Microwave noninvasive blood glucose prediction method, device, equipment, medium and product

Abstract

The invention discloses a microwave noninvasive blood glucose prediction method, device, equipment, medium and product, and relates to the field of microwave sensing, the method comprises the following steps: constructing a data set according to a blood glucose time sequence collected by a microwave noninvasive blood glucose sensing system; based on a convolution block attention module and an improved quantum particle swarm optimization algorithm, constructing an optimal bidirectional gating loop network model according to the data set; the optimal two-way gating circulation network model is a two-way gating circulation network model fusing a convolution block attention module and an improved quantum particle swarm optimization algorithm; and predicting the blood sugar of the to-be-tested person according to the optimal bidirectional gating circulation network model. According to the application, accurate prediction and effective management of the blood sugar level of the human body are realized.

Description

Technical Field

[0001] This application relates to the field of microwave sensing, and in particular to a microwave non-invasive blood glucose prediction method, device, equipment, medium, and product. Background Technology

[0002] Microwave non-invasive blood glucose prediction is a non-invasive method that uses the dielectric properties of the interaction between microwave signals and blood to measure blood glucose concentration. Compared with traditional blood glucose monitoring methods, such as finger prick blood sampling or implantable blood glucose monitoring devices, microwave non-invasive blood glucose prediction technology has advantages such as comfort, portability, no risk of infection, and low cost, making it easy to achieve continuous non-invasive blood glucose monitoring.

[0003] However, microwave non-invasive blood glucose prediction technology also faces some challenges. Since many factors such as diet, exercise status, and psychological state can directly affect the blood glucose level of the human body, it is necessary to establish a more accurate blood glucose trend prediction model. However, the existing blood glucose trend prediction models have certain limitations in dealing with these complex dynamic changes and cannot accurately predict and effectively manage the blood glucose level of the human body. Summary of the Invention

[0004] The purpose of this application is to provide a microwave non-invasive blood glucose prediction method, device, equipment, medium and product to solve the problem that existing blood glucose trend prediction models have limitations in handling these complex dynamic changes and cannot accurately predict and effectively manage human blood glucose levels.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] In a first aspect, this application provides a microwave non-invasive blood glucose prediction method, including:

[0007] A dataset was constructed based on the blood glucose time series collected by the microwave non-invasive blood glucose sensing system;

[0008] Based on the convolutional block attention module and the improved quantum particle swarm optimization algorithm, an optimal bidirectional gated recurrent network model is constructed according to the dataset; the optimal bidirectional gated recurrent network model is a bidirectional gated recurrent network model that integrates the convolutional block attention module and the improved quantum particle swarm optimization algorithm.

[0009] Predict blood glucose levels in the tested individuals based on the optimal bidirectional gated recurrent network model.

[0010] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the microwave non-invasive blood glucose prediction method described in any one of the above descriptions.

[0011] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the microwave non-invasive blood glucose prediction method described above.

[0012] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the microwave non-invasive blood glucose prediction method described above.

[0013] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0014] This application utilizes a bidirectional gated recurrent unit (BiGRU) model, which integrates a convolutional block attention module (CBAM) and an improved quantum particle swarm optimization algorithm (IQPSO), to perform time-series prediction of human blood glucose data collected by a microwave non-invasive blood glucose sensing system. It processes these complex dynamic changes in real time, enabling accurate prediction and effective management of human blood glucose levels, and further promoting the application of microwave sensing technology in the field of health monitoring. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of a microwave non-invasive blood glucose prediction method provided in this application;

[0017] Figure 2 A schematic diagram illustrating the sliding window processing effect of the IQPSO-CBAM-BiGRU input provided in this application;

[0018] Figure 3 A comparison chart of actual and predicted blood glucose values ​​input into the IQPSO-CBAM-BiGRU provided in this application; wherein, Figure 3 (a) in the graph shows the predicted trend of blood glucose levels after 3 minutes. Figure 3 (b) in the figure shows the predicted trend of blood glucose levels after 9 minutes; Figure 3(c) in the figure shows the predicted trend of blood glucose levels after 60 minutes.

[0019] Figure 4 A flowchart of another microwave non-invasive blood glucose prediction method provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] This application provides a microwave non-invasive blood glucose prediction method. This method is executed by a computer device, specifically a terminal or server, or both. In this application embodiment, for example... Figure 1 As shown, the method includes the following steps.

[0023] S1: Construct a dataset based on the blood glucose time series collected by the microwave non-invasive blood glucose sensing system.

[0024] S2: Based on the convolutional block attention module and the improved quantum particle swarm optimization algorithm, construct the optimal bidirectional gated recurrent network model according to the dataset; the optimal bidirectional gated recurrent network model is a bidirectional gated recurrent network model that integrates the convolutional block attention module and the improved quantum particle swarm optimization algorithm.

[0025] S3: Predict the blood glucose level of the person being tested based on the optimal bidirectional gated recurrent network model.

[0026] In an exemplary embodiment, S1 can be replaced by the following steps.

[0027] S11: Using a microwave non-invasive blood glucose sensing system, electrical signals returned from the volunteer's skin are collected at set time intervals within a set time period, and the electrical signals are converted into blood glucose values.

[0028] S12: Generate a blood glucose time series based on all blood glucose values ​​within a set time period.

[0029] S13: Use the blood glucose time series as a dataset, and divide the dataset into a training set, a validation set, and a test set according to the time order.

[0030] In practical application, a microwave non-invasive blood glucose sensing system was used to monitor the blood glucose levels of three volunteers for half a month. The system was attached to the volunteers' wrists via a strap to collect blood glucose data. During the experiment, the microwave non-invasive blood glucose sensing system acquired an electrical signal every 3 minutes and converted it into a blood glucose value. All data was transmitted and stored in real time on a host computer system. Ultimately, over 7000 blood glucose time series data were obtained as a dataset. The first 70% of the continuous sequences were used as the training set, the next 15% as the validation set, and the remaining 15% as the test set.

[0031] In an exemplary embodiment, a preprocessing process is further included after S1, which is as follows:

[0032] S1.1: Use spline interpolation to clean the dataset and determine the cleaned dataset.

[0033] S1.2: The cleaned dataset is normalized using the min-max normalization method to determine the normalized dataset.

[0034] S1.3: Construct a sliding window and use a direct strategy method to transform the normalized dataset to determine the processed dataset.

[0035] In practical applications, the preprocessing process includes data cleaning of the original dataset, data normalization, and the construction of a sliding window.

[0036] 1) Use spline interpolation for data imputation. Spline interpolation constructs a piecewise polynomial function to generate a smooth curve using adjacent data points, thereby estimating missing values. The Savitzky-Golay method is then used to filter the data to remove noise and smooth it further. The filtering formula is as follows:

[0037]

[0038] Among them, X j X j 'Represents the original data and the smoothed data respectively, C i represents the polynomial fitting coefficients, and m is the width of the sliding window.

[0039] 2) Normalize the data using the min-max normalization method, scaling the data values ​​to the range [0,1]. The min-max normalization formula is as follows:

[0040]

[0041] Where, x' ix is the normalized data value. i x represents the original data value. max and x min These represent the maximum and minimum values ​​in the data, respectively.

[0042] 3) By constructing a sliding window, the data is transformed into a form suitable for supervised learning. The direct policy method is applied for time series data transformation, and the specific formula is as follows:

[0043] y t+h =f h (y t ,…,y t-n+1 )+w

[0044] Where t∈{n,...,NH} represents the time step, h∈{1,...,H} represents the prediction step number, and H is the prediction time step size. The input step size is set to 10, and the output step size to 3. In microwave non-invasive blood glucose prediction, data is recorded every 3 minutes.

[0045] Using blood glucose data from the first 30 minutes, the model can predict blood glucose levels for the next 9 minutes. Figure 2 This demonstrates that the sliding window method successfully transforms the raw data into input-output pairs for supervised learning, enabling the model to accurately predict future blood glucose levels based on historical data.

[0046] In an exemplary embodiment, S2 can be replaced by the following steps.

[0047] S21: Using a bidirectional gated recurrent network model as the basic prediction model, a convolutional block attention module is introduced during the training process of the bidirectional gated recurrent network model to capture key features in the processed dataset, and an improved quantum particle swarm optimization algorithm is used to perform a global search in the search space to determine the optimal hyperparameters; the key features include forward and backward information of blood glucose time series; the improved quantum particle swarm optimization algorithm introduces a contraction-expansion factor to adjust the search range of particles.

[0048] S22: Based on the optimal hyperparameters, construct the bidirectional gated recurrent network model and determine the optimal bidirectional gated recurrent network model.

[0049] In an exemplary embodiment, S21 can be replaced by the following steps.

[0050] S211: The convolutional block attention module includes a channel attention module and a spatial attention module connected in sequence.

[0051] S212: Convert the blood glucose time series into an original feature map; the original feature map includes feature information from multiple channels.

[0052] S213: Based on the bidirectional gated recurrent network model, the original feature map is input into the channel attention module, and a one-dimensional channel attention map is output.

[0053] S214: Perform dot product of the one-dimensional channel attention map and the original feature map on a channel-by-channel basis to determine the weighted feature map.

[0054] S215: Input the weighted feature map into the spatial attention module and output a two-dimensional spatial attention map.

[0055] S216: Multiply the two-dimensional spatial attention map and the original feature map to determine the feature map after the multiplication.

[0056] In an exemplary embodiment, the original feature map is input to the channel attention module, and a one-dimensional channel attention map is output, specifically including:

[0057] The channel attention module includes a max pooling layer, an average pooling layer, and a multilayer perceptron.

[0058] The original feature maps are input into the max pooling layer and the average pooling layer, respectively, and the max pooling feature map and the average pooling feature map are output.

[0059] The max pooling feature map and the average pooling feature map are input into the multilayer perceptron to determine two different channel attention weights.

[0060] Using the Sigmoid activation function, a one-dimensional channel attention map is determined based on two different channel attention weights.

[0061] In practical applications, BiGRU is selected as the basic prediction model to capture forward and backward information of sequence data, thereby improving the understanding and prediction ability of sequence patterns. The hyperparameters of the BiGRU network are set as follows: optimizer (Adam), learning rate range [0.001, 0.1], sample number range [32, 256], number of iterations 100, and loss function mean squared error (MSE).

[0062] During training, a convolutional block attention module is introduced. The data processing of the convolutional block attention module specifically includes: given the original feature map of the input F∈R C ×H×W, where R CLet M be a channel-dimensional tensor space, which is actually equal to C, where C is the number of channels (i.e., depth) of the original feature map, H is the height of the original feature map, and W is the width of the original feature map. That is, the original feature map has C channels, and each channel is a feature map of size H×W. This original feature map usually comes from the preceding convolutional layers and contains multi-dimensional feature information of the image. A one-dimensional channel attention map M is obtained through the channel attention module. C ∈R C ×1×1, then perform a dot product with the original feature map to obtain a weighted feature map F′. This result is used as the input to the spatial attention module to obtain a two-dimensional spatial attention map M. S ∈R1×H×W, where R1 is a single-channel dimension tensor space. The output result is then multiplied with the original feature map to obtain the output dot product feature map F″.

[0063] In practical applications, the data processing of the channel attention module specifically includes: the input original feature map is transformed from C×H×W to C×1×1 shape through two parallel max pooling layers and average pooling layers, resulting in two pooling results. These pooling results are then passed through a shared multilayer perceptron to generate two attention weights of size C×1×1. These two weights are then added together and passed through a sigmoid activation function to obtain the final channel attention output, i.e., a C×1×1 channel attention map M. C .

[0064] Finally, the channel attention map M C The original input feature map F is multiplied channel by channel to obtain the output feature map F′. This weighted feature map F′ is adjusted according to the different weights of each channel, so that important channels are enhanced and unimportant channels are suppressed, thus helping the model to focus more on useful features.

[0065] In one exemplary embodiment, an improved quantum particle swarm optimization algorithm is used to perform a global search within the search space to determine the optimal hyperparameters, specifically including:

[0066] The population is initialized based on the initial hyperparameters of the bidirectional gated recurrent network model.

[0067] The fitness values ​​of individuals in the population were assessed using the shrinkage-expansion factor.

[0068] Based on the fitness value, the bidirectional gated recurrent network model is trained, and the training result is determined.

[0069] The mean squared error of the loss function is calculated based on the training results, and the mean squared error of the loss function is used as the fitness value.

[0070] The bidirectional gated recurrent network model is trained iteratively, recording all optimal fitness values ​​and individual optimal fitness values, and the hyperparameters corresponding to the global optimal fitness value are taken as the best hyperparameters.

[0071] In practical applications, the population consists of N particles, each representing a combination of hyperparameters.

[0072] The specific parameters included in hyperparameters (i.e., individual parameters) are as follows.

[0073] 1) BiGRU structure-related parameters.

[0074] Hidden units: Controls the number of neurons in the BiGRU and affects the model's expressive power.

[0075] Number of layers: The number of layers in BiGRU determines the depth of the network.

[0076] Dropout rate: The proportion of neurons that are randomly dropped to prevent overfitting.

[0077] 2) Optimizer-related parameters.

[0078] Learning rate: Affects the convergence speed of the model, and is usually optimized between 0.0001 and 0.01.

[0079] Gradient clipping threshold: Prevents gradient explosion by setting a maximum gradient threshold.

[0080] 3) Training strategy related parameters.

[0081] Batch size: The number of samples input during each training session, such as 16, 32, 64, etc.

[0082] Training epochs: The number of times the dataset is completely traversed during the training process.

[0083] 4) IQPSO related parameters.

[0084] Population size (N): Sets the initial number of particles, for example, N=30.

[0085] Contraction-expansion factor (α): controls the search range and affects the convergence speed and global search capability of the algorithm.

[0086] Individual optimal fitness value (pbest): Records the best hyperparameters found by each individual during the search process.

[0087] Global optimal fitness value (gbest): Records the optimal combination of hyperparameters found in the entire population.

[0088] Number of iterations: For example, 50 iterations, to ensure that the algorithm converges to a better solution.

[0089] In practical applications, the improved quantum particle swarm optimization algorithm is used to perform a global search within the search space. Compared with the traditional particle swarm optimization algorithm, IQPSO optimizes the global search capability by introducing a contraction-expansion factor and enhancing information sharing and cooperation among particles, thereby effectively reducing the risk of getting trapped in local optima and improving the algorithm's ability to escape local optima.

[0090] In IQPSO, the purpose of global search is to find combinations of hyperparameters for the model, that is, to improve the model's performance by optimizing hyperparameters (such as the weights and learning rate of a BiGRU network).

[0091] The algorithm uses a global search to find the optimal combination of hyperparameters, which will then be used to train the BiGRU model. In other words, the global search result within the search space represents the best combination of hyperparameters.

[0092] The search results after the global search are the hyperparameter combinations corresponding to the optimal particle gbest, obtained after 50 iterations of the IQPSO algorithm. These hyperparameters enable the BiGRU model to achieve the lowest mean squared error of the loss function during training. In short, the final search results are the optimized hyperparameter configurations, which will be applied to the training of the BiGRU model to improve its performance.

[0093] The data processing procedure of IQPSO specifically includes: First, initializing the population with a population size of N=30 and setting an initial contraction-expansion factor α=1.0, evaluating fitness, and training a BiGRU algorithm incorporating a convolutional block attention module. The mean squared error of the loss function is calculated based on the training results and used as the fitness value f(Xi). The globally optimal fitness value gbest and the individual optimal fitness value pbest are recorded, iterated 50 times, and the hyperparameter combination corresponding to the globally optimal fitness value gbest (i.e., the globally optimal particle) is returned.

[0094] Compared with the traditional Quantum Particle Swarm Optimization Algorithm (QPSO), the IQPSO provided in this application has the following main improvements:

[0095] 1) Introduction of the contraction-expansion factor:

[0096] In IQPSO, the contraction-expansion factor (α) is used to adjust the search range of particles. In traditional QPSO, particle updates can easily get trapped in local optima, while by adjusting the value of α, IQPSO can flexibly control the search range, thereby improving the algorithm's ability to escape local optima. α = 1.0 is usually used as the initial value, but during the algorithm iteration process, α may be dynamically adjusted to adapt to different stages of the search space, ensuring the efficiency of the global search.

[0097] 2) Particle swarm cooperation and information sharing:

[0098] IQPSO further enhances cooperation and information sharing among particles, enabling them to communicate and influence each other better, thereby improving search efficiency and reducing the risk of getting trapped in local optima. Particles not only rely on their own historical best position (i.e., pbest) and global best position (i.e., gbest), but the mutual influence between particles also helps to explore the search space more broadly.

[0099] 3) Improved update mechanism:

[0100] In traditional QPSO, the particle velocity update method is relatively simple, usually relying on the particle's current position and historical information. IQPSO, however, introduces a new update mechanism that allows particles to jump within a wider search space and avoids premature convergence through certain strategies.

[0101] The IQPSO provided in this application enables particles to perform global searches more effectively by introducing a contraction-expansion factor and enhancing cooperation and information sharing among particles, thus avoiding getting trapped in local optima.

[0102] This application integrates a bidirectional gated recurrent network model with a convolutional block attention module, which facilitates more effective capture of key features in the data and enhances the model's ability to focus on important information. Furthermore, hyperparameter optimization is a crucial step in improving model performance. To find better parameter combinations in the complex hyperparameter space, this application employs an improved quantum particle swarm optimization algorithm. This algorithm, by improving the particle swarm optimization strategy, can more efficiently explore the hyperparameter space and optimize the model's learning process, thereby constructing a more accurate and robust prediction model, namely the bidirectional gated recurrent network model (IQPSO-CBAM-BiGRU) that integrates the convolutional block attention module and the improved quantum particle swarm optimization algorithm.

[0103] In practical applications, this application trains and tests the BiGRU in IQPSO-CBAM-BiGRU, specifically including:

[0104] 1) Train the BiGRU model using the training set.

[0105] 2) Use the trained BiGRU model to predict blood glucose values ​​in the test set.

[0106] Figure 3 The results show the blood glucose prediction results for different output step sizes when the input duration Δt is 30 minutes. The results indicate that when the prediction duration is 3 minutes, the predicted curve almost perfectly matches the actual blood glucose value curve, demonstrating excellent prediction performance. When the prediction duration is 9 minutes, although the predicted curve shows slight fluctuations, it still predicts blood glucose changes relatively well. However, when the prediction duration is 1 hour, the predicted curve differs significantly from the actual blood glucose value curve, the model error increases, and it can only predict a general trend.

[0107] 3) Calculate the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) between the predicted and actual blood glucose values ​​in the test set. 2 As an evaluation metric for the model:

[0108]

[0109] Where, x i and These are the actual value and the predicted value of the i-th data point, respectively. , where n is the mean of the original data and n is the sample size. This is used to measure the model's predictive performance on non-invasive blood glucose data. The final results for the IQPSO-CBAM-BiGRU model are RMSE = 0.268, MAE = 0.188, and R0 = 0. 2 =0.963.

[0110] In another exemplary embodiment, such as Figure 4 As shown, another microwave non-invasive blood glucose prediction method provided in this application specifically includes:

[0111] Step 1: Data preprocessing, which includes data cleaning, data normalization, and building a sliding window.

[0112] Step 2: Initialize model parameters, specifically including: constructing a BiGRU model that incorporates CBAM.

[0113] Step 3: Hyperparameter optimization, which includes: initializing the IQPSO algorithm, determining the individual particle and global optimal solution, updating the particle position, and determining whether the termination condition is met. If yes, proceed to step 4; otherwise, return to "determine the individual particle and global optimal solution, update the particle position".

[0114] Step 4: Input the global optimal result into the BiGRU model, then input the training set samples into the BiGRU model for training, and determine whether the test set error meets the requirements. If yes, save IQPSO-CBAM-BiGRU to predict blood glucose values; otherwise, return to "Determine individual particles and global optimal solutions, and update particle positions".

[0115] The IQPSO-CBAM-BiGRU provided in this application uses sliding window technology to segment sequence data, enabling flexible adjustment of input and output step sizes.

[0116] The IQPSO-CBAM-BiGRU provided in this application introduces a convolutional block attention module, which more effectively captures dependencies in sequence data by dynamically adjusting the weights of each feature in the channel and time dimensions.

[0117] The IQPSO-CBAM-BiGRU provided in this application uses an improved quantum particle swarm optimization algorithm. By improving the particle swarm optimization strategy, it can explore the hyperparameter space more efficiently, optimize the model's learning process, and thus build a more accurate and robust prediction model.

[0118] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores microwave non-invasive blood glucose prediction data. The input / output interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a microwave non-invasive blood glucose prediction method.

[0119] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0120] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the methods described above.

[0121] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods described above.

[0122] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0123] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0124] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0125] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A microwave non-invasive blood glucose prediction method, characterized in that, The microwave non-invasive blood glucose prediction method includes: A dataset was constructed based on the blood glucose time series collected by the microwave non-invasive blood glucose sensing system; Based on the convolutional block attention module and the improved quantum particle swarm optimization algorithm, an optimal bidirectional gated recurrent network model is constructed according to the dataset; the optimal bidirectional gated recurrent network model is a bidirectional gated recurrent network model that integrates the convolutional block attention module and the improved quantum particle swarm optimization algorithm. Predict blood glucose levels in the tested individuals based on the optimal bidirectional gated recurrent network model.

2. The microwave non-invasive blood glucose prediction method according to claim 1, characterized in that, A dataset was constructed based on the blood glucose time series collected by the microwave non-invasive blood glucose sensing system, specifically including: A microwave non-invasive blood glucose sensing system is used to collect electrical signals returned from the skin of volunteers at set time intervals within a set time period, and the electrical signals are converted into blood glucose values. Generate a blood glucose time series based on all blood glucose values ​​within a set time period; The blood glucose time series is used as a dataset, and the dataset is divided into a training set, a validation set, and a test set according to the time order.

3. The microwave non-invasive blood glucose prediction method according to claim 1, characterized in that, A dataset is constructed based on the blood glucose time series collected by the microwave non-invasive blood glucose sensing system, and the following steps are also included: The dataset was cleaned using spline interpolation to determine the cleaned dataset. The cleaned dataset is normalized using the min-max normalization method to determine the normalized dataset. A sliding window is constructed, and a direct strategy method is used to transform the normalized dataset to determine the processed dataset.

4. The microwave non-invasive blood glucose prediction method according to claim 3, characterized in that, Based on the convolutional block attention module and an improved quantum particle swarm optimization algorithm, an optimal bidirectional gated recurrent network model is constructed according to the dataset, specifically including: A bidirectional gated recurrent network model is used as the basic prediction model. During the training process of the bidirectional gated recurrent network model, a convolutional block attention module is introduced to capture key features in the processed dataset. An improved quantum particle swarm optimization algorithm is then used to perform a global search in the search space to determine the optimal hyperparameters. The key features include forward and backward information of the blood glucose time series. The improved quantum particle swarm optimization algorithm introduces a contraction-expansion factor to adjust the search range of the particles. Based on the optimal hyperparameters, the bidirectional gated recurrent network model is constructed, and the optimal bidirectional gated recurrent network model is determined.

5. The microwave non-invasive blood glucose prediction method according to claim 4, characterized in that, A bidirectional gated recurrent neural network (BRNN) model is used as the basic prediction model. During the training process of the BRNN model, a convolutional block attention module is introduced to capture key features in the processed dataset, specifically including: The convolutional block attention module includes a channel attention module and a spatial attention module connected in sequence. The blood glucose time series is converted into a raw feature map; the raw feature map includes feature information from multiple channels. Based on the bidirectional gated recurrent network model, the original feature map is input into the channel attention module, and a one-dimensional channel attention map is output. The weighted feature map is determined by multiplying the one-dimensional channel attention map and the original feature map channel by channel; The weighted feature map is input into the spatial attention module, and a two-dimensional spatial attention map is output. The dot product of the two-dimensional spatial attention map and the original feature map is used to determine the feature map after the dot product.

6. The microwave non-invasive blood glucose prediction method according to claim 5, characterized in that, The original feature map is input into the channel attention module, and a one-dimensional channel attention map is output, specifically including: The channel attention module includes a max pooling layer, an average pooling layer, and a multilayer perceptron. The original feature map is input into the max pooling layer and the average pooling layer respectively, and the max pooling feature map and the average pooling feature map are output. The max pooling feature map and the average pooling feature map are input into the multilayer perceptron to determine two different channel attention weights; Using the Sigmoid activation function, a one-dimensional channel attention map is determined based on two different channel attention weights.

7. The microwave non-invasive blood glucose prediction method according to claim 4, characterized in that, An improved quantum particle swarm optimization algorithm is used to perform a global search within the search space to determine the optimal hyperparameters, specifically including: Initialize the population based on the initial hyperparameters of the bidirectional gated recurrent network model; The fitness values ​​of individuals in the population were assessed using the contraction-expansion factor. Based on the fitness value, train the bidirectional gated recurrent network model and determine the training result; The mean squared error of the loss function is calculated based on the training results, and the mean squared error of the loss function is used as the fitness value. The bidirectional gated recurrent network model is trained iteratively, recording all optimal fitness values ​​and individual optimal fitness values, and the hyperparameters corresponding to the global optimal fitness value are taken as the best hyperparameters.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the microwave non-invasive blood glucose prediction method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the microwave non-invasive blood glucose prediction method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the microwave non-invasive blood glucose prediction method according to any one of claims 1-6.

Images