Blood glucose health dynamic regulation method and system based on AI algorithm

Abstract

The embodiment of the application discloses a blood glucose health dynamic regulation method and system based on an AI algorithm. The method comprises the following steps: acquiring first blood glucose data of a monitoring target in a current time window; acquiring at least one first feature map of blood glucose fluctuation of the monitoring target according to the first blood glucose data and second blood glucose data of at least one historical time window; generating a second feature map according to the first feature map and related information; obtaining a second prediction probability group according to the second feature map and a first prediction probability group of the at least one historical time window; the second prediction probability group comprises probabilities corresponding to different risk levels and confidence degrees corresponding to the probabilities; determining a risk level of the monitoring target in a next time window according to the second prediction probability group; the risk level comprises a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia; and storing the second prediction probability group into a first queue.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health, in particular to a blood glucose health dynamic regulation method and system based on an AI algorithm. BACKGROUND

[0002] With the increasing emphasis on health management, especially the continuous expansion of the target group of diabetes monitoring, accurate monitoring and risk prediction of blood glucose are increasingly critical. Blood glucose fluctuations not only reflect the current health status of the human body, but are also closely related to long-term diabetes complications. For example, hypoglycemia can cause dizziness, syncope, and even endanger life; hyperglycemia can gradually damage blood vessels, nerves, and other body tissues. Therefore, real-time and accurate prediction of blood glucose fluctuation risk can provide a basis for timely adjustment of diet, exercise, and treatment regimen for monitoring targets, which has great clinical and life guidance significance.

[0003] Currently, blood glucose monitoring mainly relies on regular fingertip blood testing or continuous dynamic blood glucose monitoring devices. Traditional fingertip blood testing can only obtain blood glucose values at a certain moment, and blood glucose value testing is not comprehensive, and has limited auxiliary effect on guiding the diet of users and the diagnosis of blood glucose-related diseases. SUMMARY

[0004] Therefore, the embodiments of the present application provide a blood glucose health dynamic regulation method and system based on an AI algorithm. The technical solution of the present application is as follows:

[0005] The first aspect provides a blood glucose health dynamic regulation method based on an AI algorithm, comprising:

[0006] obtaining first blood glucose data of a current time window of a monitoring target;

[0007] obtaining at least one first feature map of blood glucose fluctuation of the monitoring target according to the first blood glucose data and second blood glucose data of at least one historical time window, wherein each pixel point in the first feature map includes a first feature vector; one first feature vector corresponds to a time period; the first feature vector includes time information, scene information, and at least one blood glucose measurement value arranged in sequence; the scene information at least includes spatial information;

[0008] obtaining relevant information related to blood glucose monitoring of the monitoring target; the relevant information includes physiological data and dietary data;

[0009] generating a second feature map according to the first feature map and the relevant information;

[0010] obtaining a second prediction probability group according to the second feature map and a first prediction probability group of at least one historical time window; the second prediction probability group includes probabilities corresponding to different risk levels and confidence degrees corresponding to the probabilities;

[0011] determining, according to the second prediction probability group, a risk level of the monitoring target in a next time window; the risk level comprises a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia;

[0012] storing the second prediction probability group into a first queue; the first queue stores risk levels of multiple time windows; the first queue is used for risk prompt and / or diet prompt.

[0013] The second aspect provides an AI algorithm-based blood glucose health dynamic regulation system, comprising:

[0014] a first acquisition module, configured to acquire first blood glucose data of a monitoring target in a current time window;

[0015] a second acquisition module, configured to acquire at least one first feature map of blood glucose fluctuation of the monitoring target according to the first blood glucose data and second blood glucose data of at least one historical time window, wherein each pixel point in the first feature map comprises a first feature vector; one first feature vector corresponds to one time period; the first feature vector comprises time information, scene information and at least one blood glucose measurement value arranged in sequence; the scene information at least comprises spatial information;

[0016] a third acquisition module, configured to acquire relevant information related to blood glucose monitoring of the monitoring target; the relevant information comprises physiological data and diet data;

[0017] a generation module, configured to generate a second feature map according to the first feature map and the relevant information;

[0018] a fourth acquisition module, configured to obtain a second prediction probability group according to the second feature map and a first prediction probability group of at least one historical time window; the second prediction probability group comprises probabilities corresponding to different risk levels and confidence degrees corresponding to the probabilities;

[0019] a determination module, configured to determine a risk level of the monitoring target in a next time window according to the second prediction probability group; the risk level comprises a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia;

[0020] a storage module, configured to store the second prediction probability group into a first queue; the first queue stores risk levels of multiple time windows; the first queue is used for risk prompt and / or diet prompt.

[0021] Based on the above scheme, the method fuses current and historical blood glucose, scene, physiological and dietary data to build a personalized model; dynamic iterative prediction (including probability and confidence) is used to improve the reliability and accurately classify high and low sugar risks; historical risk sequences are stored to assist real-time early warning and long-term management, improving the comprehensiveness of blood glucose analysis, prediction reliability and intervention accuracy, and effectively avoiding blood glucose risks. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions serve to explain the present application, and do not constitute improper limitations on the present application. In the drawings:

[0023] Figure 1 A flowchart of an AI algorithm-based blood glucose health dynamic regulation method provided for an embodiment of the present application is shown in FIG. 1.

[0024] Figure 2 A flowchart of another AI algorithm-based blood glucose health dynamic regulation method provided for an embodiment of the present application is shown in FIG. 2.

[0025] Figure 3 A structural diagram of an AI algorithm-based blood glucose health dynamic regulation system provided for an embodiment of the present application is shown in FIG. 3.

[0026] Figure 4 A structural diagram of an electronic device provided for an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0027] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", and the like in the description and in the claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0029] As Figure 1 indicated, the present disclosure provides an AI algorithm-based blood glucose health dynamic regulation method, comprising:

[0030] S1101: acquiring first blood glucose data of a current time window of a monitoring target;

[0031] S1102: acquiring at least one first feature map of blood glucose fluctuation of the monitoring target according to the first blood glucose data and second blood glucose data of at least one historical time window, wherein each pixel point in the first feature map includes a first feature vector; one first feature vector corresponds to a time period; the first feature vector includes time information, scene information and at least one blood glucose measurement value arranged in sequence; the scene information at least includes spatial information;

[0032] S1103: acquiring relevant information related to blood glucose monitoring of the monitoring target; the relevant information includes physiological data and dietary data;

[0033] S1104: generating a second feature map according to the first feature map and the relevant information;

[0034] S1105: obtaining a second prediction probability group according to the second feature map and a first prediction probability group of at least one historical time window; the second prediction probability group includes probabilities corresponding to different risk levels and confidence degrees corresponding to the probabilities;

[0035] S1106: determining a risk level of the monitoring target in a next time window according to the second prediction probability group; the risk level includes a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia;

[0036] S1107: storing the second prediction probability group into a first queue; the first queue stores risk levels of multiple time windows; the first queue is used for risk prompt and / or dietary prompt.

[0037] The AI algorithm-based blood glucose health dynamic regulation method can be used in various electronic devices. The electronic device can be connected to a big data service platform such as a health monitoring platform, and information is interacted with the server of the platform.

[0038] Exemplarily, the AI algorithm-based blood glucose health dynamic regulation method can be used in one or more electronic devices. Exemplarily, the method can be used in a mobile device, which can be a handheld device and / or a wearable device written for the monitoring target, and can specifically include but is not limited to a mobile phone, a bracelet, a foot ring, and the like. In other embodiments, the AI algorithm-based blood glucose health dynamic regulation method can also be used in a medical auxiliary device. Exemplarily, the medical auxiliary device can be a health status monitoring device for the monitoring target in a hospital or a clinic. The holder of the electronic device can be the monitoring target himself / herself or a friend or guardian of the monitoring target, and the like.

[0039] The electronic device runs an application or a mini-program for health monitoring. In some embodiments, the application or the mini-program can monitor various health indicators of the health target. For example, while monitoring the blood glucose of the monitoring target, the blood lipid of the monitoring target is also monitored.

[0040] In some embodiments, the monitoring target can be a monitoring target or a potential monitoring target with various diseases caused by abnormal blood glucose. Of course, in the specific implementation process, the monitoring target can be any object, not limited to a monitoring target or a potential monitoring target with abnormal blood lipid, and can also be a healthy person who is concerned about his / her own blood glucose status.

[0041] In some embodiments, the monitoring target includes a person but is not limited to a person, and can even be a pet or an animal in a zoo, and the like.

[0042] In some embodiments, the current time window can be a time window for this time blood glucose fluctuation. Exemplarily, the first blood glucose data measured by periodically acquiring blood glucose can be obtained, and at this time, the current time window is the current period. In some embodiments, the time window can be determined according to the eating and / or medication of the monitoring target. For example, the closer the current time window is to the eating or medication time of the monitoring target, the shorter the duration of the current time window, and otherwise, the longer the duration.

[0043] In some embodiments, the first blood glucose data can be any data reflecting the blood glucose status of the monitoring target. Exemplarily, the blood glucose data of the monitoring target is collected in real time, including but not limited to continuous glucose monitoring (CGM) data.

[0044] In some embodiments, the at least one historical time window can be any one or more historical time windows before the current time window. Illustratively, the at least one historical time window includes at least one historical time window before the current time window. Further illustratively, the at least one historical time window can be before the current time window. In some embodiments, the blood glucose values of different positions are collected at multiple time points as needed within the current time window. In this way, the first blood glucose data can be an array or a matrix of blood glucose values.

[0045] In some embodiments, the first feature map can be a two-dimensional feature map or a three-dimensional feature map. If it is a multi-dimensional feature map, for example, a three-dimensional feature map or a four-dimensional feature map. Each pixel point in the first feature map includes a first feature vector; one of the first feature vectors corresponds to a time period; the first feature vector arranges the time information, the scene information, and at least one blood glucose measurement value in order; and the scene information includes at least spatial information. In some embodiments, the spatial information can be specific position information or spatial scene information, such as a restaurant, etc.

[0046] In order to accurately measure the blood glucose fluctuation of the monitoring target, the second blood glucose data measured in the historical time window is combined to construct the first feature map, so that when the artificial intelligence (AI) model is used for risk prediction, the second blood glucose data measured in the historical time window and the first blood glucose data in the current time window can be combined to obtain a first feature map with both time and space characteristics.

[0047] In some embodiments, the related information includes physiological data and dietary data.

[0048] In some embodiments, the physiological data includes at least one of the following:

[0049] Insulin level: Insulin is secreted by pancreatic beta cells and can promote cells to uptake, utilize, and store glucose, thereby lowering blood glucose. When insulin secretion is insufficient or defective, blood glucose is difficult to be normally regulated, which can lead to elevated blood glucose, such as in type 1 diabetes, in which the monitoring target has absolute insufficient insulin secretion and needs to rely on exogenous insulin to control blood glucose.

[0050] Glycated hemoglobin (HbA1c): It reflects the average blood glucose level in the past 2-3 months. The normal reference range is generally 4%-6%. The higher the HbA1c level, the less ideal the long-term control of blood glucose, and the higher the risk of diabetic complications. For example, for every 1% increase in HbA1c, the risk of diabetic microvascular complications increases by about 40%.

[0051] Blood pressure: Long-term high blood pressure can damage the endothelial cells of blood vessels, leading to increased insulin resistance, affecting the action of insulin, causing abnormal glucose regulation, and thus high blood sugar. At the same time, high blood sugar can also cause damage to blood vessels and kidneys, etc., aggravating high blood pressure, and the two influence each other, forming a vicious cycle.

[0052] Body mass index (BMI): BMI = weight (kg) ÷ height ² (m 2 ). The normal range is generally 18.5 -23.9 kg / m 2. Overweight or obesity (BMI≥24 kg / m 2 ) is an important risk factor for insulin resistance, which can reduce the sensitivity of body cells to insulin, leading to elevated blood glucose that cannot be effectively utilized.

[0053] In some embodiments, the dietary data includes, but is not limited to, at least one of the following:

[0054] Carbohydrate intake: Carbohydrates are the main factor affecting blood sugar. Different types of carbohydrates have different effects on blood sugar. For example, refined grains and added sugars are quickly digested and absorbed, causing rapid blood sugar rise; while whole grains, legumes and other complex carbohydrates rich in dietary fiber are relatively slow in digestion and absorption, causing a relatively slow rise in blood sugar. For example, after eating a bowl of white rice, blood sugar will rise rapidly in the short term, but if you replace white rice with brown rice, the blood sugar rise will be relatively small.

[0055] Dietary fiber intake: Dietary fiber can delay the digestion and absorption of carbohydrates, reducing the glycemic index of food. Increasing dietary fiber intake helps control blood sugar fluctuations and keep blood sugar relatively stable. It is generally recommended that adults consume 25-35 grams of dietary fiber per day. Eating more foods rich in dietary fiber such as vegetables, fruits, whole grains, legumes, etc. can reduce large fluctuations in blood sugar.

[0056] Glycemic index (GI) of food: It is an index to measure the speed and amplitude of blood sugar rise caused by food. High GI foods (GI≥70) such as watermelon, white bread, etc. are quickly digested and absorbed after entering the intestine, causing rapid blood sugar rise; low GI foods (GI≤55) such as oats, legumes, etc. are slowly digested and absorbed in the intestine, causing a slow rise in blood sugar. Understanding the GI value of food helps to choose food reasonably and control blood sugar.

[0057] Meal time and frequency: Regular meal time and reasonable meal frequency are important for blood glucose stability. Regular and quantitative meal can make the body's insulin secretion regular, which is conducive to the control of blood glucose. For example, breakfast at 7-8 o'clock, lunch at 12-13 o'clock, dinner at 18-19 o'clock, and appropriate snacks, such as fruits and nuts at 10 o'clock in the morning and 4 o'clock in the afternoon, can help maintain stable blood glucose and avoid blood glucose fluctuations caused by long-term fasting or overeating.

[0058] In some embodiments, acquiring physiological data can include connecting the interface between the application and the hospital platform or the physical examination center to read the physiological data.

[0059] In some other embodiments, the physiological data received from the UI interface is user input.

[0060] In some embodiments, the dietary information can be obtained from various sensors or by connecting various applications or mini-programs.

[0061] In some embodiments, the physiological data can be quasi-static data and can be pre-stored.

[0062] In some embodiments, a second feature map is generated according to the first feature map and the related information. For example, the related information is numerically valued and combined with the first feature map to form the second feature map according to the components of the second feature map.

[0063] In some embodiments, a second prediction probability group is obtained according to the second feature map and the first prediction probability group of at least one historical time window.

[0064] In some embodiments, the second prediction probability group includes probabilities corresponding to different risk levels and confidence degrees corresponding to the probabilities.

[0065] In some embodiments, the risk level includes a first type of value and a second type of value; the first type of value is related to hypoglycemia; and the second type of value is related to hyperglycemia. The hyperglycemia and hypoglycemia herein can be determined according to medical standards. In some embodiments, the hyperglycemia and hypoglycemia can be determined by comparing the blood glucose value with a threshold value.

[0066] In some embodiments, the second prediction probability group is stored in a first queue; and the first queue stores risk levels of multiple time windows. For example, the first queue can be a variety of types of queues. For example, the first queue can be a first-in-first-out queue. In some embodiments, the first queue is used for risk prompt and / or dietary prompt.

[0067] In some embodiments, the risk level can include at least one of the following:

[0068] a safety level, the level indicating a health of the monitoring target;

[0069] a first risk level, the level corresponding to a sub-health of the blood glucose status;

[0070] a second risk level, the level corresponding to a health of the blood glucose status;

[0071] a third risk level, the level corresponding to an unhealth of the blood glucose status.

[0072] In some embodiments, the risk prompt can be output when the blood glucose of the monitoring target is abnormal (e.g., too high or too low). In some embodiments, the risk prompt is used to indicate the risk level of the blood glucose.

[0073] In some embodiments, the diet prompt is used to prompt the monitoring target whether to perform a corresponding diet.

[0074] In some embodiments, for a blood glucose patient, the diet prompt can also be used to prompt the monitoring target to take medicine.

[0075] Through the above dynamic blood glucose precise health, the monitoring target can be well assisted to control the blood glucose, and accurate auxiliary data is provided for subsequent diagnosis of blood glucose related diseases of the monitoring target by the doctor.

[0076] In some embodiments, the second prediction probability group can include probability values corresponding to the risk level, and also include probability values corresponding to a blood glucose change trend. The probability values corresponding to the change trend include probability values of an upward trend and probability values of a downward trend. Further, the probability values corresponding to the change trend also have a confidence degree. At this time, the diet prompt can be given based on the change trend probability values in the second prediction probability group. For example, for a monitoring target with a high blood glucose symptom, a medicine taking prompt is given according to the probability value of the upward change trend. For a monitoring target with low blood glucose, a meal taking prompt is given according to the probability value of the downward change trend. That is, in some embodiments, the diet prompt is given according to the second probability value and information such as the blood glucose history of the monitoring target.

[0077] In some embodiments, the first information of the monitoring target is obtained, including at least one of:

[0078] a third feature map of a previous time point is obtained; the third feature map is a feature map of a previous historical time window;

[0079] a fourth feature map is obtained from the third feature map according to a time variation rule of blood glucose data in the third feature map, by using a clipping window; a size of the fourth feature map is related to the clipping window and is smaller than a size of the third feature map;

[0080] According to the time variation law, the fourth feature map is filled according to the first blood glucose data, to obtain the first feature map; the size of the first feature map is the same as that of the third feature map.

[0081] In this way, the first feature map can be quickly constructed, and the feature map includes blood glucose data of one or more historical time windows, thereby reducing the generation overhead of the feature map, improving the efficiency of the first feature map, and improving the generation efficiency of the promotion information and the diagnosis information used for assistance.

[0082] To solve the problem that the traditional blood glucose fluctuation prediction method is difficult to fully utilize the spatial and temporal characteristics of blood glucose data. The previous method may only simply consider the blood glucose data at the current time point or within a short time, and cannot effectively capture the complex correlation between data. The embodiments of the present disclosure fully mine the characteristics of blood glucose data through specific models and processing methods, and improve the prediction effect.

[0083] The problem of inaccurate prediction under complex blood glucose fluctuation is solved. When the blood glucose fluctuation is relatively complex, it is difficult to accurately predict only by using the data in the current time window. The embodiments of the present disclosure can more accurately predict the blood glucose fluctuation by combining the historical time window data and performing various fusion processing, thereby providing a more reliable basis for blood glucose management of the diabetes monitoring target.

[0084] In some embodiments, as shown in FIG. 11B, S1105 can include: Figure 2

[0085] S11051: processing the second feature map based on a convolutional spiking neural network model based on an attention mechanism to obtain a spatial correlation feature;

[0086] S11052: processing the second feature map based on an iTransformer model to obtain a first time feature;

[0087] S11053: if the first time feature does not meet a first condition, reading a second time feature of N historical windows;

[0088] S11054: fusing the first time feature and the second time feature in a first manner to obtain a third time feature; the third time feature includes a blood glucose fluctuation feature indicating a time sequence;

[0089] S11055: fusing the first time feature and the second time feature in a second manner to obtain a fourth time feature; the fourth time feature includes a blood glucose fluctuation feature describing different time windows;

[0090] S11056: obtaining a second prediction probability group according to the third time feature, the fourth time feature, and a first prediction probability group of at least one historical time window. ​

[0091] For example, a convolutional spiking neural network model based on an attention mechanism processes the second feature map to obtain spatial correlation features:

[0092] A convolutional spiking neural network (CSNN) model with an attention mechanism (e.g., setting different convolutional kernels to extract feature information at different scales) is used to process the second feature map. During processing, the attention mechanism automatically focuses on regions in the blood glucose data that show significant changes and are potentially more critical for predicting blood glucose fluctuations (such as data regions during rapid rises or falls in blood glucose). After processing by this model, spatial correlation features are obtained, which describe the correlation between blood glucose data at different time points and different feature dimensions, such as the correlation between blood glucose values ​​at adjacent time points and the correlation of blood glucose changes under different physical states (e.g., after exercise, after eating). The CSNN model based on the attention mechanism processes the second feature map to obtain spatial correlation features: the attention mechanism can focus on key information, and the CSNN can effectively extract the spatial features of blood glucose data. The combination of the two allows the model to better capture the correlation of blood glucose data at different dimensions and time points, providing more valuable feature information for subsequent predictions.

[0093] For example, the first-time features are obtained by processing the second feature map based on the iTransformer model:

[0094] The second feature map is input into the iTransformer model. The iTransformer model utilizes its unique multi-head self-attention mechanism and feedforward neural network structure to model blood glucose data in the time dimension. For example, the model captures the dependencies between blood glucose data at different time intervals, such as the relationship between blood glucose values ​​at specific time intervals like one hour or two hours apart. After processing, the first temporal feature is obtained, which reflects the temporal characteristics of the blood glucose data within the current time window.

[0095] The first temporal feature is obtained by processing the second feature map based on the iTransformer model: The iTransformer model utilizes a self-attention mechanism and a feedforward neural network to fully explore the dependencies of blood glucose data in the time dimension and obtain the temporal characteristics within the current time window, which is an important foundation for accurately predicting blood glucose fluctuations.

[0096] If the first time feature does not meet the first condition, read the second time feature of N historical windows:

[0097] The first condition is set as follows: if the standard deviation of blood glucose fluctuation in the first time feature is less than a certain threshold (e.g., 0.5 mmol / L), then blood glucose fluctuation within the current time window is considered relatively stable. If the first time feature does not meet this condition (i.e., the standard deviation of blood glucose fluctuation is greater than 0.5 mmol / L), then the second time features of N historical time windows (assuming N = 3, i.e., the first three time windows are read) are read. These second time features of historical time windows contain blood glucose data information from previous time periods. By judging whether the first time feature meets the first condition, unnecessary data processing can be reduced. For example, the first time feature could be whether the blood glucose limit value within the corresponding time range is outside the specified range. For example, hypoglycemia below the lower limit of blood glucose and hyperglycemia above the upper limit of blood glucose can be considered as not meeting the first condition. Another example is if the overall blood glucose fluctuation reflected by the first time feature exceeds the reasonable range of fluctuation, which can also be considered as not meeting the first condition. Yet another example is if the rate of increase or decrease of blood glucose within the first time feature exceeds the reasonable range of fluctuation rate, which can also be considered as not meeting the first condition. For example, based on the variance of blood glucose fluctuations corresponding to the first-time feature, it can be determined whether the first condition is met. For instance, if the fluctuation variance is less than a reasonable range associated with the target object, it can be considered to meet the first condition; otherwise, it can be considered not to meet the first condition. Of course, the above are just examples; in specific implementations, other situations can be considered to meet the first condition.

[0098] The first and second methods are different feature fusion methods, and different fusion methods can yield features that are manifested at different levels.

[0099] The third time feature is obtained by fusing the first and second time features using the first method: This method employs a fusion approach that combines features after concatenation and then performs time-series convolution. First, the first and second time features are concatenated along their feature dimensions, and then processed through a one-dimensional convolutional layer (time-series convolution). In this way, the third time feature can capture the temporal sequence of blood glucose fluctuations at different time scales. For example, it can indicate whether the current blood glucose fluctuation is a continuation or a turning point of the previous trend, highlighting the temporal sequence-related characteristics of blood glucose fluctuations.

[0100] A fourth time feature is obtained by fusing the first and second time features using a second method: the second method employs a weighted fusion followed by feature transformation. Based on the importance of the first and second time features in describing blood glucose fluctuations, weights are assigned to them (e.g., weights are dynamically generated using an attention module), and then a weighted sum is performed. The weighted result is then transformed using a fully connected layer. This fourth time feature, in this way, can comprehensively describe the blood glucose fluctuation characteristics of different time windows, such as differences in the amplitude and frequency of blood glucose fluctuations within different time windows.

[0101] Two fusion methods yield the third and fourth temporal features: the second method, weighted summation fusion, can integrate the importance of current and historical features, reflecting the temporal order-related blood glucose fluctuation features; the first method, feature splicing fusion, can compare blood glucose fluctuation features across different time windows, enriching feature representation from different perspectives. It is worth noting that in practical implementation, the first and second methods can correspond to different approaches, not limited to direct splicing and weighted fusion.

[0102] Based on the third and fourth time features and the first predicted probability set from at least one historical time window, a second predicted probability set is obtained. This is achieved by inputting the third and fourth time features, along with the first predicted probability set from previous historical time windows (such as the probabilities of blood glucose rising, falling, or remaining stable predicted for each of the first three historical time windows), into a multilayer perceptron model. After training, the multilayer perceptron model outputs the second predicted probability set based on these input features and probability information; this second set represents the predicted probability of blood glucose fluctuations (rising, falling, or remaining stable) at the current time point.

[0103] Based on the first prediction probability group using the third time feature, the fourth time feature, and at least one historical time window, the second prediction probability group is obtained:

[0104] The third and fourth time features, along with the first set of predicted probabilities obtained from previous historical time windows (e.g., the probabilities of blood glucose rising, falling, or remaining stable predicted for each of the previous three historical time windows), are input into a multilayer perceptron model. After training, the multilayer perceptron model outputs a second set of predicted probabilities based on these input features and probability information; that is, the predicted probability of blood glucose fluctuations (rising, falling, or remaining stable) at the current time point.

[0105] Determine if the first time feature meets the first condition and read the historical time window features: Judge the current time feature status by setting conditions. If the stability condition is not met, read the historical features. This can combine more historical information to assist in the prediction of current blood glucose fluctuations and improve the prediction accuracy.

[0106] A second prediction probability group is obtained based on multiple features and historical prediction probability groups: by integrating feature information from different perspectives and historical prediction probabilities into the multilayer perceptron, more reliable blood glucose fluctuation prediction probabilities can be obtained by making full use of existing information.

[0107] In summary, using this method to predict the second probability group improves the accuracy of blood glucose fluctuation prediction. By employing multiple models and fusion methods, the spatial and temporal characteristics of blood glucose data, as well as historical data information, are fully utilized, enabling predictions to more accurately reflect real blood glucose fluctuations. This enhances the model's adaptability to complex blood glucose fluctuations. For complex situations such as unstable blood glucose fluctuations, reading and fusing historical features allows for better handling and more effective predictions.

[0108] In some embodiments, the third time feature is used to indicate blood glucose fluctuation patterns exceeding a specified time range; and / or, the first time feature or the second time feature is used to indicate blood glucose fluctuation patterns less than or equal to the specified time range; and / or, the fourth time feature is used to indicate the range of blood glucose fluctuations at the same time point within different time windows.

[0109] In some embodiments, the third temporal feature is primarily designed to indicate patterns of blood glucose fluctuations over a specified time range. This means that this feature focuses on uncovering patterns of blood glucose changes over long time spans. For example, using days or weeks as time scales, it analyzes trends in blood glucose changes at the same time on different dates, or observes the periodic patterns of blood glucose fluctuations in monitored targets over a period of time. By fusing the first and second temporal features (combining historical time window information), more data can be integrated to capture patterns that are not easily noticeable in a short period, such as patterns of blood glucose fluctuations in certain monitored targets that may occur at specific times each month due to physiological cycles.

[0110] In other embodiments, a first time feature or a second time feature is primarily used to indicate blood glucose fluctuation patterns within a specified time range. The first time feature is extracted from the second feature map by the iTransformer model, reflecting the temporal characteristics of blood glucose data within the current time window; the second time feature is derived from historical time windows. These features focus on the immediate changes in blood glucose over a relatively short period (e.g., hourly), such as the rapid rise in blood glucose shortly after a meal or the short-term decline in blood glucose after exercise. This provides a basis for immediate assessment of blood glucose status.

[0111] In some embodiments, the fourth temporal feature focuses on indicating the range of blood glucose fluctuations at the same time point within different time windows. By comparing blood glucose data from different time windows (such as the same time on different days), the dispersion and range of blood glucose fluctuations at that time point can be clearly presented. For example, comparing blood glucose values ​​at 8 a.m. each day can help analyze whether the range of blood glucose fluctuations at that time point is gradually decreasing or increasing, which helps to assess the stability of blood glucose control.

[0112] When processing information such as second-order probability group predictions, the model can characterize fluctuations across long and short time scales and specific time points, comprehensively covering blood glucose fluctuation information. These rich features, as input, allow the prediction model to more accurately grasp blood glucose trends, reduce misjudgments, and improve prediction accuracy. For example, when predicting whether a monitored target's blood glucose will rise, fall, or remain stable at the next moment, these features can provide more accurate results. Doctors can use these features to gain a deeper understanding of the monitored target's blood glucose fluctuation characteristics. For monitored targets with complex blood glucose fluctuation patterns, diet and exercise plans can be developed based on long-term fluctuation patterns; medication dosages can be adjusted promptly based on short-term fluctuation patterns; and the effectiveness of treatment plans can be evaluated based on fluctuation ranges at specific time points, achieving personalized precision medicine. For the monitored target's own blood glucose management, understanding fluctuation patterns at different time scales can help them rationally arrange their daily activities. For example, knowing the short-term postprandial blood glucose fluctuation pattern allows for adjustments to dietary structure; understanding the long-term fluctuation pattern allows for early prevention of abnormal blood glucose fluctuations, improving self-management of blood glucose.

[0113] In some embodiments, in addition to blood glucose data, physiological data such as heart rate, blood pressure, and exercise intensity can be incorporated into the feature extraction process. For example, analyzing the correlation between heart rate changes and blood glucose fluctuations during exercise can be used as a new feature dimension to further enrich the description of physical condition and make the analysis of blood glucose fluctuation patterns more comprehensive and accurate. Environmental factors such as ambient temperature, humidity, and dietary type can also be included. For instance, studies have found that blood glucose fluctuations in some monitored targets are exacerbated when the ambient temperature is low. Combining these environmental factors with blood glucose features can more accurately explain and predict blood glucose fluctuations. When analyzing blood glucose data within different time windows, the DTW algorithm is used to calculate the similarity between different time features (e.g., third and fourth time features) and extract relevant features. This helps to more accurately compare blood glucose fluctuations within different time windows and uncover potential patterns, even when the timeframes are not perfectly aligned.

[0114] In some embodiments, the iTransformer model is a modified iTransformer model; the modified iTransformer model includes: a first branch, a second branch, and a connection module; the connection module connects the first branch; the first branch and the second branch have a first connection;

[0115] The first branch is used to extract the first time feature;

[0116] The connection module determines whether the first time feature meets the first condition based on the output of the first branch, and modifies the weight of the first connection according to whether the first time meets the first condition.

[0117] When the first connection has a first weight, the first connection is used to input the first time feature into the second branch;

[0118] When the first connection has a second weight, the first connection is not used to input the first time feature to the second branch.

[0119] In embodiments of this disclosure, a modified iTransformer model is used to process a second feature map derived from the blood glucose data of the monitored target to obtain a first temporal feature.

[0120] For example, the modified iTransformer model structure may include:

[0121] The first branch contains multiple attention-based sub-modules and convolutional layers. When processing the second feature map, the attention mechanism sub-module focuses on the parts of the second feature map that are closely related to the changes in blood glucose time series, such as the region of rapid rise and fall in blood glucose within a specific time period. The convolutional layers then perform further feature extraction and transformation on the focused features. Through these operations, the first branch extracts the first temporal feature from the second feature map that reflects the temporal characteristics of blood glucose data within the current time window. For example, if the current time window is set to 1 hour, the first branch can capture information such as the initial value, peak value, trough value, and overall trend of blood glucose within this 1 hour.

[0122] The second branch mainly consists of a recurrent neural network (RNN) structure. Its function is to perform further time series modeling and feature processing on the input features, in preparation for the subsequent reception and fusion processing of the first time features.

[0123] Connection Module: The connection module is connected to the first branch and acquires the first-time features output by the first branch in real time. Simultaneously, the connection module internally stores the judgment rules regarding the first condition (e.g., in the first-time features set above, a standard deviation of blood glucose fluctuation less than 0.5 mmol / L is considered to meet the first condition). When the first branch outputs the first-time features, the connection module calculates the standard deviation of blood glucose fluctuation based on the relevant data in the first-time features and compares it with the set threshold to determine whether the first condition is met.

[0124] Weight Adjustment and Connection Function: When the first time-varying feature meets the first condition, the connection module modifies the weight of the first connection to the first weight. At this time, the first connection passes the first time-varying feature extracted by the first branch to the second branch. Upon receiving the first time-varying feature, the second branch uses its own RNN structure to further process and fuse it, for example, by updating the hidden layer state to integrate the first time-varying feature, thereby further mining the potential information of the blood glucose data in the time dimension. When the first time-varying feature does not meet the first condition, the connection module modifies the weight of the first connection to the second weight. In this case, the first connection will not input the first time-varying feature into the second branch, and the second branch will continue to process other inputs (if any) based on its existing processing logic.

[0125] In some embodiments, when obtaining the second prediction probability set based on the second feature map and the first prediction probability set of at least one historical time window, the operation is performed as follows:

[0126] Generating the third and fourth time features: When the system successfully generates the third and fourth time features using the fusion methods described above (the first method yields the third time feature, and the second method yields the fourth time feature), it inputs the third time feature (indicating the blood glucose fluctuation pattern exceeding a specified time range) and the fourth time feature (indicating the blood glucose fluctuation range at the same time point within different time windows) into the prediction module. The prediction module can be a multilayer perceptron (MLP) model trained on a large amount of blood glucose data. The neurons within this MLP model perform weighted calculations and activation operations on the input third and fourth time features. Through feature transformation and information integration across multiple hidden layers, it ultimately outputs a second prediction probability set, used to predict the probability of blood glucose fluctuation at the current time point (elevation, decrease, stability, etc.).

[0127] In the absence of third and fourth time features: the prediction module will also process the first time feature, using its internal network structure to analyze and calculate the blood glucose fluctuation information within the current time window contained in the first time feature, and finally output the second prediction probability group.

[0128] In some cases, the third and fourth time features may need to be generated, i.e., they do not meet the first condition. However, there may be anomalies that prevent the third and fourth time features from being generated. Even so, the prediction of the second prediction probability group will still be made based on the first time feature.

[0129] In this embodiment, the first branch of the modified iTransformer model is specifically responsible for extracting first-time features. Through a combination of attention mechanisms and convolutional layers, it can accurately focus on and extract the temporal characteristics of blood glucose data, making the acquired first-time features more targeted and effective. The second branch provides the infrastructure for subsequent possible feature fusion and further processing. The two branches work together to improve the model's ability to process blood glucose data. The technical effect is to improve the accuracy and efficiency of feature extraction, laying a good foundation for subsequent accurate analysis of blood glucose fluctuation patterns.

[0130] The connection module can dynamically adjust the weights of the first connection based on whether the first time-time feature meets the first condition, thus flexibly controlling whether the first time-time feature is passed to the second branch. This dynamic control mechanism allows the model to adaptively adjust according to the actual fluctuations in blood glucose data. When blood glucose fluctuations are stable (meeting the first condition), the first time-time feature is passed to the second branch for further processing, which can uncover more potential information; when blood glucose fluctuations are unstable, the feature is not passed to avoid invalid or interfering information entering the second branch, ensuring the accuracy of model processing. The technical effect is to enhance the model's adaptive ability and improve the processing accuracy for different blood glucose fluctuation scenarios. Furthermore, different features are input to the prediction module depending on whether third and fourth time-time features are generated. This approach can fully utilize the effective feature information obtained by the system. When generating third and fourth time-time features, inputting these features that more comprehensively reflect the pattern of blood glucose fluctuations allows the prediction module to make predictions based on richer information; even when not generated, it ensures that predictions are made using the first time-time features, without wasting the acquired feature data. The technical effect is to improve the rationality and effectiveness of the input data to the prediction module, thereby improving the generation quality and prediction accuracy of the second prediction probability group, providing more reliable results for blood glucose fluctuation prediction.

[0131] In some embodiments, the method further includes:

[0132] If the monitored target is detected to be about to eat or is eating, a first pop-up window is displayed and the image acquisition module is activated; the first pop-up window displays a first prompt to take a picture of the eating content.

[0133] The image acquisition module is used to acquire images of the eating target;

[0134] The dietary data is obtained based on the analysis of the images.

[0135] Monitor whether the target is purchasing food through shopping apps or mini-programs to determine whether the target is about to eat or is currently eating;

[0136] Monitor whether the target's payment application or mini-program has made any expenditures for food purchases, and determine whether the target is about to eat or is currently eating.

[0137] For example, the monitoring function is implemented through an indoor camera and an application associated with the target's mobile phone. Also for example, it is identified whether the target is eating through a wearable device worn by the target. When the action recognition algorithm detects hand movements (such as picking up utensils) and / or mouth movements (such as opening the mouth to prepare to eat) of the target, or by analyzing order information from shopping apps or mini-programs on the target's mobile phone and discovering that the target has purchased food, or by detecting expenditure records for food purchases in the target's payment apps or mini-programs, it is determined that the target is about to eat or is currently eating.

[0138] For example, monitoring whether the monitored target is about to eat or is eating includes at least one of the following: detecting the monitored target's hand movements and / or mouth movements to determine whether the monitored target is about to eat or is eating; monitoring whether the monitored target's shopping app or mini-program purchases food to determine whether the monitored target is about to eat or is eating; monitoring whether the monitored target's payment app or mini-program makes expenditures for food purchases to determine whether the monitored target is about to eat or is eating. In some embodiments, to achieve accurate monitoring, two or more of the above may be combined.

[0139] In some embodiments, once it is determined that the monitored target is about to eat or is eating, the system will display a first pop-up window on the health monitoring application interface of the monitored target's mobile phone, with the message "Please take a picture to record your eating content." This is the first prompt. Simultaneously, the system automatically activates the phone's camera (i.e., the image acquisition module). The monitored target follows the prompt to take a picture of the food they are eating, and the camera captures an image of the eating target. The system analyzes the captured image, using image recognition technology to identify information such as the type and quantity of food, thereby obtaining dietary data. For example, it identifies that the monitored target is eating a bowl of rice, a stir-fry dish, and an apple, and estimates the weight of the rice, the approximate calories of the stir-fry dish, and other data.

[0140] In some embodiments, multiple methods are used to monitor eating, such as detecting hand and mouth movements, monitoring orders from shopping apps or mini-programs, and monitoring spending from payment apps or mini-programs, to determine whether the monitored target is about to eat or is currently eating. This multi-dimensional monitoring approach improves the accuracy and reliability of the judgment and avoids misjudgments that may occur with a single method.

[0141] In some embodiments, after determining the eating status, the first pop-up window is automatically displayed and the image acquisition module is started, realizing the automation and convenience of dietary data acquisition. It eliminates the need for manual operation of the monitored target, thereby improving user experience and data acquisition efficiency.

[0142] In some embodiments, image recognition technology is used to analyze the collected images of the eating target to obtain dietary data, which can intuitively and relatively accurately obtain the actual dietary situation of the monitored target, providing basic data for subsequent health management.

[0143] By accurately capturing eating behavior, the system combines multiple monitoring methods to precisely capture the eating moments of the monitored target, ensuring no eating instance is missed and providing a comprehensive understanding of the target's dietary patterns. Furthermore, by simplifying the data collection process—automatic pop-up windows and image acquisition functions reduce the operational costs of recording the monitored target's diet, increasing cooperation and making it easier to obtain continuous and complete dietary data. In addition, it provides reliable dietary information: the image analysis method for acquiring dietary data provides relatively accurate raw data for subsequent system analysis of the monitored target's dietary nutritional components, calorie intake, etc., contributing to more scientific health assessments.

[0144] In some embodiments, the method further includes:

[0145] A second prompt is displayed based on the dietary data and the first queue; the second prompt is related to the eating habits of the monitored target.

[0146] After acquiring dietary data, a second prompt is generated by combining it with a pre-set first queue (which could be a dietary recommendation queue based on the monitored target's health status, diet plan, etc., such as specifying daily intake ranges for carbohydrates, protein, and fat). If the monitored target's food intake contains excessively high amounts of carbohydrates, exceeding the daily recommended carbohydrate intake value set in the first queue, the second prompt may display: "Your current diet contains a high amount of carbohydrates. Please pay attention to your future dietary combinations and appropriately increase your protein intake." This prompt will be displayed on a specific interface of the mobile health monitoring app to remind the monitored target of the relationship between this meal and their healthy diet plan, helping them to adjust their diet accordingly.

[0147] In some embodiments, a second prompt is generated in conjunction with a dietary cohort: based on dietary data and a pre-set first cohort, the actual dietary situation of the monitored target is combined with a healthy eating plan, making the prompt more targeted and scientific, and better guiding the monitored target to eat a reasonable diet.

[0148] In some embodiments, the eating-related prompts are as follows: The second prompts revolve around the eating situation of the monitored target, are directly related to the eating behavior of the monitored target, help the monitored target understand the relationship between their diet and health in a timely manner, and encourage the monitored target to actively adjust their diet.

[0149] Personalized dietary guidance: Based on dietary data and dietary cohorts, second-level prompts provide individualized suggestions tailored to the actual dietary and health needs of each monitored individual, meeting their differentiated health management requirements. Promoting healthy eating awareness: Timely eating-related prompts enable monitored individuals to quickly understand the rationality of their diet during or after meals, enhancing their awareness and understanding of healthy eating, and gradually guiding them to develop healthy eating habits.

[0150] In some embodiments, the voice interaction function adds a voice broadcast function when displaying the first pop-up window and the second prompt. For monitoring targets who are unable to conveniently view their mobile phone screens (such as those who are busy), voice prompts allow them to promptly understand the system's reminders, further enhancing the user experience. For example, when displaying the first pop-up window, a voice broadcast can simultaneously announce, "Please take a photo to record your food intake"; when displaying the second prompt, specific dietary recommendations can be broadcast. In some embodiments, data fusion with medical devices is also included: dietary data is fused and analyzed with data measured by medical devices such as blood glucose meters and body fat scales. For example, by combining post-meal blood glucose changes, the impact of the monitoring target's meal on blood glucose can be more accurately assessed, providing more comprehensive and targeted dietary and health management recommendations. Simultaneously, this also helps doctors gain a deeper understanding of the monitoring target's health status and develop more appropriate treatment plans.

[0151] In some embodiments, for monitored individuals who do indeed have a disease, social interaction features are provided: a social sharing and interaction module is added. The monitored individual's device can share their dietary records and health tips with family, friends, or health management groups they have joined. Family and friends can offer encouragement and supervision, and other members within the group can also share dietary experiences and health insights, creating a positive social atmosphere for health management and motivating the monitored individual to better adhere to a healthy diet and lifestyle.

[0152] In some embodiments, the method further includes:

[0153] Based on the signed information of the monitoring targets, participate in communities related to blood glucose health;

[0154] Output community information, for example, display community information.

[0155] This monitoring approach will involve participants in communities related to blood glucose health, leveraging the power of the team to better maintain and implement blood glucose monitoring.

[0156] For example, the management platform will create a health management community. A blood glucose management community is one such community, where users can exchange blood glucose management experiences, share diet and exercise tips, and encourage and support each other. The platform includes features such as topic discussion areas, experience sharing sections, and user check-in challenges. For instance, a "Weekly Healthy Eating Challenge" can be launched, where users upload their daily diet records to participate. Completing the challenge earns points, which can be redeemed for health-related gifts or platform value-added services. Through social interaction, user engagement and adherence are enhanced, fostering a positive health management atmosphere. Simultaneously, the platform can collect user data and feedback from these exchanges to further optimize service content and algorithm models.

[0157] In some cases, to protect privacy, identity authentication and access control are employed, which may include:

[0158] By leveraging blockchain's public-key cryptography and digital signature mechanisms, identity authentication and access control for monitored targets are achieved. When a monitored target registers on the platform, a public and private key pair is generated. The private key is kept secret by the target, while the public key is stored on the blockchain. During login and data operations, the target uses its private key to sign the operation information. The platform verifies the signature to confirm the authenticity and legitimacy of the monitored target's identity. Simultaneously, based on the monitored target's role and needs, corresponding access rules are set on the blockchain. Only monitored targets with the appropriate permissions can access and operate on specific data, effectively ensuring the privacy and security of the monitored target's data.

[0159] Establish a comprehensive access control system, assigning different access permissions based on the roles (e.g., general monitoring targets, medical personnel, health managers, platform administrators, etc.) and responsibilities of the monitored targets. For example, general monitoring targets can only access their own health data and personal settings; medical personnel can view and manage the health data of the monitoring targets under their responsibility, and make diagnostic and treatment recommendations; platform administrators have the highest privileges and are responsible for the overall management and maintenance of the system, but their operations on the monitored targets' data are also subject to strict auditing and supervision.

[0160] When conducting data analysis and research, the data of monitored targets is anonymized to remove information that can directly or indirectly identify the monitored targets (such as names, ID numbers, contact information, etc.). Anonymous identifiers are used to replace the monitored targets' identities, ensuring that the privacy of the monitored targets is not leaked. At the same time, the "minimum necessary" principle is followed during data use, using only the minimum amount of data relevant to the analysis purpose to protect the data security of the monitored targets to the greatest extent.

[0161] Establish a data security audit mechanism to monitor and record all data operations within the platform in real time, including data access, querying, modification, and deletion. Regularly analyze audit logs to promptly identify potential data security risks and violations, and take appropriate measures to address them. Simultaneously, set up a security early warning system to immediately issue alerts and notify relevant personnel for emergency response when abnormal data access or operation behavior is detected.

[0162] like Figure 3 The present disclosure provides a dynamic blood glucose health regulation system based on an AI algorithm, comprising:

[0163] The first acquisition module 3101 is used to acquire the first blood glucose data of the monitoring target in the current time window;

[0164] The second acquisition module 3102 is used to acquire at least one first feature map of the blood glucose fluctuation of the monitored target based on the first blood glucose data and the second blood glucose data of at least one historical time window, wherein each pixel in the first feature map includes a first feature vector; one first feature vector corresponds to a time period; the first feature vector is arranged in sequence with time information, scene information and at least one blood glucose measurement value; the scene information includes at least spatial information.

[0165] The third acquisition module 3103 is used to acquire relevant information related to blood glucose monitoring for the monitoring target; the relevant information includes physiological data and dietary data.

[0166] The generation module 3104 is used to generate a second feature map based on the first feature map and the relevant information;

[0167] The fourth acquisition module 3105 is used to obtain a second prediction probability group based on the second feature map and a first prediction probability group of at least one historical time window; the second prediction probability group includes the probability corresponding to different risk levels and the confidence level corresponding to the probability.

[0168] The determination module 3106 is used to determine the risk level of the monitored target in the next time window based on the second predicted probability group; the risk level includes a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia;

[0169] Storage module 3107 is used to store the second predicted probability group into a first queue; the first queue stores the risk levels of multiple time windows; the first queue is used for risk warnings and / or dietary warnings.

[0170] In some embodiments, the second acquisition module is specifically used to acquire the third feature map of the previous time step; the third feature map is the feature map of the previous historical time window.

[0171] Based on the time variation pattern of blood glucose data in the third feature image, a fourth feature image is extracted from the third feature image using a cropping window; wherein, the size of the fourth feature image is related to the cropping window and is smaller than the size of the third feature image;

[0172] Based on the time-varying pattern and the first blood glucose data, the fourth feature map is filled to obtain the first feature map; the size of the first feature map is the same as the size of the third feature map.

[0173] In some embodiments, the fourth acquisition module is configured to process the second feature map using a convolutional spiking neural network model based on an attention mechanism to obtain spatial correlation features; process the second feature map using an iTransformer model to obtain first temporal features; if the first temporal features do not meet a first condition, read second temporal features from N historical windows; fuse the first temporal features and the second temporal features in a first manner to obtain a third temporal feature; the third temporal feature includes blood glucose fluctuation features indicating chronological order; fuse the first temporal features and the second temporal features in a second manner to obtain a fourth temporal feature; the fourth temporal feature includes blood glucose fluctuation features describing different time windows; and obtain a second prediction probability group based on the third temporal feature, the fourth temporal feature, and a first prediction probability group of at least one historical time window.

[0174] In some embodiments, the third time feature is used to indicate patterns of blood glucose fluctuations exceeding a specified time range; and / or,

[0175] The first time feature or the second time feature is used to indicate the pattern of blood glucose fluctuations within a time range that is less than or equal to the specified time range; and / or,

[0176] The fourth time feature is used to indicate the range of blood glucose fluctuations at the same time point within different time windows.

[0177] In some embodiments, the iTransformer model is a modified iTransformer model; the modified iTransformer model includes: a first branch, a second branch, and a connection module; the connection module connects the first branch; the first branch and the second branch have a first connection;

[0178] The first branch is used to extract the first time feature;

[0179] The connection module determines whether the first time feature meets the first condition based on the output of the first branch, and modifies the weight of the first connection according to whether the first time meets the first condition.

[0180] When the first connection has a first weight, the first connection is used to input the first time feature into the second branch;

[0181] When the first connection has a second weight, the first connection is not used to input the first time feature to the second branch.

[0182] In some embodiments, the fifth acquisition module is further configured to input the third time feature and the fourth time feature into the prediction module when the third time feature and the fourth time feature are generated;

[0183] When the third and fourth time features are not generated, the first time feature is input into the prediction module;

[0184] The prediction module is used to generate the second prediction probability group.

[0185] In some embodiments, the device further includes:

[0186] The display module is used to display a first pop-up window and start the image acquisition module when the monitored target is detected to be about to eat or is eating; the first pop-up window displays a first prompt to take a picture of the eating content.

[0187] The acquisition module is used to acquire images of the eating target using the image acquisition module;

[0188] The fifth acquisition module is used to acquire the dietary data based on the analysis of the image.

[0189] In some embodiments, the display module is further configured to display a second prompt based on the dietary data and the first queue; the second prompt is related to the eating of the monitored target.

[0190] In some embodiments, the monitoring module is configured to perform at least one of the following: detect hand and / or mouth movements of the monitored target to determine that the monitored target is about to eat or is eating; monitor whether the monitored target's shopping app or mini-program purchases food to determine that the monitored target is about to eat or is eating; monitor whether the monitored target's payment app or mini-program makes expenditures on food purchases to determine that the monitored target is about to eat or is eating.

[0191] Combination Figure 4As shown, this application embodiment provides an electronic device including a processor 10 and a memory 11. Optionally, the device may further include a communication interface 12 and a bus 9. The processor 10, communication interface 12, and memory 11 can communicate with each other via the bus 9. The communication interface 12 can be used for information transmission. The processor 10 can call logical instructions in the memory 11 to execute the queue-based voiceprint data processing method of the above embodiment.

[0192] Furthermore, the logical instructions in the aforementioned memory 11 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0193] The memory 11, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 10 executes functional applications and data processing by running the program instructions / modules stored in the memory 11, thereby realizing the monitoring method of the queue-based oil and gas transportation system in the above embodiments.

[0194] The memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 11 may include high-speed random access memory and may also include non-volatile memory.

[0195] This electronic device can be used as an edge controller, central controller, or edge control terminal, etc.

[0196] This application provides a computer program product, which includes a computer program stored on a storage medium. The computer program includes program instructions, which, when executed by a computer, cause the computer to perform the above-described monitoring method for an oil and gas transportation system.

[0197] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0198] The technical solutions of this application embodiment can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this application embodiment. The aforementioned storage medium can be a non-transitory storage medium, including various media capable of storing program code such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks, or it can be a transient storage medium.

[0199] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0200] The embodiments or examples disclosed in this application are not exhaustive, but merely illustrative of some embodiments or examples, and are not intended to limit the scope of protection of this disclosure. Unless contradictory, each step in a particular embodiment or example can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment or example can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment or example can be arbitrarily interchanged. Furthermore, optional methods or examples in a particular embodiment or example can be arbitrarily combined; moreover, embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a particular embodiment or example can be arbitrarily combined with optional methods or examples of other embodiments or examples.

[0201] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0202] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0203] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0204] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0205] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for dynamic blood glucose health regulation based on AI algorithms, characterized in that, include; Obtain the first blood glucose data of the monitored target within the current time window; Based on the first blood glucose data and the second blood glucose data of at least one historical time window, at least one first feature map of the blood glucose fluctuation of the monitored target is obtained, wherein each pixel in the first feature map includes a first feature vector; one first feature vector corresponds to a time period; the first feature vector includes time information, scene information and at least one blood glucose measurement value arranged in sequence; the scene information includes at least spatial information. Obtain relevant information related to blood glucose monitoring for the monitoring target; the relevant information includes physiological data and dietary data; A second feature map is generated based on the first feature map and the relevant information; A second prediction probability group is obtained based on the second feature map and a first prediction probability group of at least one historical time window, including: processing the second feature map using a convolutional spiking neural network model based on an attention mechanism to obtain spatial correlation features; processing the second feature map using an improved iTransformer model to obtain first temporal features; if the first temporal features do not meet a first condition, reading second temporal features from N historical windows; fusing the first temporal features and the second temporal features in a first manner to obtain a third temporal feature; the third temporal feature includes blood glucose fluctuation features indicating chronological order; fusing the first temporal features and the second temporal features in a second manner to obtain a fourth temporal feature; the fourth temporal feature includes blood glucose fluctuation features describing different time windows; and obtaining the second prediction probability group based on the third temporal feature, the fourth temporal feature, and the first prediction probability group of at least one historical time window; the second prediction probability group includes probabilities corresponding to different risk levels and confidence levels corresponding to those probabilities; the first prediction probability group corresponds to the second prediction probability group. The risk level of the monitored target in the next time window is determined based on the second predicted probability group; the risk level includes a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia; The second predicted probability group is stored in a first queue; the first queue stores the risk levels for multiple time windows; the first queue is used for risk alerts and / or dietary alerts.

2. The method according to claim 1, characterized in that, Obtain initial information about the monitored target, including at least one of the following: Obtain the third feature map of the previous time point; the third feature map is the feature map of the previous historical time window. Based on the time variation pattern of blood glucose data in the third feature image, a fourth feature image is extracted from the third feature image using a cropping window; wherein, the size of the fourth feature image is related to the cropping window and is smaller than the size of the third feature image; Based on the time-varying pattern and the first blood glucose data, the fourth feature map is filled to obtain the first feature map; the size of the first feature map is the same as the size of the third feature map.

3. The method according to claim 1 or 2, characterized in that, The third time feature is used to indicate the pattern of blood glucose fluctuations exceeding a specified time range; and / or, The first time feature or the second time feature is used to indicate the pattern of blood glucose fluctuations within a time range that is less than or equal to the specified time range; and / or, The fourth time feature is used to indicate the range of blood glucose fluctuations at the same time point within different time windows.

4. The method according to claim 1 or 2, characterized in that, The iTransformer model is a modified iTransformer model; the modified iTransformer model includes: a first branch, a second branch, and a connection module; the connection module connects the first branch; the first branch and the second branch have a first connection; The first branch is used to extract the first time feature; The connection module determines whether the first time feature meets the first condition based on the output of the first branch, and modifies the weight of the first connection according to whether the first time meets the first condition. When the first connection has a first weight, the first connection is used to input the first time feature into the second branch; When the first connection has a second weight, the first connection is not used to input the first time feature to the second branch.

5. The method according to claim 1 or 2, characterized in that, A second prediction probability set is obtained based on the second feature map and a first prediction probability set for at least one historical time window, including: When the third time feature and the fourth time feature are generated, the third time feature and the fourth time feature are input into the prediction module; When the third and fourth time features are not generated, the first time feature is input into the prediction module; The prediction module is used to generate the second prediction probability group.

6. The method according to claim 1 or 2, characterized in that, The method further includes: If the monitored target is detected to be about to eat or is eating, a first pop-up window is displayed and the image acquisition module is activated; the first pop-up window displays a first prompt to take a picture of the eating content. The image acquisition module is used to acquire images of the eating target; The dietary data is obtained based on the analysis of the images.

7. The method according to claim 1 or 2, characterized in that, The method further includes: A second prompt is displayed based on the dietary data and the first queue; the second prompt is related to the eating habits of the monitored target.

8. The method according to claim 7, characterized in that, Monitoring the target to be about to eat or to be eating includes at least one of the following: Detect the hand movements and / or mouth movements of the monitored target to determine whether the monitored target is about to eat or is eating. Monitor whether the target is purchasing food through shopping apps or mini-programs to determine whether the target is about to eat or is currently eating; Monitor whether the target's payment application or mini-program has made any expenditures for food purchases, and determine whether the target is about to eat or is currently eating.

9. A dynamic blood glucose health regulation system based on AI algorithms, characterized in that, include; The first acquisition module is used to acquire the first blood glucose data of the monitoring target in the current time window. The second acquisition module is used to acquire at least one first feature map of the blood glucose fluctuation of the monitored target based on the first blood glucose data and the second blood glucose data of at least one historical time window, wherein each pixel in the first feature map includes a first feature vector; one first feature vector corresponds to a time period; the first feature vector includes time information, scene information and at least one blood glucose measurement value arranged in sequence; the scene information includes at least spatial information. The third acquisition module is used to acquire relevant information related to blood glucose monitoring for the monitoring target; the relevant information includes physiological data and dietary data. A generation module is used to generate a second feature map based on the first feature map and the relevant information; The fourth acquisition module is used to obtain a second prediction probability group based on the second feature map and a first prediction probability group of at least one historical time window, including: processing the second feature map using a convolutional spiking neural network model based on an attention mechanism to obtain spatial correlation features; processing the second feature map using an improved iTransformer model to obtain first temporal features; if the first temporal features do not meet a first condition, reading the second temporal features of N historical windows; fusing the first temporal features and the second temporal features in a first manner to obtain a third temporal feature; the third temporal feature includes blood glucose fluctuation features indicating chronological order; fusing the first temporal features and the second temporal features in a second manner to obtain a fourth temporal feature; the fourth temporal feature includes blood glucose fluctuation features describing different time windows; obtaining the second prediction probability group based on the third temporal feature, the fourth temporal feature, and the first prediction probability group of at least one historical time window; the second prediction probability group includes probabilities corresponding to different risk levels and confidence levels corresponding to the probabilities; the first prediction probability group corresponds to the second prediction probability group. The determination module is used to determine the risk level of the monitored target in the next time window based on the second predicted probability group; the risk level includes a first type of value and a second type of value; the first type of value is related to hypoglycemia; the second type of value is related to hyperglycemia; A storage module is used to store the second predicted probability group into a first queue; the first queue stores the risk levels of multiple time windows; the first queue is used for risk alerts and / or dietary alerts.

Images