A blood glucose concentration prediction method combined with neural network

Abstract

The present invention discloses a blood glucose concentration prediction method combined with a neural network, and relates to the technical field of blood glucose concentration prediction. The method includes: a continuous blood glucose monitoring device for interactive target users to perform blood glucose monitoring, extracting a dynamic blood glucose concentration sequence within a preset monitoring window; obtaining a set of exercise data, a set of food intake data, a set of insulin use records, and a set of environmental monitoring data of the target user within the preset monitoring window; obtaining a topological framework sequence of blood glucose data; obtaining a blood glucose feature recognition model; performing cross-scale mapping interactive analysis to obtain blood glucose associated features; training a blood glucose concentration prediction network layer; performing blood glucose concentration prediction on blood glucose associated features to obtain blood glucose concentration prediction results. The present invention solves the technical problem of low accuracy of prediction results in a single dimension of blood glucose concentration prediction in the prior art, and achieves the technical effect of improving the accuracy and timeliness of blood glucose concentration prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of blood sugar concentration prediction, and in particular to a blood sugar concentration prediction method combined with a neural network. Background Art

[0002] Existing blood glucose concentration prediction technologies usually rely on a single data source or a single-dimensional analysis, such as relying solely on dynamic blood glucose concentration data for prediction. This method ignores other key factors, such as exercise data, food intake, insulin usage records, and environmental changes, which results in low accuracy of the prediction results. For example, if predictions are made based only on changes in blood glucose concentrations of blood glucose monitoring devices, it is impossible to take into account changes in the user's exercise intensity in different time periods, and the user's blood glucose concentration fluctuations may be misjudged. In addition, the timing and type of food intake, the amount of insulin used, etc. can also significantly affect blood glucose levels, but if only relying on blood glucose concentration data and ignoring these factors, the prediction accuracy will be difficult to improve. Therefore, the single-dimensional prediction method cannot fully reflect the complexity of blood glucose fluctuations, resulting in poor results in practical applications and low reliability and accuracy of the prediction results. Summary of the invention

[0003] The present application provides a blood glucose concentration prediction method combined with a neural network, which is used to solve the technical problem of low accuracy of prediction results in single-dimensional blood glucose concentration prediction in the prior art.

[0004] In view of the above problems, the present application provides a method for predicting blood glucose concentration in combination with a neural network, the method comprising:

[0005] A continuous blood glucose monitoring device for interactive target users to monitor blood glucose, extracting a dynamic blood glucose concentration sequence within a preset monitoring window;

[0006] Acquire a set of exercise data, a set of food intake data, a set of insulin usage records, and a set of environmental monitoring data of the target user within the preset monitoring window;

[0007] Calling a pre-constructed blood glucose data topology framework sequence, inputting the dynamic blood glucose concentration sequence, the exercise data set, the food intake data set, the insulin use record set and the environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence, and obtaining a blood glucose data topology framework sequence;

[0008] A blood glucose feature recognition model is obtained by modeling based on a pre-constructed blood glucose data topology framework sequence, multiple sample data sets, corresponding multiple sample blood glucose recognition features, and a fully connected neural network;

[0009] Using the blood sugar feature recognition model to perform feature recognition on the blood sugar data topology frame sequence respectively, and performing cross-scale mapping interactive analysis on the recognition results to obtain blood sugar related features;

[0010] Training a blood sugar concentration prediction network layer, wherein the blood sugar concentration prediction network layer uses a support vector machine for binary classification training;

[0011] The trained blood glucose concentration prediction network layer is used to predict the blood glucose concentration of the blood glucose associated features to obtain a blood glucose concentration prediction result.

[0012] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0013] The present application uses a continuous blood glucose monitoring device for interactive target user blood glucose monitoring to extract a dynamic blood glucose concentration sequence within a preset monitoring window, and then obtains the target user's exercise data set, food intake data set, insulin usage record set and environmental monitoring data set within the preset monitoring window, and then calls a pre-constructed blood glucose data topology framework sequence, inputs the dynamic blood glucose concentration sequence, exercise data set, food intake data set, insulin usage record set and environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence to obtain the blood glucose data topology framework sequence, and then models the pre-constructed blood glucose data topology framework sequence and multiple sample data sets and corresponding multiple sample blood glucose recognition features, as well as a fully connected neural network to obtain a blood glucose feature recognition model, and uses the blood glucose feature recognition model to perform feature recognition on the blood glucose data topology framework sequence respectively, and performs cross-scale mapping interactive analysis on the recognition results to obtain blood glucose associated features, and trains a blood glucose concentration prediction network layer, wherein the blood glucose concentration prediction network layer uses a support vector machine for binary classification training, and then uses the trained blood glucose concentration prediction network layer to predict blood glucose concentration based on blood glucose associated features to obtain a blood glucose concentration prediction result. The technical effect of improving the accuracy of blood sugar prediction by conducting cross-scale multidimensional data comprehensive analysis through multi-dimensional data fusion and neural network modeling has been achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 A schematic diagram of a flow chart of a method for predicting blood glucose concentration in combination with a neural network provided in an embodiment of the present application;

[0016] Figure 2A schematic diagram of a flow chart for obtaining blood glucose associated features in a blood glucose concentration prediction method combined with a neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application provides a blood glucose concentration prediction method combined with a neural network, which is used to solve the technical problem of low accuracy of prediction results in a single dimension of blood glucose concentration prediction in the prior art.

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0020] Examples, such as Figure 1 As shown, the present application provides a method for predicting blood glucose concentration in combination with a neural network, wherein the method comprises:

[0021] S100: The continuous blood glucose monitoring device for the interactive target user to perform blood glucose monitoring extracts a dynamic blood glucose concentration sequence within a preset monitoring window;

[0022] S200: Acquire a set of exercise data, a set of food intake data, a set of insulin usage records, and a set of environmental monitoring data of the target user within the preset monitoring window;

[0023] In a possible embodiment, the target user is any user who needs to monitor blood glucose concentration. The preset monitoring window is a continuous blood glucose monitoring time window (e.g., 24 hours, 72 hours, etc.) set by a technician in this field according to the health status and needs of the target user. Optionally, a continuous blood glucose monitoring device (CGM device) is used to continuously monitor the blood glucose concentration of the target user within the preset monitoring window. The device can monitor the user's blood glucose level in real time, collect blood glucose concentration data through sensors, and summarize the collected data in chronological order to obtain the dynamic blood glucose concentration sequence. Among them, the dynamic blood glucose concentration sequence reflects the changes in blood glucose concentration monitored by the device worn by the target user within the preset monitoring window.

[0024] Preferably, the motion data of the target user within a preset monitoring window is obtained through a motion tracking device (such as a smart bracelet, fitness watch, etc.) to obtain the motion data set. Each motion data is the recorded data corresponding to a type of exercise performed by the target user, including data such as exercise duration, exercise start time, exercise type, intensity, etc. The user's food intake data set is collected through manual input, smart food tracking application or intake sensor, and each food intake data includes food type, intake amount, intake time, etc. Through the interconnection of the blood glucose monitoring device and the insulin pump system, or manual recording by the user, the time, dosage and other data of insulin injection or use within the preset monitoring window are collected to obtain the insulin use record set. The environmental data within the preset monitoring window is obtained through an environmental monitoring device (such as a temperature and humidity sensor) to obtain the environmental monitoring data set.

[0025] By obtaining the dynamic blood glucose concentration sequence, exercise data set, food intake data set, insulin usage record set and environmental monitoring data set, the technical effect of providing basic data support for subsequent multi-dimensional blood glucose concentration prediction and the construction of a blood glucose data topological framework sequence is achieved, thereby improving the accuracy and reliability of blood glucose concentration prediction.

[0026] S300: calling a pre-constructed blood glucose data topology framework sequence, inputting the dynamic blood glucose concentration sequence, the exercise data set, the food intake data set, the insulin use record set and the environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence, and obtaining a blood glucose data topology framework sequence;

[0027] Further, calling a pre-constructed blood glucose data topology framework sequence, inputting the dynamic blood glucose concentration sequence, the exercise data set, the food intake data set, the insulin usage record set and the environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence, and obtaining the blood glucose data topology framework sequence, step S300 of the embodiment of the present application further includes:

[0028] Traverse the dynamic blood glucose concentration, exercise data, food intake data, insulin use records and environmental monitoring data to extract data indicators, and obtain a dynamic blood glucose concentration indicator set, an exercise data indicator set, a food intake data indicator set, an insulin use indicator set and an environmental monitoring indicator set;

[0029] The dynamic blood glucose concentration, exercise data, food intake data, insulin use record and environmental monitoring data are respectively used as first-level topological nodes to obtain a first-level topological node set;

[0030] Taking each indicator in the dynamic blood glucose concentration indicator set, the exercise data indicator set, the food intake data indicator set, the insulin use indicator set and the environmental monitoring indicator set as a secondary topological node, and obtaining multiple secondary topological node sets;

[0031] A blood glucose data topology framework is constructed based on the primary topology node set and the multiple secondary topology node sets to obtain a pre-constructed blood glucose data topology framework.

[0032] Furthermore, step S300 of the embodiment of the present application further includes:

[0033] Obtaining the number of dynamic blood glucose concentrations and corresponding multiple monitoring time points in the dynamic blood glucose concentration sequence;

[0034] The pre-constructed blood glucose data topology framework is copied according to the number of dynamic blood glucose concentrations, and the multiple copied pre-constructed blood glucose data topology frameworks are sorted according to the multiple monitoring time points to obtain a sequence of the pre-constructed blood glucose data topology frameworks;

[0035] The dynamic blood glucose concentration sequence is mapped into the pre-constructed blood glucose data topology framework sequence, and the data in the exercise data set, food intake data set, insulin usage record set and environmental monitoring data set are respectively input into the copied pre-constructed blood glucose data topology framework corresponding to the minimum time difference according to the data collection time points of the data in the exercise data set, food intake data set, insulin usage record set and environmental monitoring data set, so as to obtain the blood glucose data topology framework sequence, wherein the minimum time difference is the minimum time difference between the data collection time point corresponding to the data and multiple monitoring time points.

[0036] In one embodiment of the present application, the pre-constructed blood glucose data topology framework sequence is a framework obtained by sorting the data topology frameworks generated by multiple actions that affect blood glucose concentration of the target user at different times in chronological order. Preferably, the framework storage data in the pre-constructed blood glucose data topology framework sequence is empty. The number of nodes of each pre-constructed blood glucose data topology framework sequence in the pre-constructed blood glucose data topology framework sequence is consistent. After the dynamic blood glucose concentration sequence, exercise data set, food intake data set, insulin use record set and environmental monitoring data set are respectively input into the pre-constructed blood glucose data topology framework sequence according to the time node of data acquisition, each blood glucose data topology framework in the obtained blood glucose data topology framework sequence reflects the blood glucose impact data situation at different monitoring time points. By obtaining the blood glucose data topology framework sequence, data support is provided for the subsequent analysis of the lag impact relationship between data.

[0037] In one embodiment, respective key indicators are extracted from dynamic blood glucose concentration, exercise data, food intake data, insulin use records and environmental monitoring data. For example, the range of change of blood glucose concentration, exercise intensity of exercise data, carbohydrate content of food intake, insulin dosage, environmental factors (such as temperature and humidity), etc., so as to obtain the dynamic blood glucose concentration index set, exercise data index set, food intake data index set, insulin use index set and environmental monitoring index set. Wherein, the dynamic blood glucose concentration index set, exercise data index set, food intake data index set, insulin use index set and environmental monitoring index set are used to quantify the actual situation of different data.

[0038] Optionally, the dynamic blood glucose concentration indicator set includes blood glucose concentration, blood glucose concentration change rate, blood glucose fluctuation amplitude, blood glucose peak, blood glucose trough, etc. The exercise data indicator set includes exercise intensity, exercise duration, number of steps, exercise type, energy consumption, etc. The food intake data indicator set includes food intake, carbohydrate intake, dietary fiber intake, protein intake, etc. The insulin use indicator set includes insulin dosage, insulin use time, insulin type, insulin sensitivity, insulin duration, etc. The environmental monitoring indicator set includes ambient temperature, ambient humidity, air pressure, ultraviolet intensity, etc.

[0039] The dynamic blood glucose concentration, exercise data, food intake data, insulin use records and environmental monitoring data are respectively used as first-level topological nodes, that is, the data types are used as first-level topological nodes to construct the first-level topological node set. These first-level nodes will serve as the basic nodes of the blood glucose data topological framework, representing different data sources or data types.

[0040] Each indicator in the dynamic blood glucose concentration indicator set, exercise data indicator set, food intake data indicator set, insulin use indicator set and environmental monitoring indicator set is used as a secondary topological node to create multiple secondary topological node sets. Each secondary topological node set corresponds to a primary topological node, which is used to further refine the specific data features in each primary topological node.

[0041] Based on the first-level topological node set and multiple second-level topological node sets, the topological framework of blood glucose data is constructed. By establishing the connection relationship between nodes, a framework for data flow and interaction is formed. This topological framework effectively organizes and associates multiple data types (such as blood glucose concentration, exercise, diet, insulin, and environment), achieving the technical effect of providing structured data support for subsequent data modeling and analysis.

[0042] In a possible embodiment, the number of pre-constructed blood glucose data topology frame sequences and the corresponding time in the pre-constructed blood glucose data topology frame sequence are determined by obtaining the number of dynamic blood glucose concentrations in the dynamic blood glucose concentration sequence and the corresponding multiple monitoring time points. The pre-constructed blood glucose data topology frame is copied according to the number of dynamic blood glucose concentrations, and each copied topology frame will be used to associate the dynamic blood glucose concentration data of the corresponding monitoring time point, as well as the exercise data set, food intake data set, insulin use record set and environmental monitoring data set near the monitoring time point.

[0043] The multiple replicated pre-built blood glucose data topology frameworks are sorted according to multiple monitoring time points in the dynamic blood glucose concentration series. The purpose of sorting is to ensure the temporal consistency of the topology framework and the dynamic blood glucose concentration series, so as to perform accurate analysis.

[0044] Preferably, the dynamic blood glucose concentration sequence is mapped to a pre-constructed blood glucose data topology framework sequence, and the data indicators in the dynamic blood glucose concentration sequence are filled into the corresponding first-level topology node and second-level topology node sets in the blood glucose data topology framework sequence, ensuring that each dynamic blood glucose concentration data corresponds to the correct topology framework.

[0045] Then, according to the data collection time points corresponding to the motion data set, food intake data set, insulin use record set and environmental monitoring data set, the data therein are respectively mapped to the corresponding pre-built blood sugar data topology framework. Optionally, according to the data collection time points of the data in the motion data set, food intake data set, insulin use record set and environmental monitoring data set, the time difference comparison is carried out with the monitoring time point of the dynamic blood glucose concentration data. For each motion data, food intake data, insulin use record and each environmental monitoring data in the motion data set, food intake data set, insulin use record set and environmental monitoring data set, select the data with the minimum time difference with the monitoring time point to be input into the corresponding pre-built blood sugar data topology framework, thereby ensure the time synchronization and correlation of the data. After time synchronization and minimum time difference matching, the generation of the blood sugar data topology framework sequence is completed.

[0046] Through the above step S300, various types of data (such as dynamic blood glucose concentration, exercise, diet, insulin, and environmental monitoring) can be effectively associated and integrated. The constructed blood glucose data topology framework sequence can provide structured, time-synchronized multi-dimensional data support for the subsequent blood glucose concentration prediction model, thereby improving the accuracy and reliability of the prediction results.

[0047] S400: Modeling is performed according to a pre-constructed blood glucose data topology framework sequence, a plurality of sample data sets and corresponding plurality of sample blood glucose recognition features, and a fully connected neural network to obtain a blood glucose feature recognition model;

[0048] Furthermore, step S400 in the embodiment of the present application further includes:

[0049] Acquire multiple sample data sets and corresponding multiple sample blood glucose identification features, wherein each sample data set includes sample dynamic blood glucose concentration, sample exercise data, sample food intake data, sample insulin use record and sample environmental monitoring data;

[0050] Inputting the plurality of sample data sets into any one of the pre-constructed blood glucose data topology frames in the pre-constructed blood glucose data topology frame sequence respectively to obtain a plurality of sample blood glucose data topology frames;

[0051] The framework constructed based on the fully connected neural network is supervised by using the multiple sample blood glucose data topological frameworks and the multiple sample blood glucose identification features to learn the mapping relationship between the blood glucose data topological framework and the blood glucose identification features until the output of the objective function reaches the minimum value, and the training is stopped to obtain the trained blood glucose feature recognition model.

[0052] Furthermore, the objective function is:

[0053] ;

[0054] in, is the objective function of the blood glucose feature recognition model, Identify features for multiple sample blood glucose, Identify features for multiple output blood glucose outputs of the model, is the i-th sample blood glucose identification feature among multiple sample blood glucose identification features, is the ith output blood glucose identification feature among the multiple output blood glucose identification features output by the model, N is the number of the multiple sample blood glucose identification features, and N is an integer greater than or equal to 3.

[0055] In one embodiment, it is necessary to collect blood glucose identification features and sample data within a historical period from the historical data of the target user to obtain multiple sample data sets and corresponding multiple sample blood glucose identification features. Each sample data set includes sample dynamic blood glucose concentration, sample exercise data, sample food intake data, sample insulin use records, and sample environmental monitoring data. These data provide detailed multi-dimensional information such as blood glucose changes, environmental impacts, and exercise impacts for each sample.

[0056] Each sample data set is input into any pre-constructed blood glucose data topology framework in the pre-constructed blood glucose data topology framework sequence. Optionally, the dynamic blood glucose concentration, exercise data, food intake data, insulin usage record and environmental monitoring data of each sample are used as different first-level topology nodes, and the indicators in the data are used as corresponding second-level topology node sets, which are filled into the pre-constructed blood glucose data topology framework to form a sample blood glucose data topology framework, thereby obtaining the multiple sample blood glucose data topology frameworks. The construction of these topology frameworks is based on different data sets, and each sample data set will generate an independent topology framework. These blood glucose data topology frameworks are sent to the neural network for further modeling and training.

[0057] The fully connected neural network (FCNN) is used to model the topological framework of blood glucose data. The fully connected neural network learns the relationship between input data and identification features by inputting sample data sets and corresponding blood glucose identification features. During the training process, the supervised learning method is used, that is, learning is performed based on the sample data set input and the known sample blood glucose identification features. The goal of the training is to minimize the difference between the network's output blood glucose identification features (predicted values) and the actual blood glucose identification features (true values), and to optimize the network's weights and biases.

[0058] Optionally, the objective function is mean square error, specifically:

[0059] ;

[0060] in, is the objective function of the blood glucose feature recognition model, Identify features for multiple sample blood glucose, Identify features for multiple output blood glucose outputs of the model, is the i-th sample blood glucose identification feature among multiple sample blood glucose identification features, is the ith output blood glucose identification feature among the multiple output blood glucose identification features output by the model, N is the number of the multiple sample blood glucose identification features, and N is an integer greater than or equal to 3.

[0061] Through the training process, the neural network will continuously optimize the weights and biases until the objective function value (error) reaches the minimum and the model converges. After the training is completed, the blood glucose feature recognition model is used for blood glucose feature recognition. At this point, the model is able to predict the impact of the corresponding blood glucose concentration, exercise, food intake, insulin use records and environmental data on blood glucose based on the input blood glucose data topology framework. Finally, the obtained blood glucose feature recognition model can be used for new data sets for real-time blood glucose concentration prediction and optimization.

[0062] The key to step S400 is to train the pre-constructed blood glucose data topology framework through a fully connected neural network, learn the law of blood glucose concentration changes based on sample data and identification features, and finally generate a blood glucose feature recognition model for blood glucose concentration prediction.

[0063] S500: using the blood sugar feature recognition model to perform feature recognition on the blood sugar data topology frame sequence respectively, and performing cross-scale mapping interactive analysis on the recognition results to obtain blood sugar related features;

[0064] Further, such as Figure 2 As shown, the blood glucose feature recognition model is used to perform feature recognition on the blood glucose data topological frame sequence respectively, and the recognition results are subjected to cross-scale mapping interactive analysis to obtain blood glucose related features. Step S500 of the embodiment of the present application also includes:

[0065] Using the blood sugar feature recognition model to perform feature recognition on the blood sugar data topology framework sequence respectively to obtain a blood sugar recognition feature sequence;

[0066] Extracting a first blood sugar identification feature and a second blood sugar identification feature from the blood sugar identification feature sequence in a front-to-back order;

[0067] Performing cross-scale mapping interaction analysis on the first blood glucose identification feature and the second blood glucose identification feature to obtain a first mapping interaction blood glucose association feature;

[0068] extracting a third blood sugar identification feature from the blood sugar identification feature sequence again, performing cross-scale mapping interaction analysis on the third blood sugar identification feature and the first mapping interactive blood sugar association feature, and obtaining a second mapping interactive blood sugar association feature;

[0069] The cross-scale mapping interactive analysis is performed in sequence until the end of the blood glucose identification feature sequence is reached to obtain the blood glucose associated feature.

[0070] Further, a cross-scale mapping interaction analysis is performed on the first blood glucose identification feature and the second blood glucose identification feature to obtain a first mapping interaction blood glucose association feature. Step S500 of the embodiment of the present application further includes:

[0071] Performing feature mapping inner product calculation on the first blood sugar identification feature and the second blood sugar identification feature to determine a feature similarity matrix;

[0072] Normalizing the feature similarity matrix to obtain a mapping interaction matrix;

[0073] A convolution operation is performed according to the mapping interaction matrix and the second blood glucose identification feature to obtain the first mapping interaction blood glucose association feature.

[0074] In a possible embodiment, the blood glucose feature recognition model is used to intelligently recognize the blood glucose features represented by each blood glucose data topology frame in the blood glucose data topology frame sequence, and obtain a blood glucose recognition feature sequence. The blood glucose recognition feature sequence reflects the changes in the blood glucose characteristics of the target user at different monitoring time points within the preset monitoring window. Since the user's blood glucose concentration is affected by many factors, and the influence process often takes a certain amount of time, the data fluctuations monitored by the continuous blood glucose monitoring device lag behind the actual blood glucose concentration fluctuations. Therefore, by obtaining the blood glucose recognition feature sequence, the potential influence of various influencing factors can be associated and identified, and more accurate blood glucose association features can be obtained after mapping interactive analysis. Furthermore, the user's blood glucose concentration is predicted based on the blood glucose association features, achieving the technical effect of improving the accuracy and reliability of blood glucose concentration prediction.

[0075] Preferably, the first blood sugar identification feature and the second blood sugar identification feature are extracted from the blood sugar identification feature sequence in a front-to-back order. The first blood sugar identification feature reflects the feature related to the influence on the blood sugar concentration at the first monitoring time point. The second blood sugar identification feature reflects the feature related to the influence on the blood sugar concentration at the second monitoring time point.

[0076] Then, the first blood sugar identification feature and the second blood sugar identification feature are subjected to cross-scale mapping interaction analysis to analyze the correlation influence degree of the influencing feature in the first blood sugar identification feature on the second blood sugar identification feature, thereby obtaining the first mapping interaction blood sugar correlation feature. Through cross-scale mapping interaction analysis, it is possible to reveal how different factors (such as exercise, food, insulin, etc.) affect blood sugar concentration on a time scale.

[0077] Based on the same principle as that of obtaining the first mapping interactive blood sugar associated feature, a third blood sugar identification feature is extracted from the blood sugar identification feature sequence again, and the third blood sugar identification feature and the first mapping interactive blood sugar associated feature are subjected to cross-scale mapping interactive analysis to obtain a second mapping interactive blood sugar associated feature. The cross-scale mapping interactive analysis is performed sequentially until the end of the blood sugar identification feature sequence is reached to obtain the blood sugar associated feature.

[0078] Optionally, the feature mapping inner product calculation is performed on the first blood sugar identification feature and the second blood sugar identification feature, that is, according to the feature type in the first blood sugar identification feature, the second blood sugar identification feature is mapped, and the similarity calculation is performed on each mapping result using the cosine similarity formula, and the calculation result is embedded in a matrix with an initial value of null to determine a feature similarity matrix. The feature similarity matrix reflects the feature similarity between the first blood sugar identification feature and the second blood sugar identification feature.

[0079] Preferably, the feature similarity matrix is ​​normalized using a normalization formula. The normalization process is completed by respectively calculating the difference between any feature similarity in the feature similarity matrix and the minimum feature similarity in the matrix, and then comparing the calculated result with a preset difference, wherein the preset difference is the difference between the maximum feature similarity and the minimum feature similarity in the matrix, to obtain a mapping interaction matrix. The mapping interaction matrix reflects the final influence of the first blood glucose recognition feature on the second blood glucose recognition feature after interactive recognition.

[0080] Then, the mapping interaction matrix and the second blood glucose identification feature are convolved using the trained convolutional neural network to obtain the first mapping interaction blood glucose association feature. Optionally, multiple sample mapping interaction matrices, multiple sample blood glucose identification features, and multiple sample mapping interaction blood glucose association features are obtained as training data, and the training data are used to perform supervised learning on a framework built based on a convolutional neural network to learn the two-to-one mapping relationship between the mapping interaction matrix, the blood glucose identification feature, and the mapping interaction blood glucose association feature until the training converges to obtain the trained convolutional neural network.

[0081] By utilizing the blood glucose feature recognition model to perform feature recognition on the topological framework sequence of blood glucose data, and through cross-scale mapping interactive analysis, the complex association between blood glucose changes and multidimensional data (such as exercise, food intake, insulin use, etc.) is revealed, thereby obtaining more accurate and comprehensive blood glucose association features, achieving the technical effect of providing accurate analysis data for subsequent blood glucose concentration prediction models.

[0082] S600: training a blood glucose concentration prediction network layer, wherein the blood glucose concentration prediction network layer performs binary classification training using a support vector machine;

[0083] S700: Predicting the blood glucose concentration of the blood glucose associated features using the trained blood glucose concentration prediction network layer to obtain a blood glucose concentration prediction result.

[0084] Furthermore, step S600 of the embodiment of the present application further includes:

[0085] Acquire a class label set, wherein the class label set includes a positive class label and a negative class label, the positive class label indicates a normal blood glucose concentration state, and the negative class label indicates an abnormal blood glucose concentration state;

[0086] Performing sample data retrieval based on the positive class label and the negative class label to obtain a plurality of blood sugar associated features of a plurality of positive class samples and a plurality of blood sugar associated features of a plurality of negative class samples;

[0087] The blood glucose concentration prediction network layer is trained to convergence based on the support vector machine according to the multiple blood glucose associated features of the multiple positive class samples and the multiple blood glucose associated features of the multiple negative class samples, wherein the expression of the blood glucose concentration prediction network layer is:

[0088] ;

[0089] in, For blood sugar prediction results, is the Lagrange multiplier of the support vector machine, is the category label, is the blood sugar related feature corresponding to the positive class sample, is the blood sugar related feature corresponding to the negative category sample, for and The kernel function of For bias.

[0090] In one embodiment of the present application, the blood glucose concentration prediction network layer will use a support vector machine (SVM) for binary classification training to identify and classify normal and abnormal states of blood glucose concentration. A set of category labels is obtained, which contains two types of labels: a positive category label indicates that the blood glucose concentration is in a normal state (for example, the blood glucose concentration is within a reasonable range, the blood glucose fluctuation amplitude is within a reasonable range, etc.). A negative category label indicates that the blood glucose concentration is in an abnormal state (for example, the blood glucose concentration is lower or higher than the normal range, the blood glucose fluctuation amplitude is not within a reasonable range, etc.).

[0091] Sample data is retrieved using the positive category label and the negative category label as indexes respectively, samples with normal blood glucose concentration are retrieved from the big data, the blood glucose-related features of these samples are summarized to obtain multiple blood glucose-related features of the multiple positive category samples, and samples with abnormal blood glucose concentration are retrieved from the big data, the blood glucose-related features of these samples are summarized to obtain multiple blood glucose-related features of the multiple negative category samples.

[0092] Multiple blood glucose associated features of multiple positive class samples and multiple blood glucose associated features of multiple negative class samples are used as training data with corresponding class labels (positive class labels or negative class labels). The blood glucose concentration prediction network layer is trained by a support vector machine algorithm using the blood glucose associated features of the positive class and the negative class to learn how to distinguish between normal blood glucose concentration and abnormal blood glucose concentration. The goal of the support vector machine is to find an optimal hyperplane to separate positive class samples from negative class samples, while maximizing the interval from the hyperplane to the sample point as much as possible to improve classification accuracy.

[0093] The expression of the blood glucose concentration prediction network layer is:

[0094] ;

[0095] in, For blood sugar prediction results, is the Lagrange multiplier of the support vector machine, is the category label, is the blood sugar related feature corresponding to the positive class sample, is the blood sugar related feature corresponding to the negative category sample, for and The kernel function of is a bias. Optionally, the It is used to measure the similarity between samples. The b is used to adjust the offset of the model.

[0096] Through the training process of the support vector machine, the above-mentioned blood sugar-related features and category labels are used for learning until the training process converges and the training is completed. The blood sugar concentration prediction network layer can perform blood sugar concentration classification prediction based on the input blood sugar-related features.

[0097] Then, the trained blood glucose concentration prediction network layer is used to predict blood glucose concentration to obtain the blood glucose concentration prediction result. The blood glucose concentration prediction result is abnormal blood glucose concentration or normal blood glucose concentration, which reflects the blood glucose concentration of the target user. The technical effect of reliable blood glucose concentration prediction is achieved.

[0098] In summary, the embodiments of the present application have at least the following technical effects:

[0099] This application uses a continuous blood glucose monitoring device to extract the target user's dynamic blood glucose concentration sequence within a preset monitoring window, and combines exercise data, food intake data, insulin use records and environmental monitoring data to build a blood glucose data topology framework sequence. Based on the topology framework sequence and sample data set, a fully connected neural network is used for modeling, a blood glucose feature recognition model is built, and cross-scale mapping interactive analysis is performed to extract blood glucose-related features. These features are predicted by the blood glucose concentration prediction network layer trained by a support vector machine, and finally the prediction result of blood glucose concentration is obtained. The technical effect of accurately predicting blood glucose concentration is achieved.

[0100] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0101] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0102] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for predicting blood glucose concentration using a neural network, characterized in that: The method comprises: A continuous blood glucose monitoring device for interactive target users to monitor blood glucose, extracting a dynamic blood glucose concentration sequence within a preset monitoring window; Acquire a set of exercise data, a set of food intake data, a set of insulin usage records, and a set of environmental monitoring data of the target user within the preset monitoring window; Calling a pre-constructed blood glucose data topology framework sequence, inputting the dynamic blood glucose concentration sequence, the exercise data set, the food intake data set, the insulin use record set and the environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence, and obtaining a blood glucose data topology framework sequence; A blood glucose feature recognition model is obtained according to a pre-constructed blood glucose data topology framework sequence, multiple sample data sets, corresponding multiple sample blood glucose recognition features, and a fully connected neural network, wherein each sample data set includes a sample dynamic blood glucose concentration, a sample exercise data, a sample food intake data, a sample insulin use record, and a sample environmental monitoring data; Using the blood sugar feature recognition model to perform feature recognition on the blood sugar data topology frame sequence respectively, and performing cross-scale mapping interactive analysis on the recognition results to obtain blood sugar related features; Training a blood sugar concentration prediction network layer, wherein the blood sugar concentration prediction network layer uses a support vector machine for binary classification training; Using the trained blood glucose concentration prediction network layer to predict the blood glucose concentration of the blood glucose association feature, to obtain a blood glucose concentration prediction result; The method of calling a pre-constructed blood glucose data topology framework sequence, inputting the dynamic blood glucose concentration sequence, the exercise data set, the food intake data set, the insulin use record set and the environmental monitoring data set into the pre-constructed blood glucose data topology framework sequence, and obtaining the blood glucose data topology framework sequence includes: Traverse the dynamic blood glucose concentration, exercise data, food intake data, insulin use records and environmental monitoring data to extract data indicators, and obtain a dynamic blood glucose concentration indicator set, an exercise data indicator set, a food intake data indicator set, an insulin use indicator set and an environmental monitoring indicator set; The dynamic blood glucose concentration, exercise data, food intake data, insulin use record and environmental monitoring data are respectively used as first-level topological nodes to obtain a first-level topological node set; Taking each indicator in the dynamic blood glucose concentration indicator set, the exercise data indicator set, the food intake data indicator set, the insulin use indicator set and the environmental monitoring indicator set as a secondary topological node, and obtaining multiple secondary topological node sets; A blood glucose data topology framework is constructed based on the primary topology node set and the multiple secondary topology node sets to obtain a pre-constructed blood glucose data topology framework.

2. The method according to claim 1, characterized in that include: Obtaining the number of dynamic blood glucose concentrations and corresponding multiple monitoring time points in the dynamic blood glucose concentration sequence; The pre-constructed blood glucose data topology framework is copied according to the number of dynamic blood glucose concentrations, and the multiple copied pre-constructed blood glucose data topology frameworks are sorted according to the multiple monitoring time points to obtain a sequence of the pre-constructed blood glucose data topology frameworks; The dynamic blood glucose concentration sequence is mapped into the pre-constructed blood glucose data topology framework sequence, and the data in the exercise data set, food intake data set, insulin usage record set and environmental monitoring data set are respectively input into the copied pre-constructed blood glucose data topology framework corresponding to the minimum time difference according to the data collection time points of the data in the exercise data set, food intake data set, insulin usage record set and environmental monitoring data set, so as to obtain the blood glucose data topology framework sequence, wherein the minimum time difference is the minimum time difference between the data collection time point corresponding to the data and multiple monitoring time points.

3. The method according to claim 1, characterized in that The blood glucose feature recognition model is used to perform feature recognition on the blood glucose data topological frame sequence respectively, and the recognition results are subjected to cross-scale mapping interactive analysis to obtain blood glucose related features, including: Using the blood sugar feature recognition model to perform feature recognition on the blood sugar data topology framework sequence respectively to obtain a blood sugar recognition feature sequence; Extracting a first blood sugar identification feature and a second blood sugar identification feature from the blood sugar identification feature sequence in a front-to-back order; Performing cross-scale mapping interaction analysis on the first blood glucose identification feature and the second blood glucose identification feature to obtain a first mapping interaction blood glucose association feature; extracting a third blood sugar identification feature from the blood sugar identification feature sequence again, performing cross-scale mapping interaction analysis on the third blood sugar identification feature and the first mapping interactive blood sugar association feature, and obtaining a second mapping interactive blood sugar association feature; The cross-scale mapping interactive analysis is performed in sequence until the end of the blood glucose identification feature sequence is reached to obtain the blood glucose associated feature.

4. The method according to claim 3, characterized in that Performing cross-scale mapping interaction analysis on the first blood sugar identification feature and the second blood sugar identification feature to obtain a first mapping interaction blood sugar association feature includes: Performing feature mapping inner product calculation on the first blood sugar identification feature and the second blood sugar identification feature to determine a feature similarity matrix; Normalizing the feature similarity matrix to obtain a mapping interaction matrix; A convolution operation is performed according to the mapping interaction matrix and the second blood glucose identification feature to obtain the first mapping interaction blood glucose association feature.

5. The method according to claim 1, characterized in that include: Acquire multiple sample data sets and corresponding multiple sample blood glucose identification features; Inputting the plurality of sample data sets into any one of the pre-constructed blood glucose data topology frames in the pre-constructed blood glucose data topology frame sequence respectively to obtain a plurality of sample blood glucose data topology frames; The framework constructed based on the fully connected neural network is supervised by using the multiple sample blood glucose data topological frameworks and the multiple sample blood glucose identification features to learn the mapping relationship between the blood glucose data topological framework and the blood glucose identification features until the output of the objective function reaches the minimum value, and the training is stopped to obtain the trained blood glucose feature recognition model.

6. The method according to claim 5, characterized in that The objective function is: in, is the objective function of the blood glucose feature recognition model, y is the blood glucose recognition feature of multiple samples, Identify the multiple output blood glucose features output by the model, y i is the i-th sample blood glucose identification feature among multiple sample blood glucose identification features, is the ith output blood glucose identification feature among the multiple output blood glucose identification features output by the model, N is the number of the multiple sample blood glucose identification features, and N is an integer greater than or equal to 3.

7. The method according to claim 1, characterized in that include: Acquire a class label set, wherein the class label set includes a positive class label and a negative class label, the positive class label indicates a normal blood glucose concentration state, and the negative class label indicates an abnormal blood glucose concentration state; Performing sample data retrieval based on the positive class label and the negative class label to obtain a plurality of blood sugar associated features of a plurality of positive class samples and a plurality of blood sugar associated features of a plurality of negative class samples; The blood glucose concentration prediction network layer is trained to convergence based on the support vector machine according to the multiple blood glucose associated features of the multiple positive class samples and the multiple blood glucose associated features of the multiple negative class samples, wherein the expression of the blood glucose concentration prediction network layer is: Among them, M(F) is the blood sugar prediction result, a i is the Lagrange multiplier of the support vector machine, y i is the category label, F is the blood sugar related feature corresponding to the positive category sample, and F i is the blood glucose related feature corresponding to the negative class sample, K(F,F i ) for F and F i The kernel function of is and b is the bias.

Images