Heart rate monitoring method and system based on neural network

Abstract

The invention belongs to the technical field of heart rate monitoring, and discloses a heart rate monitoring method and system based on a neural network. The method comprises the following steps: constructing a model by using a neural network algorithm based on a cloud computing center; acquiring real-time postoperative clinical data and real-time heart rate acquisition data based on a user terminal; generating a user portrait based on the cloud computing center; carrying out heart rate abnormity identification; heart rate monitoring optimization is carried out; the real-time heart rate monitoring optimization strategy is adjusted; postoperative rehabilitation suggestions are generated; sending a real-time heart rate abnormity identification result, an adjusted real-time heart rate monitoring optimization strategy and a real-time postoperative rehabilitation suggestion; generating a real-time alarm signal based on a user terminal, and visualizing the real-time alarm signal; a real-time heart rate monitoring instruction is generated and sent to the heart rate collecting device. According to the invention, the problems of complex data management, low practicability, large labor cost investment, poor accuracy, poor functionality and poor user experience in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of heart rate monitoring, and in particular relates to a heart rate monitoring method and system based on a neural network. Background Art

[0002] Heart rate monitoring is an important technology in the medical field. It is used to evaluate heart function and monitor the health status of users, and is applied to postoperative health management. With the advancement of technology, heart rate monitoring methods are also constantly developing. Traditional heart rate monitoring requires users to wear various heart rate acquisition devices. The heart rate data collected by these heart rate acquisition devices are complex to manage and have low practicality. They need to be analyzed with the help of manpower, which has high labor cost investment, poor accuracy, and poor functionality. They only achieve simple heart rate acquisition and often use preset heart rate monitoring modes. They lack customized functions for different rehabilitation situations and different users, and the user experience is poor. Summary of the invention

[0003] In order to solve the problems of complex data management, low practicality, high manpower cost, poor accuracy, poor functionality and poor user experience in the prior art, the present invention aims to provide a heart rate monitoring method and system based on a neural network.

[0004] The technical solution adopted by the present invention is:

[0005] A heart rate monitoring method based on a neural network comprises the following steps:

[0006] Based on the cloud computing center, using neural network algorithms, we build user portrait generation models, knowledge graphs, heart rate abnormality recognition models, and heart rate monitoring optimization models;

[0007] Based on the user terminal, the real-time postoperative clinical data and the real-time heart rate data sent by the heart rate collection device are received and uploaded to the cloud computing center;

[0008] Based on the cloud computing center, the user portrait generation model is used to generate user portraits according to real-time postoperative clinical data to obtain real-time user portraits;

[0009] Based on the cloud computing center, the heart rate abnormality recognition model is used to identify the heart rate abnormality according to the real-time postoperative clinical data, real-time heart rate collection data and real-time user portraits, and the real-time heart rate abnormality recognition results are obtained;

[0010] Based on the cloud computing center, according to the real-time heart rate abnormality recognition results and real-time user portraits, the heart rate monitoring optimization model is used to optimize the heart rate monitoring and obtain the real-time heart rate monitoring optimization strategy;

[0011] Based on the cloud computing center, the real-time heart rate monitoring optimization strategy is adjusted using the knowledge graph to obtain the adjusted real-time heart rate monitoring optimization strategy;

[0012] Based on the cloud computing center, postoperative rehabilitation suggestions are generated based on real-time user portraits and real-time heart rate abnormality recognition results using knowledge graphs to obtain real-time postoperative rehabilitation suggestions;

[0013] Based on the cloud computing center, the real-time heart rate abnormality recognition results, the adjusted real-time heart rate monitoring optimization strategy and the real-time postoperative rehabilitation suggestions are sent to the corresponding user terminals;

[0014] Based on the user terminal, a real-time alarm signal is generated according to the real-time abnormal heart rate recognition result, and the real-time alarm signal, the real-time abnormal heart rate recognition result and the real-time postoperative rehabilitation advice are visualized;

[0015] Based on the user terminal, a real-time heart rate monitoring instruction is generated according to the adjusted real-time heart rate monitoring optimization strategy, and the real-time heart rate monitoring instruction is sent to the heart rate collection device.

[0016] Furthermore, based on the cloud computing center, a neural network algorithm is used to construct a user portrait generation model, a knowledge graph, a heart rate abnormality recognition model, and a heart rate monitoring optimization model, including the following steps:

[0017] Based on the cloud computing center, some heart rate monitoring and postoperative rehabilitation knowledge, some historical postoperative clinical data and some historical heart rate collection data are collected;

[0018] Perform preprocessing to obtain some post-preprocessing heart rate monitoring and post-operative rehabilitation knowledge, some post-preprocessing historical post-operative clinical data, and some post-preprocessing historical heart rate collection data;

[0019] Performing data dimensionality reduction on a number of pre-processed historical postoperative clinical data to obtain a number of dimensionality-reduced historical postoperative clinical data and key clinical indicator sets;

[0020] Based on several historical postoperative clinical data after dimensionality reduction and several historical heart rate collection data after preprocessing, a user portrait generation model is constructed using a neural network algorithm, and several historical user portraits are generated;

[0021] Based on the pre-processed heart rate monitoring and post-operative rehabilitation knowledge, a knowledge graph was constructed using a natural language processing algorithm;

[0022] Based on several historical postoperative clinical data after dimensionality reduction, several historical heart rate collection data after preprocessing, and several historical user portraits, a neural network algorithm is used to build a heart rate abnormality recognition model and generate several historical heart rate abnormality recognition results;

[0023] Based on several historical heart rate abnormality recognition results and several historical user portraits, a heart rate monitoring optimization model is constructed using a reinforcement learning algorithm.

[0024] Furthermore, based on the pre-processed heart rate monitoring and post-operative rehabilitation knowledge, a natural language processing algorithm is used to construct a knowledge graph, including the following steps:

[0025] Using a pre-trained named entity extraction model, extract several heart rate monitoring named entities and several postoperative rehabilitation named entities of the pre-processed heart rate monitoring and postoperative rehabilitation knowledge;

[0026] Use a pre-trained entity relationship extraction model to extract heart rate monitoring entity relationships between several heart rate monitoring named entities and postoperative rehabilitation entity relationships between several postoperative rehabilitation named entities;

[0027] A knowledge graph is constructed based on several heart rate monitoring named entities, several heart rate monitoring entity relationships, several postoperative rehabilitation named entities, and several postoperative rehabilitation entity relationships.

[0028] Furthermore, the user portrait generation model is built based on the N-GAN-Attention-LSTM algorithm;

[0029] The abnormal heart rate recognition model is built based on the RNN-GNN-MLP algorithm;

[0030] The heart rate monitoring optimization model is built based on the DQN algorithm.

[0031] Furthermore, based on the cloud computing center, according to the real-time postoperative clinical data, the user portrait generation model is used to generate the user portrait to obtain the real-time user portrait, including the following steps:

[0032] Based on the cloud computing center, the real-time postoperative clinical data is reduced in dimension according to the key clinical indicator set to obtain the real-time postoperative clinical data after dimension reduction;

[0033] Input the reduced-dimensional real-time postoperative clinical data into the user portrait generation model to extract N real-time key dimension features;

[0034] According to the preset attention weights, N real-time key dimension features are weighted and spliced ​​to obtain real-time weighted splicing features;

[0035] According to the real-time weighted splicing features, user label prediction is performed to obtain a number of real-time user labels, and user portrait generation is performed based on the several real-time user labels to obtain a real-time user portrait.

[0036] Furthermore, based on the cloud computing center, according to the real-time postoperative clinical data, the real-time heart rate collection data and the real-time user portrait, the heart rate abnormality recognition model is used to perform heart rate abnormality recognition to obtain the real-time heart rate abnormality recognition result, including the following steps:

[0037] Based on the cloud computing center, the real-time postoperative clinical data after dimensionality reduction, the real-time heart rate collection data and the real-time user portrait are input into the heart rate abnormality recognition model;

[0038] Extract the real-time postoperative clinical data features of the real-time postoperative clinical data after dimensionality reduction, the real-time heart rate collection data features of the real-time heart rate collection data, and the real-time graph features of the real-time user portrait;

[0039] Perform feature fusion on real-time postoperative clinical data features, real-time heart rate collection data features and real-time graph features to obtain real-time fusion features;

[0040] According to the real-time fusion features, abnormal heart rate recognition is performed to obtain a real-time abnormal heart rate prediction label, that is, a real-time abnormal heart rate recognition result.

[0041] Furthermore, based on the cloud computing center, according to the real-time heart rate abnormality recognition results and the real-time user portrait, the heart rate monitoring optimization model is used to optimize the heart rate monitoring, and the real-time heart rate monitoring optimization strategy is obtained, including the following steps:

[0042] Based on the cloud computing center, according to the real-time user portrait, several historical heart rate monitoring optimization experiences matching in the experience playback pool of the heart rate monitoring optimization model are extracted;

[0043] According to several historical heart rate monitoring optimization experiences and real-time heart rate abnormality recognition results, the model parameters of the heart rate monitoring optimization model are updated to obtain updated model parameters;

[0044] According to the updated model parameters, the heart rate monitoring optimization model is used to optimize the heart rate monitoring, and a number of real-time execution actions are obtained. Based on the number of real-time execution actions, a real-time heart rate monitoring optimization strategy is obtained.

[0045] Further, based on the cloud computing center, the real-time heart rate monitoring optimization strategy is adjusted using the knowledge graph to obtain the adjusted real-time heart rate monitoring optimization strategy, including the following steps;

[0046] Based on the cloud computing center, according to the action named entities corresponding to the real-time execution actions in the real-time heart rate monitoring optimization strategy, the heart rate monitoring named entities in the knowledge graph are searched and matched to obtain several matching heart rate monitoring named entities;

[0047] According to a number of matching heart rate monitoring named entities and the heart rate monitoring entity relationships between the matching heart rate monitoring named entities and other heart rate monitoring named entities, the real-time heart rate monitoring optimization strategy is adjusted to obtain an adjusted real-time heart rate monitoring optimization strategy.

[0048] Furthermore, based on the cloud computing center, according to the real-time user portrait and the real-time abnormal heart rate recognition results, the knowledge graph is used to generate postoperative rehabilitation suggestions, and the real-time postoperative rehabilitation suggestions are obtained, including the following steps;

[0049] Based on the cloud computing center, according to the user named entity corresponding to the real-time user label in the real-time user portrait and the abnormal named entity corresponding to the real-time heart rate abnormality prediction label in the real-time heart rate abnormality recognition result, the postoperative rehabilitation named entity in the knowledge graph is searched and matched to obtain several matching postoperative rehabilitation named entities;

[0050] According to a number of matching postoperative rehabilitation named entities and the postoperative rehabilitation entity relationship between the matching postoperative rehabilitation named entity and other postoperative rehabilitation named entities, postoperative rehabilitation suggestions are generated to obtain real-time postoperative rehabilitation suggestions.

[0051] A heart rate monitoring system based on a neural network is used to implement a heart rate monitoring method. The system includes a cloud computing center, a plurality of user terminals and a plurality of heart rate acquisition devices. The cloud computing center is respectively connected to the plurality of user terminals in communication, and each user terminal is connected to a heart rate acquisition device in communication in a one-to-one correspondence.

[0052] The cloud computing center includes a model building unit, a user portrait generating unit, a heart rate abnormality identifying unit, a heart rate monitoring optimization unit, an optimization strategy adjusting unit, a postoperative rehabilitation suggestion generating unit and a data sending unit which are connected in sequence.

[0053] The beneficial effects of the present invention are:

[0054] The present invention discloses a heart rate monitoring method and system based on a neural network, which combines real-time postoperative clinical data and real-time heart rate collection data sent by a heart rate collection device, more comprehensively reflects the user's health status information, and improves the comprehensiveness of the data; uniformly manages and analyzes the data based on a cloud computing center, and improves practicality; the constructed user portrait generation model and heart rate monitoring optimization model can generate personalized heart rate monitoring plans according to the user's specific health status and heart rate data, realizes customization functions for different rehabilitation conditions and different users, and improves user experience; the constructed heart rate abnormality recognition model can combine real-time postoperative clinical data, real-time heart rate collection data and real-time user portraits to identify heart rate abnormalities, reduce labor cost investment, and improve the accuracy of heart rate abnormality recognition; the constructed knowledge graph can generate real-time postoperative rehabilitation suggestions according to real-time user portraits and real-time heart rate abnormality recognition results, and improves functionality.

[0055] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is the process flow of the heart rate monitoring method based on neural network in the present invention. Figure 1 .

[0057] Figure 2 This is the process flow of the heart rate monitoring method based on neural network in the present invention. Figure 2 .

[0058] Figure 3 It is a structural block diagram of the heart rate monitoring system based on neural network in the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0060] Embodiment 1:

[0061] like Figure 1 and Figure 2 As shown together, this embodiment provides a heart rate monitoring method based on a neural network, comprising the following steps:

[0062] S1: Based on the cloud computing center, using the neural network algorithm, construct the user portrait generation model, knowledge graph, heart rate abnormality recognition model and heart rate monitoring optimization model, including the following steps:

[0063] S1-1: Based on the cloud computing center, collect some heart rate monitoring and postoperative rehabilitation knowledge, some historical postoperative clinical data and some historical heart rate collection data;

[0064] Heart rate monitoring and postoperative rehabilitation knowledge includes heart rate-related medical knowledge and postoperative rehabilitation-related medical knowledge. Heart rate-related medical knowledge includes the definition of heart rate, symptoms of abnormal heart rate, heart rate monitoring rules, etc. Postoperative rehabilitation-related medical knowledge includes abnormal heart rate range, postoperative rehabilitation rules, etc.;

[0065] Postoperative clinical data includes the user's basic information, health status, living habits, exercise preferences, physiological data, clinical consultation, surgical items, etc.;

[0066] Heart rate data collection includes electrocardiogram data, photoplethysmography data, motion sensor data, pulse waveform data, phonocardiogram data, etc.

[0067] S1-2: Perform preprocessing to obtain some post-preprocessing heart rate monitoring and post-operative rehabilitation knowledge, some post-preprocessing historical post-operative clinical data, and some post-preprocessing historical heart rate collection data;

[0068] Preprocessing includes noise removal, data screening, data cleaning and normalization of the collected data in order to eliminate noise interference, remove duplicate and erroneous data, and eliminate data magnitude differences to improve data quality;

[0069] S1-3: Using the principal component analysis (PCA) method, a number of pre-processed historical postoperative clinical data are subjected to data dimensionality reduction to obtain a number of reduced-dimensional historical postoperative clinical data and key clinical indicator sets, including the following steps:

[0070] S1-3-1: Perform matrix conversion on some pre-processed historical postoperative clinical data to obtain the corresponding historical postoperative clinical data matrix X = [x1, x2, ...x p ,...,x n ] T , where x p is the pth row vector of historical postoperative clinical data after pretreatment, p is the row vector indicator, and n is the total number of historical postoperative clinical data after pretreatment; the initial row vector of the historical postoperative clinical data matrix is ​​the historical postoperative clinical data after pretreatment, and the initial column vector of the historical postoperative clinical data matrix is ​​the historical postoperative clinical index data, and the historical postoperative clinical index data includes postoperative clinical indicators including basic information indicators, health status indicators, living habit indicators, exercise preference indicators, physiological data indicators, clinical consultation indicators, surgical project indicators, etc., and each indicator data corresponds to the data value of each column in the historical postoperative clinical data;

[0071] S1-3-2: Standardize the historical postoperative clinical data matrix to obtain the corresponding standardized historical postoperative clinical data matrix;

[0072] The formula is:

[0073]

[0074] Where, X' is the historical postoperative clinical data matrix after standardization; μ is the mean of the historical postoperative clinical data matrix; σ is the variance of the historical postoperative clinical data matrix;

[0075] S1-3-3: Obtain the covariance matrix of the standardized historical postoperative clinical data matrix, and obtain the corresponding candidate principal component matrix based on the standardized historical postoperative clinical data matrix and the covariance matrix; the candidate row vectors of the candidate principal component matrix are the preprocessed historical postoperative clinical data, and the candidate column vectors of the candidate principal component matrix are the historical postoperative clinical candidate indicator data;

[0076] The formula is:

[0077]

[0078] Where D is the covariance matrix of the historical postoperative clinical data matrix after standardization; Y is the candidate principal component matrix; P is the transformation matrix; E is the unit eigenvector matrix; n is the total number of historical postoperative clinical data after preprocessing;

[0079] Y=PX'

[0080] Where, Y is the candidate principal component matrix; P is the transformation matrix; X' is the historical postoperative clinical data matrix after standardization;

[0081] S1-3-4: Set the candidate principal component matrix Y = [y1, y2, ...y l ,...,y L ] Several candidate column vectors y' of the top 90% of the cumulative contribution rate of variance l As the corresponding principal component column vectors, the reduced dimension historical postoperative clinical data matrix Y'=[y'1,y'2,...y' l ,...,y K ]; the key row vector of the historical postoperative clinical data matrix after dimensionality reduction is the historical postoperative clinical data after dimensionality reduction, and the data index corresponding to the key column vector of the historical postoperative clinical data matrix after dimensionality reduction is the key clinical index, and a set of key clinical indicators is obtained;

[0082] The formula is:

[0083]

[0084] In the formula, G is the cumulative contribution rate of variance; l The lth candidate principal component y l The variance of ; l is the indicator of the candidate principal component; L is the total number of candidate principal components; K is the total number of principal components;

[0085] S1-3-5: Perform inverse matrix transformation on the dimension-reduced historical postoperative clinical data matrix to obtain a number of dimension-reduced historical postoperative clinical data;

[0086] S1-4: Based on several historical postoperative clinical data after dimensionality reduction and several historical heart rate collection data after preprocessing, the N-GAN-Attention-LSTM algorithm is used to build a user portrait generation model and generate several historical user portraits;

[0087] The user portrait generation model includes N key dimension feature extraction modules based on the Generative Adversarial Networks (GAN) algorithm, an attention weight generation module based on the Attention mechanism, and a user portrait generation module based on the Long Short-Term Memory (LSTM) network, where N is the total number of key dimensions, including the surgical level dimension, the rehabilitation difficulty dimension, the heart rate risk dimension, and the postoperative status dimension.

[0088] The key dimension feature extraction module includes a generator and a discriminator. The goal of the generator is to generate sufficiently real data, while the goal of the discriminator is to distinguish between real data and data generated by the generator. In the key dimension feature extraction, GAN can be used to generate data that matches the real data features to achieve feature extraction of key dimensions. The attention weight generation module uses the preset attention weights to achieve splicing of key dimension features and integrate scattered features. The LSTM network is a special recurrent neural network. The special cell state design makes it perform well in processing and predicting time series data. It can capture long-term dependencies and is suitable for data with time correlation. It is used to predict spliced ​​data features.

[0089] S1-5: Based on the pre-processed heart rate monitoring and post-operative rehabilitation knowledge, a knowledge graph is constructed using a natural language processing algorithm, including the following steps:

[0090] S1-5-1: Use the pre-trained named entity extraction model to extract several heart rate monitoring named entities and several postoperative rehabilitation named entities of the pre-processed heart rate monitoring and postoperative rehabilitation knowledge;

[0091] The named entity extraction model is built based on the BERT-BiLSTM-CRF algorithm. The pre-trained language sub-model of bidirectional encoder representation from Transformers (BERT) is used to vectorize the pre-processed heart rate monitoring and postoperative rehabilitation knowledge to obtain the knowledge data vector. The bidirectional long short-term memory network (BiLSTM) is used to extract the semantic features of the knowledge data vector. The linear chain conditional random field module (CRF) is used to annotate the named entities according to the semantic features to obtain the corresponding heart rate monitoring named entities and postoperative rehabilitation named entities.

[0092] S1-5-2: Use the pre-trained entity relationship extraction model to extract the heart rate monitoring entity relationships between several heart rate monitoring named entities and the postoperative rehabilitation entity relationships between several postoperative rehabilitation named entities;

[0093] The entity relationship extraction model is built based on the BERT-BiGRU-Attention algorithm. The BERT model is used to vectorize the preprocessed heart rate monitoring and postoperative rehabilitation knowledge to obtain the knowledge data vector. The bidirectional recurrent neural network (BiGRU) is used to extract the semantic features of the knowledge data vector. The Attention mechanism is used to assign attention weights to each channel of the bidirectional recurrent neural network to reduce the impact of knowledge with incorrect relationship annotations. Based on the BiGRU network, the heart rate monitoring entity relationships between several heart rate monitoring named entities and the postoperative rehabilitation entity relationships between several postoperative rehabilitation named entities are extracted according to several heart rate monitoring named entities and several postoperative rehabilitation named entities, the knowledge data vector and the attention weights.

[0094] S1-5-3: Construct a knowledge graph based on a number of heart rate monitoring named entities, a number of heart rate monitoring entity relationships, a number of postoperative rehabilitation named entities, and a number of postoperative rehabilitation entity relationships;

[0095] S1-6: Based on several historical postoperative clinical data after dimensionality reduction, several historical heart rate collection data after preprocessing, and several historical user portraits, the RNN-GNN-MLP algorithm is used to build a heart rate abnormality recognition model and generate several historical heart rate abnormality recognition results;

[0096] The heart rate abnormality recognition model includes a data feature extraction module based on the Recurrent Neural Network (RNN) algorithm, a graph feature extraction module based on the Graph Neural Network (GNN) algorithm, and a fusion prediction module based on the Multi-Layer Perceptron (MLP) algorithm.

[0097] The RNN network is a special neural network architecture that allows information to flow in time series data and can process and predict sequence data. It is used to extract postoperative clinical data features. The GNN network is a neural network specifically used to process graph data, where the graph consists of nodes and edges. It is used to extract graph features of user portraits, including node information of user labels and edge information between user labels. The MLP network is a fully connected layer, which is characterized by all neurons between layers being interconnected. It is used to fuse different features to obtain corresponding fused features, improve data representation capabilities, and perform feature fusion. Based on the fused features, abnormal heart rate recognition is performed.

[0098] S1-7: Based on several historical heart rate abnormality recognition results and several historical user portraits, a heart rate monitoring optimization model is constructed using the Deep Q Network (DQN) algorithm, including the following steps:

[0099] S1-7-1: Generate the simulation environment of the DQN algorithm using the heart rate monitoring optimization solution to build the intelligent agent and experience replay pool;

[0100] S1-7-2: According to each heart rate state type corresponding to the historical heart rate abnormality recognition results, define the state space of the DQN algorithm, and the parameters of the state space correspond to each heart rate state, for example, too fast heart rate, normal heart rate, too slow heart rate, etc.;

[0101] S1-7-3: Define the action space of the DQN algorithm according to the actions that the heart rate monitoring optimization strategy needs to output; for example, reduce the monitoring frequency, increase the monitoring time, handle abnormal situations, etc.;

[0102] S1-7-4: Based on the possible impact of each action in the action space, define the reward function of the DQN algorithm to evaluate the quality or impact of the action;

[0103] S1-7-5: Construct the input layer, several hidden layers, and output layer of the deep Q network, connect the input layer to the state space, and connect the output layer to the action space;

[0104] S1-7-6: Based on the state space, action space and reward function, according to several historical heart rate abnormality recognition results, the deep Q network and the agent are optimized and trained to build a heart rate monitoring optimization model, and several historical heart rate monitoring optimization experiences are generated;

[0105] S1-7-7: Use historical user portraits as retrieval tags for historical heart rate monitoring optimization experiences, and store several historical heart rate monitoring optimization experiences with retrieval tags in an experience playback pool;

[0106] S2: Based on the user terminal, receive the input real-time postoperative clinical data and the real-time heart rate data sent by the heart rate collection device, and upload them to the cloud computing center;

[0107] S3: Based on the cloud computing center, according to the real-time postoperative clinical data, the user portrait generation model is used to generate the user portrait to obtain the real-time user portrait, including the following steps:

[0108] S3-1: Based on the cloud computing center, the real-time postoperative clinical data is reduced in dimension according to the key clinical indicator set to obtain the real-time postoperative clinical data after dimension reduction;

[0109] S3-2: Input the reduced-dimensional real-time postoperative clinical data into the user portrait generation model, and use the key dimension feature extraction module to extract the real-time key dimension features of the reduced-dimensional real-time postoperative clinical data in N key dimensions;

[0110] S3-3: According to the preset attention weights, use the attention weight generation module to weightedly splice N real-time key dimension features to obtain real-time weighted splicing features;

[0111] S3-4: Based on the real-time weighted splicing features, a user portrait generation module is used to predict user tags to obtain a number of real-time user tags, and a user portrait is generated based on the real-time user tags to obtain a real-time user portrait;

[0112] S4: Based on the cloud computing center, according to the real-time postoperative clinical data, real-time heart rate collection data and real-time user portrait, the heart rate abnormality recognition model is used to perform heart rate abnormality recognition to obtain the real-time heart rate abnormality recognition result, including the following steps:

[0113] S4-1: Based on the cloud computing center, the real-time postoperative clinical data after dimensionality reduction, the real-time heart rate collection data and the real-time user portrait are input into the heart rate abnormality recognition model;

[0114] S4-2: using a data feature extraction module to extract real-time postoperative clinical data features of the real-time postoperative clinical data after dimensionality reduction, using a data feature extraction module to extract real-time heart rate collection data features of the real-time heart rate collection data, and using a graph feature extraction module to extract real-time graph features of the real-time user portrait;

[0115] S4-3: Use the fusion prediction module to fuse the real-time postoperative clinical data features, real-time heart rate collection data features and real-time graph features to obtain real-time fusion features;

[0116] S4-4: Perform heart rate abnormality recognition based on the real-time fusion features to obtain a real-time heart rate abnormality prediction label, i.e., a real-time heart rate abnormality recognition result;

[0117] S5: Based on the cloud computing center, according to the real-time heart rate abnormality recognition results and the real-time user portrait, the heart rate monitoring optimization model is used to optimize the heart rate monitoring and obtain the real-time heart rate monitoring optimization strategy, including the following steps:

[0118] S1-3-1: Based on the cloud computing center, according to the real-time user portrait, a number of search tags are matched, and a number of historical heart rate monitoring optimization experiences matched in the experience playback pool of the heart rate monitoring optimization model are extracted;

[0119] S1-3-2: Based on some historical heart rate monitoring optimization experiences and real-time heart rate abnormality recognition results, the model parameters of the heart rate monitoring optimization model are updated to obtain updated model parameters, including the following steps:

[0120] S1-3-2-1: Based on some historical heart rate monitoring optimization experiences, update the action space of the heart rate monitoring optimization model to obtain the updated action space A' = [a'1, ..., a' j" ,...,a' I ], where a' j" is the updated j-th "action value", j" is the action indicator; I is the total number of action space dimensions;

[0121] S1-3-2-2: According to the real-time heart rate abnormality recognition result, the state space of the heart rate monitoring optimization model is updated to obtain the updated state space S'=[s'1,...,s' i" ,...,s' I' ], where s' i" is the updated i'th state value, i' is the state indicator, and I' is the total number of state space dimensions;

[0122] S1-3-3: According to the updated action space A' = [a'1, ..., a' j" ,...,a' I ] and the updated state space S'=[s'1,...,s' i" ,...,s' I' ], using the heart rate monitoring optimization model, performing heart rate monitoring optimization, obtaining a number of real-time execution actions, and obtaining a real-time heart rate monitoring optimization strategy based on the number of real-time execution actions, including the following steps:

[0123] S1-3-3-1: Update the state space S' = [s'1, ..., s' i' ,...,s' I ] is used as the input of the heart rate monitoring optimization model, using a deep Q network to generate an updated action space A'=[a'1,...,a' j' ,...,a' I ] is the Q value of each possible action;

[0124] S1-3-3-2: Use the reward function to obtain the reward value of each possible action in the updated action space, and update the Q value of the possible action according to the reward value to obtain the updated Q value of the possible action;

[0125] The formula is:

[0126] Q(s' p' ,a' p' )=(1-α")·Q(s p' ,a p' )+α"·(R(s p' ,a p' ,s' p' )+γ·Q max (s p' ,a p' ))

[0127] In the formula, Q(s' p' ,a' p' ) is the updated state value s' p' and the updated action value a' p' The corresponding updated Q value; Q(s p' ,a p' ) is the state value s p' and action value a p' The corresponding predicted Q value; α" is the learning rate; Q max (s p' ,a p' ) is the highest predicted Q value;

[0128] S1-3-3-3: Repeat the above steps until the iteration threshold is reached, use the greedy strategy, take the possible action corresponding to the highest updated Q value as the execution action, and output the execution action as the real-time heart rate monitoring optimization strategy, including reducing the monitoring frequency and increasing the monitoring time;

[0129] S6: Based on the cloud computing center, using the knowledge graph, adjusting the real-time heart rate monitoring optimization strategy to obtain the adjusted real-time heart rate monitoring optimization strategy, including the following steps;

[0130] S6-1: Based on the cloud computing center, according to the action named entities corresponding to the real-time execution actions in the real-time heart rate monitoring optimization strategy, including "reduce", "monitor", "frequency", "increase", and "duration", search and match the heart rate monitoring named entities in the knowledge graph, and obtain several matching heart rate monitoring named entities, including "reduce", "reduce", "monitoring equipment", "collection frequency", "monitor", "frequency", "increase", "sampling duration", etc.;

[0131] S6-2: According to a number of matching heart rate monitoring named entities, and the heart rate monitoring entity relationship between the matching heart rate monitoring named entity and other heart rate monitoring named entities, the real-time heart rate monitoring optimization strategy is adjusted to obtain an adjusted real-time heart rate monitoring optimization strategy, including reducing the trigger frequency of electrocardiogram data collection in the heart rate collection device, increasing the sampling time of the motion sensor in the heart rate collection device, enhancing motion collection, reducing electrocardiogram collection, and avoiding the influence of motion on the accuracy of electrocardiogram data;

[0132] S7: Based on the cloud computing center, according to the real-time user portrait and the real-time heart rate abnormality recognition results, the knowledge graph is used to generate postoperative rehabilitation suggestions, and the real-time postoperative rehabilitation suggestions are obtained, including the following steps;

[0133] S7-1: Based on the cloud computing center, according to the user named entity corresponding to the real-time user label in the real-time user portrait and the abnormal named entity corresponding to the real-time heart rate abnormality prediction label in the real-time heart rate abnormality recognition result, search and match the postoperative rehabilitation named entity in the knowledge graph to obtain several matching postoperative rehabilitation named entities;

[0134] S7-2: generating postoperative rehabilitation suggestions based on a number of matching postoperative rehabilitation named entities and postoperative rehabilitation entity relationships between the matching postoperative rehabilitation named entities and other postoperative rehabilitation named entities to obtain real-time postoperative rehabilitation suggestions;

[0135] S8: Based on the cloud computing center, the real-time heart rate abnormality recognition results, the adjusted real-time heart rate monitoring optimization strategy and the real-time postoperative rehabilitation suggestions are sent to the corresponding user terminal;

[0136] S9: Based on the user terminal, a real-time alarm signal is generated according to the real-time abnormal heart rate recognition result, and the real-time alarm signal, the real-time abnormal heart rate recognition result, and the real-time postoperative rehabilitation suggestion are visualized;

[0137] If the real-time heart rate abnormality recognition result is that the heart rate is too low, a real-time alarm signal is generated, marked in red, highlighted, and distributed to the mobile terminals of medical staff to achieve automatic reminders;

[0138] S10: Based on the user terminal, a real-time heart rate monitoring instruction is generated according to the adjusted real-time heart rate monitoring optimization strategy, and the real-time heart rate monitoring instruction is sent to the heart rate collection device.

[0139] Embodiment 2:

[0140] like Figure 3As shown, this embodiment provides a heart rate monitoring system based on a neural network, which is used to implement a heart rate monitoring method. The system includes a cloud computing center, a plurality of user terminals, and a plurality of heart rate acquisition devices. The cloud computing center is respectively connected to the plurality of user terminals in communication, and each user terminal is connected to a heart rate acquisition device in communication in a one-to-one correspondence.

[0141] The user terminal is used to receive the input real-time postoperative clinical data and the real-time heart rate data sent by the heart rate collection device, and upload them to the cloud computing center; generate a real-time alarm signal according to the real-time heart rate abnormality recognition result, and visualize the real-time alarm signal, the real-time heart rate abnormality recognition result and the real-time postoperative rehabilitation suggestion; generate a real-time heart rate monitoring instruction according to the adjusted real-time heart rate monitoring optimization strategy, and send the real-time heart rate monitoring instruction to the heart rate collection device;

[0142] The heart rate collection device collects the user's real-time heart rate data and sends it to the user terminal; receives the real-time heart rate monitoring instruction sent by the user terminal and executes the real-time heart rate monitoring instruction;

[0143] The cloud computing center includes a model building unit, a user portrait generating unit, a heart rate abnormality identifying unit, a heart rate monitoring optimization unit, an optimization strategy adjusting unit, a postoperative rehabilitation suggestion generating unit, and a data sending unit, which are connected in sequence;

[0144] A model building unit, which is used to build a user portrait generation model, a knowledge graph, a heart rate abnormality recognition model, and a heart rate monitoring optimization model using a neural network algorithm;

[0145] A user portrait generation unit, used to generate a user portrait based on real-time postoperative clinical data using a user portrait generation model to obtain a real-time user portrait;

[0146] The heart rate abnormality recognition unit is used to identify the heart rate abnormality based on the real-time postoperative clinical data, the real-time heart rate collection data and the real-time user portrait, using the heart rate abnormality recognition model to obtain the real-time heart rate abnormality recognition result;

[0147] A heart rate monitoring optimization unit is used to optimize the heart rate monitoring based on the real-time heart rate abnormality recognition result and the real-time user portrait using the heart rate monitoring optimization model to obtain a real-time heart rate monitoring optimization strategy;

[0148] An optimization strategy adjustment unit, used to use the knowledge graph to adjust the real-time heart rate monitoring optimization strategy to obtain an adjusted real-time heart rate monitoring optimization strategy;

[0149] A postoperative rehabilitation suggestion generation unit is used to generate postoperative rehabilitation suggestions based on the real-time user portrait and the real-time heart rate abnormality recognition results using the knowledge graph to obtain real-time postoperative rehabilitation suggestions;

[0150] The data sending unit is used to send the real-time heart rate abnormality recognition results, the adjusted real-time heart rate monitoring optimization strategy and the real-time postoperative rehabilitation suggestions to the corresponding user terminal.

[0151] The present invention discloses a heart rate monitoring method and system based on a neural network, which combines real-time postoperative clinical data and real-time heart rate collection data sent by a heart rate collection device, more comprehensively reflects the user's health status information, and improves the comprehensiveness of the data; uniformly manages and analyzes the data based on a cloud computing center, and improves practicality; the constructed user portrait generation model and heart rate monitoring optimization model can generate personalized heart rate monitoring plans according to the user's specific health status and heart rate data, realizes customization functions for different rehabilitation conditions and different users, and improves user experience; the constructed heart rate abnormality recognition model can combine real-time postoperative clinical data, real-time heart rate collection data and real-time user portraits to identify heart rate abnormalities, reduce labor cost investment, and improve the accuracy of heart rate abnormality recognition; the constructed knowledge graph can generate real-time postoperative rehabilitation suggestions according to real-time user portraits and real-time heart rate abnormality recognition results, and improves functionality.

[0152] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A heart rate monitoring method based on a neural network, characterized in that: The steps include: Based on the cloud computing center, using neural network algorithms, we build user portrait generation models, knowledge graphs, heart rate abnormality recognition models, and heart rate monitoring optimization models; Based on the user terminal, the real-time postoperative clinical data and the real-time heart rate data sent by the heart rate collection device are received and uploaded to the cloud computing center; Based on the cloud computing center, the user portrait generation model is used to generate user portraits according to real-time postoperative clinical data to obtain real-time user portraits; Based on the cloud computing center, the heart rate abnormality recognition model is used to identify the heart rate abnormality according to the real-time postoperative clinical data, real-time heart rate collection data and real-time user portraits, and the real-time heart rate abnormality recognition results are obtained; Based on the cloud computing center, according to the real-time heart rate abnormality recognition results and real-time user portraits, the heart rate monitoring optimization model is used to optimize the heart rate monitoring and obtain the real-time heart rate monitoring optimization strategy; Based on the cloud computing center, the real-time heart rate monitoring optimization strategy is adjusted using the knowledge graph to obtain the adjusted real-time heart rate monitoring optimization strategy; Based on the cloud computing center, postoperative rehabilitation suggestions are generated based on real-time user portraits and real-time heart rate abnormality recognition results using knowledge graphs to obtain real-time postoperative rehabilitation suggestions; Based on the cloud computing center, the real-time heart rate abnormality recognition results, the adjusted real-time heart rate monitoring optimization strategy and the real-time postoperative rehabilitation suggestions are sent to the corresponding user terminals; Based on the user terminal, a real-time alarm signal is generated according to the real-time abnormal heart rate recognition result, and the real-time alarm signal, the real-time abnormal heart rate recognition result and the real-time postoperative rehabilitation advice are visualized; Based on the user terminal, a real-time heart rate monitoring instruction is generated according to the adjusted real-time heart rate monitoring optimization strategy, and the real-time heart rate monitoring instruction is sent to the heart rate collection device.

2. A heart rate monitoring method based on a neural network according to claim 1, characterized in that: Based on the cloud computing center, using the neural network algorithm, we build a user portrait generation model, knowledge graph, heart rate abnormality recognition model, and heart rate monitoring optimization model, including the following steps: Based on the cloud computing center, some heart rate monitoring and postoperative rehabilitation knowledge, some historical postoperative clinical data and some historical heart rate collection data are collected; Perform preprocessing to obtain some post-preprocessing heart rate monitoring and post-operative rehabilitation knowledge, some post-preprocessing historical post-operative clinical data, and some post-preprocessing historical heart rate collection data; Performing data dimensionality reduction on a number of pre-processed historical postoperative clinical data to obtain a number of dimensionality-reduced historical postoperative clinical data and key clinical indicator sets; Based on several historical postoperative clinical data after dimensionality reduction and several historical heart rate collection data after preprocessing, a user portrait generation model is constructed using a neural network algorithm, and several historical user portraits are generated; Based on the pre-processed heart rate monitoring and post-operative rehabilitation knowledge, a knowledge graph was constructed using a natural language processing algorithm; Based on several historical postoperative clinical data after dimensionality reduction, several historical heart rate collection data after preprocessing, and several historical user portraits, a neural network algorithm is used to build a heart rate abnormality recognition model and generate several historical heart rate abnormality recognition results; Based on several historical heart rate abnormality recognition results and several historical user portraits, a heart rate monitoring optimization model is constructed using a reinforcement learning algorithm.

3. A heart rate monitoring method based on a neural network according to claim 2, characterized in that: Based on the pre-processed heart rate monitoring and post-operative rehabilitation knowledge, a knowledge graph is constructed using a natural language processing algorithm, including the following steps: Using a pre-trained named entity extraction model, extract several heart rate monitoring named entities and several postoperative rehabilitation named entities of the pre-processed heart rate monitoring and postoperative rehabilitation knowledge; Use a pre-trained entity relationship extraction model to extract heart rate monitoring entity relationships between several heart rate monitoring named entities and postoperative rehabilitation entity relationships between several postoperative rehabilitation named entities; A knowledge graph is constructed based on several heart rate monitoring named entities, several heart rate monitoring entity relationships, several postoperative rehabilitation named entities, and several postoperative rehabilitation entity relationships.

4. A heart rate monitoring method based on a neural network according to claim 3, characterized in that: The user portrait generation model is built based on the N-GAN-Attention-LSTM algorithm; The abnormal heart rate recognition model is constructed based on the RNN-GNN-MLP algorithm; The heart rate monitoring optimization model is constructed based on the DQN algorithm.

5. A heart rate monitoring method based on a neural network according to claim 4, characterized in that: Based on the cloud computing center, according to the real-time postoperative clinical data, the user portrait generation model is used to generate the user portrait to obtain the real-time user portrait, including the following steps: Based on the cloud computing center, the real-time postoperative clinical data is reduced in dimension according to the key clinical indicator set to obtain the real-time postoperative clinical data after dimension reduction; Input the reduced-dimensional real-time postoperative clinical data into the user portrait generation model to extract N real-time key dimension features; According to the preset attention weights, N real-time key dimension features are weighted and spliced ​​to obtain real-time weighted splicing features; According to the real-time weighted splicing features, user label prediction is performed to obtain a number of real-time user labels, and user portrait generation is performed based on the several real-time user labels to obtain a real-time user portrait.

6. A heart rate monitoring method based on a neural network according to claim 5, characterized in that: Based on the cloud computing center, according to the real-time postoperative clinical data, real-time heart rate collection data and real-time user portrait, the heart rate abnormality recognition model is used to perform heart rate abnormality recognition and obtain the real-time heart rate abnormality recognition result, including the following steps: Based on the cloud computing center, the real-time postoperative clinical data after dimensionality reduction, the real-time heart rate collection data and the real-time user portrait are input into the heart rate abnormality recognition model; Extract the real-time postoperative clinical data features of the real-time postoperative clinical data after dimensionality reduction, the real-time heart rate collection data features of the real-time heart rate collection data, and the real-time graph features of the real-time user portrait; Perform feature fusion on real-time postoperative clinical data features, real-time heart rate collection data features and real-time graph features to obtain real-time fusion features; According to the real-time fusion features, abnormal heart rate recognition is performed to obtain a real-time abnormal heart rate prediction label, that is, a real-time abnormal heart rate recognition result.

7. A heart rate monitoring method based on a neural network according to claim 6, characterized in that: Based on the cloud computing center, according to the real-time heart rate abnormality recognition results and real-time user portraits, the heart rate monitoring optimization model is used to optimize the heart rate monitoring and obtain the real-time heart rate monitoring optimization strategy, which includes the following steps: Based on the cloud computing center, according to the real-time user portrait, several historical heart rate monitoring optimization experiences matching in the experience playback pool of the heart rate monitoring optimization model are extracted; According to several historical heart rate monitoring optimization experiences and real-time heart rate abnormality recognition results, the model parameters of the heart rate monitoring optimization model are updated to obtain updated model parameters; According to the updated model parameters, the heart rate monitoring optimization model is used to optimize the heart rate monitoring, and a number of real-time execution actions are obtained. Based on the number of real-time execution actions, a real-time heart rate monitoring optimization strategy is obtained.

8. The heart rate monitoring method based on a neural network according to claim 7, characterized in that: Based on the cloud computing center, the real-time heart rate monitoring optimization strategy is adjusted using the knowledge graph to obtain the adjusted real-time heart rate monitoring optimization strategy, including the following steps; Based on the cloud computing center, according to the action named entities corresponding to the real-time execution actions in the real-time heart rate monitoring optimization strategy, the heart rate monitoring named entities in the knowledge graph are searched and matched to obtain several matching heart rate monitoring named entities; According to a number of matching heart rate monitoring named entities and the heart rate monitoring entity relationships between the matching heart rate monitoring named entities and other heart rate monitoring named entities, the real-time heart rate monitoring optimization strategy is adjusted to obtain an adjusted real-time heart rate monitoring optimization strategy.

9. A heart rate monitoring method based on a neural network according to claim 8, characterized in that: Based on the cloud computing center, according to the real-time user portrait and the real-time heart rate abnormality recognition results, the knowledge graph is used to generate postoperative rehabilitation suggestions, and the real-time postoperative rehabilitation suggestions are obtained, including the following steps; Based on the cloud computing center, according to the user named entity corresponding to the real-time user label in the real-time user portrait and the abnormal named entity corresponding to the real-time heart rate abnormality prediction label in the real-time heart rate abnormality recognition result, the postoperative rehabilitation named entity in the knowledge graph is searched and matched to obtain several matching postoperative rehabilitation named entities; According to a number of matching postoperative rehabilitation named entities and the postoperative rehabilitation entity relationship between the matching postoperative rehabilitation named entity and other postoperative rehabilitation named entities, postoperative rehabilitation suggestions are generated to obtain real-time postoperative rehabilitation suggestions.

10. A heart rate monitoring system based on a neural network, used to implement the heart rate monitoring method according to any one of claims 1 to 9, characterized in that: The system includes a cloud computing center, a plurality of user terminals and a plurality of heart rate collection devices. The cloud computing center is respectively connected to the plurality of user terminals in communication, and each of the user terminals is connected to a heart rate collection device in communication in a one-to-one correspondence. The cloud computing center includes a model building unit, a user portrait generating unit, a heart rate abnormality identifying unit, a heart rate monitoring optimizing unit, an optimizing strategy adjusting unit, a postoperative rehabilitation suggestion generating unit and a data sending unit which are connected in sequence.

Images