data corresponding to each retest time point. These features are then categorized, normalized in scale, and aligned across modalities according to the physical evidence dimensions in a unified physical evidence space. This ensures that physiological signal features, image representation features, and medical history text features from different sources form comparable retest physical evidence features at the same retest time point. The retest physical evidence features are then summarized according to the physical evidence dimensions, and principal component dimensionality reduction is used to compress redundant dimensions, retaining the main evidence components that can characterize the differences in the retest physical state of the subjects. This process reconstructs the actual physical state vector corresponding to each retest time point.

[0102] It should be noted that the retesting cycle is predetermined based on the records of improvement in physical imbalance in historical samples, the recommended duration range in the conditioning dimension library, the subject's physical imbalance level, and the feasibility of collecting actual retesting data. Actual retesting data refers to the actual test data that is re-collected by the subject at each retesting time point during the conditioning process and is consistent with the initial collection type, including pulse signals, heart rate variability signals, skin conductance signals, tongue images, facial images, and medical records.

[0103] Based on the actual physical condition vector, the data is serialized and organized according to the retest time sequence to form the actual physical condition sequence.

[0104] The specific process includes reading the actual physical condition vector, subject identifier, and collection time corresponding to each retest time point, verifying the consistency of the source according to the subject identifier, and then arranging the actual physical condition vectors in the order of retest time. When there is duplicate data at the same retest time point, data with a higher degree of comparability are retained based on collection quality, time consistency, and feature integrity, and finally, the data are linked together to form an actual physical condition sequence.

[0105] The expected improvement baseline curve is mapped to the same retest time axis using the actual physical condition sequence, and the dynamic deviation residual between the actual physical condition sequence and the expected improvement baseline curve is calculated. The expression is as follows: ; in, Indicates the first The dynamic deviation residuals corresponding to each retest time point The actual physical state sequence is represented in the th order. The actual physical condition vector corresponding to each retest time point Indicates the first Each retest time point This indicates the expected improvement in the baseline curve at the 1st. The baseline state vector corresponding to each retest time point.

[0106] The specific process includes: using the actual physical condition sequence to read the subject identifier, retest time point, and actual physical condition vector, and establishing the same retest time axis with the retest time point; reading the projected time point and expected physical condition evolution state from the expected improvement baseline curve, and mapping the expected physical condition evolution state to the retest time point. When the retest time point and the projected time point are the same, the corresponding expected physical condition evolution state is directly read. When the retest time point is between adjacent projected time points, time interpolation is performed based on the expected physical condition evolution state of adjacent projected time points to obtain the baseline state vector corresponding to each retest time point; comparing the actual physical condition vector corresponding to each retest time point with the baseline state vector with the same dimensions, calculating the degree of difference between the actual physical condition vector and the baseline state vector, and normalizing it in combination with the amplitude of the baseline state vector to obtain the dynamic deviation residual corresponding to each retest time point.

[0107] like Figure 6 The graph showing the decrease in dynamic deviation residuals and the change in the feedback correction trigger ratio demonstrates the dynamic feedback correction capability of this invention in a continuous retest scenario. The graph uses three curves—average dynamic deviation residuals, correction trigger ratio, and tolerance threshold—to show how the deviation between the actual physical condition sequence and the expected improvement baseline curve changes over retesting time. It also shows that when the dynamic deviation residuals exceed the tolerance threshold, feedback correction is triggered, and subsequently, the correction trigger ratio and dynamic deviation residuals gradually decrease. This indicates that the edge weight reset reliability of the physical condition-physiological parameter-syndrome characteristic correlation graph and the parameters of the continuous physical condition evolution calculation unit can be adaptively calibrated based on real retest data.

[0108] The dynamic deviation residual is used to screen target physical state components whose deviation exceeds the preset tolerance threshold. According to the preset mapping relationship between the target physical state components and physiological parameter nodes and syndrome feature nodes, the target physiological parameter nodes and target syndrome feature nodes that cause the deviation are determined. Based on the deviation direction and deviation magnitude, the edge weights of the relevant connecting edges in the physical condition-physiological parameter-syndrome feature association graph are adjusted and the reliability is reset. The dynamic deviation residual is used as a calibration error signal to adaptively calibrate the parameters of the continuous physical condition evolution calculation unit.

[0109] The specific process includes: using dynamic deviation residuals to read the magnitude and direction of the residuals for each component of the actual physical state vector; comparing the magnitude of the residuals with a tolerance threshold; identifying the physical state component whose residual exceeds the tolerance threshold as the target physical state component; determining the direction of the residual as the deviation direction; and determining the difference between the residual magnitude and the tolerance threshold as the deviation magnitude. Based on the target physical state component, according to the mapping relationship between the physical state component and the physiological parameter node and the syndrome feature node pre-established in the unified physical evidence space, the corresponding target physiological parameter node and target syndrome feature node are found, and the relevant connection edges between the target physiological parameter node, the target syndrome feature node and the candidate physical state node are read. An adaptive correction step size is generated based on the deviation magnitude.

[0110] The adaptive correction step size increases with the deviation magnitude but is limited to a preset step size range. When the deviation direction is consistent with the physical bias direction supported by the relevant connecting edge, and the deviation magnitude at multiple consecutive retest time points exceeds the tolerance threshold, the edge weight reset reliability of the relevant connecting edge is reduced according to the adaptive correction step size. When the deviation direction is opposite to the physical bias direction supported by the relevant connecting edge, and the actual physical state sequence shows an improving trend relative to the expected improvement baseline curve, the edge weight reset reliability of the relevant connecting edge is increased according to the adaptive correction step size, and the corrected edge weight reset reliability is limited to a preset reliability range. Simultaneously, the dynamic deviation residual is used as a calibration error signal to construct a system based on the actual physical state vector and... The calibration loss function aims to minimize the sum of squared residuals between predicted physical state vectors. Based on the calibration loss function, the conditioning response coefficient, individual difference correction coefficient, and time decay coefficient in the continuous physical evolution calculation unit are iteratively updated. After each update, the predicted physical evolution trajectory and dynamic deviation residuals are reacquired. If the updated dynamic deviation residuals decrease, the parameter update is retained. If the updated dynamic deviation residuals increase, the adaptive correction step size is reduced or the parameter update is rolled back until the dynamic deviation residuals are below the tolerance threshold or the preset maximum number of iterations is reached. This achieves adaptive calibration of the physical-physiological parameter-syndrome feature correlation diagram and the parameters of the continuous physical evolution calculation unit.

[0111] It should be noted that the tolerance threshold is pre-determined based on the distribution of dynamic deviation residuals in historical samples, the range of repeated measurement fluctuations in real retest data, and the stability of acquisition quality. For example, after normalization of the dynamic deviation residuals, the tolerance threshold can be set to 0.05 to 0.15. A value below this range indicates that fluctuations are allowed, while a value exceeding this range indicates that the target physical condition component needs to be determined. The step size range is preset based on the minimum effective correction amplitude and the maximum stable correction amplitude of the dynamic deviation residuals in historical retest samples. The confidence level range is preset based on the normalized value range of the confidence level of the weighted resetting in the physical condition-physiological parameter-syndrome feature correlation graph. The maximum number of iterations is preset based on the average number of iterations or the upper limit number of iterations required for the dynamic deviation residuals in historical calibration samples to reach convergence.

[0112] In summary, this invention achieves improved accuracy in constitution identification, stability in constitution pattern classification, and reliability in judging the degree of constitution bias by: integrating the constitution state vector and individual physiological prior vector into the constitution-physiological parameter-syndrome feature association graph and combining it with the constitution feature pattern classification results; determining the expected improvement baseline curve by using the constitution evolution prediction trajectory set and reconstructing the actual constitution state sequence by combining it with real retest data; and realizing dynamic tracking of the constitution improvement process, updating of constitution state pattern classification, and deviation feedback correction, thereby improving individual adaptability and continuous monitoring stability in the conditioning and improvement process.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying, regulating, and improving constitution in Traditional Chinese Medicine, characterized in that, include: Multimodal physiological signals and medical records of subjects were collected and feature alignment and dimensionality reduction mapping were performed to obtain physical state vectors and individual physiological prior vectors. Input the physical condition vector and the individual physiological prior vector into the pre-constructed physical condition-physiological parameter-syndrome feature association graph, and output the physical condition identification data and physical condition bias data corresponding to the subject; Based on the physical constitution identification data and the physical constitution bias data, the core monitoring dimensions are obtained, and the core monitoring dimensions are constrained, trimmed and standardized according to the preset safety boundary to obtain a set of standardized conditioning and assessment vectors that match the physical condition of the subjects. The standardized conditioning assessment vector set and the physical condition state vector are input together into the continuous physical condition evolution calculation unit, and the set of physical condition evolution prediction trajectories corresponding to each conditioning suggestion is output. The expected improvement baseline curve is determined based on the set of predicted physical constitution evolution trajectories, and the actual physical constitution state sequence is reconstructed based on continuously collected real retest data. The dynamic deviation residual between the actual physical constitution state sequence and the expected improvement baseline curve is calculated. The edge weight reset reliability of the physical constitution-physiological parameter-syndrome characteristic correlation diagram and the parameters of the continuous physical constitution evolution calculation unit are corrected based on the dynamic deviation residual.

2. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The multimodal physiological signals include pulse signals, heart rate variability signals, electrodermal signals, tongue images, and facial images.

3. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The specific steps for obtaining the physical state vector and the individual physiological prior vector are as follows: Multimodal physiological signals and medical records of subjects were collected and synchronously labeled and standardized according to subject identification and collection time to obtain a standardized multimodal dataset. Physiological signal features, image representation features, and medical history text features are extracted from a standardized multimodal dataset and mapped to a unified physical evidence space for alignment, resulting in an aligned set of physical evidence features. The dimensionality reduction of the aligned physical evidence feature set is fused to obtain the physical state vector, and the individual physiological prior vector is generated by combining the resting stable physiological features.

4. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The specific steps for outputting the physical constitution identification data and physical constitution bias data corresponding to the subjects are as follows: The physical state vector is used as the current physical evidence anchor point, and the individual physiological prior vector is used as the individual baseline anchor point. The current physical evidence anchor point and the individual baseline anchor point are respectively connected to the physical node, physiological parameter node and syndrome feature node in the physical constitution-physiological parameter-syndrome feature association graph to form an individualized physical constitution relationship graph. Based on the individualized constitution relationship diagram, the evidence matching strength between the current constitution evidence anchor point and each physiological parameter node is calculated, and the target physiological parameter node and target syndrome feature node related to the subject's current state are activated according to the evidence matching strength. Based on the target physiological parameter node and the target syndrome feature node, read the node relationship and edge weight between the target physiological parameter node, the target syndrome feature node and each candidate constitution node, and convert the offset between the individual baseline anchor point and the target physiological parameter node into the individual offset evidence edge. Individual offset evidence edges are used to individually correct node relationships and edge weights, and combined with the constitution bias discrimination rule to suppress conflicting propagation paths, forming a set of constitution relationship diffusion paths for subjects. The set of physical constitution relationship diffusion paths is used to perform multi-hop relationship-aware reasoning on each candidate physical constitution node, and the relationship diffusion confidence of each candidate physical constitution node is calculated. The candidate physical constitution nodes are then sorted from high to low according to the relationship diffusion confidence. When at least two candidate physical nodes have the same relationship diffusion confidence, they are sorted from most to least effective propagation paths. When the number of effective propagation paths is still the same, the candidate physical nodes are sorted from high to low according to the average edge weight to obtain the sorting sequence. Based on the candidate constitution node ranking sequence, the dominant constitution node and the secondary constitution node are determined to form the constitution identification data corresponding to the subject. Then, according to the relationship diffusion confidence, individual deviation evidence edge strength and constitution bias discrimination rules corresponding to the dominant constitution node and the secondary constitution node, the bias degree value and bias level corresponding to each constitution node are determined to form the constitution bias degree data corresponding to the subject.

5. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The specific steps for obtaining the core monitoring dimensions are as follows: The dominant constitution node, the secondary constitution node, and the relationship diffusion confidence are read from the constitution identification data, and the bias degree value and bias level corresponding to each constitution node are read from the constitution bias degree data. The dominant constitution nodes, secondary constitution nodes, relationship diffusion confidence, bias degree value and bias level are sorted out to form a constitution bias contribution data table. Using the constitution bias contribution data table, the dominant constitution nodes and the secondary constitution nodes are matched to the preset conditioning dimension library, and the bias contribution matching value of each conditioning dimension is calculated to form a set of candidate intervention dimensions. Based on the set of candidate intervention dimensions, the bias contribution matching value and bias level corresponding to each candidate intervention dimension are read respectively, and the candidate intervention dimensions are sorted according to the rule that the higher the bias contribution matching value and the more severe the bias level, the higher the priority. The candidate intervention dimensions that are ranked first and meet the preset selection requirements are determined as the core monitoring dimensions.

6. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1 or 5, characterized in that, The specific steps for obtaining the standardized conditioning assessment vector set that matches the subject's physical condition are as follows: Extract the bias contribution matching value, bias level and priority from the core monitoring dimensions, and call the corresponding suggestion intensity range, suggestion frequency range and suggestion duration range from the adjustment dimension library to form the core monitoring dimension parameter table; Based on the core monitoring dimension parameter table, conditioning suggestion parameters are generated according to the bias contribution matching value and bias level, and the conditioning suggestion parameters are mapped to the safety boundary for constraint trimming. After the constraint trimming conditioning suggestion parameters are coded with unified dimensions, the priority of each core monitoring dimension is determined according to the bias contribution matching value from high to low and the bias level from severe to mild. Based on the priority, a standardized conditioning evaluation vector set that can characterize the conditioning direction, conditioning intensity and conditioning cycle is formed.

7. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The specific steps for outputting the set of predicted physical evolution trajectories corresponding to each conditioning suggestion are as follows: The intensity, frequency, and duration of each treatment suggestion are read from the standardized treatment assessment vector set and converted into a continuous treatment driving sequence according to the time axis of the simulation. By using a continuous conditioning driving sequence, each conditioning suggestion is concatenated with the physical condition vector, and combined with the observation period, continuous inference input data corresponding to each conditioning suggestion is generated. Based on continuous inference input data, the continuous physical evolution calculation unit continuously infers the changes in the physical state of the subjects on the inference time axis, forming a physical evolution state sequence. The trajectory is then organized according to the conditioning suggestion identifier and the inference time order to obtain the set of physical evolution prediction trajectories corresponding to each conditioning suggestion.

8. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The specific steps for determining the expected improvement baseline curve based on the set of predicted physical evolution trajectories are as follows: Read the physical evolution state sequence corresponding to each conditioning suggestion from the set of physical evolution prediction trajectories, and calculate the decrease in physical bias and trajectory smoothness corresponding to each physical evolution state sequence; Based on the magnitude of the decrease in physical bias and the smoothness of the trajectory corresponding to each physical evolution state sequence, the physical evolution state sequences are sorted and screened, and the physical evolution state sequences that meet the preset baseline determination requirements are determined as the expected improvement baseline curves.

9. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The process involves reconstructing the actual physical condition sequence based on continuously collected real retest data, and calculating the dynamic deviation residual between the actual physical condition sequence and the expected improvement baseline curve. The specific steps are as follows: Based on the expected improvement of the baseline curve, real retest data are continuously collected according to the preset retest cycle, and the actual physical condition vector corresponding to each retest time point is reconstructed through multimodal feature extraction, cross-modal alignment and physical evidence space dimensionality reduction mapping. Based on the actual physical condition vector, the data is serialized and organized according to the retest time sequence to form the actual physical condition sequence. The expected improvement baseline curve was mapped to the same retest time axis using the actual physical condition sequence, and the dynamic deviation residual between the actual physical condition sequence and the expected improvement baseline curve was calculated.

10. The method for identifying, regulating, and improving traditional Chinese medicine constitution as described in claim 1, characterized in that, The method of resetting the edge weights of the correlation graph of constitution-physiological parameters-syndrome characteristics based on dynamic deviation residuals and the parameters of the continuous constitution evolution calculation unit refers to using dynamic deviation residuals to screen target constitution state components whose deviation exceeds the preset tolerance threshold, and determining the target physiological parameter nodes and target syndrome characteristic nodes that cause deviations according to the preset mapping relationship between the target constitution state components and physiological parameter nodes and syndrome characteristic nodes. Based on the direction and magnitude of deviation, the edge weights of the relevant connecting edges in the constitution-physiological parameters-syndrome characteristic correlation graph are reset to improve reliability. The dynamic deviation residual is used as a calibration error signal to adaptively calibrate the parameters of the continuous constitution evolution calculation unit.

Images