A blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression and an electronic device

Abstract

The application provides a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression and an electronic device. The method comprises the following steps: first, obtaining historical blood glucose concentration data of a user and time series data related to blood glucose change, and preprocessing the obtained data; then, performing feature extraction based on the preprocessed data, constructing a feature vector, and taking corresponding blood glucose concentration as a label, so as to construct a sample set; subsequently, establishing a nonlinear relationship between the feature vector and the label by using a mixed kernel function Gaussian process regression model, optimizing parameters of a blood glucose concentration prediction model by using a swarm intelligence algorithm, and obtaining an optimal blood glucose concentration prediction model; and finally, inputting a current feature vector into the optimal blood glucose concentration prediction model, and taking a corresponding prediction label output by the optimal blood glucose concentration prediction model as a prediction result of the blood glucose concentration. The application can improve the accuracy of blood glucose concentration prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of blood glucose concentration prediction, and particularly relates to a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression and an electronic device. BACKGROUND

[0002] Accurate prediction of blood glucose concentration is a core technology for intelligent management of diabetes, directly affecting clinical decisions such as insulin dose adjustment and hypoglycemia warning. Current blood glucose prediction methods are mainly divided into two categories: one is a calculation method based on a physiological model, which predicts by simulating the physiological mechanism in the human metabolic process, and the other is a data-driven method based on machine learning, which trains a model using a large amount of historical data to achieve prediction.

[0003] However, the existing blood glucose concentration prediction method still has significant limitations, and the prediction accuracy needs to be improved. SUMMARY

[0004] The purpose of the present application is to provide a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression and an electronic device, which can improve the accuracy of blood glucose concentration prediction.

[0005] The technical scheme adopted by the present application is:

[0006] A blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression, comprising:

[0007] Obtaining historical blood glucose concentration data of a user and time series data related to blood glucose changes, and preprocessing the obtained data;

[0008] Performing feature extraction based on the preprocessed data, constructing a feature vector, and constructing a training sample set with the corresponding blood glucose concentration as a label;

[0009] A mixed kernel function Gaussian process regression model is used to establish a blood glucose concentration prediction model; the blood glucose concentration prediction model is used to describe the non-linear relationship between the feature vector and the label; specifically: the blood glucose concentration prediction model calculates the covariance between the current feature vector and each training sample feature vector through a mixed kernel function, and combines the contribution weight of each training sample to the prediction result to aggregate the information of all training samples to generate the prediction label corresponding to the current feature vector; wherein the contribution weight of each training sample to the prediction result is equal to the element at the corresponding position after the inverse matrix of the actual observation covariance matrix is multiplied by a vector composed of the labels of all training samples in the training sample set; Based on the training sample set and the mixed kernel function, first, the covariance between the feature vectors of the training sample set is calculated based on the mixed kernel function; then, based on the calculated covariance, a covariance matrix between the feature vectors of the training sample set is constructed ; adding noise to ;

[0010] wherein the mixed kernel function comprises a Gaussian kernel function and a rational quadratic kernel function;

[0011] Based on the training sample set, a swarm intelligence algorithm is used to optimize parameters of the blood glucose concentration prediction model to obtain an optimal blood glucose concentration prediction model.

[0012] The current feature vector is input into the optimal blood glucose concentration prediction model, and a corresponding prediction label output by the optimal blood glucose concentration prediction model is taken as a prediction result of the blood glucose concentration.

[0013] In a possible implementation, the time series data related to blood glucose change includes: insulin injection dose / time, carbohydrate intake / time, exercise type / intensity / time, and heart rate.

[0014] In a possible implementation, the pre-processing of the obtained data includes: missing value processing, outlier detection and correction, time alignment, and normalization.

[0015] In a possible implementation, feature extraction is performed based on the pre-processed data to construct a feature vector, including:

[0016] Features are extracted from historical blood glucose concentration data, including: blood glucose change rate, blood glucose change acceleration, blood glucose mean value, variance, and extreme value in a past period of time;

[0017] Features are extracted from the time series data related to blood glucose change, including: insulin activity curve estimation, carbohydrate absorption curve estimation, and exercise intensity quantification index.

[0018] The feature vector is constructed based on the extracted features.

[0019] In a possible implementation, the training sample set is denoted as ; is a feature vector of the i-th training sample, is a true label of the i-th training sample, is a number of training samples; The blood glucose concentration prediction model is:

[0020]

[0021] ;

[0022] wherein, is a current feature vector corresponding prediction label;​​​ denotes the covariance between the and , which is calculated based on the hybrid kernel function; ;

[0023] denotes the contribution weight of the th training sample to the prediction result, which is equal to the th element of ;

[0024] The formula for calculating the covariance between each pair of feature vectors of the training sample set based on the hybrid kernel function is:

[0025] ;

[0026] wherein: denotes the covariance between the and , is the feature vector of the th training sample, ; and are the weights of the Gaussian kernel function and the rational quadratic kernel function, respectively, and the hyperparameters is the control parameter of the amplitude of the Gaussian kernel, and the hyperparameter is the scale factor for distance calculation; is the shape parameter, which controls the weighting between different scales; is the length scale parameter, which controls the smoothness of the variation; , , , , , are all obtained by optimizing the swarm intelligence algorithm;

[0027] The formula for constructing the covariance matrix between the feature vectors of the training sample set based on the calculated covariance is:

[0028] ;

[0029] Add noise to to obtain The formula is: ; wherein, is the identity matrix of , is the variance of the Gaussian white noise;

[0030] , denotes the vector composed of the labels of all training samples in the training sample set.​

[0031] In a possible implementation, the swarm intelligence algorithm adopts a fishing optimization algorithm (CFOA algorithm).

[0032] In a possible implementation, based on a training sample set, parameters of a blood glucose concentration prediction model are optimized by using a fishing optimization algorithm to obtain an optimal blood glucose concentration prediction model, including:

[0033] 1) Parameters of the fishing optimization algorithm are set, including: the number of fishermen, the maximum number of iterations, the learning factor and the cooperation factor; positions of each fisherman are initialized; each position of the fisherman corresponds to a set of parameter values of the blood glucose concentration prediction model;

[0034] 2) The current position of each fisherman is substituted into the blood glucose concentration prediction model, and a feature vector of each training sample is input into the blood glucose concentration prediction model as a current feature vector to obtain a corresponding prediction label; the fitness value of the current position of each fisherman is calculated based on the real label and the prediction label of the training sample:

[0035] ;

[0036] In the formula, is the fitness value, is the prediction label corresponding to the i th training sample, is the real label of the i th sample;

[0037] 3) The optimal fisherman position and the best fitness value are obtained;

[0038] 4) The position of each fisherman is updated according to the optimal individual and the best fitness value, combined with the learning factor and the cooperation factor, and step 2) is returned for iterative optimization until the maximum number of iterations is reached, if so, the parameter values of the blood glucose concentration prediction model corresponding to the global optimal fisherman position and the global best fitness value are output as the optimal parameter values; the optimal parameter values are substituted into the blood glucose concentration prediction model to obtain the optimal blood glucose concentration prediction model.

[0039] In a second aspect, the present application provides an electronic device, comprising: a memory and a processor;

[0040] The memory is configured to store a computer program.

[0041] The processor is configured to call the computer program to execute the method described above.

[0042] ​​​In a third aspect, the present application provides a computer readable storage medium, having stored therein a computer program, which, when executed on an electronic device, causes the electronic device to implement the method as described above.

[0043] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed on an electronic device, causes the electronic device to implement the method as described above.

[0044] The specific implementation manners of the above-mentioned second to fourth aspects of the present application can refer to the implementation manners of the above-mentioned first aspect, which will not be described here.

[0045] Advantages:

[0046] The present application provides a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression (GPR) and an electronic device, which effectively combines the advantages of a Gaussian kernel function in capturing local fine changes and a rational quadratic kernel function in depicting multi-scale features. For example, the Gaussian kernel function can capture local sensitive features such as postprandial blood glucose spikes, and the rational quadratic kernel function can fit the long-term trend of slow changes in basal metabolism, thereby significantly enhancing the representation ability of the Gaussian process regression model for complex nonlinear dynamics of blood glucose. In addition, a swarm intelligence algorithm is introduced for global automatic optimization, which greatly improves the model performance and overcomes the limitations and inefficiency of manual parameter tuning, significantly improves the accuracy and generalization ability of the blood glucose concentration prediction model, and provides more reliable technical support for personalized blood glucose management of diabetic patients. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the present application, the technical solutions of the present application will be further described in detail below with reference to the embodiments of the present application and the accompanying drawings.

[0049] The application provides a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression (GPR) and an electronic device, which provides more reliable technical support for personalized blood glucose management of diabetic patients. First, historical blood glucose concentration data and time series data related to blood glucose change of a user are acquired, and the acquired data is preprocessed; then, feature extraction is performed based on the preprocessed data, a feature vector is constructed, and the corresponding blood glucose concentration is taken as a label, so as to construct a sample set; subsequently, a mixed kernel function Gaussian process regression model is used to establish a nonlinear relationship between the feature vector and the label, a population intelligence algorithm is used to optimize parameters of a blood glucose concentration prediction model, and an optimal blood glucose concentration prediction model is obtained; a current feature vector is input into the optimal blood glucose concentration prediction model, and a prediction label output by the optimal blood glucose concentration prediction model is taken as a prediction result of the blood glucose concentration.

[0050] The specific embodiments according to the application will be described below with reference to the drawings. Figure 1 The specific embodiments according to the application will be described below with reference to the drawings.

[0051] Embodiment one:

[0052] The application embodiment provides a blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression, comprising:

[0053] S1, acquiring historical blood glucose concentration data and time series data related to blood glucose change of a user, and preprocessing the acquired data.

[0054] In some embodiments, the time series data related to blood glucose change includes insulin injection dose / time, carbohydrate intake amount / time, exercise type / intensity / time, heart rate.

[0055] In the application embodiment, the character " / " represents a "and" relationship between the objects before and after it. For example, A / B means A and B.

[0056] Subsequently, the acquired data is preprocessed, including missing value processing, outlier detection and correction, time alignment, and normalization.

[0057] S2, performing feature extraction based on the preprocessed data, constructing a feature vector, and taking the corresponding blood glucose concentration as a label to construct a sample set.

[0058] In some embodiments, features are extracted from the historical blood glucose concentration data, including blood glucose change rate, blood glucose change acceleration, blood glucose mean value, variance and extreme value in a past period of time;

[0059] Features are extracted from the time series data related to blood glucose change, including insulin activity curve estimation, carbohydrate absorption curve estimation, and exercise intensity quantification index.

[0060] Based on the features extracted above, a feature vector is constructed, and the corresponding blood glucose concentration is used as a label.

[0061] Considering that blood glucose changes are influenced by multiple factors such as insulin activity, carbohydrate absorption, and exercise intensity, this application extracts these features to construct feature vectors, trains the model, and makes predictions, which can improve prediction performance.

[0062] In this embodiment of the application, the sample set is divided into a training sample set and a test sample set. For example, the sample set and the test sample set are divided in an 8:2 ratio.

[0063] The training sample set can be described as The test sample set can be described as follows: ; For the first The feature vectors of each training sample ; For the first The feature vector of each test sample ; For the first The true labels of each training sample; For the first The true label of each test sample. and These represent the number of training and testing samples, respectively. , , This represents a vector composed of the feature vectors of all training samples in the training sample set. This represents a vector consisting of the true labels of all training samples in the training sample set; , , This represents a vector composed of the feature vectors of all test samples in the test sample set. It is a vector consisting of the true labels of all test samples in the test sample set.

[0064] S3. A blood glucose concentration prediction model is established using a Gaussian process regression model with mixed kernel functions; the blood glucose concentration prediction model is used to describe the nonlinear relationship between the feature vector and the label (blood glucose concentration); wherein, the mixed kernel function includes a Gaussian kernel function and a rational quadratic kernel function.

[0065] Considering that blood glucose changes are affected by multiple factors such as insulin activity, carbohydrate absorption, exercise intensity, etc., and that the changes present complex characteristics of local rapid fluctuations (such as postprandial peaks) and long-term gradual changes (such as basal metabolism) coexisting, a single kernel function (such as a Gaussian kernel) is difficult to capture such multi-scale spatiotemporal characteristics at the same time, resulting in limited prediction accuracy. The existing methods are not fine enough in extracting and fusing dynamic physiological characteristics such as insulin activity curve, carbohydrate absorption rate, and exercise intensity, which limits the model's ability to analyze the blood glucose driving mechanism. Gaussian process regression (GPR) has the advantages of providing prediction uncertainty quantification and strong modeling ability for nonlinear relationships. Therefore, the present application uses a mixed kernel function Gaussian process regression for modeling to improve the model's expression ability.

[0066] The obtained feature vector is used as the input of model training, and the label is used as the output of the model.

[0067] In some embodiments, the present step specifically comprises:

[0068] S3.1, a mixed kernel function is constructed, and the expression is as follows:

[0069] ;

[0070] Wherein: represents the covariance between and , which can quantify the similarity degree of the two feature vectors, and are the weights of the Gaussian kernel function and the rational quadratic kernel function respectively, and the hyperparameter is the control amplitude of the Gaussian kernel, and the hyperparameter is the scale factor of distance calculation; is the shape parameter, which controls the weighting between different scales; is the length scale parameter, which controls the smoothness of the change.

[0071] S3.2, the covariance matrix between the feature vectors of the training sample set is calculated using the constructed mixed kernel function :

[0072] ;

[0073] Wherein, is the ideal noise-free covariance matrix, is calculated based on the feature vectors of the training sample set, and is used to construct the Gaussian process prior distribution.

[0074] Considering the influence of noise, noise is added to to obtain the actual observation covariance matrix , as follows:

[0075] ;

[0076] wherein, is an identity matrix, is a Gaussian white noise variance of the Gaussian process regression.

[0077] S3.3, constructing a blood glucose concentration prediction model:

[0078] ;

[0079] wherein, is a current feature vector corresponding to a predicted label; denotes a covariance between , which is calculated based on a mixed kernel function; denotes a contribution weight of an i-th training sample to a prediction result, which is a value of an i-th element of . S4, based on the training sample set, a blood glucose concentration prediction model is optimized by using a swarm intelligence algorithm to obtain an optimal blood glucose concentration prediction model.

[0080] Considering that the mixed kernel function has a large number of hyperparameters and strong coupling, a traditional optimization method faces a dimension disaster, is easy to fall into a local optimum, and has huge calculation overhead, and it is difficult to meet the real-time demand of blood glucose prediction. The embodiment of the application uses a swarm intelligence algorithm to optimize the parameters of the blood glucose concentration prediction model to improve the parameter optimization effect.

[0081] In some embodiments, the swarm intelligence algorithm uses a catch fish optimization algorithm CFOA, an ant colony algorithm ACO, a particle swarm algorithm PSO, or the like.

[0082] In some embodiments, the step specifically includes:

[0083] In some embodiments, the step specifically includes:

[0084] S4.1, setting parameters of the catch fish optimization algorithm, including: a number of fishermen, a maximum number of iterations, a learning factor, and a cooperation factor; initializing a position of each fisherman; the position of each fisherman corresponds to a set of parameter values of the blood glucose concentration prediction model;

[0085] S4.2, respectively substituting a current position of each fisherman into the blood glucose concentration prediction model, and inputting a feature vector of each training sample as a current feature vector into the blood glucose concentration prediction model to obtain a corresponding predicted label; calculating an adaptability value of the current position of each fisherman based on a real label and a predicted label of the training sample:

[0086] ; ​

[0087] wherein is a fitness value, is the th training sample corresponding predicted label, is the th sample's true label;

[0088] S4.3, obtaining the optimal fisherman position and the best fitness value;

[0089] S4.4, updating the position of each fisherman according to the optimal individual and the best fitness value, combining the learning factor and the cooperation factor, returning to step S4.2 for iteration optimization until the maximum iteration number is reached, if so, outputting the global optimal fisherman position and the parameter value of the blood glucose concentration prediction model corresponding to the global best fitness value as the optimal parameter value; substituting the optimal parameter value into the blood glucose concentration prediction model to obtain the optimal blood glucose concentration prediction model.

[0090] After obtaining the optimal blood glucose concentration prediction model, the performance of the optimal blood glucose concentration prediction model can be tested using the training sample set.

[0091] In the test phase:

[0092] First, the constructed mixed kernel function is used to calculate the covariance matrix between the feature vectors of the training sample set and the test sample set :

[0093] ;

[0094] is a matrix. For posterior prediction, the trained model parameters are called.

[0095] Then, the optimal blood glucose concentration prediction model outputs the predicted value;

[0096] ;

[0097] wherein, is a vector composed of the mean (optimal estimate) of the predicted labels of all test samples in the test sample set.

[0098] Next, the prediction uncertainty is evaluated.

[0099] First, the constructed mixed kernel function is used to calculate the covariance matrix between the feature vectors of the test sample set , is a matrix.

[0100] ;

[0101] It is The matrix.

[0102] Reconstruct the true label vectors of the training samples and the predicted label vector of the test sample The joint prior distribution is as follows:

[0103] ;

[0104] In the formula With a mean of 0 and a covariance of The normal distribution.

[0105] Through the By defining a Gaussian joint prior distribution, the predicted label vector can be analytically derived. The posterior distribution can be written as ;in Let be a vector consisting of the mean (optimal estimate) of the predicted labels of all test samples in the test sample set. The covariance of the predicted labels for all test samples in the test sample set is calculated as follows:

[0106] , ;

[0107] Therefore, the model's prediction error and uncertainty can be assessed.

[0108] For example, the root mean square error (RMSE) is used to assess prediction error, and the calculation formula is as follows:

[0109] ;

[0110] In the formula, express The Middle The nth element represents the nth element in the test sample set. The mean of the predicted labels for each test sample.

[0111] The uncertainty of the model is assessed based on the coverage ratio, and the calculation formula is as follows:

[0112] ;

[0113] in, yes The Middle The diagonal element represents the nth element in the test sample set. The prediction standard deviation of each test sample; is an indicator function, count 1 if the inner condition is satisfied, otherwise do not count; 1.96 corresponds to the 95% confidence interval of the normal distribution; the proportion of samples satisfying the condition is counted.

[0114] The proportion of true labels in the test sample set falling within the predicted confidence interval is calculated by the coverage calculation formula. If the coverage is greater than the coverage threshold (such as 95%), it means that the model is uncertain and has high reliability. Otherwise, if the coverage is less than the coverage threshold, it means that the mixed kernel function parameters or noise terms need to be adjusted .

[0115] S5, input the current feature vector into the optimal blood glucose concentration prediction model, and take the corresponding prediction label output by the optimal blood glucose concentration prediction model as the prediction result of the blood glucose concentration.

[0116] The present application effectively combines the advantages of Gaussian kernel function capturing local fine changes and rational quadratic kernel function depicting multi-scale features, significantly enhancing the representation ability of Gaussian process regression model for complex nonlinear dynamics of blood glucose; In addition, the global automatic optimization is introduced by introducing the swarm intelligence algorithm, which greatly improves the model performance and overcomes the limitations and inefficiency of manual parameter adjustment, significantly improves the accuracy and generalization ability of the blood glucose concentration prediction model, and provides more reliable technical support for personalized blood glucose management of diabetic patients.

[0117] It should be understood that the above numbers S1-S5, S3.1-S3.3, S4.1-S4.4 are only used to distinguish and facilitate the expression of each different step, and do not necessarily constitute a limitation on the execution order between each step.

[0118] Example two:

[0119] The embodiment provides an electronic device, comprising: a memory and a processor;

[0120] The memory is used for storing a computer program;

[0121] The processor is used for calling the computer program to execute the method as described in example one.

[0122] Example three:

[0123] The embodiment provides a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program runs on an electronic device, so that the electronic device implements the method as described in example one.

[0124] Example four:

[0125] The embodiment provides a computer program product comprising a computer program which, when running on an electronic device, causes the electronic device to implement the method according to the first embodiment.

[0126] The specific implementation of the system, the electronic device, the computer readable storage medium and the computer program product provided by the embodiment of the present application can refer to the specific embodiments of the above method, and details are not described here.

[0127] The technical contents of the above embodiments can be referred to each other, and for the same or similar technical features, part of the embodiments has been omitted appropriately.

[0128] Obviously, those skilled in the art should understand that each unit or each step of the above-mentioned present application can be realized by using general computing device, and they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be realized by using program code executable by the computing device, so that they can be stored in the storage device and executed by the computing device, or they can be made into each integrated circuit module respectively, or multiple modules or steps among them can be made into a single integrated circuit module to realize. Thus, the present application is not limited to any specific combination of hardware and software.

[0129] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A blood glucose concentration prediction method based on a mixed kernel function Gaussian process regression, characterized by, The method comprises the following steps: Obtain historical blood glucose concentration data and time series data related to blood glucose changes of a user, and preprocess the obtained data; Based on the preprocessed data, feature extraction is performed, a feature vector is constructed, and the corresponding blood glucose concentration is taken as a label to construct a training sample set; A hybrid kernel function Gaussian process regression model is used to establish a blood glucose concentration prediction model; the blood glucose concentration prediction model is used to describe the nonlinear relationship between the feature vector and the label; specifically: The blood glucose concentration prediction model calculates the covariance of the current feature vector and the feature vector of each training sample by a mixed kernel function, combines the contribution weight of each training sample to the prediction result, and aggregates the information of all training samples to generate a prediction label corresponding to the current feature vector; wherein the contribution weight of each training sample to the prediction result is equal to the element at the corresponding position after the inverse matrix of the actual observed covariance matrix multiplied by the vector composed of the labels of all training samples in the training sample set. Let the training sample set be ; is the feature vector of the th training sample, is the true label of the th training sample, is the number of training samples; The blood glucose concentration prediction model is: ； in, For the current feature vector Corresponding predicted labels; express and The covariance between them is calculated based on the mixture kernel function; Indicates the first The contribution weights of each training sample to the prediction result are set to a value of . The One element; wherein, , represents a vector consisting of the labels of all training samples in the training sample set; wherein, based on the training sample set and a mixed kernel function, the mixed kernel function comprising a Gaussian kernel function and a rational quadratic kernel function, the constructing step comprising: First, based on the hybrid kernel function, the covariance between the feature vectors of the training sample set is calculated, and the formula is: ； wherein: denotes the covariance between and , is the feature vector of the th training sample, ; and are the weights of the Gaussian kernel and the Rational Quadratic kernel, respectively, and the hyperparameters is the amplitude controlling the Gaussian kernel, and the hyperparameter is the scale factor for distance computation; is the shape parameter, controlling the weighting between different scales; is the length scale parameter, controlling the degree of smoothness of the variation; , , , , , are all obtained by optimizing the swarm intelligence algorithm; Then, based on the calculated covariance, a covariance matrix between the feature vectors of the training sample set is constructed The formula is: ； Again add noise on top to get , the formula is: ; in the formula, is unit matrix, is Gaussian white noise variance; Based on the training sample set, a swarm intelligence algorithm is used to optimize the parameters of the blood glucose concentration prediction model to obtain an optimal blood glucose concentration prediction model; The current feature vector is input into the optimal blood glucose concentration prediction model, and the corresponding predicted label output by the optimal blood glucose concentration prediction model is taken as the prediction result of the blood glucose concentration.

2. The method of claim 1, wherein, The time series data related to blood glucose changes include: insulin injection dose / time, carbohydrate intake / time, exercise type / intensity / time, and heart rate.

3. The method of claim 2, wherein, The preprocessing of the obtained data includes: missing value processing, outlier detection and correction, time alignment, and normalization.

4. The method of claim 2, wherein, Based on the preprocessed data, feature extraction is performed, a feature vector is constructed, including: Features are extracted from historical blood glucose concentration data, including: blood glucose change rate, blood glucose change acceleration, blood glucose mean value, variance, and extreme value in a certain period of time in the past; Features are extracted from time series data related to blood glucose changes, including: insulin activity curve estimation, carbohydrate absorption curve estimation, and exercise intensity quantification index; The feature vector is constructed based on the extracted features.

5. The method of claim 4, wherein, The swarm intelligence algorithm uses the fishing optimization algorithm.

6. The method of claim 5, wherein, Based on the training sample set, the fishing optimization algorithm is used to optimize the parameters of the blood glucose concentration prediction model to obtain an optimal blood glucose concentration prediction model, including: 1) Set the parameters of the fishing optimization algorithm, including: the number of fishermen, the maximum number of iterations, the learning factor, and the cooperation factor; initialize the position of each fisherman; the position of each fisherman corresponds to a set of parameter values of the blood glucose concentration prediction model; 2) Substitute the current position of each fisherman into the blood glucose concentration prediction model, and input the feature vector of each training sample into the blood glucose concentration prediction model as the current feature vector to obtain the corresponding predicted label; calculate the fitness value of the current position of each fisherman based on the true label and the predicted label of the training sample: ； In the formula is a fitness value, is the training sample corresponding predicted label, is the true label of the sample; 3) Obtain the optimal fisherman position and the best fitness value; 4) Update the position of each fisherman according to the optimal individual and the best fitness value, combine the learning factor and the cooperation factor, and return to step 2) for iterative optimization until the maximum number of iterations is reached; if so, output the parameter values of the blood glucose concentration prediction model corresponding to the global optimal fisherman position and the global best fitness value as the optimal parameter values; substitute the optimal parameter values into the blood glucose concentration prediction model to obtain the optimal blood glucose concentration prediction model.

7. An electronic device, comprising: The method comprises the following steps: A memory and a processor; The memory is used to store a computer program; The processor is configured to invoke the computer program to perform the method of any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium and, when running on the electronic device, enables the electronic device to implement the method of any one of claims 1-6.

9. A computer program product comprising a computer program, characterized in that, The computer program, when running on the electronic device, enables the electronic device to implement the method of any one of claims 1-6.

Images