Time sequence knowledge graph-based senile chronic disease risk prediction method and system

Abstract

The invention discloses an old people chronic disease risk prediction method and system based on a time sequence knowledge graph, and belongs to the technical field of medical health information, and the method comprises the following steps: S1, obtaining a health data set of an old people group; s2, on the basis of an ESOA-XGBoost algorithm, performing feature selection and sorting on each piece of health data in the health data set so as to determine correlation among variables, and generating a health data feature set; s3, constructing and training a time sequence knowledge graph prediction model including an STKGR-PR model and a T i PNN model, and predicting a relationship among the target entity, other entities and time sequence information through the time sequence knowledge graph prediction model; s4, predicting the old chronic disease risk of the old people group according to the trained time sequence knowledge graph prediction model; through the method, an elderly chronic disease risk prediction result with a good prediction effect and high accuracy can be obtained.

Description

Technical Field

[0001] The present invention relates to the field of medical health information technology, and in particular to a method and system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph. Background Art

[0004] Existing methods for predicting the risks of chronic diseases in the elderly are mainly divided into the following categories: prediction methods based on statistical models, prediction methods based on machine learning, prediction methods based on time series analysis, and prediction methods based on knowledge graphs.

[0005] These methods can predict the chronic disease risks of the elderly population to a certain extent, but they have certain limitations: the prediction methods based on statistical models are simple to use and easy to interpret, but have limited effects when dealing with high-dimensional and nonlinear data, and it is difficult to capture complex nonlinear relationships and interaction effects; the prediction methods based on machine learning can achieve good prediction results in some cases, but are sensitive to parameter selection, have high computational complexity, and may perform poorly when dealing with large-scale data sets; the prediction methods based on time series analysis can effectively capture the time dependence of data, but have limitations in feature selection and complex relationship modeling, and cannot fully utilize multi-dimensional health data for comprehensive risk prediction; the prediction methods based on knowledge graphs capture complex relationships in the data by constructing and inferring knowledge graphs, but there are still challenges in dealing with sparse data and reasoning paths, and require a large amount of labeled data for training.

[0006] In summary, the prediction accuracy of the above-mentioned methods for predicting the risk of chronic diseases in the elderly is not high. Therefore, it is necessary to propose a method for predicting the risk of chronic diseases in the elderly with good prediction effect. Summary of the invention

[0007] In view of the above-mentioned prior art, the present invention provides a method and system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, which mainly solves the technical problems existing in the above-mentioned background technology.

[0008] To achieve the above object, the technical solution of the embodiment of the present invention is implemented as follows:

[0009] In a first aspect, the present invention provides a method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the method comprising the following steps:

[0010] Step S1: Obtain the health data set of the elderly population;

[0011] Step S2: Based on the ESOA-XGBoost algorithm, feature selection and sorting are performed on each health data in the health data set to determine the correlation between variables and generate a health data feature set;

[0012] Step S3: construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, and predict the relationship between the target entity, other entities, and time series information through the time series knowledge graph prediction model;

[0013] Step S4: Predict the risk of chronic diseases in the elderly population based on the trained time series knowledge graph prediction model.

[0014] Optionally, the step S2 specifically includes:

[0015] Using the ESOA algorithm to update the position of the health data set, after each position update, filtering the health data set according to the discrimination condition to obtain a health data subset;

[0016] Inputting the healthy data subset into the XGBoost model for training, and adjusting the healthy data subset according to the training results;

[0017] The SHAP method is used to quantify the contribution of each feature of the healthy data subset to the output of the XGBoost model to obtain a healthy data feature set.

[0018] Optionally, the ESOA algorithm is used to update the position of the health data set, specifically including:

[0019] A waiting strategy is implemented to perform a preliminary position update on the health data set. The specific formula is as follows:

[0020] x a,i =x i +exp(-t / (0.1×t max ))×0.1×hop×g i

[0021] Among them, x a,i Updated position, x i represents the current position, exp represents the natural exponential function, t represents the current iteration number, and t max is the maximum number of iterations, hop represents the feasible range of the independent variable, g i is a random number;

[0022] According to the implementation result of the waiting strategy, the random walk strategy is further implemented to further update the position of the health data set. The specific formula is as follows:

[0023]

[0024] Among them, r b,i is a random number in the range of (π / 2,π / 2);

[0025] Based on the implementation results of the random walk strategy, the encirclement strategy is further implemented to achieve position optimization adjustment; the specific formula is as follows:

[0026] x c,i =(1-r i -r j )×x i +r i ×D h +r j ×D g

[0027] Among them, D h represents the difference matrix between the current position and the optimal position of the Egret team, D g Represents the difference matrix between the current position and the optimal position of the egret population, r i and r j is a random number in the range [0,0.5];

[0028] The specific formula of the discrimination condition is as follows:

[0029] x s,i =[x a,i ,x b,i ,x c,i ]

[0030] c i =argmin(y s,i )

[0031]

[0032] Among them, x s,i Represents a vector containing candidate eigenvalues ​​generated from three different sources, namely x a,i ,x b,i ,x c,i ;c i Indicates the index of selecting a specific source. The feature source that produces the smallest error or the best result is selected. s,i and i is the fitness value, and r is a random number.

[0033] Optionally, the STKGR-PR model is used to map the relationship between the target entity and other entities into a multidimensional space through multi-hop path reasoning and a reward mechanism, thereby vectorizing the relationship between the target entity and other entities in the temporal knowledge graph prediction model, thereby predicting the relationship between the target entity and other entities; the TiPNN model predicts the risk of chronic diseases in the elderly population based on the dependency relationship between temporal information and the target entity.

[0034] Optionally, construct a STKGR-PR model, specifically including:

[0035] Set the state representation and action space, the set state representation is:

[0036] s t =[r q ;e t ;h t ;tim]

[0037] Among them, s t is the current state at time t, e t is the current entity, r q For query relations, h t is the historical path information, tim is the timestamp;

[0038] The set action space is expressed as:

[0039] A t ={(r,e′,tim)|(e t ,r,e′,tim)∈TKG}∪{(self-loop:e t )};

[0040] Among them, A t is the action space, which means from the current entity e t All possible quadruples (e t ,r,e',tim), self-loop is a self-loop action, which means staying at the current node;

[0041] Set the conversion strategy, reward function, policy network, and dynamic path completion strategy, where the conversion strategy is expressed as:

[0042] s t+1 =[r q ;e t+1 ;h t+1 ;tim]

[0043] Among them, s t+1 is the state at time t+1, e t+1 is the entity at time t+1, r qFor query relations, h t+1 is the historical path information at time t+1, and tim is the timestamp;

[0044] The reward function is expressed as:

[0045]

[0046] Among them, R p represents the reward function, E(e o ,r q ,e T ) represents the expected score function of path embedding, and D(·) represents the cosine similarity function;

[0047] The policy network is expressed as:

[0048] h t =LSTM(h t-1 ,a t-1 )

[0049] π θ (a t ∣s t )=σ(A t ×W1RELU(W2|s t ))

[0050] Among them, h t is the hidden state at the current moment, indicating the path reasoning history information so far, LSTM is the long short-term memory network, h t-1 is the hidden state of the previous moment, including the historical information of the previous reasoning path, a t-1 is the action selected at the last moment, σ represents the softmax operator, W1 and W2 are the parameters of the two policy networks, π θ (a t |s t ) is the policy network in state s t Next select action a t The probability distribution of A t is the current state t The corresponding action embedding matrix, each row corresponds to a possible action, s t is the current state, including the current entity, the relationship of the query, and the historical path information from reasoning to the current step, and RELU is the activation function;

[0051] The dynamic path completion strategy is expressed as:

[0052]

[0053] A t =A t +Aadd

[0054] p((r p ,e p ,tim p )|s t )=p((r p ,tim p )|s t )p(e p |(r p ,tim p ),s t )

[0055] in, Represents the combination of the jth relation and timestamp, w is the current state s t The probability of each action being selected is SoftMax, which is used to convert the input value into a probability distribution to ensure that the sum of all output values ​​is 1. Each value represents the probability of selecting the corresponding action in the current state. MLP(s t ) is the current state s t Processing, s t is the current state, including the current entity's location and historical information; A t represents the original action space, A add represents a high-quality potential action set; p((r p ,tim p )|s t ) means in state s t Next select relation r p and timestamp tim p The probability, p(e p ∣(r p ,tim p ),s t ) means that in a given relationship r p and timestamp tim p Next select the tail entity e p probability.

[0056] Optionally, construct a TiPNN model, specifically including:

[0057] Build a historical time graph, define the time path, and initialize the time path representation. The constructed historical time graph is represented as follows:

[0058]

[0059] in, represents the historical time graph, which contains the subgraph information of all timestamps from time t-m+1 to t; ← represents the construction or update operation, which is used to generate the historical time graph, (s,rτ ,o) represents an edge in the historical time graph, where s is the subject entity and r τ is a relation type with time attribute τ, and o is an object entity; (s, r, o) represents a fact in the subgraph of timestamp τ, including subject entity s, relation r, and object entity o; G τ represents the subgraph at timestamp τ, which contains all facts at that specific point in time; t represents the timestamp of the query; m represents the length of the historical information used for reasoning, that is, how many historical timestamps are considered;

[0060] The defined time path is expressed as:

[0061]

[0062] Among them, ⊕ represents the path aggregation operator, represents the set of paths from s to o in the historical temporal graph, Represents the cardinality of the path set;

[0063] The initialization time path is expressed as:

[0064]

[0065] A time path aggregation layer is set, wherein the time path aggregation layer includes a time edge merging function and a path aggregation function, wherein the time edge merging function is expressed as:

[0066]

[0067] w r (z,p τ ,o)=g(Ψ r (p)||Y(Δτ))

[0068] Δτ=|τ-t q |

[0069]

[0070] Among them, TMSG(·) is the temporal edge merging function, represents the temporal path representation of entity z in the l-1th iteration, w r (z,p τ ,o) represents the time edge (z,p τ ,o), where r is the relation type, τ is the time attribute, z is the starting entity, and o is the ending entity; r (p) represents the query relation-aware basic static representation of edge type p, and Y(Δτ) represents the temporal edge type p r is the temporal embedding of , g(·) is a feed-forward neural network, tq represents the query timestamp, Δτ represents the time interval between query timestamps, and w * and φ * is a learnable parameter and is the dimension of the vector representation; d: represents the dimension of the vector representation, that is, the dimension of the output time encoding vector;

[0071] The path aggregation function is expressed as:

[0072]

[0073] Among them, AGG(·) is the path aggregation function, w r Query-aware temporal representations representing temporal edge types, is the time relation p in the historical time graph r A temporal edge between entity z and candidate entity o occurring at historical time t;

[0074] A scoring function and a loss function are set, and the conditional probability of future object candidates is calculated through the scoring function. The time series knowledge graph prediction model is guided to perform iterative optimization through the loss function, where the scoring function is expressed as:

[0075]

[0076] in, is a feedforward neural network, σ(·) is the sigmoid function;

[0077] By minimizing the negative log-likelihood loss of positive and negative samples, a regularization term is introduced to obtain the final loss function;

[0078] The loss function is expressed as:

[0079]

[0080] Among them, L TKG is the negative log-likelihood loss, L REG is the regularization parameter, L is the final loss function, p(s,r,o) represents the probability of the existence of the target object o given the subject s and the relation r. is a constant term used for normalization, is a negative sample triplet The probability of is the negative log-likelihood treatment; R T R: represents the matrix R and its transpose R T The product of , α is a scalar hyperparameter, and I is the identity matrix.

[0081] Optionally, the STKGR-PR model is used to map the relationship between the target entity and other entities into a multidimensional space through multi-hop path reasoning and a reward mechanism, so as to vectorize the relationship between the target entity and other entities in the temporal knowledge graph, thereby predicting the relationship between the target entity and other entities, specifically including:

[0082] Selecting an action according to the health data feature set, and performing a state transition based on the action;

[0083] Calculate the reward based on the converted state;

[0084] Optimizing a policy network based on the reward to generate a probability of selecting the action;

[0085] A dynamic path completion strategy is implemented based on the probability of generating and selecting the action, and high-probability relationship-timestamp combinations and tail entities are selected to generate a high-quality action set.

[0086] Optionally, the TiPNN model predicts the chronic disease risk of the elderly population based on the dependency relationship between the time series information and the target entity, specifically including:

[0087] Based on the time series information, build a historical time graph, define the time path and initialize it to obtain the time path characteristics;

[0088] Based on the time path feature, in the time path aggregation layer, time edge information is calculated using a time edge merging function, and the time edge information is integrated using a path aggregation function to obtain a new time path feature;

[0089] The new time path features are used to calculate the conditional probability of future object candidates through a scoring function, thereby predicting the chronic disease risks of the elderly population. In this process, the parameters in the TiPNN model are continuously iteratively adjusted through a loss function.

[0090] Optionally, the step S3 also includes: evaluating the time series knowledge graph prediction model to determine the reliability of the time series knowledge graph prediction model, and the evaluation includes error analysis and accuracy analysis.

[0091] In a second aspect, the present invention also provides a system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the system is used to implement the method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, and the system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph includes a data acquisition and processing module, a feature selection module, and a data prediction module;

[0092] The data acquisition and processing module is used to obtain the health data of the elderly patient and process the health data to obtain a health data set; the health data comes from the patient's medical records, examination reports, and living habits; the processing of the health data includes removing outliers, filling missing values, and standardizing the health data;

[0093] The feature selection module performs feature selection and sorting on the health data set output by the data acquisition and processing module based on the ESOA-XGBoost algorithm to generate a health data feature set;

[0094] The data prediction module is used to construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, input the health data feature set generated by the feature selection module into the STKGR-PR model for training, use multi-hop path reasoning and reward mechanism to map the relationship between the target entity and other entities into a multidimensional space, learn the complex relationship between the target entity and other entities, input the output result of the STKGR-PR model into the TiPNN model for deep learning and serial learning, and predict the chronic disease risk of the elderly population by predicting the relationship between time series information and target entities.

[0095] The beneficial effects of the present invention are as follows: the present invention first obtains a health data set, and based on the ESOA algorithm combined with the XGBoost model, performs feature selection and sorting on the health data of the elderly population to generate a health data feature set, optimizes the feature selection process, and improves the accuracy and robustness of the model; based on the STKGR-PR model, multi-hop path reasoning and reward mechanism are used to quantify the complex relationship between health data indicators. Through multi-hop path reasoning and reward mechanism, the problems of path sparsity and reward sparsity in the knowledge graph can be solved, and the reasoning ability and prediction accuracy can be improved; based on the TiPNN model combined with the time series information in the trained time series knowledge graph prediction model, deep learning and serialized learning are performed, which can capture the dynamic health data characteristics that change over time, further improve the learning ability and prediction accuracy of the model, and reduce dependence on large-scale labeled data; finally, a trained time series knowledge graph prediction model is obtained, and the risk of chronic diseases in the elderly population is predicted based on the trained time series knowledge graph prediction model, thereby obtaining a prediction result of chronic diseases in the elderly with good prediction effect and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 Schematic diagram of a method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph in an embodiment of the present invention;

[0097] Figure 2Schematic diagram of a system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph in an embodiment of the present invention. DETAILED DESCRIPTION

[0098] The technical solution of the present invention is further elaborated in detail below in conjunction with the drawings and specific embodiments of the specification. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, the expression "some embodiments" is related to a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0099] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0100] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments proposed herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and the scope of the present invention will be fully conveyed to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not intended to be a limitation of the present invention. When used herein, the singular forms of "one", "one" and "said / the" are also intended to include plural forms, unless the context clearly indicates another way. It should also be understood that the terms "compose" and / or "include" when used in this specification determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0101] It should also be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "inside", "outside", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only implementation method.

[0102] In order to fully understand the present invention, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present invention. The optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementations.

[0103] ESOA algorithm: Egret Swarm Optimization Algorithm;

[0104] XGBoost algorithm: distributed gradient boosting algorithm;

[0105] STKGR-PR model: a multi-hop path reasoning model for sparse temporal knowledge graphs based on path completion and reward shaping;

[0106] TiPNN model: based on time inductive path neural network;

[0107] Example 1

[0108] Please refer to the attached Figure 1 The present application provides a method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the method comprising the following steps:

[0109] Step S1: Obtain the health data set of the elderly population;

[0110] First, the health data of the elderly population is obtained, where the health data includes physical examination data, medical history data and living habit data; then the health data is cleaned, which includes removing outliers, missing values ​​or duplicate values ​​to ensure the accuracy and reliability of the data, and obtain a health data set for the elderly population.

[0111] Step S2: Based on the ESOA-XGBoost algorithm, feature selection and sorting are performed on each health data in the health data set to determine the correlation between variables and generate a health data feature set;

[0112] Step S3: construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, and predict the relationship between the target entity, other entities, and time series information through the time series knowledge graph prediction model;

[0113] Step S4: Predict the risk of chronic diseases in the elderly population based on the trained time series knowledge graph prediction model.

[0114] The health data is processed first to obtain a health data set, and then the health data set is input into ESOA-XGBoost for feature screening to obtain a health data feature set, and the health data features are optimized. The health data feature set is then input into the STKGR-PR model for training. The relationship and time series relationship between the target entity and other entities are used to capture the changes of each health data feature in the health data feature set over time, thereby improving the accuracy of model prediction. The output results of the STKGR-PR model are then input into the TiPNN model for optimization. The dynamic health data features that change over time are further captured in the TiPNN model, and the model is able to learn complex time dependencies. Finally, a time series knowledge graph prediction model with high learning ability and prediction accuracy is obtained, which can accurately predict the risk of chronic diseases in the elderly population.

[0115] As a further explanation of the above embodiment, step S2 specifically includes:

[0116] The settings include the egret population n, the maximum number of iterations t max , basic parameters of the ESOA algorithm including the discrimination conditions;

[0117] Using the ESOA algorithm to update the position of the health data set, after each position update, filtering the health data set according to the discrimination condition to obtain a health data subset;

[0118] Specifically, a waiting strategy is implemented to perform a preliminary position update on the health data set. The specific formula is as follows:

[0119] x a,i =x i +exp(-t / (0.1×t max ))×0.1×hop×g i

[0120] Among them, x a,i Updated position, x i represents the current position, exp represents the natural exponential function, t represents the current iteration number, and t max is the maximum number of iterations, hop represents the feasible range of the independent variable, g i is a random number;

[0121] According to the implementation result of the waiting strategy, the random walk strategy is further implemented to further update the position of the health data set. The specific formula is as follows:

[0122]

[0123] Among them, r b,i is a random number in the range of (π / 2,π / 2);

[0124] Based on the implementation results of the random walk strategy, the encirclement strategy is further implemented to achieve position optimization adjustment; the specific formula is as follows:

[0125] x c,i =(1-r i -r j )×x i +r i ×D h +r j ×D g

[0126] Among them, D h represents the difference matrix between the current position and the optimal position of the Egret team, D g Represents the difference matrix between the current position and the optimal position of the egret population, r i and r j is a random number in the range [0,0.5];

[0127] It should be noted that in the implementation of the waiting strategy, a preliminary position update is performed based on the current state in the early stage of the prediction process, and random numbers are used to simulate the random fluctuations of variables and adjust the state of the model; in the implementation of the random walk strategy, a further random walk is performed after the preliminary position update, so that the model can explore different possible paths or states and get closer to the optimal solution; in the implementation of the encirclement strategy, the position is further optimized and adjusted through the encirclement strategy to simulate a more accurate search near the target area;

[0128] The specific formula of the discrimination condition is as follows:

[0129] x s,i =[x a,i ,x b,i ,x c,i ]

[0130] c i =argmin(y s,i )

[0131]

[0132] Among them, x s,i Represents a vector containing candidate feature values ​​generated from three different sources (different models or methods), respectively x a,i ,x b,i ,x c,i ;c i Indicates the index of selecting a specific source. The feature source that produces the smallest error or the best result is selected. s, i and y i is the fitness value, r is a random number;

[0133] Inputting the healthy data subset into the XGBoost model for training, and adjusting the healthy data subset according to the training results;

[0134] Specifically, define the objective function and perform training based on XGBoost; the specific formula is as follows:

[0135]

[0136] Among them, l is the loss function, Ω is the regularization term, is the predicted value; φ is the predicted score function, and F is the function space of the decision tree; represents the prediction result, K represents the total number of trees, φ(|xi|) represents the prediction score of sample xi, F represents the function space associated with the decision tree, fk represents the Kth decision tree, and Ω(fj) represents the regularization term of the jth tree;

[0137] quantifying the contribution of each feature of the healthy data subset to the output of the XGBoost model using the SHAP method to obtain a healthy data feature set;

[0138] Specifically, the SHAP method is used to quantify the contribution of each feature to the model output. The specific formula is as follows:

[0139]

[0140] Among them, φ i is the SHAP value, z′ is the feature subset, M is the total number of features; |z′| represents the number of non-zero entries in z′, f x(z') represents the output of model f when only the features in subset z′ are used for prediction; f x (z′\i) represents the output of model f after removing feature i from subset z′, which is used to measure the impact of feature i on the model output when added to the feature subset; |z′|! (M-|z′|-1)! / M! This is a combination coefficient, which indicates how to weight subsets of different sizes when averaging the contribution of feature i over all subsets;

[0141] It should be noted that by enumerating subsets of all features, quantifying the impact of each feature on the model output in different combinations, and then calculating the average edge contribution of the feature, we can determine which health indicator is more important. Based on this, we can determine the features that play a key role in the model, and then choose which important features to retain and which unimportant features to remove, thereby optimizing the model's computational efficiency and prediction accuracy and determining the final health data feature set.

[0142] By inputting the health data set into ESOA and updating the position, after each position update, the parameters can be continuously adjusted to filter the features to obtain the health data subset, and then the health data subset is input into the XGBoost model for training, and the input of the XGBoost model is continuously optimized, thereby improving the prediction performance of the XGBoost model and obtaining a more relevant health data feature set.

[0143] As a further illustration of the above embodiment, the STKGR-PR model is used to map the relationship between the target entity and other entities into a multi-dimensional space through multi-hop path reasoning and a reward mechanism, thereby vectorizing the relationship between the target entity and other entities in the temporal knowledge graph prediction model, thereby predicting the relationship between the target entity and other entities;

[0144] The construction of the STKGR-PR model specifically includes:

[0145] Set the state representation and action space, the set state representation is:

[0146] s t =[r q ;e t ;h t ;tim]

[0147] Among them, s t is the current state at time t, e t is the current entity, r q For query relations, h t is the historical path information, tim is the timestamp;

[0148] The set action space is expressed as:

[0149] A t ={(r,e′,tim)|(e t ,r,e′,tim)∈TKG}∪{(self-loop:e t )};

[0150] Among them, A t is the action space, which means from the current entity e t All possible quadruples (e t ,r,e',tim), self-loop is a self-loop action, which means staying at the current node;

[0151] Set the conversion strategy, reward function, policy network, and dynamic path completion strategy, where the conversion strategy is expressed as:

[0152] s t+1 =[r q ;et+1 ;h t+1 ;tim]

[0153] Among them, s t+1 is the state at time t+1, e t+1 is the entity at time t+1, r q For query relations, h t+1 is the historical path information at time t+1, and tim is the timestamp;

[0154] The reward function is expressed as:

[0155]

[0156] Among them, R p represents the reward function, E(e o ,r q ,e T ) represents the expected score function of path embedding, and D(·) represents the cosine similarity function;

[0157] The policy network is expressed as:

[0158] h t =LSTM(h t-1 ,a t-1 )

[0159] π θ (a t ∣s t )=σ(A t ×W1RELU(W2|s t ))

[0160] Among them, h t is the hidden state at the current moment, indicating the path reasoning history information so far, LSTM is the long short-term memory network, h t-1 is the hidden state of the previous moment, including the historical information of the previous reasoning path, a t-1 is the action selected at the last moment, σ represents the softmax operator, W1 and W2 are the parameters of the two policy networks, π θ (a t |s t ) is the policy network in state s t Next select action a t The probability distribution of A t is the current state t The corresponding action embedding matrix, each row corresponds to a possible action (such as the relationship of other entities that can be jumped from the current entity), s t is the current state, including the current entity, the relationship of the query, and the historical path information from reasoning to the current step, and RELU is the activation function;

[0161] The dynamic path completion strategy is expressed as:

[0162]

[0163] A t =A t +A add

[0164] p((r p ,e p ,tim p )|s t )=p((r p ,tim p )|s t )p(e p |(r p ,tim p ),s t )

[0165] in, Represents the combination of the jth relation and timestamp, w is the current state s t The probability of each action (relationship) being selected is SoftMax, which is used to convert the input value into a probability distribution to ensure that the sum of all output values ​​is 1. Each value represents the probability of selecting the corresponding action in the current state. MLP (s t ) is the current state s t Processing, s t is the current state, including the current entity's location and historical information; A t represents the original action space, A add represents a high-quality potential action set; p((r p ,tim p )|s t ) means in state s t Next select relation r p and timestamp tim p The probability, p(e p ∣(r p ,tim p ),s t ) means that in a given relationship r p and timestamp tim p Next select the tail entity e p probability;

[0166] The STKGR-PR model is used to map the relationship between the target entity and other entities into a multi-dimensional space through multi-hop path reasoning and reward mechanism, and realize the vectorization of the relationship between the target entity and other entities in the temporal knowledge graph, so as to predict the relationship between the target entity and other entities, specifically including:

[0167] Selecting an action according to the health data feature set, and performing a state transition based on the action;

[0168] Calculate the reward based on the converted state;

[0169] Optimizing a policy network based on the reward to generate a probability of selecting the action;

[0170] Implementing a dynamic path completion strategy based on the probability of generating and selecting the action, selecting a high-probability relationship-timestamp combination and tail entity, and generating a high-quality action set;

[0171] It should be noted that action refers to the quadruple of relationship, entity, path information and timestamp in the time series knowledge graph, which are used for reasoning and jumping in the time series knowledge graph prediction model. The time series knowledge graph prediction model needs to select a suitable action (i.e., associated health data features and time information) from the current state (i.e., a specific health data point in the health data feature set) for the next step of reasoning. The selection of these actions is based on the health data feature set. The time series knowledge graph prediction model determines which features are the most important through the health data feature set, and selects the next action based on its time series information. The content of the health data feature set directly affects the decision of the time series knowledge graph prediction model in multi-hop path reasoning, and finally obtains the relevant health data features.

[0172] High-quality action sets refer to the best actions (the most relevant health data features in the health data feature set) selected through multi-hop path reasoning and reward mechanisms. By using time series information, the model can capture the most relevant health data features in a specific time period, thereby associating these health data features with the risk of chronic diseases. These action sets include not only disease-related features, but also the time dependency of the complex relationship between health indicators. In this way, the model can accurately predict the risk of chronic diseases in the elderly.

[0173] The STKGR-PR model is responsible for capturing the complex dependencies between the target entity and other entity relationships in the temporal knowledge graph prediction model, that is, capturing the complex dependencies between different health data features; since the STKGR-PR model needs to select a suitable action (that is, the associated health features and time information) from the current state (that is, a specific health data point) for the next reasoning, the selection of these actions is based on the health data feature set, so that the STKGR-PR model uses the health data feature set to determine which features are most important and select the next action based on its time series information; therefore, in the process of continuous training of the STKGR-PR model, the time factor is taken into account, so that the STKGR-PR model can track and understand the changes of various health indicators over time, and then evaluate the quality of the reasoning path through the reward mechanism, associate the time changes between different health indicators, and improve the prediction ability of the model.

[0174] As a further illustration of the above embodiment, the TiPNN model predicts the risk of chronic diseases in the elderly population based on the dependency relationship between the temporal information and the target entity;

[0175] The construction of the TiPNN model specifically includes:

[0176] Build a historical time graph, define the time path, and initialize the time path representation. The constructed historical time graph is represented as follows:

[0177]

[0178] in, represents the historical time graph, which contains the subgraph information of all timestamps from time t-m+1 to t; ← represents the construction or update operation, which is used to generate the historical time graph, (s,r τ ,o) represents an edge in the historical time graph (or called time edge), where s is the main entity and r τ is a relation type with time attribute τ, and o is an object entity; (s, r, o) represents a fact (or triple) in the subgraph of timestamp τ, including subject entity s, relation r, and object entity o; G τ represents the subgraph at timestamp τ, which contains all facts at that specific point in time; t: represents the timestamp of the query; m: represents the length of the historical information used for reasoning, that is, how many historical timestamps are considered;

[0179] The defined time path is expressed as:

[0180]

[0181] Among them, ⊕ represents the path aggregation operator, represents the set of paths from s to o in the historical temporal graph, Represents the cardinality of the path set;

[0182] The initialization time path is expressed as:

[0183]

[0184] A time path aggregation layer is set, wherein the time path aggregation layer includes a time edge merging function and a path aggregation function, wherein the time edge merging function is expressed as:

[0185]

[0186] w r (z,p τ ,o)=g(Ψ r (p)||Y(Δτ))

[0187] Δτ=|τ-t q |

[0188]

[0189] Among them, TMSG(·) is the temporal edge merging function, represents the temporal path representation of entity z in the l-1th iteration, w r (z,p τ ,o) represents the time edge (z,p τ ,o), where r is the relation type, τ is the time attribute, z is the starting entity, and o is the ending entity; r (p) represents the query relation-aware basic static representation of edge type p, and Y(Δτ) represents the temporal edge type p r is the temporal embedding of , g(·) is a feed-forward neural network, t q represents the query timestamp, Δτ represents the time interval between query timestamps, and w * and φ * is a learnable parameter and is the dimension of the vector representation; d: represents the dimension of the vector representation, that is, the dimension of the output time encoding vector;

[0190] The path aggregation function is expressed as:

[0191]

[0192] Among them, AGG(·) is the path aggregation function, w r Query-aware temporal representations representing temporal edge types, is the time relation p in the historical time graph r A temporal edge between entity z and candidate entity o occurring at historical time t;

[0193] A scoring function and a loss function are set, and the conditional probability of future object candidates is calculated through the scoring function. The time series knowledge graph prediction model is guided to perform iterative optimization through the loss function, where the scoring function is expressed as:

[0194]

[0195] in, is a feedforward neural network, σ(·) is the sigmoid function;

[0196] By minimizing the negative log-likelihood loss of positive and negative samples, a regularization term is introduced to obtain the final loss function;

[0197] The loss function is expressed as:

[0198]

[0199] Among them, L TKG is the negative log-likelihood loss, L REG is the regularization parameter, L is the final loss function, p(s,r,o) represents the probability of the existence of the target object o given the subject s and the relation r. is a constant term used for normalization, is a negative sample triplet The probability of is the negative log-likelihood treatment; R T R: represents the matrix R and its transpose R T The product of , α is a scalar hyperparameter, and I is the identity matrix.

[0200] The TiPNN model predicts the risk of chronic diseases in the elderly population based on the dependency relationship between time series information and target entities, specifically including:

[0201] Based on the time series information, build a historical time graph, define the time path and initialize it to obtain the time path characteristics;

[0202] Based on the time path feature, in the time path aggregation layer, time edge information is calculated using a time edge merging function, and the time edge information is integrated using a path aggregation function to obtain a new time path feature;

[0203] The new time path features are used to calculate the conditional probability of future object candidates through a scoring function, thereby predicting the chronic disease risks of the elderly population. In this process, the parameters in the TiPNN model are continuously iteratively adjusted through a loss function.

[0204] It should be noted that the time path aggregation layer in the TiPNN model is responsible for learning the characteristics of each health indicator changing over time through the layer message passing mechanism. The aggregation of the time path combines time information and path information, iteratively expands the path length, and captures the time path characteristics between the query target entity and other entities. This serialized learning method further optimizes the TiPNN model, enabling the TiPNN model to learn complex time dependencies;

[0205] The time edge merging function in the Ti PNN model is used to process the edges in the time path and merge the time information with the edge features. This ensures that the model has a high expressive ability in capturing time-related dynamics and can reflect the dynamic changes of various health indicators at different time nodes.

[0206] As a further illustration of the above embodiment, the step S3 also includes: evaluating the time series knowledge graph prediction model to determine the reliability of the time series knowledge graph prediction model, and the evaluation includes error analysis and accuracy analysis.

[0207] After obtaining the health data, data cleaning is performed to remove outliers, fill missing values ​​and standardize data to obtain a health data set, which can ensure the quality and consistency of the health data. Then, the health data set is combined with the XGBoost algorithm based on ESOA to select features for each health data, remove redundant features, and retain key features to generate a health data feature set. ESOA can effectively select important features, reduce the impact of noise features, and improve the accuracy and robustness of the model. In the step of establishing the temporal knowledge graph model, based on the health data feature set output by ESOA combined with XGBoost, a temporal knowledge graph prediction model is constructed for training. The STKGR-PR model is combined with multi-hop path reasoning and reward mechanism to capture the complex dependencies between features to predict the relationship between the target entity (chronic diseases of the elderly) and other entities (healthy features). The TiPNN model is used to predict the risk of chronic diseases in the elderly according to the dependency between the predicted target entity and other entities. Finally, the model is evaluated, and the prediction results are analyzed for error and accuracy to determine the reliability of the prediction model. In addition, the present invention adopts the TiPNN model for deep optimization and serialized learning, which can highlight the potential semantic information and multivariate information between health indicators, and effectively mine their dependencies using serialized learning. The Ti PNN model makes up for the limitations of the low accuracy of the traditional link prediction model, and improves the accuracy of risk prediction combining feature selection and time series knowledge graph. Therefore, the prediction method provided by the present invention can effectively improve the accuracy and reliability of risk prediction of chronic diseases in the elderly, reduce the impact of noise data, capture complex time dependencies, and through deep optimization, reduce the dependence on large-scale annotated data and computing resources, thereby improving the learning ability and prediction accuracy of the model.

[0208] Example 2

[0209] Please refer to the attached Figure 2 The present invention also provides a system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the system is used to implement a method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph includes a data acquisition and processing module, a feature selection module, and a data prediction module;

[0210] The data acquisition and processing module is used to obtain the health data of the elderly patient and process the health data to obtain a health data set; the health data comes from the patient's medical records, examination reports, and living habits; the processing of the health data includes removing outliers, filling missing values, and standardizing the health data;

[0211] The feature selection module performs feature selection and sorting on the health data set output by the data acquisition and processing module based on the ESOA-XGBoost algorithm to generate a health data feature set;

[0212] The data prediction module is used to construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, input the health data feature set generated by the feature selection module into the STKGR-PR model for training, use multi-hop path reasoning and reward mechanism to map the relationship between the target entity and other entities into a multidimensional space, learn the complex relationship between the target entity and other entities, input the output result of the STKGR-PR model into the TiPNN model for deep learning and serial learning, and predict the chronic disease risk of the elderly population by predicting the relationship between time series information and target entities.

[0213] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph, characterized in that: The method comprises the following steps: Step S1: Obtain the health data set of the elderly population; Step S2: Based on the ESOA-XGBoost algorithm, feature selection and sorting are performed on each health data in the health data set to determine the correlation between variables and generate a health data feature set; Step S3: construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, and predict the relationship between the target entity, other entities, and time series information through the time series knowledge graph prediction model; Step S4: Predict the risk of chronic diseases in the elderly population based on the trained time series knowledge graph prediction model.

2. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 1 is characterized in that: The step S2 specifically includes: Using the ESOA algorithm to update the position of the health data set, after each position update, filtering the health data set according to the discrimination condition to obtain a health data subset; Inputting the healthy data subset into the XGBoost model for training, and adjusting the healthy data subset according to the training results; The SHAP method is used to quantify the contribution of each feature of the healthy data subset to the output of the XGBoost model to obtain a healthy data feature set.

3. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 2 is characterized in that: The ESOA algorithm is used to update the position of the health data set, specifically including: A waiting strategy is implemented to perform a preliminary position update on the health data set. The specific formula is as follows: x a,i =x i +exp(-t / (0.1×t max ))×0.1×hop×g i Among them, x a,i Updated position, x i represents the current position, exp represents the natural exponential function, t represents the current iteration number, and t max is the maximum number of iterations, hop represents the feasible range of the independent variable, g i is a random number; According to the implementation result of the waiting strategy, the random walk strategy is further implemented to further update the position of the health data set. The specific formula is as follows: Among them, r b,i is a random number in the range of (π / 2,π / 2); Based on the implementation results of the random walk strategy, the encirclement strategy is further implemented to achieve position optimization adjustment; the specific formula is as follows: x c,i =(1-r i -r j )×x i +r i ×D h +r j ×D g Among them, D h represents the difference matrix between the current position and the optimal position of the Egret team, D g Represents the difference matrix between the current position and the optimal position of the egret population, r i and r j is a random number in the range [0,0.5]; The specific formula of the discrimination condition is as follows: x s,i =[x a,i ,x b,i ,x c,i ] c i =argmin(y s,i ) Among them, x s,i Represents a vector containing candidate eigenvalues ​​generated from three different sources, namely x a,i ,x b,i ,x c,i ;c i Indicates the index of selecting a specific source. The feature source that produces the smallest error or the best result is selected. s,i and i is the fitness value, and r is a random number.

4. The method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph according to claim 3 is characterized in that: The STKGR-PR model is used to map the relationship between the target entity and other entities into a multi-dimensional space through multi-hop path reasoning and a reward mechanism, so as to vectorize the relationship between the target entity and other entities in the temporal knowledge graph prediction model, thereby predicting the relationship between the target entity and other entities; The TiPNN model predicts the risk of chronic diseases in the elderly population based on the dependency between temporal information and target entities.

5. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 4 is characterized in that: Construct the STKGR-PR model, including: Set the state representation and action space, the set state representation is: s t =[r q ;e t ;h t ;tim] Among them, s t is the current state at time t, e t is the current entity, r q For query relations, h t is the historical path information, tim is the timestamp; The set action space is expressed as: A t ={(r,e′,tim)∣(e t ,r,e′,tim)∈TKG}∪{(self-loop:e t )}; Among them, A t is the action space, which means from the current entity e t All possible quadruples (e t ,r,e',tim), self-loop is a self-loop action, which means staying at the current node; Set the conversion strategy, reward function, policy network, and dynamic path completion strategy, where the conversion strategy is expressed as: s t+1 =[r q ;e t+1 ;h t+1 ;tim] Among them, s t+1 is the state at time t+1, e t+1 is the entity at time t+1, r q For query relations, h t+1 is the historical path information at time t+1, and tim is the timestamp; The reward function is expressed as: Among them, R p represents the reward function, E(e o ,r q ,e T ) represents the expected score function of path embedding, and D(·) represents the cosine similarity function; The policy network is expressed as: h t =LSTM(h t-1 ,a t-1 ) p θ (a t ∣s t )=σ(A t ×W1RELU(W2∣s t )) Among them, h t is the hidden state at the current moment, indicating the path reasoning history information so far, LSTM is the long short-term memory network, h t-1 is the hidden state of the previous moment, including the historical information of the previous reasoning path, a t-1 is the action selected at the last moment, σ represents the softmax operator, W1 and W2 are the parameters of the two policy networks, π θ (a t |s t ) is the policy network in state s t Next select action a t The probability distribution of A t is the current state t The corresponding action embedding matrix, each row corresponds to a possible action, s t is the current state, including the current entity, the relationship of the query, and the historical path information from reasoning to the current step, and RELU is the activation function; The dynamic path completion strategy is expressed as: is the entity with the highest probability} A t =A t +A add p((r p ,e p ,tim p )|s t )=p((r p ,tim p )|s t )p(e p |(r p ,tim p ),s t ) in, Represents the combination of the jth relation and timestamp, w is the current state s t The probability of each action being selected is SoftMax, which is used to convert the input value into a probability distribution to ensure that the sum of all output values ​​is 1. Each value represents the probability of selecting the corresponding action in the current state. MLP(s t ) is the current state s t Processing, s t is the current state, including the current entity's location and historical information; A t represents the original action space, A add represents a high-quality potential action set; p((r p ,tim p )|s t ) means in state s t Next select relation r p and timestamp tim p The probability, p(e p ∣(r p ,tim p ),s t ) means that in a given relationship r p and timestamp tim p Next select the tail entity e p probability.

6. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 5 is characterized in that: Build the TiPNN model, including: Build a historical time graph, define the time path, and initialize the time path representation. The constructed historical time graph is represented as follows: in, represents the historical time graph, which contains the subgraph information of all timestamps from time t-m+1 to t; ← represents the construction or update operation, which is used to generate the historical time graph, (s,r τ ,o) represents an edge in the historical time graph, where s is the subject entity and r τ is a relation type with time attribute τ, and o is an object entity; (s, r, o) represents a fact in the subgraph of timestamp τ, including subject entity s, relation r, and object entity o; G τ represents the subgraph at timestamp τ, which contains all facts at that specific time point; t represents the timestamp of the query; m represents the length of the historical information used for reasoning, that is, how many historical timestamps are considered; The defined time path is expressed as: in, represents the path aggregation operator, represents the set of paths from s to o in the historical temporal graph, Represents the cardinality of the path set; The initialization time path is expressed as: A time path aggregation layer is set, wherein the time path aggregation layer includes a time edge merging function and a path aggregation function, wherein the time edge merging function is expressed as: w r (z,p τ ,o)g(Ψ r (p)||Y(Δτ)) Δτ=|τ-t q | Among them, TMSG(·) is the temporal edge merging function, represents the temporal path representation of entity z in the l-1th iteration, w r (z,p τ ,o) represents the time edge (z,p τ ,o), where r is the relation type, τ is the time attribute, z is the starting entity, and o is the ending entity; r (p) represents the query relation-aware basic static representation of edge type p, and Y(Δτ) represents the temporal edge type p r is the temporal embedding of , g(·) is a feed-forward neural network, t q represents the query timestamp, Δτ represents the time interval between query timestamps, and w * and φ * is a learnable parameter and is the dimension of the vector representation; d: represents the dimension of the vector representation, that is, the dimension of the output time encoding vector; The path aggregation function is expressed as: Among them, AGG(·) is the path aggregation function, w r Query-aware temporal representations representing temporal edge types, is the time relation p in the historical time graph r A temporal edge between entity z and candidate entity o occurring at historical time t; A scoring function and a loss function are set, and the conditional probability of future object candidates is calculated through the scoring function. The time series knowledge graph prediction model is guided to perform iterative optimization through the loss function, where the scoring function is expressed as: in, is a feedforward neural network, σ(·) is the sigmoid function; By minimizing the negative log-likelihood loss of positive and negative samples, a regularization term is introduced to obtain the final loss function; The loss function is expressed as: Among them, L TKG is the negative log-likelihood loss, L REG is the regularization parameter, L is the final loss function, p(s,r,o) represents the probability of the existence of the target object o given the subject s and the relation r. is a constant term used for normalization, is a negative sample triplet The probability of is the negative log-likelihood treatment; R T R: represents the matrix R and its transpose R T The product of , α is a scalar hyperparameter, and I is the identity matrix.

7. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 6 is characterized in that: The STKGR-PR model is used to map the relationship between the target entity and other entities into a multi-dimensional space through multi-hop path reasoning and reward mechanism, and realize the vectorization of the relationship between the target entity and other entities in the temporal knowledge graph, so as to predict the relationship between the target entity and other entities, specifically including: Selecting an action according to the health data feature set, and performing a state transition based on the action; Calculate the reward based on the converted state; Optimizing a policy network based on the reward to generate a probability of selecting the action; A dynamic path completion strategy is implemented based on the probability of generating and selecting the action, and high-probability relationship-timestamp combinations and tail entities are selected to generate a high-quality action set.

8. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 7 is characterized in that: The TiPNN model predicts the risk of chronic diseases in the elderly population based on the dependency relationship between time series information and target entities, specifically including: Based on the time series information, build a historical time graph, define the time path and initialize it to obtain the time path characteristics; Based on the time path feature, in the time path aggregation layer, time edge information is calculated using a time edge merging function, and the time edge information is integrated using a path aggregation function to obtain a new time path feature; The new time path features are used to calculate the conditional probability of future object candidates through a scoring function, thereby predicting the chronic disease risks of the elderly population. In this process, the parameters in the TiPNN model are continuously iteratively adjusted through a loss function.

9. The method for predicting the risk of chronic diseases in the elderly based on time series knowledge graph according to claim 1 is characterized in that: The step S3 also includes: evaluating the time series knowledge graph prediction model to determine the reliability of the time series knowledge graph prediction model, and the evaluation includes error analysis and accuracy analysis.

10. A system for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph, the system being used to implement the method for predicting the risk of chronic diseases in the elderly based on a time series knowledge graph as claimed in any one of claims 1 to 9, characterized in that: The system includes a data acquisition and processing module, a feature selection module, and a data prediction module; The data acquisition and processing module is used to obtain the health data of the elderly patient and process the health data to obtain a health data set; the health data comes from the patient's medical records, examination reports, and living habits; The processing of the health data includes removing outliers, filling missing values ​​and standardizing the health data; The feature selection module performs feature selection and sorting on the health data set output by the data acquisition and processing module based on the ESOA-XGBoost algorithm to generate a health data feature set; The data prediction module is used to construct and train a time series knowledge graph prediction model including the STKGR-PR model and the TiPNN model, input the health data feature set generated by the feature selection module into the STKGR-PR model for training, use multi-hop path reasoning and reward mechanism to map the relationship between the target entity and other entities into a multidimensional space, learn the complex relationship between the target entity and other entities, input the output result of the STKGR-PR model into the TiPNN model for deep learning and serial learning, and predict the chronic disease risk of the elderly population by predicting the relationship between time series information and target entities.

Images