User health state management method and system based on intelligent wearable device

Abstract

The invention discloses a user health state management method and system based on intelligent wearable equipment. The method comprises the following steps: firstly, transmitting an electromagnetic wave signal with a specific wavelength to a preset body area of a user through a radio frequency transmitting module of the intelligent wearable equipment, and receiving a corresponding feedback signal; and performing glucose feedback signal judgment on the received feedback signal by using a pre-trained glucose feedback signal judgment model. And after the existence of the glucose feedback signal is confirmed, further calling the target blood glucose state evaluation model to process the signal, and outputting an autoregression feature of the feedback signal. And the blood sugar health state of the user is evaluated by comparing the blood sugar health state characteristics with pre-stored blood sugar health state characteristics. By means of the design, real-time monitoring and health state evaluation of biochemical indexes such as the blood glucose of the user are achieved, and the application efficiency of the intelligent wearable device in the field of health management is improved.

Description

Technical Field

[0001] The present invention relates to the field of smart wearable devices, and more particularly, to a method and system for managing a user's health status based on a smart wearable device. Background Art

[0002] With the increasing emphasis on health by people, smart wearable devices are increasingly widely used in the field of health monitoring. However, the existing technologies have deficiencies in monitoring the user's health status, especially the blood glucose health status. For example, the existing monitoring methods are not accurate enough and have poor real-time performance, making it difficult to accurately judge the user's blood glucose health status and unable to meet people's higher requirements for health management. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for managing a user's health status based on a smart wearable device.

[0004] In a first aspect, an embodiment of the present invention provides a method for managing a user's health status based on a smart wearable device, including:

[0005] In response to a health status evaluation instruction triggered by a preset period, initializing a radio frequency transmission module and a feedback signal reception module of the smart wearable device;

[0006] Transmitting an electromagnetic wave signal with a preset wavelength to a preset body area of a target user through the radio frequency transmission module;

[0007] Receiving a feedback signal to be determined corresponding to the electromagnetic wave signal with the preset wavelength through the feedback signal reception module;

[0008] Invoking a pre-trained glucose feedback signal determination model to determine a glucose feedback signal for the feedback signal to be determined, and obtaining a determination result for characterizing whether the feedback signal to be determined includes a glucose feedback signal;

[0009] When the determination result indicates the existence of a glucose feedback signal, obtaining a glucose feedback signal to be processed;

[0010] Invoking a pre-trained target blood glucose status evaluation model to process the glucose feedback signal and output an autoregressive feature of the feedback signal;

[0011] Performing a comparison operation on the autoregressive feature of the feedback signal and a pre-stored blood glucose health status feature to obtain a comparison result;

[0012] When the comparison result indicates that the feature distance between the autoregressive feature of the feedback signal and the blood glucose health status feature is less than a preset feature distance threshold, obtaining the blood glucose health status corresponding to the glucose feedback signal.

[0013] In a second aspect, an embodiment of the present invention provides a server system, including a server for executing the method described in the first aspect.

[0014] Compared with the prior art, the beneficial effects provided by the present invention include: By using a user health status management method and system based on an intelligent wearable device disclosed in the present invention, an electromagnetic wave signal with a specific wavelength is transmitted to a preset body area of the user through the radio frequency transmission module of the intelligent wearable device, and the corresponding feedback signal is received. Using a pre-trained glucose feedback signal determination model, the received feedback signal is subjected to glucose feedback signal determination. After confirming the existence of a glucose feedback signal, a target blood glucose status evaluation model is further called to process the signal and output the autoregressive features of the feedback signal. By comparing with the pre-stored blood glucose health status features, the blood glucose health status of the user is evaluated. With such a design, real-time monitoring of biochemical indicators such as the user's blood glucose and health status evaluation are realized, and the application efficiency of intelligent wearable devices in the field of health management is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic flowchart of the steps of the user health status management method based on an intelligent wearable device provided by an embodiment of the present invention;

[0017] Figure 2 It is a schematic block diagram of the structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0019] The following will describe in detail the specific embodiments of the present invention in conjunction with the drawings.

[0020] To solve the technical problems in the aforementioned background art, Figure 1It is a schematic flowchart of a user health status management method based on a smart wearable device provided by an embodiment of the present disclosure. The user health status management method based on the smart wearable device will be introduced in detail below.

[0021] Step S201: In response to a health status assessment instruction triggered by a preset period, initialize the radio frequency transmission module and the feedback signal reception module of the smart wearable device.

[0022] Step S202: Transmit an electromagnetic wave signal with a preset wavelength to a preset body area of a target user through the radio frequency transmission module.

[0023] Step S203: Receive a feedback signal to be determined corresponding to the electromagnetic wave signal with the preset wavelength through the feedback signal reception module.

[0024] Step S204: Invoke a pre-trained glucose feedback signal determination model to determine a glucose feedback signal for the feedback signal to be determined, and obtain a determination result for characterizing whether the feedback signal to be determined includes a glucose feedback signal.

[0025] Step S205: When the determination result indicates the existence of a glucose feedback signal, obtain the glucose feedback signal to be processed.

[0026] Step S206: Invoke a pre-trained target blood glucose status assessment model to process the glucose feedback signal and output an autoregressive feature of the feedback signal.

[0027] Step S207: Perform a comparison operation on the autoregressive feature of the feedback signal and a pre-stored blood glucose health status feature to obtain a comparison result.

[0028] Step S208: When the comparison result indicates that the feature distance between the autoregressive feature of the feedback signal and the blood glucose health status feature is less than a preset feature distance threshold, obtain the blood glucose health status corresponding to the glucose feedback signal.

[0029] In an embodiment of the present invention, by way of example, for instance, the preset period is 8:00 am every day. When it reaches 8:00 am, the server receives a health status assessment instruction. The server sends an initialization instruction to the smart wearable device. After receiving the instruction, the smart wearable device performs an initialization operation on its internal radio frequency transmission module and feedback signal reception module. For example, the smart wearable device can be a smart bracelet. When initializing the radio frequency transmission module, the device will reset parameters such as the transmission frequency and power to ensure that the emitted electromagnetic wave signal is stable and meets the preset standards. For the feedback signal reception module, the previously received cached data will be cleared, and parameters such as the reception sensitivity will be adjusted to the initial state to prepare for receiving new feedback signals. For example, the preset body area is the inner side of the wrist, and the electromagnetic wave signal of the preset wavelength is a radio wave of a specific frequency. The server controls the radio frequency transmission module of the smart wearable device to emit the electromagnetic wave signal of this specific frequency to the inner side of the user's wrist. For example, the radio frequency transmission module in the smart bracelet will generate and emit an electromagnetic wave signal with a frequency of 1 GHz. These electromagnetic wave signals can penetrate the skin to a certain depth and interact with the tissues and body fluids on the inner side of the wrist. After the radio frequency transmission module emits the electromagnetic wave signals, these signals will undergo phenomena such as reflection, refraction, and scattering when penetrating the tissues and body fluids on the inner side of the user's wrist. The feedback signal reception module will receive these electromagnetic wave feedback signals after the interaction. For example, the signals received by the feedback signal reception module can include changes in frequency and amplitude caused by factors such as tissue composition and blood flow. These changed signals are the feedback signals to be determined. A glucose feedback signal determination model trained with a large amount of data is pre-stored and loaded in the server. After receiving the feedback signals to be determined, the server inputs these signals into the model for determination. For example, the model will analyze and calculate multiple features such as the spectral characteristics, amplitude changes, and phase changes of the feedback signals. If the model determines that these features match the relevant feature patterns of glucose, the determination result is that the feedback signal contains glucose; otherwise, it is determined that it does not contain glucose. For example, after analyzing the feedback signals to be determined, the model finds that the intensity changes in certain frequency bands in its spectral characteristics conform to the expected pattern of the interaction between glucose molecules and electromagnetic waves, so the determination result is that the feedback signal contains glucose. If the determination result indicates that there is a glucose feedback signal in the feedback signals to be determined, the server will further obtain these glucose feedback signals to be processed. For example, the server will receive more detailed and accurate data about the glucose feedback signals from the smart wearable device, which can include information such as the duration of the signal and the intensity change curve. The server calls the pre-trained target blood glucose status assessment model to process the obtained glucose feedback signals. For example, the model will perform complex mathematical operations and feature extraction on the glucose feedback signals, and finally output the autoregressive features of the feedback signals. This autoregressive feature can reflect the change trend and pattern of the blood glucose level.For example, the autoregressive feature is a vector containing multiple numerical values, where each value represents the change trend and correlation of blood glucose levels at different time points. Feature data of various blood glucose health states are pre-stored in the server. The autoregressive feature of the just-obtained feedback signal is compared with these pre-stored features. For example, the comparison operation can be calculating the Euclidean distance or cosine similarity between two feature vectors, etc. For example, the pre-stored feature patterns of blood glucose health states include those of normal blood glucose, high blood glucose, and low blood glucose. Through comparison, the similarity between the autoregressive feature of the feedback signal and the normal blood glucose feature pattern is calculated to be 0.8, the similarity with the high blood glucose feature pattern is 0.2, and the similarity with the low blood glucose feature pattern is 0.1. If the comparison result shows that the feature distance between the autoregressive feature of the feedback signal and the blood glucose health state feature is less than a preset threshold, the server determines the current blood glucose health state of the user. For example, the preset feature distance threshold is 0.5. Since the similarity between the autoregressive feature of the feedback signal and the normal blood glucose feature pattern is 0.8 (greater than 0.5), the server determines that the user's current blood glucose is in a normal and healthy state.

[0030] In the embodiment of the present invention, the glucose feedback signal determination model is obtained in the following manner.

[0031] Call the glucose feedback signal determination model to perform glucose feedback signal determination on the invalid instantaneous feedback signal instances and the valid instantaneous feedback signal instances respectively, and obtain the invalid instantaneous determination result and the valid instantaneous result correspondingly;

[0032] Among them, the invalid instantaneous feedback signal instances are obtained by adding an instantaneous feedback signal to the invalid instances, and the valid instantaneous feedback signal instances are obtained by adding an instantaneous feedback signal to the valid instances corresponding to the invalid instances;

[0033] Based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid instantaneous determination result, and the valid instantaneous result, determine the error parameter of the first error function;

[0034] Call the glucose feedback signal determination model to perform glucose feedback signal determination on the invalid continuous feedback signal instances and the valid continuous feedback signal instances respectively, and obtain the invalid continuous determination result and the valid continuous determination result correspondingly;

[0035] Among them, the invalid continuous feedback signal instances are obtained by adding a continuous feedback signal to the invalid instances, and the valid continuous feedback signal instances are obtained by adding a continuous feedback signal to the valid instances;

[0036] Determine a first error parameter of a second error function based on the instance target value of the invalid duration feedback signal instance, the instance target value of the valid duration feedback signal instance, the invalid duration determination result, and the valid duration determination result;

[0037] Optimize the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function and the first error parameter of the second error function.

[0038] In an embodiment of the present invention, exemplarily, first, the server prepares a large number of sample data, including invalid instances and valid instances. These instances are obtained through screening and classification from a large amount of user data collected in the past. For the acquisition of invalid instantaneous feedback signal instances, for example, the server selects a blood glucose monitoring data instance originally marked as invalid. This invalid instance may be due to data deviation caused by equipment failure, strong external interference, or other abnormal conditions. Then, the server adds a simulated instantaneous feedback signal to this invalid instance. This instantaneous feedback signal may be a short-term and high-intensity abnormal signal, such as a large voltage fluctuation occurring within a very short period of time. Next, the server calls the glucose feedback signal determination model to determine the glucose feedback signal for this invalid instance with the added instantaneous feedback signal. After analysis and calculation by the model, an invalid instantaneous determination result is given. For example, the model determines that this invalid instance with the added instantaneous feedback signal still does not conform to the characteristics of the glucose feedback signal, and the determination result is "invalid". For valid instantaneous feedback signal instances, the server selects a blood glucose monitoring data instance originally marked as valid corresponding to the above-mentioned invalid instance. This valid instance may be accurate data from a healthy individual or an individual with a stable blood glucose level under normal monitoring conditions. Then, a similar instantaneous feedback signal is also added to this valid instance. The glucose feedback signal determination model is called again for determination, and the model gives a valid instantaneous determination result, such as "valid", indicating that the model believes that this valid instance with the added instantaneous feedback signal conforms to the characteristics of the glucose feedback signal. Next, the server determines the error parameter of the first error function based on the instance target value of the invalid instance, the instance target value of the valid instance, and the just-obtained invalid instantaneous determination result and valid instantaneous result. For example, the instance target value of the invalid instance is set to 0 (indicating invalid), and the instance target value of the valid instance is set to 1 (indicating valid). If the invalid instantaneous determination result is 0 and the valid instantaneous determination result is 1, which is completely matched with the instance target value, the error parameter of the first error function will be smaller and may be close to 0. However, if the invalid instantaneous determination result is 1 or the valid instantaneous determination result is 0, a larger error will occur, and the error parameter of the first error function will increase accordingly. Then, the server continues to obtain invalid continuous feedback signal instances and valid continuous feedback signal instances. For example, for invalid continuous feedback signal instances, the server selects the previous invalid instance again. This time, a feedback signal with a longer duration is added to it. This continuous feedback signal may simulate continuous abnormal conditions over a period of time, such as continuous low-intensity interference or data changes caused by chronic equipment failures. For valid continuous feedback signal instances, a similar continuous feedback signal is added to the corresponding valid instance.Subsequently, the server calls the glucose feedback signal determination model to determine these two continuous feedback signal instances, obtaining an invalid continuous determination result and a valid continuous determination result. For example, the determination result of the model for the invalid continuous feedback signal instance is "invalid", and the determination result for the valid continuous feedback signal instance is "valid". Then, the server determines the first error parameter of the second error function based on the instance target value of the invalid continuous feedback signal instance (set to 0), the instance target value of the valid continuous feedback signal instance (set to 1), the invalid continuous determination result, and the valid continuous determination result. If the determination result is exactly the same as the instance target value, the first error parameter of the second error function will be smaller; conversely, if there is a deviation, the error parameter will increase. Finally, the server optimizes the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function and the first error parameter of the second error function. For example, the server uses an optimization algorithm (such as the gradient descent algorithm) to calculate the adjustment direction and amplitude of the network parameters according to these two error parameters. If the error parameter is large, it indicates that the determination result of the model deviates greatly from the actual target value, and the server will make a large adjustment to the network parameters to improve the accuracy and reliability of the model. By continuously repeating the above process until the error parameter reaches an acceptable small value, the glucose feedback signal determination model can effectively and accurately determine various instantaneous and continuous feedback signal instances.

[0039] In the embodiment of the present invention, the following implementation manners are further provided.

[0040] Call the glucose feedback signal determination model to respectively determine the glucose feedback signals for the invalid instance and the valid instance corresponding to the invalid instance, and obtain the invalid instance determination result and the valid instance determination result correspondingly;

[0041] Based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid instance determination result, and the valid instance determination result, determine the first error parameter of the third error function;

[0042] The optimizing the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function and the first error parameter of the second error function includes:

[0043] Optimize the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function, the first error parameter of the second error function, and the first error parameter of the third error function.

[0044] In an embodiment of the present invention, exemplarily, first, the server obtains a series of invalid instances and corresponding valid instances. For example, the invalid instances can be a set of data from individuals with serious diseases resulting in abnormal blood glucose monitoring data, while the corresponding valid instances are normal blood glucose data from healthy individuals under the same monitoring conditions. The server calls the glucose feedback signal determination model to determine these invalid instances. For example, for a certain invalid instance, after analyzing and processing its data, the model obtains a determination result, which can be determined as "invalid". Similarly, for the corresponding valid instance, the model also gives a determination result, such as being determined as "valid". Then, the server determines the first error parameter of the third error function based on the instance target value of the invalid instance (for example, 0 indicates invalid) and the instance target value of the valid instance (for example, 1 indicates valid), as well as the determination results of the invalid instance and the valid instance just obtained. For example, there are multiple pairs of invalid instances and valid instances. For the first invalid instance, its target value is 0, and the model determination result is also 0, which are consistent, and the error in this part is relatively small. For the second valid instance, the target value is 1, and the model determination result is also 1, and there is no error either. However, if for the third invalid instance, the target value is 0, but the model determination result is 1, then a relatively large error occurs. By comprehensively calculating the target values and determination results of all these instances, the first error parameter of the third error function can be obtained. Next, when optimizing the network parameters of the glucose feedback signal determination model, the server will comprehensively consider the error parameter of the first error function, the first error parameter of the second error function, and the first error parameter of the third error function. For example, the server first normalizes these three error parameters so that they can be compared and calculated on the same order of magnitude. For example, the error parameter of the first error function is 0.1, the first error parameter of the second error function is 0.2, and the first error parameter of the third error function is 0.15. Then, the server sets weights for these three error functions respectively. The setting of the weights can be determined according to experience, data characteristics, or through methods such as cross-validation. For example, the weight set for the first error function is 0.3, the weight of the second error function is 0.4, and the weight of the third error function is 0.3. Next, the server calculates the comprehensive error: 0.1×0.3 + 0.2×0.4 + 0.15×0.3 = 0.165. Based on this comprehensive error, the server uses an optimization algorithm, such as the stochastic gradient descent algorithm, to adjust the network parameters of the glucose feedback signal determination model. If the comprehensive error is large, it indicates that the performance of the model is not good enough, and the server will make a relatively large adjustment to the network parameters, such as increasing the learning rate, to converge to better parameter values faster. If the comprehensive error is small, the server will appropriately reduce the learning rate for a more refined adjustment to avoid over-adjustment resulting in model instability.During the process of adjusting the network parameters, the server will continuously repeat the above steps. For each round of new data, the server will recalculate the error parameters, update the comprehensive error, and accordingly adjust the network parameters again. For example, after the first round of adjustment, the server uses a new batch of invalid instances and valid instances to test the model again, recalculates the parameters of the three error functions, and obtains a new comprehensive error of 0.12. Compared with the previous 0.165, it has decreased, indicating that the performance of the model is gradually improving. However, the server will still continue to adjust until the comprehensive error reaches a preset minimum value or after a certain number of iterations, the performance of the model reaches the expected level. By continuously optimizing the network parameters in this way, the glucose feedback signal determination model can determine invalid instances and valid instances more and more accurately, improving its reliability and accuracy in practical applications.

[0045] In the embodiment of the present invention, the error parameters of the first error function, the first error parameters of the second error function, and the first error parameters of the third error function are used to optimize the network parameters of the glucose feedback signal determination model, which can be implemented through the following examples.

[0046] Obtain the weighting coefficients of the first error function, the weighting coefficients of the second error function, and the weighting coefficients of the third error function respectively;

[0047] Based on the error parameters and weighting coefficients of the first error function, the first error parameters and weighting coefficients of the second error function, and the first error parameters and weighting coefficients of the third error function, obtain the error parameters of the total error function;

[0048] Optimize the network parameters of the glucose feedback signal determination model according to the error parameters of the total error function.

[0049] In an embodiment of the present invention, exemplarily, the server has first obtained the error parameters of the first error function, the first error parameter of the second error function, and the first error parameter of the third error function through a series of operations. In order to more precisely optimize the network parameters of the glucose feedback signal determination model, the server needs to obtain the weighting coefficients of these three error functions respectively. For example, through the analysis of historical data and multiple experimental verifications, the server determines that the weighting coefficient of the first error function is 0.2, which means that when comprehensively considering the total error, the influence degree of the first error function accounts for 20%. Similarly, the server determines that the weighting coefficient of the second error function is 0.5, and the weighting coefficient of the third error function is 0.3. Next, the server obtains the error parameters of the total error function based on these error parameters and weighting coefficients. For example, the error parameter of the first error function is 0.1, the first error parameter of the second error function is 0.15, and the first error parameter of the third error function is 0.08. Then the error parameter of the total error function is calculated as follows: Total error = 0.1×0.2 + 0.15×0.5 + 0.08×0.3 = 0.02 + 0.075 + 0.024 = 0.119. After obtaining the error parameter 0.119 of the total error function, the server optimizes the network parameters of the glucose feedback signal determination model according to this value. The server uses an optimization algorithm, such as the Stochastic Gradient Descent (SGD) algorithm. In each iteration, the server calculates the gradient of the network parameters according to the error parameter of the total error function. For example, for example, the network parameters of the current model are a set of values [w1, w2, w3,..., wn], and the server calculates the gradient [dw1, dw2, dw3,..., dwn] of each parameter with respect to the total error through derivation. Then, the server updates the network parameters according to the learning rate and the gradient. The learning rate is an important factor controlling the update step size of the parameters. For example, the learning rate is 0.01. New network parameters = original network parameters - learning rate × gradient. That is, the new network parameters are [w1 - 0.01×dw1, w2 - 0.01×dw2, w3 - 0.01×dw3,..., wn - 0.01×dwn]. After updating the network parameters, the server tests with new data again, recalculates the error parameter of the first error function, the first error parameter of the second error function, and the first error parameter of the third error function, and obtains the error parameter of the new total error function in the above manner. If the error parameter of the new total error function is smaller than before, it means that the model is developing in a better direction, and the server continues to update the network parameters in this way. But if the error parameter of the total error function does not decrease or even increases, the server can adjust the learning rate or recheck the data and model settings. For example, after the first round of optimization, the error parameter of the total error function is reduced from 0.119 to 0.105.The server continues with the next round of optimization, repeating this process continuously until the error parameter of the total error function reaches a stable minimum value or meets the preset stop condition. During this process, the server also performs some additional operations. For example, if it is found that the influence of a certain data subset on the total error is too large or too small, the server will re-evaluate the weighting coefficient of the corresponding error function to ensure the rationality and effectiveness of the optimization process. Another example is that the server will simultaneously monitor the performance of the model on the validation set to prevent overfitting of the model. If the total error on the training set continues to decrease, but the performance on the validation set does not improve or even deteriorates, the server will take some regularization measures, such as L1 and L2 regularization, to limit the size of the network parameters and avoid the model being too complex. Through such repeated calculations, updates, and adjustments, the server continuously optimizes the network parameters of the glucose feedback signal determination model, enabling the model to more accurately determine the glucose feedback signal.

[0050] In the embodiment of the present invention, before the glucose feedback signal determination model is called to determine the glucose feedback signal for the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance respectively, the following implementation manners are further provided in the embodiment of the present invention.

[0051] Swap the instance target value of the invalid instance and the instance target value of the valid instance to obtain the enhanced instance target value of the invalid instance and the enhanced instance target value of the valid instance;

[0052] Based on the enhanced instance target value of the invalid instance, the enhanced instance target value of the valid instance, the determination result of the invalid instance, and the determination result of the valid instance, determine the second error parameter of the third error function;

[0053] Generate an instantaneous feedback signal based on the second error parameter of the third error function;

[0054] Add the instantaneous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance correspondingly.

[0055] In an embodiment of the present invention, exemplarily, first, the server obtains invalid instances and valid instances. For example, the invalid instances are a set of data obtained by a blood glucose monitoring device in an extremely interfering environment, while the valid instances are accurate data obtained under ideal monitoring conditions. Before invoking the glucose feedback signal determination model to determine the invalid instantaneous feedback signal instances and the valid instantaneous feedback signal instances, the server swaps the instance target values of the invalid instances and the instance target values of the valid instances. The instance target value of the original invalid instance can be marked as 0, and the instance target value of the valid instance is marked as 1. After swapping, the invalid instance obtains an enhanced instance target value of 1, and the valid instance obtains an enhanced instance target value of 0. Next, the server determines the second error parameter of the third error function based on the enhanced instance target value of the invalid instance, the enhanced instance target value of the valid instance, and the previous determination results of the invalid instances and the valid instances. For example, the server first calculates the error of the invalid instance. For example, for a certain invalid instance, the previous determination result is "invalid", but due to the swapping of the instance target value, an error will occur at this time. The server calculates the error value of this part by a specific error calculation method, comprehensively considering all relevant data and the enhanced instance target value of this invalid instance. Similarly, for the valid instance, the error is calculated in the same way. Then, the server combines these errors to obtain the second error parameter of the third error function. Based on this second error parameter, the server generates an instantaneous feedback signal. This instantaneous feedback signal can be an electrical signal containing specific frequency and amplitude variations. The server adds the generated instantaneous feedback signal to the invalid instance and the valid instance respectively. For example, for the data sequence [d1, d2, d3,..., dn] of a certain invalid instance, the server adds the instantaneous feedback signal [f1, f2, f3,..., fm] according to a certain rule to obtain the data sequence [d1 + f1, d2 + f2, d3 + f3,..., dn + fm] of the new invalid instantaneous feedback signal instance. For the valid instance, a similar operation is performed to obtain the valid instantaneous feedback signal instance. Through such processing, the server enriches the data for training the glucose feedback signal determination model, enabling the model to better learn and identify the glucose feedback signal characteristics in different situations, and improving the accuracy and robustness of the model. In actual operation, the server processes a large number of invalid instances and valid instances, and finely adjusts and optimizes the generated instantaneous feedback signal to ensure that the added invalid instantaneous feedback signal instances and valid instantaneous feedback signal instances can effectively improve the training effect of the model. For example, in one round of processing, the server performs the above operations on 100 invalid instances and 100 valid instances. When calculating the error, the server considers multiple feature dimensions of the data, such as the peak value, mean value, variance, etc. of the signal. For the generated instantaneous feedback signal, the server adjusts the frequency range and amplitude size of the signal according to the previous performance of the model training and the characteristics of the data.If it is found that the training effect of the model does not improve significantly after adding the instantaneous feedback signal, the server will re-analyze the error parameters and the generated instantaneous feedback signal, search for possible problems and make improvements. For another example, the server will also perform data visualization on the instances after adding the instantaneous feedback signal to visually observe the distribution and changes of the data, so as to better understand and optimize the entire processing process. By continuously looping through the above steps, the server gradually improves the training data of the model, enabling the glucose feedback signal determination model to work more accurately and reliably.

[0056] In the embodiment of the present invention, generating the instantaneous feedback signal based on the second error parameter of the third error function can be implemented through the following examples.

[0057] According to the second error parameter of the third error function, determine the parameter update direction of the third error function with respect to the network parameter, and obtain the optimization guidance amount of the glucose feedback signal determination model;

[0058] Determine the optimization guidance amount of the glucose feedback signal determination model as the instantaneous feedback signal.

[0059] In an embodiment of the present invention, exemplarily, the server has obtained the second error parameter of the third error function. This error parameter reflects the deviation degree of the current glucose feedback signal determination model when processing invalid instances and valid instances. First, the server determines the parameter update direction of the third error function with respect to the network parameters according to this second error parameter. The server will achieve this process through a series of mathematical calculations and analyses. For example, the third error function is a complex mathematical expression about network parameters, such as E = f(w1, w2, w3,..., wn), where w1 to wn are network parameters. By taking the partial derivative of this function with respect to each parameter, the server can obtain the gradient direction of each parameter. If the second error parameter is large, it indicates that the deviation of the model is large, then the gradient value calculated by the server will also be relatively large, which means that the network parameters need to be adjusted by a large margin. And if the second error parameter is small, the gradient value will also be correspondingly small, and the adjustment amplitude of the network parameters will be small. For example, for a certain network parameter w1, the gradient calculated by the server is -0.5, which means that w1 needs to be updated in the decreasing direction to reduce the error. For another network parameter w2, the gradient is 0.3, which indicates that w2 needs to be updated in the increasing direction. The server comprehensively considers the gradient directions of all network parameters, thereby determining the parameter update direction of the third error function with respect to the network parameters. This update direction is the optimization guidance quantity of the glucose feedback signal determination model. Then, the server determines this optimization guidance quantity as the instantaneous feedback signal. For example, the server arranges the above-mentioned calculated parameter update direction, including the adjustment amplitude and direction of each parameter, into a specific data format or signal form. This signal can be recognized and applied by subsequent processing procedures. For example, this instantaneous feedback signal is represented in the form of a set of values [dw1, dw2, dw3,..., dwn], where dw1 to dwn respectively correspond to the adjustment amounts of each network parameter. In actual operations, the server will process a large amount of data and complex network structures simultaneously. For each calculation and update, the server will conduct strict accuracy and rationality checks to ensure that the generated instantaneous feedback signal can effectively guide the optimization of the model. For example, when the server processes a neural network with thousands of nodes and connections, it will use parallel computing technology to accelerate the calculation of gradients and the determination of update directions. At the same time, the server will also compare and analyze the current calculation results with historical data to promptly detect possible abnormal situations, such as problems like gradient explosion or disappearance. Also, for example, the server will dynamically adjust the algorithms and parameters for calculating gradients and determining update directions according to different data distributions and model complexities. If the data has high noise or uncertainty, the server will adopt a more robust optimization algorithm, or increase the data preprocessing steps to improve the accuracy and effectiveness of the instantaneous feedback signal.In this way, the server can generate an effective instantaneous feedback signal based on the second error parameter of the third error function, providing crucial guidance for the optimization of the glucose feedback signal determination model.

[0060] In the embodiment of the present invention, adding the instantaneous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance can be implemented through the following examples.

[0061] Determine the weighting coefficient of the instantaneous feedback signal, and perform a weighting operation on the instantaneous feedback signal based on the weighting coefficient to obtain a weighted instantaneous feedback signal;

[0062] Add the weighted instantaneous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance.

[0063] In an embodiment of the present invention, exemplarily, the server has generated an instantaneous feedback signal and next needs to add it to invalid instances and valid instances. First, the server determines the weighting coefficient of the instantaneous feedback signal. The determination of this weighting coefficient can be based on various factors, such as the previous model training history, the feature distribution of the data, or calculated through preset rules and algorithms. For example, the server determines the weighting coefficient to be 0.2 according to the previous training experience and the characteristics of the current data. Then, the server performs a weighting operation on the instantaneous feedback signal based on this weighting coefficient. For example, if the original value of the instantaneous feedback signal is [10, 20, 30, 40, 50], by multiplying it by the weighting coefficient 0.2, the weighted instantaneous feedback signal is [2, 4, 6, 8, 10]. Next, the server adds this weighted instantaneous feedback signal to the invalid instances and valid instances respectively. For example, the invalid instance is a set of blood glucose monitoring data, such as [120, 130, 125, 118, 122]. By adding the weighted instantaneous feedback signal correspondingly, the invalid instantaneous feedback signal instance obtained is [122, 134, 131, 126, 132]. Similarly, for the valid instance, for example, originally [90, 88, 92, 89, 91], after adding the weighted instantaneous feedback signal, the valid instantaneous feedback signal instance obtained is [92, 92, 98, 97, 101]. In the actual processing, the server can process a large number of invalid instances and valid instances simultaneously. For example, the server may process 1000 invalid instances and 1000 valid instances at one time. For each instance, the weighting and addition operations are precisely performed according to the above steps. When determining the weighting coefficient, the server takes into account the importance and sensitivity of different types of data. For example, if the data of the invalid instance usually has large fluctuations, the server will choose a smaller weighting coefficient to avoid overly interfering with the characteristics of the original data. For the valid instance, if the data is relatively stable, the weighting coefficient will be slightly larger to enhance the impact of the feedback signal on model training. At the same time, the server also monitors and evaluates the instances after adding the weighted instantaneous feedback signal in real time. If it is found that the added instances cause deviations or instability in model training, the server will readjust the weighting coefficient or re-examine the generation method and characteristics of the instantaneous feedback signal. For example, after the server adds the weighted instantaneous feedback signal in the first round, it is found that the model is overly sensitive to certain types of blood glucose changes in subsequent training. After analysis, the server finds that it is caused by an overly large weighting coefficient. So, the server adjusts the weighting coefficient to 0.1, regenerates and adds the weighted instantaneous feedback signal, and conducts model training again to observe whether the effect is improved. Also, for example, the server will dynamically adjust the weighting coefficient and the characteristics of the instantaneous feedback signal according to different time periods or the characteristics of user groups.For example, for the blood glucose monitoring data in the morning period, due to the particularity of the human physiological state, the server will adopt different weighting strategies and feedback signal generation methods to improve the accuracy and adaptability of the model in this specific period. Through such delicate operations and continuous optimization and adjustment, the server can accurately add the weighted instantaneous feedback signal to the invalid instances and valid instances, providing richer and more effective data for the training of the glucose feedback signal determination model, thereby continuously improving the performance and accuracy of the model.

[0064] In an embodiment of the present invention, the glucose feedback signal determination invokes a discrimination component in the glucose feedback signal determination model for implementation, and the glucose feedback signal determination model further includes a synthesis component;

[0065] Before the glucose feedback signal determination model is called to perform glucose feedback signal determination on the invalid continuous feedback signal instances and the valid continuous feedback signal instances respectively, the embodiments of the present invention also provide the following implementation manners.

[0066] Invoke the synthesis component in the glucose feedback signal determination model to generate a continuous feedback signal based on the input preset continuous signal;

[0067] Add the continuous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid continuous feedback signal instance and the valid continuous feedback signal instance correspondingly.

[0068] In an embodiment of the present invention, exemplarily, during the process of determining the glucose feedback signal, the server uses a glucose feedback signal determination model including a discrimination component and a synthesis component. Before determining the invalid continuous feedback signal instances and the valid continuous feedback signal instances, the server first invokes the synthesis component in the model. For example, the preset continuous signal input to the synthesis component is a series of analog electrical signal values collected at specific time intervals, such as [5, 10, 15, 20, 25], and each value represents the signal strength at a specific moment. After receiving these preset continuous signals, the synthesis component starts to generate a continuous feedback signal through internal algorithms and logic. This generation process involves operations such as filtering, amplifying, and modulating the input signal. For example, the synthesis component can perform low-pass filtering on the input signal to remove high-frequency noise, and then linearly amplify the filtered signal to increase its amplitude by a certain multiple. Then, through a specific modulation method, such as frequency modulation or phase modulation, the processed signal is converted into a continuous feedback signal with specific characteristics. For example, the continuous feedback signal generated after the above processing is [30, 40, 50, 60, 70]. Next, the server adds this continuous feedback signal to the invalid instances and the valid instances respectively. For example, the invalid instance is a set of measured blood glucose concentration values, such as [8.0, 9.0, 8.5, 7.5, 8.2]. By adding the continuous feedback signal correspondingly, the invalid continuous feedback signal instance obtained is [38.0, 49.0, 58.5, 67.5, 78.2]. For the valid instance, for example, it was originally [5.0, 4.8, 5.2, 4.9, 5.1]. After adding the continuous feedback signal, the valid continuous feedback signal instance obtained is [35.0, 44.8, 55.2, 64.9, 75.1]. In actual operation, the server can process a large number of different invalid instances and valid instances simultaneously. For example, there may be thousands or even more invalid instances and valid instances waiting to be processed. For each invalid instance and valid instance, the server strictly follows the above steps to generate and add the continuous feedback signal to ensure the accuracy and consistency of the data. During the process of generating the continuous feedback signal, the server will adjust the parameters of the synthesis component according to different requirements and data characteristics. For example, if the input preset continuous signal has large fluctuations, the server will increase the intensity of filtering to obtain a smoother continuous feedback signal. At the same time, the server will also perform a quality check on the instances after adding the continuous feedback signal. If it is found that some instances have outliers or do not conform to the expected data pattern after adding, the server will re-examine the operations of the synthesis component and the adding process, search for possible problems and make corrections. For example, during the inspection process, the server finds that the data in a certain invalid continuous feedback signal instance significantly deviates from the normal range. After analysis, it is found that there is a parameter error in the modulation process of the synthesis component. The server promptly adjusts the parameters, regenerates and adds the continuous feedback signal to ensure the rationality of the instance data.In addition, the server may also dynamically change the preset continuous signal input to the synthesis component according to different application scenarios and user requirements. For example, for blood glucose monitoring of patients with specific diseases, a preset continuous signal with specific characteristics will be adopted to generate a continuous feedback signal that better conforms to the characteristics of the disease, improving the determination accuracy of the model for such special cases. Through such precise and accurate operations, the server can provide rich and accurate instances of invalid continuous feedback signals and valid continuous feedback signals for subsequent determination of glucose feedback signals, thereby continuously optimizing and enhancing the performance and reliability of the glucose feedback signal determination model.

[0069] In the embodiment of the present invention, the step of adding the continuous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid continuous feedback signal instance and the valid continuous feedback signal instance can be implemented through the following examples.

[0070] Determine the weighting coefficient of the continuous feedback signal, and perform a weighting operation on the continuous feedback signal based on the weighting coefficient to obtain a weighted continuous feedback signal;

[0071] Add the weighted continuous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid continuous feedback signal instance and the valid continuous feedback signal instance.

[0072] In an embodiment of the present invention, exemplarily, when the server processes the combination of the continuous feedback signal with the invalid instances and valid instances, it first needs to determine the weighting coefficient of the continuous feedback signal. For example, based on the previous model training effect and the characteristics of the data, the server determines through a series of calculations and analyses that the weighting coefficient of the continuous feedback signal is 0.3. Next, the server performs a weighting operation on the continuous feedback signal based on this weighting coefficient. For example, if the original value of the continuous feedback signal is [100, 200, 300, 400, 500], after multiplying by the weighting coefficient 0.3, the weighted continuous feedback signal is [30, 60, 90, 120, 150]. Then, the server adds this weighted continuous feedback signal to the invalid instances and valid instances respectively. For example, the invalid instance is a set of data on blood glucose concentration changes, such as [12.5, 13.0, 12.8, 12.2, 12.6]. By adding the weighted continuous feedback signal correspondingly, the invalid continuous feedback signal instance obtained is [42.5, 73.0, 102.8, 132.2, 142.6]. For the valid instance, for example, it was originally [5.5, 5.8, 6.0, 5.2, 5.6]. After adding the weighted continuous feedback signal, the valid continuous feedback signal instance obtained is [35.5, 65.8, 96.0, 115.2, 125.6]. In actual operation, the server will process a large number of invalid instances and valid instances simultaneously. For example, the server may have to process tens of thousands of invalid instances and the same number of valid instances at one time. For each invalid instance and valid instance, the server will precisely perform operations according to the above steps. When determining the weighting coefficient, the server will comprehensively consider multiple factors. For example, if the data of the invalid instances fluctuates greatly and the intensity of the continuous feedback signal is also high, in order to avoid excessive influence on the original data, the server will select a smaller weighting coefficient. On the contrary, if the data of the valid instances is relatively stable and it is desired that the continuous feedback signal can have a more significant effect on model training, the weighting coefficient will be appropriately increased. At the same time, the server will monitor and evaluate the instances after adding the weighted continuous feedback signal in real time. For example, the server will check whether the data distribution of the newly generated invalid continuous feedback signal instances and valid continuous feedback signal instances is reasonable and whether it conforms to the expected pattern. For example, during the monitoring process, the server finds that the data of some invalid continuous feedback signal instances shows abnormal aggregation or dispersion after adding the weighted continuous feedback signal. The server will immediately re-evaluate the rationality of the weighting coefficient and may adjust the weighting coefficient, and regenerate and add the weighted continuous feedback signal again. In addition, the server will also dynamically adjust the weighting coefficient and processing method according to different data sources and application scenarios. For example, for data from a specific age group or a specific disease group, the server will adopt different strategies to determine the weighting coefficient to ensure that the model can accurately adapt to different situations.Through such precise and rigorous operations, the server can ensure that the generated invalid continuous feedback signal instances and valid continuous feedback signal instances are accurate and effective, providing reliable data support for the subsequent training and optimization of the glucose feedback signal determination model.

[0073] In an embodiment of the present invention, after optimizing the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function and the first error parameter of the second error function, the embodiment of the present invention further provides the following implementation manners.

[0074] Call the optimized glucose feedback signal determination model to respectively determine the glucose feedback signals for the invalid continuous feedback signal instances and the valid continuous feedback signal instances, and obtain the invalid continuous optimization determination result and the valid continuous optimization determination result correspondingly;

[0075] Based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid continuous optimization determination result, and the valid continuous optimization determination result, determine the second error parameter of the second error function;

[0076] Based on the error parameter of the first error function and the second error parameter of the second error function, optimize the network parameters of the optimized glucose feedback signal determination model.

[0077] In an embodiment of the present invention, exemplarily, after the server completes the preliminary optimization of the network parameters of the glucose feedback signal determination model based on the error parameters of the first error function and the first error parameters of the second error function, it continues with subsequent operations. First, the server calls the preliminarily optimized glucose feedback signal determination model to respectively determine the glucose feedback signals for invalid continuous feedback signal instances and valid continuous feedback signal instances. For example, an invalid continuous feedback signal instance is a series of blood glucose-related data with specific characteristics over a relatively long period, such as continuously low-intensity fluctuations that do not conform to the normal blood glucose change pattern. A valid continuous feedback signal instance, on the other hand, is data that exhibits normal and stable blood glucose changes over the same time length. After the optimized model makes determinations on these instances, the server obtains the invalid continuous optimization determination results and the valid continuous optimization determination results. For example, for a certain invalid continuous feedback signal instance, the model determines it as "invalid"; for a certain valid continuous feedback signal instance, the model determines it as "valid". Next, the server determines the second error parameters of the second error function based on the instance target values of the invalid instances, the instance target values of the valid instances, the just-obtained invalid continuous optimization determination results, and the valid continuous optimization determination results. For example, the instance target value of an invalid instance is set to 0 (indicating invalid), and the instance target value of a valid instance is set to 1 (indicating valid). If for a series of invalid continuous feedback signal instances, the determination results of the model are exactly the same as the instance target values, that is, all are determined as "invalid", then the error in this part is relatively small. However, if there are some instances misjudged as "valid", corresponding errors will occur. Similarly, for valid continuous feedback signal instances, if the determination results of the model are all consistent with the instance target value ("valid"), the error is relatively small; if some are misjudged as "invalid", errors are generated. The server comprehensively considers the differences between all these determination results and the instance target values, and calculates the second error parameters of the second error function through a specific algorithm. Then, based on the error parameters of the first error function and the second error parameters of the second error function, the server re-optimizes the network parameters of the glucose feedback signal determination model that has already been preliminarily optimized. For example, the server will perform a weighted sum of the error parameters of the first error function and the second error parameters of the second error function, and adjust the network parameters of the model according to the sum result. For example, the error parameter of the first error function is 0.1, and the weight is 0.4; the second error parameter of the second error function is 0.05, and the weight is 0.6. The weighted sum result is 0.4×0.1 + 0.6×0.05 = 0.07. Based on this result, the server uses an optimization algorithm, such as the Stochastic Gradient Descent (SGD) algorithm, to adjust the network parameters of the model.For example, if the gradient of a certain network parameter is calculated to be -0.02 and the learning rate is set to 0.01, then the new network parameter value will be updated to the original parameter value + learning rate × gradient. In actual operation, the server will process a large number of invalid continuous feedback signal instances and valid continuous feedback signal instances simultaneously to obtain a more comprehensive and accurate error assessment. For instance, the server processes 1000 invalid continuous feedback signal instances and 1000 valid continuous feedback signal instances. For the determination result of each instance, the server will carefully record and analyze it. If it is found during the optimization process that the performance of the model does not meet the expectation, the server will further adjust the weights of the error function, the learning rate, or re-examine the data preprocessing steps to ensure that the model can more accurately determine the glucose feedback signal. Another example is that the server will compare the current optimization result with the previous optimization result and observe the changing trend of the error parameter. If it is found that the error parameter still does not decrease significantly after multiple optimizations, the server will consider increasing the quantity of training data or adopting a more complex model structure. Through such continuous iterative optimization, the server gradually improves the accuracy and reliability of the glucose feedback signal determination model.

[0078] In an embodiment of the present invention, before the glucose feedback signal determination model is called to determine the glucose feedback signal for invalid instantaneous feedback signal instances and valid instantaneous feedback signal instances respectively, the embodiment of the present invention further provides the following implementation manners.

[0079] Obtain basic invalid instances of multiple information sources, and obtain basic valid instances of the multiple information sources corresponding to the basic invalid instances;

[0080] Perform an integration operation on the basic invalid instances of the multiple information sources to obtain the invalid instances;

[0081] Perform an integration operation on the basic valid instances of the multiple information sources to obtain the valid instances.

[0082] In an embodiment of the present invention, by way of example, before invoking the glucose feedback signal determination model to determine invalid instantaneous feedback signal instances and valid instantaneous feedback signal instances, the server first needs to obtain the basic invalid instances and basic valid instances from multiple information sources and perform an integration operation on them. The server is connected to multiple different data sources, which may include different medical institutions, health monitoring device manufacturers, or research institutions. From the first information source, the server obtains a batch of basic invalid instances. For example, from the database of a large hospital, a series of blood glucose monitoring data are obtained, which are marked as inaccurate or invalid blood glucose measurement results due to equipment failures, special physiological states of patients, or other abnormal conditions. For example, these data are collected at different times, in different departments, for different patients, and contain various possible abnormal situations. Then, from the second information source, such as the cloud platform of a professional health technology company, the server obtains another set of basic invalid instances. This set of data can be collected by a new type of wearable health device, with different measurement accuracies and data formats, but is also marked as invalid. The server obtains basic invalid instances from multiple information sources in the same way. After obtaining the basic invalid instances from multiple information sources, the server starts to perform the integration operation. For example, the server first performs data cleaning on these basic invalid instances from different information sources to remove possible duplicate data, error data, or data with inconsistent formats. Then, the server performs standardization processing on these data, converting data with different measurement units and measurement time intervals into a standard format. For example, among the data from different information sources, some blood glucose values are in millimoles per liter (mmol / L) and some are in milligrams per deciliter (mg / dL), and the server will convert them all to a unified unit. For cases where the measurement time intervals are inconsistent, the server will perform interpolation or sampling processing so that all data have the same time interval. After cleaning and standardization processing, the server combines and summarizes these basic invalid instances to obtain a comprehensive invalid instance that includes a rich variety of invalid situations. Similarly, the server also obtains basic valid instances corresponding to the basic invalid instances from multiple information sources. These basic valid instances may come from the blood glucose monitoring data that have undergone strict quality control and verification in other medical institutions, or are accurate data collected in a controlled experimental environment. The server performs data cleaning, standardization processing, and integration operations similar to those of the basic invalid instances on these basic valid instances from multiple information sources, and finally obtains a comprehensive valid instance that can represent accurate and valid blood glucose monitoring situations. In actual operation, the server can handle massive amounts of data and complex data format differences. For example, the server may obtain data from dozens or even hundreds of different information sources, and each information source may provide thousands of basic instances.During the data cleaning process, the server will use advanced data analysis algorithms and tools to quickly and accurately identify and remove abnormal data. During the standardization process, the server can refer to internationally common medical data standards and specifications to ensure the consistency and comparability of the data. After the integration is completed, the server will also perform quality assessment and verification on the obtained invalid instances and valid instances, such as through random sampling inspection, comparison with known standard data sets, etc., to ensure the accuracy and reliability of the integration results. Through such comprehensive and meticulous data acquisition and integration operations, the server provides a high-quality and widely representative data basis for the training and optimization of the subsequent glucose feedback signal determination model.

[0083] In the embodiment of the present invention, the blood glucose status assessment model is obtained in the following manner.

[0084] Perform a cutting operation on the glucose feedback signal instances in the glucose feedback signal instance set to obtain a first glucose feedback signal segment and a second glucose feedback signal segment;

[0085] Call the first integration unit in the blood glucose status assessment model to process the first glucose feedback signal segment to obtain a first glucose feedback signal feature, and perform autoregressive processing and feature expansion on the first glucose feedback signal feature to obtain a first extended autoregressive feature;

[0086] Call the second integration unit in the blood glucose status assessment model to process the second glucose feedback signal segment to obtain a second glucose feedback signal feature, and perform autoregressive processing and feature expansion on the second glucose feedback signal feature to obtain a second extended autoregressive feature;

[0087] Obtain a first feature vector according to the first extended autoregressive feature and the second extended autoregressive feature of the same glucose feedback signal instance, and obtain a second feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance in the glucose feedback signal instance set and the first extended autoregressive feature or the second extended autoregressive feature of other glucose feedback signal instances;

[0088] Obtain a first cost parameter according to the first feature vector, the second feature vector, and the number of glucose feedback signal instances in the glucose feedback signal instance set, and obtain a second cost parameter according to the first glucose feedback signal feature and the second glucose feedback signal feature;

[0089] Optimize the structure of the first integration unit according to the first cost parameter and the second cost parameter to obtain a target blood glucose status assessment model for assessing the blood glucose health status of the glucose feedback signal of the target user.

[0090] In an embodiment of the present invention, exemplarily, first, the server obtains a set of glucose feedback signal instances. This set contains a large number of glucose feedback signal instances from different users, at different times, and under different monitoring conditions. The server performs a cutting operation on each glucose feedback signal instance in the set. For example, the cutting is based on a time interval. The part of each instance before a specific time point is marked as the first glucose feedback signal segment, and the part after this time point is marked as the second glucose feedback signal segment. For example, for a glucose feedback signal instance that lasts for 10 minutes, the server uses the 5th minute as the cutting point. The data in the first 5 minutes is used as the first glucose feedback signal segment, and the data in the last 5 minutes is used as the second glucose feedback signal segment. Next, the server calls the first integration unit in the blood glucose status evaluation model to process the first glucose feedback signal segment. The first integration unit may include a series of feature extraction algorithms and modules. For example, the first integration unit extracts a series of numerical features through frequency domain analysis, time domain analysis, and statistical calculations of the first glucose feedback signal segment. These features together constitute the first glucose feedback signal feature. Then, autoregressive processing is performed on the first glucose feedback signal feature. This involves establishing a mathematical model to describe the variation law of these features over time and predicting possible future values. Feature expansion is to add some relevant auxiliary features, such as features related to the physiological cycle, meal time, etc., on the basis of autoregressive processing, so as to obtain the first extended autoregressive feature. At the same time, the server calls the second integration unit in the blood glucose status evaluation model to process the second glucose feedback signal segment. The second integration unit may adopt different but complementary feature extraction methods from the first integration unit.

[0091] For example, the second integration unit may focus more on analyzing the morphology and trend of signals, extracting second glucose feedback signal features that are different from but complementary to those of the first integration unit. Similarly, autoregressive processing and feature expansion are performed on the second glucose feedback signal features to obtain second extended autoregressive features. Then, the server obtains a first feature vector based on the first extended autoregressive features and the second extended autoregressive features of the same glucose feedback signal instance. For example, if the first extended autoregressive feature is a vector containing 10 values and the second extended autoregressive feature is a vector containing 8 values, the server concatenates them to form a first feature vector containing 18 values. For the first glucose feedback signal instance in the set, the server obtains a second feature vector based on its first extended autoregressive features and the first extended autoregressive features or the second extended autoregressive features of other glucose feedback signal instances. For example, the server compares and calculates the first extended autoregressive features of the first glucose feedback signal instance with the corresponding features of other 100 instances, and obtains a second feature vector containing multiple values through mathematical methods (such as calculating the mean, variance, covariance, etc.). Next, the server obtains a first cost parameter based on the first feature vector, the second feature vector, and the number of glucose feedback signal instances in the glucose feedback signal instance set. For example, each value in the first feature vector and the second feature vector is transformed and weighted by a function, and then calculations such as summation and averaging are performed in combination with the number of instances, finally obtaining a first cost parameter that reflects the performance of the model in the current training stage. At the same time, the server obtains a second cost parameter based on the first glucose feedback signal features and the second glucose feedback signal features. For example, by calculating the difference, similarity, or distance metric between these two features, the server obtains the second cost parameter, which is used to reflect the accuracy of the model in the feature extraction stage. Finally, the server optimizes the structure of the first integration unit according to the first cost parameter and the second cost parameter. This involves adjusting parameters such as the neuron connection weights, the number of layers, and the activation function in the first integration unit. For example, if the first cost parameter is large, it indicates that the overall performance of the model is not ideal, and the server will significantly adjust the structure of the first integration unit, increasing the number of layers or adjusting the connection method of neurons. If the second cost parameter is large, it indicates that the feature extraction is not accurate enough, and the server will optimize the parameters of the feature extraction algorithm or replace some feature extraction modules. In actual operation, the server will perform multiple such iterative processes. For example, after the first round of optimization, the server recalculates the cost parameters and finds that the model performance has improved, but still does not reach the ideal state, so it continues the next round of optimization, continuously adjusting the structure and parameters of the model until both the first cost parameter and the second cost parameter reach an acceptable small value. At this time, the obtained blood glucose state assessment model can accurately assess the blood glucose health state of the target user's glucose feedback signal.In addition, the server may also consider other factors during the optimization process, such as limitations of computing resources, operating efficiency of the model, adaptability to new data, etc., to ensure that the finally obtained target blood glucose status evaluation model has good performance and practicality in actual applications.

[0092] In the embodiment of the present invention, the glucose feedback signal instances in the glucose feedback signal instance set include first glucose feedback signal instances;

[0093] The operation of performing a cutting operation on the glucose feedback signal instances in the glucose feedback signal instance set to obtain a first glucose feedback signal fragment and a second glucose feedback signal fragment can be implemented through the following examples.

[0094] Performing a cutting operation on the first glucose feedback signal instance according to a first cutting interval to obtain a first glucose feedback signal fragment;

[0095] Performing a cutting operation on the first glucose feedback signal instance according to a second cutting interval to obtain a second glucose feedback signal fragment; wherein, the first cutting interval is smaller than the second cutting interval.

[0096] In an embodiment of the present invention, exemplarily, the server obtains a set containing multiple glucose feedback signal instances, including a first glucose feedback signal instance. For example, the first glucose feedback signal instance is a blood glucose data curve continuously monitored for 30 minutes. The server sets a first cutting interval as the first 10 minutes and a second cutting interval as the first 20 minutes. According to the first cutting interval, the server performs a cutting operation on the first glucose feedback signal instance. This means that the server selects the data segment of the first 10 minutes starting from the time point of the beginning of monitoring and marks it as the first glucose feedback signal segment. For example, the data within these 10 minutes may show relatively stable changes in blood glucose levels or some slight fluctuations. Then, according to the second cutting interval, the server cuts the first glucose feedback signal instance again. This time, the selected data segment is the first 20 minutes and is marked as the second glucose feedback signal segment. Compared with the first glucose feedback signal segment, the second glucose feedback signal segment contains more information, which may include some initial trend changes or short-term blood glucose fluctuation patterns. In actual operation, the server may process a large number of similar glucose feedback signal instances simultaneously. For example, for another first glucose feedback signal instance, it may be a 60-minute monitoring data. The server also cuts it according to the set first cutting interval (such as the first 15 minutes) and the second cutting interval (such as the first 30 minutes). When processing these cutting operations, the server will accurately record the timestamp of each data point and the corresponding blood glucose value to ensure the accuracy of cutting. The server will also perform further analysis and processing on the cut segments. For example, for the first glucose feedback signal segment, the server will calculate basic statistical features such as the average value, maximum value, minimum value, and change slope of the blood glucose value within this period. For the second glucose feedback signal segment, in addition to the above basic statistical features, more complex features such as the periodicity of blood glucose changes within this period and the correlation with specific events (such as diet, exercise) may also be analyzed. By performing such cutting and analysis on different first glucose feedback signal instances, the server can accumulate rich data features and patterns, providing strong support for the training and optimization of subsequent blood glucose status assessment models. In addition, the server will dynamically adjust the lengths of the first cutting interval and the second cutting interval according to different user groups, monitoring device types, or other relevant factors. For example, for a user group with a specific type of diabetes, the server will shorten the first cutting interval to 5 minutes and extend the second cutting interval to 25 minutes to focus more on capturing long-term blood glucose change trends. Another example is that when using a new type of high-precision blood glucose monitoring device that can provide denser and more accurate data, the server will accordingly fine-tune the cutting interval to make full use of the newly added data information. Throughout the process, the server will continuously monitor and evaluate the effect of the cutting operation.If it is found that the settings of certain cutting intervals lead to poor model training effects or inaccurate evaluation results, the server will make timely adjustments and optimizations. Through such a delicate and flexible processing method, the server can effectively extract valuable information from the glucose feedback signal instance set, continuously improving the accuracy and reliability of blood glucose status evaluation.

[0097] In the embodiment of the present invention, the first integration unit in the blood glucose status evaluation model is called to process the first glucose feedback signal segment to obtain the first glucose feedback signal feature, and autoregressive processing and feature expansion are performed on the first glucose feedback signal feature to obtain the first extended autoregressive feature, which can be implemented through the following examples.

[0098] Call the feature extraction component of the first integration unit in the blood glucose status evaluation model to extract features from the first glucose feedback signal segment to obtain the first glucose feedback signal feature;

[0099] Call the autoregressive component of the first integration unit to perform autoregressive processing on the first glucose feedback signal feature to obtain the first autoregressive feature;

[0100] Call the feature conversion component in the first integration unit to perform a feature conversion operation on the first autoregressive feature to obtain the first extended autoregressive feature.

[0101] In an embodiment of the present invention, exemplarily, after the server receives the first glucose feedback signal segment, it starts to call the first integration unit in the blood glucose status evaluation model for processing. First, the server calls the feature extraction component of the first integration unit to extract features from the first glucose feedback signal segment. For example, the first glucose feedback signal segment is a series of continuous blood glucose measurement data over a period of time. The feature extraction component will analyze the statistical features of these data, such as mean, variance, median, etc.; it will also analyze the frequency features of the data to find out the dominant frequency components; and it will extract the trend features of the data, such as rising, falling, or stable trends. Through these analyses, the feature extraction component obtains a series of numerical values that can describe the characteristics of the first glucose feedback signal segment, and these numerical values constitute the first glucose feedback signal features. For example, for a first glucose feedback signal segment containing 100 blood glucose measurement values, the feature extraction component calculates that the mean is 8.5 mmol / L, the variance is 1.2, the median is 8.3 mmol / L, and determines that the data shows a slightly rising trend. These numerical values together form the first glucose feedback signal features. Next, the server calls the autoregressive component of the first integration unit to perform autoregressive processing on the first glucose feedback signal features. The autoregressive component will establish a mathematical model to describe the variation law of these features over time. For example, the autoregressive component uses a second-order autoregressive model. By analyzing the historical data of the first glucose feedback signal features, it calculates the autoregressive coefficients, thus obtaining an autoregressive model that can predict future feature values. After being processed by this model, the first autoregressive features are obtained. For example, according to the previously extracted feature values, the coefficients calculated by the autoregressive model result in a slightly rising trend for future feature values. Then, the server calls the feature transformation component in the first integration unit to perform a feature transformation operation on the first autoregressive features. The feature transformation component will perform linear or non-linear transformations on the first autoregressive features to increase the diversity and expressiveness of the features. It will also combine the first autoregressive features with other relevant physiological features or environmental features to obtain a richer feature representation. After these operations, the first extended autoregressive features are obtained. For example, the feature transformation component performs a logarithmic transformation on the first autoregressive features and combines them with features such as the current user's age and weight to obtain the first extended autoregressive features containing more information. In the actual processing process, the server will process a large number of first glucose feedback signal segments simultaneously. For each segment, the above steps are strictly followed to ensure the accuracy and consistency of feature extraction and transformation. For example, the server receives 1000 first glucose feedback signal segments within one hour, and each segment is processed sequentially by the feature extraction component, the autoregressive component, and the feature transformation component. During the processing, the server will perform quality inspection and verification on the output of each component to ensure that there are no outliers or incorrect calculation results.If some first autoregressive features are found to be inconsistent with expectations during the feature transformation process, the server will roll back to the previous steps, check whether there are problems with feature extraction and autoregressive processing, and make corresponding adjustments and recalculations. At the same time, the server will also dynamically adjust the parameters and algorithms of each component according to the characteristics of different users' features and data. For example, different feature transformation methods will be used for young users and elderly users; for users with large blood glucose fluctuations and relatively stable users, the order and parameters of the autoregressive model will also be different. Through such fine and comprehensive processing, the server can extract first extended autoregressive features with rich information and high expression ability from the first glucose feedback signal segment, providing strong support for subsequent blood glucose status evaluation.

[0102] In the embodiment of the present invention, the second integration unit in the blood glucose status evaluation model is called to process the second glucose feedback signal segment to obtain second glucose feedback signal features, and autoregressive processing and feature extension are performed on the second glucose feedback signal features to obtain second extended autoregressive features, which can be implemented through the following examples.

[0103] Call the feature extraction component of the second integration unit in the blood glucose status evaluation model to extract features from the second glucose feedback signal segment to obtain second glucose feedback signal features;

[0104] Call the autoregressive component of the second integration unit to perform autoregressive processing on the second glucose feedback signal features to obtain second autoregressive features;

[0105] Call the feature transformation component in the second integration unit to perform a feature transformation operation on the second autoregressive features to obtain the second extended autoregressive features.

[0106] In an embodiment of the present invention, exemplarily, after the server receives the second glucose feedback signal segment, it starts to call the second integration unit in the blood glucose status evaluation model for processing. First, the server calls the feature extraction component of the second integration unit to extract features from the second glucose feedback signal segment. For example, the second glucose feedback signal segment is blood glucose monitoring data over a relatively long period, such as consecutive measurement values within one hour. The feature extraction component will use different methods to extract features, such as analyzing the frequency and time characteristics of the data through wavelet transform, or using a convolutional neural network in deep learning to automatically learn the hidden features in the data. Through these methods, the feature extraction component can obtain a set of numerical values that describe the unique properties of the second glucose feedback signal segment, which are the second glucose feedback signal features. For example, the feature extraction component discovers through wavelet transform that the second glucose feedback signal segment has high-frequency fluctuations during certain specific time periods, while showing a relatively stable low-frequency trend during other time periods, and converts this information into specific feature numerical values. Next, the server calls the autoregressive component of the second integration unit to perform autoregressive processing on the second glucose feedback signal features. The autoregressive component will establish a mathematical model suitable for these features, taking into account the temporal correlation and dependence relationships between the features. For example, the autoregressive component uses a high-order autoregressive model that can capture the complex dynamic patterns in the second glucose feedback signal features. Through learning from historical data and parameter estimation, an autoregressive model that can predict future feature changes is obtained, thereby generating the second autoregressive features. For example, the autoregressive model discovers a periodic change pattern in the second glucose feedback signal features and predicts the possible trend of future feature values accordingly. Then, the server calls the feature transformation component in the second integration unit to perform a feature transformation operation on the second autoregressive features. The feature transformation component will use methods such as principal component analysis (PCA) to reduce the dimension of the second autoregressive features to remove redundant information and highlight the main features. Or, by fusing and correlating the second autoregressive features with other relevant physiological indicators (such as insulin level, exercise intensity, etc.), the expansion and enhancement of features are achieved. After these processes, more representative and comprehensive second extended autoregressive features are obtained. For example, the feature transformation component reduces the high-dimensional second autoregressive features to several main components through principal component analysis, and at the same time expands the features by combining the user's insulin injection records, making the features more capable of reflecting the comprehensive situation of blood glucose status. In actual processing, the server can process a large number of second glucose feedback signal segments simultaneously, and each segment comes from different users or different monitoring periods. For each segment, the server will strictly follow the above process to ensure the accuracy and reliability of feature extraction, autoregressive processing, and feature transformation. For example, the server receives thousands of second glucose feedback signal segments in a day. During the processing, the server will conduct quality assessment and verification on the output of each step.If inaccurate feature extraction is detected, the parameters of the feature extraction component will be adjusted or the feature extraction method will be changed. If the effect of autoregressive processing is not ideal, the structure and parameters of the autoregressive model will be re-optimized. At the same time, the server will also dynamically adjust each component of the second integration unit according to different application scenarios and user requirements. For example, for a clinical research scenario that requires a more accurate assessment of blood glucose status, a more complex and refined feature transformation method will be adopted; while for a daily health monitoring scenario, more attention will be paid to computational efficiency and simplicity, and a relatively simple but effective processing method will be selected. Through such a comprehensive and refined processing flow, the server can obtain rich and valuable second extended autoregressive features from the second glucose feedback signal segment, providing strong support for accurately assessing blood glucose health status.

[0107] In the embodiment of the present invention, the glucose feedback signal instances in the glucose feedback signal instance set include a first glucose feedback signal instance and other glucose feedback signal instances;

[0108] The obtaining of the first feature vector according to the first extended autoregressive feature and the second extended autoregressive feature of the same glucose feedback signal instance, and the obtaining of the second feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance in the glucose feedback signal instance set and the first extended autoregressive feature or the second extended autoregressive feature of other glucose feedback signal instances can be implemented through the following examples.

[0109] Obtain a first feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance, the second extended autoregressive feature of the first glucose feedback signal instance, and a preset adjustment coefficient;

[0110] Obtain a second feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance, the preset adjustment coefficient, and the first extended autoregressive feature or the second extended autoregressive feature of other glucose feedback signal instances.

[0111] In an embodiment of the present invention, exemplarily, the server processes a set of glucose feedback signal instances, which includes a first glucose feedback signal instance and other glucose feedback signal instances. First, for the first glucose feedback signal instance, the server obtains its first extended autoregressive feature and second extended autoregressive feature. For example, the first extended autoregressive feature is a set of numerical values describing the long-term trend and periodicity of blood glucose changes, such as [1.2, 0.8, 1.5, 0.9, 1.1], and the second extended autoregressive feature is a set of numerical values reflecting the short-term fluctuations and specific patterns of blood glucose changes, such as [0.3, 0.2, 0.4, 0.1, 0.5]. At the same time, the server also sets a preset adjustment coefficient, such as [0.6, 0.4]. The server performs a weighted calculation on the first extended autoregressive feature and the second extended autoregressive feature according to the preset adjustment coefficient. For example, the first numerical value 1.2 of the first extended autoregressive feature is multiplied by the first numerical value 0.6 of the preset adjustment coefficient, and the first numerical value 0.3 of the second extended autoregressive feature is multiplied by the second numerical value 0.4 of the preset adjustment coefficient, and then the two products are added together to obtain the first numerical value of the first feature vector. Calculate other numerical values in the same way, and finally obtain the first feature vector, such as [0.84, 0.48, 1.02, 0.46, 0.86]. Next, the server obtains the first extended autoregressive feature of the first glucose feedback signal instance, such as [1.2, 0.8, 1.5, 0.9, 1.1], and the first extended autoregressive feature or the second extended autoregressive feature of other glucose feedback signal instances. For example, the corresponding features of other glucose feedback signal instances are [1.0, 0.7, 1.3, 0.8, 1.2] and [0.2, 0.3, 0.5, 0.1, 0.4]. Similarly, the server processes the first extended autoregressive feature of the first glucose feedback signal instance and the features of other glucose feedback signal instances according to the preset adjustment coefficient. For example, for the first numerical value, the numerical value 1.2 of the first extended autoregressive feature of the first glucose feedback signal instance is multiplied by a numerical value of the preset adjustment coefficient (for example, 0.5), and the first numerical value 1.0 of other glucose feedback signal instances is multiplied by another numerical value of the preset adjustment coefficient (for example, 0.5), and then the two products are added together to obtain the first numerical value of the second feature vector. Calculate other numerical values in the same way, and finally obtain the second feature vector, such as [1.1, 0.75, 1.4, 0.85, 1.15]. In actual processing, the server may process a large number of glucose feedback signal instances at the same time. For example, in a large-scale health monitoring project, the server receives glucose feedback signal instances from thousands of users. For the first glucose feedback signal instance of each user, the server accurately calculates the first feature vector and the second feature vector according to the above steps. The server will strictly monitor and verify the calculation process to ensure the accuracy of numerical calculation.If data anomalies are found during the calculation, such as the value of an extended autoregressive feature exceeding a reasonable range, the server will perform data cleaning and recalculation. At the same time, the server will also dynamically adjust the values of the preset adjustment coefficients according to the characteristics and monitoring requirements of different user groups. For example, for user groups with a specific disease, a higher weight will be given to the long-term trend feature, and the preset adjustment coefficient will be adjusted accordingly. Through such fine and accurate calculations, the server can obtain the first feature vector and the second feature vector that can accurately reflect the blood glucose state characteristics, providing strong data support for subsequent blood glucose state evaluation and model optimization.

[0112] In the embodiment of the present invention, obtaining the first cost parameter according to the first feature vector, the second feature vector, and the number of glucose feedback signal instances in the glucose feedback signal instance set can be implemented through the following examples.

[0113] Obtain an intermediate parameter according to the first feature vector and the second feature vector;

[0114] Obtain the first cost parameter based on the intermediate parameter, the number of glucose feedback signal instances in the glucose feedback signal instance set, and a reference value.

[0115] In an embodiment of the present invention, exemplarily, the server has obtained a first feature vector, a second feature vector, and the number of instances in the glucose feedback signal instance set. First, the server obtains an intermediate parameter based on the first feature vector and the second feature vector. For example, the first feature vector is [0.2, 0.3, 0.5, 0.4, 0.6], and the second feature vector is [0.1, 0.2, 0.4, 0.3, 0.5]. The server may obtain the intermediate parameter by calculating the inner product, Euclidean distance, or other mathematical operations of these two vectors. For example, by calculating the inner product of the two vectors, the server multiplies the values at the corresponding positions of the first feature vector and the second feature vector, and then adds these products to obtain an intermediate parameter of 0.9. Next, the server obtains a first cost parameter based on this intermediate parameter, the number of glucose feedback signal instances in the glucose feedback signal instance set, and a reference value. For example, there are 1000 instances in the glucose feedback signal instance set, and the reference value is set to 0.5. The server will first perform some mathematical transformations on the intermediate parameter, such as multiplying by a specific coefficient or performing a logarithmic operation, etc. Then, further calculations are performed in combination with the number of instances and the reference value. For example, the server multiplies the intermediate parameter by 100 to get 90. Then, calculate (90 - 0.5×1000)2 / 1000 to obtain the first cost parameter. In the actual processing process, the server can process multiple different glucose feedback signal instance sets simultaneously. For example, in a large medical data analysis project, the server will face multiple sets from people in different regions, different age groups, and different health conditions. For each set, the server will accurately calculate the intermediate parameter and the first cost parameter according to the above steps. During the calculation process, the server will perform strict quality control and verification on the data. If an abnormal value in a certain feature vector is found, the server will check the data source and processing process to ensure the accuracy and reliability of the data. At the same time, the server will also dynamically adjust the reference value and the specific algorithm for calculating the first cost parameter according to different application scenarios and analysis purposes. For example, a relatively simple calculation method can be used during the initial model training; while for the fine optimization of the model, a more complex and accurate algorithm will be adopted. Through such rigorous and meticulous calculations, the server can obtain accurate first cost parameters, providing an important basis for the optimization and improvement of the subsequent blood glucose status assessment model.

[0116] In an embodiment of the present invention, the following implementation manners are also provided.

[0117] Frame the glucose feedback signal instances in the glucose feedback signal instance set to obtain multiple glucose feedback signals of the glucose feedback signal instances;

[0118] Perform interference processing on the multiple glucose feedback signals to obtain a glucose feedback signal instance containing interference data;

[0119] Performing a cutting operation on the glucose feedback signal instances within the set of glucose feedback signal instances to obtain a first glucose feedback signal fragment and a second glucose feedback signal fragment, including:

[0120] Performing a cutting operation on the glucose feedback signal instance containing interference data to obtain a first glucose feedback signal fragment;

[0121] Performing a cutting operation on the glucose feedback signal instance not containing interference data to obtain a second glucose feedback signal fragment.

[0122] In an embodiment of the present invention, exemplarily, the server first obtains a set of glucose feedback signal instances, which contains a large number of glucose feedback signal instances. The server performs frame segmentation on each glucose feedback signal instance in the set. For example, each glucose feedback signal instance is a continuous time series data. The server divides each instance into multiple small segments according to a fixed time length, such as one second per frame, so as to obtain multiple glucose feedback signals of the glucose feedback signal instance. For example, a glucose feedback signal instance lasting for 10 seconds is segmented by the server into 10 glucose feedback signals with a length of 1 second. Next, the server performs interference processing on these multiple glucose feedback signals obtained by frame segmentation. The interference processing may include operations such as adding random noise, simulating signal attenuation, and introducing spike pulses, so as to obtain glucose feedback signal instances containing interference data. For example, for a certain glucose feedback signal after frame segmentation, the server adds a certain degree of random noise to its value, making the originally stable signal show some fluctuations and uncertainties. Then, the server performs a cutting operation on the glucose feedback signal instance containing interference data to obtain a first glucose feedback signal segment. For example, the basis for cutting is a time point or a threshold of signal strength. The server selects a specific time period or a part that meets specific conditions from the glucose feedback signal instance containing interference data as the first glucose feedback signal segment. For example, for a 5-second-long glucose feedback signal instance affected by interference, the server takes the data from the 2nd second as the starting point and selects the next 2 seconds of data as the first glucose feedback signal segment. At the same time, the server performs a cutting operation on the glucose feedback signal instance that does not contain interference data to obtain a second glucose feedback signal segment. The cutting operation here can be based on the same rules as those for the instance containing interference data, but since the data itself has no interference, the second glucose feedback signal segment obtained by cutting has clearer and more stable characteristics. For example, for a 6-second-long glucose feedback signal instance that is not affected by interference, the server takes the data from the 3rd second as the starting point and selects the next 3 seconds of data as the second glucose feedback signal segment. In actual operation, the server may process thousands of glucose feedback signal instances simultaneously. For example, in a large-scale medical research project, the server needs to process continuous monitoring data from hundreds of patients. For multiple glucose feedback signal instances of each patient, the server strictly operates according to the steps of frame segmentation, interference processing, and cutting. During the frame segmentation process, the server will precisely control the time interval to ensure that the length of each frame is consistent and can accurately reflect the signal changes. During the interference processing, the server will randomly generate interference data and add it to the original signal according to the preset interference mode and intensity to simulate various interference situations that may occur in the real environment.During the cutting operation, the server will flexibly adjust the starting point, ending point, and length of the cut according to the specific requirements and objectives of the research to ensure that the obtained first and second glucose feedback signal segments are representative and valuable for research. At the same time, the server will make detailed records and markings for each processed segment, including information such as the source of the original instance, the type and intensity of interference processing, and the cutting parameters, for subsequent analysis and traceability. If any data anomalies or inconsistencies are found during the processing, the server will conduct error troubleshooting and data cleaning and re-perform the corresponding processing steps to ensure the quality and usability of the finally obtained signal segments. Through such a comprehensive and meticulous processing flow, the server can provide a rich and diverse dataset with practical significance for subsequent blood glucose status assessment and model training.

[0123] In an embodiment of the present invention, the first integration unit in the blood glucose status assessment model is called to process the first glucose feedback signal segment to obtain the first glucose feedback signal feature, which can be implemented through the following examples.

[0124] Call the first integration unit in the blood glucose status assessment model to process the interference data to obtain the first glucose feedback signal feature;

[0125] The calling of the second integration unit in the blood glucose status assessment model to process the second glucose feedback signal segment to obtain the second glucose feedback signal feature includes:

[0126] Based on the interference data in the first glucose feedback signal segment, obtain the data at the same position in the second glucose feedback signal segment;

[0127] Call the second integration unit in the blood glucose status assessment model to process the data to obtain the second glucose feedback signal feature.

[0128] In an embodiment of the present invention, exemplarily, when the server processes the first glucose feedback signal segment and the second glucose feedback signal segment, it calls the integration unit in the blood glucose status evaluation model according to a specific process for feature extraction. For the first glucose feedback signal segment, the server calls the first integration unit in the blood glucose status evaluation model to process the interference data therein. For example, the first glucose feedback signal segment contains interference data generated due to factors such as device noise and environmental interference, such as abnormal spikes or large fluctuations at certain time points. The first integration unit may include algorithms and modules specifically designed to process interference data. For example, it can use a filtering algorithm to remove high-frequency noise, or use methods for outlier detection and correction to smooth those abnormal spikes. Through these processes, the server extracts key features from the interference data that can reflect the blood glucose status, and defines them as the first glucose feedback signal features. For example, after processing, the server extracts features such as the average change trend of blood glucose values, the main fluctuation period, and the difference from common interference patterns from the interference data as the first glucose feedback signal features. For the second glucose feedback signal segment, the server first obtains the data at the same position in the second glucose feedback signal segment based on the interference data in the first glucose feedback signal segment. For example, if the interference data in the first glucose feedback signal segment shows strong fluctuations at a specific position on the time axis, the server will look for similar data patterns at the corresponding time position in the second glucose feedback signal segment. Then, the server calls the second integration unit in the blood glucose status evaluation model to process these data. The second integration unit may adopt processing methods and feature extraction strategies that are different but complementary to those of the first integration unit. For example, the second integration unit may focus more on analyzing the local details of the data, the similarity with historical data, or the association with other physiological indicators. Through the processing of these data, the server obtains the second glucose feedback signal features. In an actual processing scenario, the server may face a large number of first and second glucose feedback signal segments simultaneously. For example, in a large-scale health monitoring system, the server needs to process a large amount of blood glucose data from thousands of users every day. For each data segment of each user, the server strictly processes it according to the above steps. When processing interference data, the server will continuously optimize and adjust the parameters of the processing algorithm to adapt to different types and intensities of interference. For example, if it is found that the interference in a certain batch of data is mainly low-frequency noise, the server will correspondingly adjust the cut-off frequency of the filtering algorithm. At the same time, the server will conduct real-time monitoring and quality assessment on the processing results of the first and second integration units. If it is found that the extracted features do not meet the expectations or are inconsistent with other relevant data, the server will trace back to the processing process, check whether the data processing steps and the application of the algorithm are correct, and make necessary adjustments and recalculations. In addition, the server will also consider the individual differences of users and the changes in their health conditions.For users suffering from specific diseases or in special physiological stages, the server will adopt specially customized processing strategies and feature extraction methods to improve the accuracy and pertinence of blood glucose status assessment. Through such a refined and comprehensive processing process, the server can make full use of the information in the first and second glucose feedback signal segments, extract valuable features, and provide strong support for accurately assessing the blood glucose status.

[0129] In the embodiment of the present invention, obtaining the second cost parameter according to the first glucose feedback signal feature and the second glucose feedback signal feature can be implemented through the following examples.

[0130] Obtain the absolute value of the difference between the first glucose feedback signal feature and the second glucose feedback signal feature;

[0131] Based on the absolute value of the difference and the reference value, obtain the second cost parameter.

[0132] In an embodiment of the present invention, exemplarily, the server has obtained the first glucose feedback signal feature and the second glucose feedback signal feature. First, the server calculates the absolute value of the difference between the first glucose feedback signal feature and the second glucose feedback signal feature. For example, the first glucose feedback signal feature is a set of numerical values, such as [1.5, 2.0, 3.5, 4.0, 5.0], and the second glucose feedback signal feature is [1.2, 2.2, 3.0, 4.5, 5.5]. The server subtracts the numerical values at the corresponding positions in these two features and takes the absolute value of the difference. For example, the absolute value of the difference at the first position is |1.5 - 1.2| = 0.3, and the absolute value of the difference at the second position is |2.0 - 2.2| = 0.2, and so on, to obtain the set of absolute values of the differences [0.3, 0.2, 0.5, 0.5, 0.5]. Next, the server obtains the second cost parameter based on these absolute values of the differences and a reference value. For example, the reference value is set to 0.4. The server first performs operations such as summing or averaging the absolute values of the differences. For example, first calculate the average value of the above set of absolute values of the differences, and get (0.3 + 0.2 + 0.5 + 0.5 + 0.5) / 5 = 0.4. Then, the server compares and calculates this average value with the reference value. For example, the square of (average value - reference value) can be calculated, that is, (0.4 - 0.4)^2 = 0, and this is used as the second cost parameter. In actual operations, the server can process a large number of pairs of the first and second glucose feedback signal features simultaneously. For example, in a large-scale health monitoring project, the server receives multiple sets of feature data from thousands of users every day. For each set of feature data, the server strictly calculates according to the above steps. When calculating the absolute value of the difference, the server uses a high-precision numerical calculation method to ensure the accuracy of the result. If it is found that the absolute values of some differences are too large or too small, the server will conduct further analysis. It will check whether there are abnormalities in the data collection process or whether the feature extraction algorithm needs to be adjusted. At the same time, the server will dynamically adjust the reference value according to different application scenarios and requirements. For example, for a group of patients with more severe conditions, the reference value will be lowered to improve the sensitivity to blood glucose changes. In addition, the server will also use the calculated second cost parameter for subsequent model optimization and evaluation. If the second cost parameter remains large, the server will consider that there are problems with the model in processing these two features and needs to improve and adjust the model. Through such precise and accurate calculations and analyses, the server can effectively utilize the differences between the first and second glucose feedback signal features to obtain a reliable second cost parameter, providing strong support for the improvement and optimization of the blood glucose status evaluation model.

[0133] In an embodiment of the present invention, the present invention embodiment also provides the following implementation manners.

[0134] Obtain a third feature vector based on the first extended autoregressive feature, the second extended autoregressive feature of the same glucose feedback signal instance, and the number of glucose feedback signal instances in the set of glucose feedback signal instances;

[0135] Obtain a correlation matrix based on the first extended autoregressive feature and the second extended autoregressive feature of the same glucose feedback signal instance, and obtain a fourth feature vector based on the correlation matrix;

[0136] Obtain a third cost parameter based on the third feature vector and the fourth feature vector;

[0137] The structural optimization of the first integration unit according to the first cost parameter and the second cost parameter includes:

[0138] Perform structural optimization on the first integration unit according to the first cost parameter, the second cost parameter, and the third cost parameter.

[0139] In an embodiment of the present invention, exemplarily, the server first processes each instance in the set of glucose feedback signal instances. For a specific glucose feedback signal instance, the server has obtained its first extended autoregressive feature and second extended autoregressive feature. For example, the first extended autoregressive feature of this glucose feedback signal instance is a set of values [1.2, 0.8, 1.5, 0.9, 1.1], and the second extended autoregressive feature is [0.7, 0.6, 0.8, 0.5, 0.9]. The server obtains a third feature vector based on the first extended autoregressive feature, the second extended autoregressive feature of the same glucose feedback signal instance, and the number of glucose feedback signal instances in the set of glucose feedback signal instances. For example, the server first calculates some statistics of these two extended autoregressive features, such as the mean, variance, etc. The mean of the first extended autoregressive feature is 1.0, and the variance is 0.1; the mean of the second extended autoregressive feature is 0.7, and the variance is 0.1. For example, there are a total of 500 instances in the set of glucose feedback signal instances. Then, the server combines and calculates these statistics and the number of instances, such as multiplying the mean by the number of instances, to obtain [500, 350]. Then, some normalization or transformation processing is performed on it, and finally a third feature vector is obtained, such as [0.5, 0.35]. Next, the server obtains a correlation matrix based on the first extended autoregressive feature and the second extended autoregressive feature of the same glucose feedback signal instance. The server calculates the correlation between these two features to obtain a correlation matrix. For example, this correlation matrix is [[1, 0.8], [0.8, 1]]. Then, the server performs further processing based on this correlation matrix, such as calculating the eigenvalues of the matrix or performing a form of decomposition, so as to obtain a fourth feature vector. For example, the fourth feature vector obtained after processing is [0.6, 0.4]. The server obtains a third cost parameter based on the third feature vector and the fourth feature vector. For example, the server can calculate the distance or difference measure between these two feature vectors, such as the Euclidean distance. For example, the third feature vector is [0.5, 0.35], and the fourth feature vector is [0.6, 0.4], and their Euclidean distance is √((0.5 - 0.6)2 + (0.35 - 0.4)2) = 0.141. The server can directly use this distance as the third cost parameter, or perform some transformation and adjustment on it. When optimizing the structure of the first integration unit, the server comprehensively considers the first cost parameter, the second cost parameter, and the third cost parameter. For example, the first cost parameter reflects the error in the overall prediction of the model, and the value is 0.2; the second cost parameter reflects the accuracy of feature extraction, and the value is 0.15; the third cost parameter reflects the degree of relationship matching between features, and the value is 0.1. The server will assign different weights to these cost parameters, such as the weight of the first cost parameter is 0.5, the weight of the second cost parameter is 0.3, and the weight of the third cost parameter is 0.2.Then, the server calculates the weighted sum: 0.2 × 0.5 + 0.15 × 0.3 + 0.1 × 0.2 = 0.145. Based on this comprehensive cost evaluation result, the server optimizes the structure of the first integration unit. This can include operations such as adjusting the connection weights of neurons, adding or deleting neurons, changing the number of network layers, etc. In actual operations, the server may process a large number of glucose feedback signal instances simultaneously, and perform the above-mentioned feature calculations and cost evaluations for each instance. For example, on a dataset containing 1000 instances, the server processes each instance in turn, accumulates the statistical information of the cost parameters, and determines the direction and magnitude of the structure optimization based on the overall evaluation result. If it is found during the optimization process that the cost parameters do not decrease significantly, the server will try different weight allocation strategies or deeply analyze the data features to find possible problems. At the same time, the server will also consider the computing resources and time costs, and select appropriate optimization algorithms and parameters on the premise of ensuring the optimization effect. Through such a comprehensive consideration and optimization process, the server can continuously improve the performance and accuracy of the blood glucose state evaluation model.

[0140] In an embodiment of the present invention, the embodiment of the present invention also provides the following implementation manners.

[0141] Perform type recognition on the first autoregressive feature obtained by autoregressive processing of the first glucose feedback signal feature to obtain a first confidence level;

[0142] Perform type recognition on the second autoregressive feature obtained by autoregressive processing of the second glucose feedback signal feature to obtain a second confidence level;

[0143] Obtain a fourth cost parameter based on the first confidence level and the second confidence level;

[0144] The optimizing the structure of the first integration unit according to the first cost parameter, the second cost parameter, and the third cost parameter includes:

[0145] Optimize the structure of the first integration unit according to the first cost parameter, the second cost parameter, the third cost parameter, and the fourth cost parameter.

[0146] In an embodiment of the present invention, exemplarily, during the process of processing the glucose feedback signal features, the server performs a series of complex operations to optimize the model structure. First, the server performs autoregressive processing on the first glucose feedback signal feature to obtain the first autoregressive feature. Then, a specific type recognition algorithm is used to analyze the first autoregressive feature. For example, the first autoregressive feature presents a relatively stable and predictable pattern. By matching and comparing its pattern, the type recognition algorithm determines that it belongs to a specific type, such as the "normal blood glucose fluctuation type". At the same time, the algorithm gives a value representing the confidence level of this judgment, that is, the first confidence level. For example, this first confidence level is 0.85, indicating that the server has a relatively high certainty about this type judgment. Similarly, the server performs autoregressive processing on the second glucose feedback signal feature to obtain the second autoregressive feature, and performs type recognition on it to obtain the second confidence level. For example, the second autoregressive feature presents a relatively complex and less common pattern, and after type recognition, it is classified as the "potential blood glucose abnormality type", and the corresponding second confidence level is 0.7. Based on the first confidence level and the second confidence level, the server obtains the fourth cost parameter. For example, the server can set a rule to convert and weight the confidence levels to obtain the fourth cost parameter. For example, the server multiplies the confidence levels by 10 and takes the absolute value of their difference, that is, |0.85×10 - 0.7×10| = 1.5, and uses this value as the fourth cost parameter. When optimizing the structure of the first integration unit, the server no longer relies solely on the first cost parameter, the second cost parameter, and the third cost parameter, but comprehensively considers the fourth cost parameter. For example, the first cost parameter reflects the deviation of the overall prediction of the model, with a value of 0.2; the second cost parameter reflects the accuracy of feature extraction, with a value of 0.15; the third cost parameter reflects the degree of relationship matching between features, with a value of 0.1; the fourth cost parameter is calculated as 1.5 as described above. The server assigns corresponding weights to each cost parameter. For example, the weight of the first cost parameter is 0.4, the weight of the second cost parameter is 0.3, the weight of the third cost parameter is 0.2, and the weight of the fourth cost parameter is 0.1. Then, the server calculates their weighted sum: 0.2×0.4 + 0.15×0.3 + 0.1×0.2 + 1.5×0.1 = 0.345. According to this comprehensive cost evaluation result, the server optimizes the structure of the first integration unit. This involves adjusting the connection weights between neurons. For example, if it is found that certain connection weights have a greater negative impact on the cost evaluation result, the server will reduce the values of these weights. Or, the server may add or delete neurons to change the processing ability and complexity of the first integration unit. For example, currently, the first integration unit has 100 neurons, and the server decides to delete 10 neurons that contribute less to the result based on the cost evaluation result to simplify the model structure and improve the calculation efficiency. In addition, the server may also change the number of network layers.If it is found that the current number of layers cannot effectively process data, the server may increase or decrease the number of layers to optimize the performance of the model. In actual operation, the server may process a large number of different glucose feedback signal data simultaneously. For example, in a large medical database, there are blood glucose monitoring data of thousands of patients. For each set of data, the server performs the above-mentioned feature processing, confidence calculation, and cost evaluation, and continuously optimizes the first integration unit according to the comprehensive results. During the optimization process, if the server finds that although the first integration unit has been adjusted multiple times, the cost parameter still does not decrease significantly, the server will re-examine the algorithms for feature processing and type recognition to check for errors or areas that need improvement. At the same time, the server will also consider the characteristics of different patient groups and the differences in data distribution. For certain special types of patients, such as children with diabetes or elderly diabetic patients, the server will analyze and optimize them separately according to their data characteristics to ensure that the model can provide accurate blood glucose status assessments for different groups. Through such a comprehensive, detailed, and dynamic optimization process, the server can continuously improve the performance of the first integration unit, thereby enhancing the accuracy and reliability of the entire blood glucose status assessment model.

[0147] An embodiment of the present invention provides a computer device 100. The computer device 100 includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the foregoing user health status management method based on an intelligent wearable device. As Figure 2 shown, Figure 2 is a structural block diagram of the computer device 100 provided by an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To achieve data transmission or interaction, the elements of the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines.

[0148] For illustrative purposes, the previous description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. The embodiments are selected and described to best illustrate the principles of the disclosure and its practical applications, so that those skilled in the art can best utilize the disclosure and use various embodiments with different modifications to suit the particular applications contemplated.

Claims

1. A method for managing the health status of users based on intelligent wearable devices, characterized in that Including: In response to a health status assessment instruction triggered by a preset period, initialize the radio frequency transmission module and the feedback signal receiving module of the smart wearable device; Transmit an electromagnetic wave signal with a preset wavelength to a preset body area of the target user through the radio frequency transmission module; Receive a to-be-determined feedback signal corresponding to the electromagnetic wave signal with the preset wavelength through the feedback signal receiving module; Call a pre-trained glucose feedback signal determination model to determine the glucose feedback signal for the to-be-determined feedback signal, and obtain a determination result for characterizing whether the to-be-determined feedback signal includes a glucose feedback signal; In the case where the determination result indicates the existence of a glucose feedback signal, obtain the glucose feedback signal to be processed; Call a pre-trained target blood glucose status assessment model to process the glucose feedback signal and output an autoregressive feature of the feedback signal; Perform a comparison operation on the autoregressive feature of the feedback signal and the pre-stored blood glucose health status feature to obtain a comparison result; When the comparison result indicates that the feature distance between the autoregressive feature of the feedback signal and the blood glucose health status feature is less than a preset feature distance threshold, obtain the blood glucose health status corresponding to the glucose feedback signal.

2. The method according to claim 1, wherein The glucose feedback signal determination model is obtained through the following method, including: Call the glucose feedback signal determination model to respectively determine the glucose feedback signal for an invalid instantaneous feedback signal instance and a valid instantaneous feedback signal instance, and correspondingly obtain an invalid instantaneous determination result and a valid instantaneous result; Among them, the invalid instantaneous feedback signal instance is obtained by adding an instantaneous feedback signal to an invalid instance, and the valid instantaneous feedback signal instance is obtained by adding an instantaneous feedback signal to a valid instance corresponding to the invalid instance; Based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid instantaneous determination result, and the valid instantaneous result, determine the error parameter of the first error function; Call the glucose feedback signal determination model to respectively determine the glucose feedback signal for an invalid continuous feedback signal instance and a valid continuous feedback signal instance, and correspondingly obtain an invalid continuous determination result and a valid continuous determination result; Among them, the invalid continuous feedback signal instance is obtained by adding a continuous feedback signal to the invalid instance, and the valid continuous feedback signal instance is obtained by adding a continuous feedback signal to the valid instance; Based on the instance target value of the invalid continuous feedback signal instance, the instance target value of the valid continuous feedback signal instance, the invalid continuous determination result, and the valid continuous determination result, determine the first error parameter of the second error function; Based on the error parameter of the first error function and the first error parameter of the second error function, optimize the network parameter of the glucose feedback signal determination model; Call the optimized glucose feedback signal determination model to respectively determine the glucose feedback signal for the invalid continuous feedback signal instance and the valid continuous feedback signal instance, and correspondingly obtain an invalid continuous optimized determination result and a valid continuous optimized determination result; Determine the second error parameter of the second error function based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid continuous optimization determination result, and the valid continuous optimization determination result; Optimize the network parameters of the optimized glucose feedback signal determination model based on the error parameter of the first error function and the second error parameter of the second error function to obtain the trained glucose feedback signal determination model.

3. The method according to claim 2, characterized in that The method further includes: Call the glucose feedback signal determination model to perform glucose feedback signal determination on the invalid instance and the valid instance corresponding to the invalid instance respectively, and obtain the invalid instance determination result and the valid instance determination result correspondingly; Determine the first error parameter of the third error function based on the instance target value of the invalid instance, the instance target value of the valid instance, the invalid instance determination result, and the valid instance determination result; The optimizing the network parameters of the glucose feedback signal determination model based on the error parameter of the first error function and the first error parameter of the second error function includes: Obtain the weighting coefficient of the first error function, the weighting coefficient of the second error function, and the weighting coefficient of the third error function respectively; Obtain the error parameter of the total error function based on the error parameter and the weighting coefficient of the first error function, the first error parameter and the weighting coefficient of the second error function, and the first error parameter and the weighting coefficient of the third error function; Optimize the network parameters of the glucose feedback signal determination model according to the error parameter of the total error function.

4. The method according to claim 3, characterized in that, Before the calling the glucose feedback signal determination model to perform glucose feedback signal determination on the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance respectively, the method further includes: Exchange the instance target value of the invalid instance and the instance target value of the valid instance to obtain the enhanced instance target value of the invalid instance and the enhanced instance target value of the valid instance; Determine the second error parameter of the third error function based on the enhanced instance target value of the invalid instance, the enhanced instance target value of the valid instance, the invalid instance determination result, and the valid instance determination result; Determine the parameter update direction of the third error function for the network parameters according to the second error parameter of the third error function to obtain the optimization guidance quantity of the glucose feedback signal determination model; Determine the optimization guidance quantity of the glucose feedback signal determination model as the instantaneous feedback signal; Determine the weighting coefficient of the instantaneous feedback signal, and perform a weight assignment operation on the instantaneous feedback signal based on the weighting coefficient to obtain the weighted instantaneous feedback signal; Add the weighted instantaneous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance; Before the calling the glucose feedback signal determination model to perform glucose feedback signal determination on the invalid instantaneous feedback signal instance and the valid instantaneous feedback signal instance respectively, the method further includes: Obtain the basic invalid instances of multiple information sources, and obtain the basic valid instances of the multiple information sources corresponding to the basic invalid instances; Perform an integration operation on the basic invalid instances of the multiple information sources to obtain the invalid instances; Perform an integration operation on the basic valid instances of the multiple information sources to obtain the valid instances.

5. The method according to claim 2, wherein The determination of the glucose feedback signal is implemented by calling the discriminant component in the glucose feedback signal determination model, and the glucose feedback signal determination model further includes a synthesis component; Before the glucose feedback signal determination model is called to perform glucose feedback signal determination on the invalid continuous feedback signal instances and the valid continuous feedback signal instances respectively, the method further includes: Call the synthesis component in the glucose feedback signal determination model to generate a continuous feedback signal based on the input preset continuous signal; Determine the weighting coefficient of the continuous feedback signal, and perform a weighting operation on the continuous feedback signal based on the weighting coefficient to obtain a weighted continuous feedback signal; Add the weighted continuous feedback signal to the invalid instance and the valid instance respectively to obtain the invalid continuous feedback signal instance and the valid continuous feedback signal instance.

6. The method according to claim 1, characterized in that, The blood glucose status evaluation model is obtained through the following methods, including: Perform a cutting operation on the glucose feedback signal instances in the glucose feedback signal instance set to obtain a first glucose feedback signal segment and a second glucose feedback signal segment; the glucose feedback signal instances in the glucose feedback signal instance set include a first glucose feedback signal instance and other glucose feedback signal instances; Call the first integration unit in the blood glucose status evaluation model to process the first glucose feedback signal segment to obtain a first glucose feedback signal feature, and perform autoregressive processing and feature expansion on the first glucose feedback signal feature to obtain a first extended autoregressive feature; Call the feature extraction component of the second integration unit in the blood glucose status evaluation model to extract features from the second glucose feedback signal segment to obtain a second glucose feedback signal feature; Call the autoregressive component of the second integration unit to perform autoregressive processing on the second glucose feedback signal feature to obtain a second autoregressive feature; Call the feature conversion component in the second integration unit to perform a feature conversion operation on the second autoregressive feature to obtain a second extended autoregressive feature; Obtain a first feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance, the second extended autoregressive feature of the first glucose feedback signal instance, and a preset adjustment coefficient; Obtain a second feature vector according to the first extended autoregressive feature of the first glucose feedback signal instance, the preset adjustment coefficient, and the first extended autoregressive feature or the second extended autoregressive feature of the other glucose feedback signal instances; Obtain an intermediate parameter according to the first feature vector and the second feature vector; A first cost parameter is obtained based on the intermediate parameter, the number of glucose feedback signal instances in the set of glucose feedback signal instances, and a reference value, and a second cost parameter is obtained according to the first glucose feedback signal feature and the second glucose feedback signal feature; Based on the first cost parameter and the second cost parameter, the structure of the first integration unit is optimized to obtain a target blood glucose status evaluation model for evaluating the blood glucose health status of the glucose feedback signal of a target user.

7. The method according to claim 6, wherein The first integration unit in the blood glucose status evaluation model is called to process the first glucose feedback signal segment to obtain a first glucose feedback signal feature, and autoregressive processing and feature expansion are performed on the first glucose feedback signal feature to obtain a first extended autoregressive feature, including: Calling a feature extraction component of the first integration unit in the blood glucose status evaluation model to extract features from the first glucose feedback signal segment to obtain a first glucose feedback signal feature; Calling the autoregressive component of the first integration unit to perform autoregressive processing on the first glucose feedback signal feature to obtain a first autoregressive feature; Calling a feature transformation component in the first integration unit to perform a feature transformation operation on the first autoregressive feature to obtain the first extended autoregressive feature.

8. The method according to claim 6, wherein Performing a cutting operation on the glucose feedback signal instances in the set of glucose feedback signal instances to obtain a first glucose feedback signal segment and a second glucose feedback signal segment, including: Performing a cutting operation on the first glucose feedback signal instance according to a first cutting interval to obtain a first glucose feedback signal segment; Performing a cutting operation on the first glucose feedback signal instance according to a second cutting interval to obtain a second glucose feedback signal segment; wherein, the first cutting interval is smaller than the second cutting interval.

9. The method according to claim 6, wherein The method further includes: Framing the glucose feedback signal instances in the set of glucose feedback signal instances to obtain a plurality of glucose feedback signals of the glucose feedback signal instances; Performing interference processing on the plurality of glucose feedback signals to obtain glucose feedback signal instances containing interference data; Performing a cutting operation on the glucose feedback signal instances in the set of glucose feedback signal instances to obtain a first glucose feedback signal segment and a second glucose feedback signal segment, including: Performing a cutting operation on the glucose feedback signal instance containing interference data to obtain a first glucose feedback signal segment; Performing a cutting operation on the glucose feedback signal instance not containing interference data to obtain a second glucose feedback signal segment; Calling the first integration unit in the blood glucose status evaluation model to process the first glucose feedback signal segment to obtain a first glucose feedback signal feature, including: Calling the first integration unit in the blood glucose status evaluation model to process the interference data to obtain a first glucose feedback signal feature; Calling the second integration unit in the blood glucose status evaluation model to process the second glucose feedback signal segment to obtain a second glucose feedback signal feature, including: Based on the interference data in the first glucose feedback signal segment, obtain the data at the same position in the second glucose feedback signal segment; Invoke the second integration unit in the blood glucose status evaluation model to process the data to obtain the second glucose feedback signal feature; The obtaining of the second cost parameter according to the first glucose feedback signal feature and the second glucose feedback signal feature includes: Obtain the absolute value of the difference between the first glucose feedback signal feature and the second glucose feedback signal feature; Based on the absolute value of the difference and the reference value, obtain the second cost parameter; The method further includes: Obtain a third feature vector according to the first extended autoregressive feature, the second extended autoregressive feature of the same glucose feedback signal instance, and the number of glucose feedback signal instances in the glucose feedback signal instance set; Obtain a correlation matrix according to the first extended autoregressive feature and the second extended autoregressive feature of the same glucose feedback signal instance, and obtain a fourth feature vector based on the correlation matrix; Obtain a third cost parameter according to the third feature vector and the fourth feature vector; The structurally optimizing the first integration unit according to the first cost parameter and the second cost parameter includes: Structurally optimize the first integration unit according to the first cost parameter, the second cost parameter, and the third cost parameter; The method further includes: Perform type recognition on the first autoregressive feature obtained by autoregressive processing of the first glucose feedback signal feature to obtain a first confidence level; Perform type recognition on the second autoregressive feature obtained by autoregressive processing of the second glucose feedback signal feature to obtain a second confidence level; Obtain a fourth cost parameter based on the first confidence level and the second confidence level; The structurally optimizing the first integration unit according to the first cost parameter, the second cost parameter, and the third cost parameter includes: Structurally optimize the first integration unit according to the first cost parameter, the second cost parameter, the third cost parameter, and the fourth cost parameter.

10. A server system, characterized in that, Includes a server, and the server is configured to execute the method according to any one of claims 1-9.

Images