Insulin pump control system infusion parameter generation method and device based on timing law

Abstract

This invention provides a method and apparatus for generating infusion parameters for an insulin pump control system based on time-series patterns, relating to the field of infusion parameter adjustment technology. The implementation process of this solution is as follows: During the generation of infusion parameters for the insulin pump control system, a time-series dual-mode adaptive basal rate selection is invoked to choose between the first-day basal rate initialization mode and the second-day basal rate iteration mode. After completing the aforementioned time-series dual-mode adaptive basal rate selection, a multi-period coupling analysis of the blood glucose time-series signal, extracting periodic components at different time scales from the blood glucose monitoring signal, is invoked. Subsequently, multi-period coupling discrimination of the blood glucose time-series signal is performed. The discrimination result determines whether to invoke blood glucose time-series coupling correction. If correction is required, blood glucose time-series coupling correction is performed first, and then infusion parameters are generated based on the corrected result. If correction is not required, infusion parameters are directly generated. This invention can improve the accuracy of infusion parameters.

Description

Technical Field

[0001] This invention relates to the field of infusion parameter adjustment technology, and in particular to a method and apparatus for generating infusion parameters for an insulin pump control system based on time-series patterns. Background Technology

[0002] With the continuous innovation of continuous glucose monitoring and insulin delivery methods, the closed-loop pancreatic system that integrates glucose monitoring, intelligent control algorithms and precise insulin infusion has become an important technological direction for improving blood glucose control and reducing related metabolic disorders. However, the inherent delay in signal transmission and drug onset between glucose monitoring signals and insulin infusion, inter-individual and intra-individual metabolic differences, and the rhythmic fluctuations of daily behavioral factors pose continuous challenges to the precise regulation of blood glucose throughout the day.

[0003] Existing insulin pump control systems, such as the Apexel Smart Insulin Pump, mainly consist of a dynamic blood glucose monitoring device, a control algorithm processing unit, and an insulin pump. The dynamic blood glucose monitoring device continuously measures the glucose concentration in the subcutaneous interstitial fluid using an electrochemical sensor and generates a blood glucose monitoring signal, which is then sent to the control algorithm processing unit. The control algorithm processing unit receives the acquired blood glucose monitoring signal, combines it with preset parameters, and executes the control algorithm to generate an infusion command. The insulin pump receives the infusion command and drives the pump mechanism via a micro-motor to continuously or on-demand perform insulin infusion operations.

[0004] Existing insulin pump control systems have implemented infusion control methods that construct infusion models based on pre-collected population clinical data and call these models for infusion control by inputting parameters such as the basal daily total dose. For example, a method for controlling basal rate infusion in an insulin pump, as disclosed in Chinese invention patent application CN103083751A, includes the following steps: establishing a basal rate infusion model based on a large amount of clinical data; storing the basal rate infusion model in an insulin pump storage device; selecting the basal rate infusion model by a control device based on signals input from the insulin pump input device; and controlling the insulin pump drive propulsion device to infuse insulin based on the insulin type, basal daily total dose, and selected basal rate infusion model input by the input device.

[0005] Existing insulin pump control systems have implemented infusion control methods that utilize models based on insulin type and basal daily dose to control insulin pump infusion. However, existing methods often rely on specific pathological features, such as absolute insulin deficiency, and lack adaptability to differences in insulin resistance and residual function. They also lack adequate compensation mechanisms for specific time periods in the circadian rhythm, such as the early morning. Furthermore, the generation of infusion parameters may involve multiple periodic rhythmic components superimposed to form a superimposed blood glucose time-series signal. Examples include a 24-hour basal circadian rhythm, a 3-5 hour postprandial rapid response rhythm, a 6-12 hour exercise-induced post-exercise effect rhythm, and a long-cycle rhythm of approximately 28 days. When nonlinear coupling exists between different cycles—meaning the amplitude, phase, or waveform shape of one cycle's rhythm is modulated by the current state of another cycle—the combined effect is not equal to the linear sum of their individual effects, typically exhibiting a non-linear relationship. Additive interaction effects, such as the coupling of post-exercise effect rhythm with circadian rhythm, mean that the timing of exercise (morning, afternoon, or night) has a nonlinear modulating effect on the depth and timing of subsequent blood glucose troughs. For example, the likelihood of nighttime blood glucose signals falling below a preset threshold after evening exercise is higher than after morning exercise, and the delay in the occurrence of blood glucose troughs relative to the end of exercise also varies nonlinearly with the exercise time. Existing methods typically use models based on a single dominant cycle, such as those considering only circadian rhythms or linear models based on simple linear superposition of multiple cycles, to generate predicted blood glucose values. This can lead to low accuracy in the generated predicted blood glucose values. When generating infusion parameters based on these low-accuracy predicted blood glucose values, such as nighttime basal rate or pre-meal bolus dose estimates, the generated infusion parameters may not match actual needs. For instance, during the nighttime period after evening exercise, the linear model may underestimate the magnitude of the blood glucose signal drop, resulting in overestimated infusion parameters and potentially causing abnormally low signal states within the predicted period, ultimately leading to insufficient accuracy in the generated infusion parameters. Summary of the Invention

[0006] Therefore, embodiments of the present invention provide a method and apparatus for generating infusion parameters of an insulin pump control system based on time-series patterns, which can improve the accuracy of infusion parameters.

[0007] The technical solution of this invention is implemented as follows: This invention provides a method for generating infusion parameters for an insulin pump control system based on time-series patterns, comprising: initiating a time-series dual-mode adaptive basal rate selection for choosing between the first-day basal rate initialization mode and the second-day basal rate iteration mode during the generation of infusion parameters for the insulin pump control system; after completing the above-mentioned time-series dual-mode adaptive basal rate selection, performing multi-period coupling analysis of blood glucose time-series signals by extracting periodic components of different time scales from blood glucose monitoring signals, followed by multi-period coupling discrimination of blood glucose time-series signals; determining whether to initiate blood glucose time-series coupling correction based on the discrimination result; if correction is required, performing blood glucose time-series coupling correction first, and then generating infusion parameters based on the corrected result; if correction is not required, directly generating infusion parameters.

[0008] This application also provides an insulin pump control system infusion parameter generation device based on time-series patterns. The device includes: a basal rate mode adaptive selection module, a multi-cycle coupling monitoring module, a multi-cycle coupling correction module, and an infusion parameter generation module. The basal rate mode adaptive selection module initiates a time-series dual-mode basal rate adaptive selection during the insulin pump control system infusion parameter generation process, selecting between the first-day basal rate initialization mode and the second-day basal rate iteration mode. The multi-cycle coupling monitoring module, after completing the aforementioned time-series dual-mode basal rate adaptive selection, performs multi-cycle coupling analysis of blood glucose time-series signals by extracting periodic components from different time scales from the blood glucose monitoring signal, followed by multi-cycle coupling discrimination of the blood glucose time-series signal. The multi-cycle coupling correction module determines whether to initiate blood glucose time-series coupling correction based on the discrimination result. The infusion parameter generation module, if correction is required, first performs blood glucose time-series coupling correction, then generates infusion parameters based on the corrected result; if correction is not required, it directly generates infusion parameters.

[0009] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: 1. By initiating a time-series dual-mode adaptive basal rate selection for choosing between the initial basal rate mode on the first day and the iterative basal rate mode on the second day, it helps to quickly match the initial basal rate curve when there is no historical data, avoiding the risk of blood glucose signal fluctuations caused by blind parameter setting. Simultaneously, when historical data is available, the daily updates of infusion parameters can accurately follow the slow drift of blood glucose time-series patterns, shortening the time for blood glucose to enter a stable control window. This adaptive selection strategy summarizes the basal rate and personalized modes based on historical data, such as the dosage requirements corresponding to exercise events, thereby achieving adaptive infusion parameter configuration without relying on additional input. Furthermore, the iterative basal rate mode on the second day uses a blood glucose sampling scheme at seven preset key time points daily to replace the dynamic blood glucose monitoring in existing technologies, achieving higher measurement accuracy than continuous blood glucose monitoring devices based on interstitial fluid measurement, thus ensuring data reliability. After completing the above time-series dual-mode adaptive basal rate selection, multi-period coupling analysis of blood glucose time-series signals, extracting periodic components at different time scales from the blood glucose monitoring signals, is conducted, followed by analysis of the blood glucose time-series signals. Multi-period coupling discrimination helps to automatically separate periodic components at different time scales from raw blood glucose monitoring data and quantitatively assess the nonlinear interaction strength between these components. This reveals blood glucose fluctuation patterns that cannot be captured by single-period or linear superposition models without relying on additional input information. Simultaneously, it can dynamically adjust model parameters according to the gradual changes in blood glucose time series. The longer the usage period, the more accurate the adaptation to metabolic rhythms and daily behavioral patterns. The discrimination result determines whether to initiate blood glucose time series coupling correction. If correction is required, it is performed first, and then infusion parameters are generated based on the corrected result. If correction is not required, infusion parameters are generated directly. This helps to compensate for additional blood glucose fluctuation offsets caused by multi-period nonlinear coupling while ensuring computational efficiency. The basal rate segment value continuously approaches the optimal infusion baseline through daily iterations, and the pre-meal high dose is continuously optimized through successive feedback to match the dose requirements under different activity levels. This adaptively approaches the optimal infusion scheme throughout the day, improving the matching accuracy between infusion parameters and actual blood glucose needs.

[0010] 2. Targeted acquisition of intrinsic mode function components and blood glucose temporal coupling feature vectors, including high-frequency temporal components, diurnal rhythm components, exercise aftereffect components, and long-period components, helps to overcome the limitations of existing technologies that rely solely on a single period, such as diurnal rhythms or simple linear superposition models. This allows for the comprehensive extraction of multi-timescale rhythmic information and its nonlinear interaction characteristics implicit in blood glucose monitoring signals, thus more precisely characterizing the temporal patterns of individual blood glucose fluctuations and avoiding feature distortion caused by mode aliasing or period omission. Based on the blood glucose temporal coupling feature vectors, multi-period coupling discrimination of blood glucose temporal signals is performed to obtain… The Euclidean distance between the blood glucose time-series coupled feature vector and the preset time-series coupled feature vector is denoted as the time-series coupled Euclidean distance. If the time-series coupled Euclidean distance is greater than the preset distance threshold, blood glucose time-series coupling correction is performed. Otherwise, infusion parameters are generated based on the blood glucose prediction value. This helps to overcome the shortcomings of existing methods that perform complex coupling calculations indiscriminately in all time periods or ignore coupling effects. Adaptive correction decisions are achieved through quantitative discrimination, reducing the risk of over-adjustment introduced by blind correction. This makes the generation of infusion parameters more robust and accurate, and adapts to the time-varying characteristics of multi-period coupling in real-life scenarios.

[0011] 3. When blood glucose monitoring signals are subject to strong high-frequency noise interference, or when sampling intervals fluctuate irregularly due to equipment limitations, online empirical mode decomposition (EMD) is sensitive to noise and its boundary effects can easily lead to mode aliasing between different frequency components. This may cause the decomposed intrinsic mode function components to lose their clear physical periodic meaning, resulting in a significant decrease in the accuracy of the blood glucose time-series coupled feature vector. Therefore, an alternative scheme for multi-period coupling analysis of blood glucose time-series signals is needed. This scheme uses the blood glucose monitoring sequence as input to a pre-defined multi-scale time-series feature extraction network, and outputs four multi-scale time-series feature maps representing local change patterns at different time scales. This helps to adaptively extract features related to rapid fluctuations, postprandial response, post-exercise effects, and daytime variations. The feature map corresponding to the nocturnal rhythm is used to maintain the stability and discriminative power of signal features under strong noise or irregular sampling environments, thereby improving the robustness of subsequent coupling analysis. This overcomes the shortcomings of existing technologies that are highly sensitive to data quality. Four multi-scale time series feature maps are spliced ​​along the channel dimension to form a multi-scale feature tensor. The multi-scale feature vector is input into a preset feedforward neural network, which outputs the blood glucose prediction value at the next sampling time. This helps to implicitly learn the nonlinear coupling relationship between features at different scales using an end-to-end deep network architecture, reducing multi-step cumulative error and maintaining high prediction accuracy under noise interference. This provides an optional and highly robust coupling analysis path for insulin pump control systems, broadening the applicable scenarios of the method. Attached Figure Description

[0012] Figure 1This is a flowchart of a method for generating infusion parameters for an insulin pump control system based on time-series patterns, provided in an embodiment of the present invention. Figure 2 This is a flowchart summarizing the method for generating infusion parameters for an insulin pump control system based on time-series patterns, provided in this embodiment of the invention. Figure 3 This is a logic diagram of the method for generating infusion parameters for an insulin pump control system based on time-series patterns, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the infusion parameter generation device for an insulin pump control system based on time-series patterns provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0015] Example 1, as Figure 1The flowchart of the insulin pump control system infusion parameter generation method based on time-series patterns shows that: Basal rate mode adaptive selection is initiated during the insulin pump control system infusion parameter generation process. This involves a time-series dual-mode adaptive basal rate selection process, firstly choosing between the first-day basal rate initialization mode and the second-day basal rate iteration mode. This ensures that both the initialization and daily iteration of infusion parameters conform to the actual time-series patterns. The first-day basal rate initialization mode is used to match the initial basal rate curve from multiple preset basal rate profiles based on the current blood glucose value when no historical blood glucose data is available, enabling a safe start-up of the insulin pump control system from zero. The second-day basal rate iteration mode is used to form a closed-loop optimization mechanism that continuously adapts to the evolution of blood glucose time-series patterns based on the previous day's total basal rate. (Multi-cycle...) Coupling monitoring: After completing the adaptive selection of the dual-mode basal rate in the time series, multi-period coupling analysis of the blood glucose time series signal is carried out to extract periodic components of different time scales from the blood glucose monitoring signal and quantify the nonlinear coupling effect between the components. Subsequently, multi-period coupling discrimination of the blood glucose time series signal is performed to determine whether there is a significant nonlinear coupling bias in the current time period. Multi-period coupling correction: The discrimination result determines whether to activate blood glucose time series coupling correction to explicitly compensate for the additional blood glucose fluctuation offset caused by nonlinear coupling into the blood glucose prediction, so as to improve the prediction accuracy. Infusion parameter generation: If it is determined that correction is needed, blood glucose time series coupling correction is performed first, and then infusion parameters are generated based on the correction result. If it is determined that no correction is needed, infusion parameters are generated directly.

[0016] It should be understood that the insulin pump control system infusion parameter generation method based on time-series patterns provided in this application embodiment can, in a specific embodiment, pre-construct an infusion parameter storage database that stores a series of key control parameters and historical blood glucose regulation operation records. The series of key control parameters are pre-calibrated by a professional team in the fields of endocrine metabolism regulation and medical device control, such as preset time period rhythm coefficients, preset blood glucose correction coefficients, preset specific time periods in the early morning, and preset fixed compensation amounts. The historical blood glucose regulation operation records cover blood glucose time series, blood glucose response curves, post-exercise blood glucose recovery waveforms, actual execution trajectories of 24-hour segmented basal rates, pre-meal high-dose trigger frequencies and their corresponding blood glucose prediction deviation sequences under various behavioral patterns. In addition, the infusion parameter storage database also saves updated data generated by the continuous evolution of the model, such as calibration records of the correspondence between intrinsic mode function components and physiological cycles in online empirical mode decomposition, dynamic correction values ​​of Euclidean distance thresholds in multi-cycle coupling discrimination, and the history of coupling coefficient adjustments in the incremental learning process of nonlinear regression models.

[0017] At the physical storage level, the infusion parameter storage database adopts a hybrid partitioning architecture: the relational database is responsible for maintaining quantitative indicators with fixed formats, such as daily fasting blood glucose target values, changes in total basal rate before and after iteration, corresponding statistical values ​​of blood glucose prediction deviation and adjustment dose, and daily change rates of basal rate segments for each time period; while the non-relational database is used to store high-dimensional or unstructured information, such as the original blood glucose monitoring signal sequence within the sliding window, the instantaneous amplitude and instantaneous phase waveforms of each intrinsic mode function component after empirical mode decomposition, the high-dimensional feature mapping of coupled feature vectors, and the time series graph of the residuals of blood glucose prediction values ​​before and after correction. Based on the multi-dimensional feedback data collected from the real-time insulin pump control process, including the execution feedback of the multi-cycle coupled monitoring module, the preset parameters stored in the infusion parameter storage database can be periodically reviewed and adaptively fine-tuned, so that the time-series regulation feature library can continuously adapt to different physiological rhythm differences, slow drift of insulin resistance levels, time-varying sensor noise characteristics, and complex and ever-changing daily behavioral disturbances.

[0018] It should also be understood that the method for generating infusion parameters for an insulin pump control system based on time-series patterns provided in this application, which involves various preset thresholds such as preset time period rhythm coefficients, preset blood glucose correction coefficients, preset specific time periods in the early morning, preset fixed compensation amounts, preset distance thresholds, preset blood glucose intervals, preset lower limit protection values ​​of basal rate, and preset upper limit protection values ​​of basal rate, follows a unified offline calibration and online verification process: First, based on the basic physiological principles of blood glucose regulation, such as the diurnal trend of insulin sensitivity changes, the typical time window for enhanced endogenous glucose output at dawn, the common fasting blood glucose distribution range, and the hardware specifications of the insulin pump, such as minimum infusion accuracy, maximum safe infusion rate, and storage capacity, the initial suggested values ​​or ranges for each preset value are initially set by experts in the field of medical device control; then, in a standard simulation environment, such as using UVa-Padova approved by the U.S. Food and Drug Administration. A virtual patient simulator for type 2 diabetes was developed, and offline simulation tests were conducted under typical lifestyle patterns and dietary / exercise parameters. Various preset values ​​were progressively adjusted, and performance indicators such as the percentage of time the blood glucose target was achieved, the incidence of hypoglycemic events, the incidence of hyperglycemic events, and the convergence speed of infusion parameters were monitored. The preset values ​​were iteratively optimized using grid search or Bayesian optimization methods until the system achieved preset performance targets, such as a blood glucose target achievement rate greater than 70%, a hypoglycemic event rate less than 5%, and a basal rate convergence period of less than seven days. The optimized parameter values ​​were stored in an infusion parameter storage database, along with corresponding simulation conditions, such as the degree of insulin resistance and residual pancreatic function. Subsequently, whenever the system was used with new hardware configurations, such as changing to a different precision insulin pump or sensor type, the above offline calibration process was repeated. Simultaneously, during actual clinical use, real-time feedback data such as the amplitude of blood glucose fluctuations, the frequency of hypoglycemic warnings, and the magnitude of infusion parameter adjustments were monitored online to periodically fine-tune and adaptively update the preset values, ensuring that the preset values ​​strictly match the current time-series patterns and system hardware characteristics.

[0019] In this embodiment, by adaptive selection of basal rate mode, multi-cycle coupling monitoring, multi-cycle coupling correction, and infusion parameter generation, the insulin pump control system's ability to adapt to multi-cycle temporal patterns such as individual circadian rhythms, short-term responses, and post-exercise effects is improved. This enables the generation of infusion parameters to respond in real time to the nonlinear interactions implied in blood glucose signals, thereby maintaining high prediction accuracy and regulatory stability in all weather conditions and multiple scenarios. This effectively reduces the risk of excessive blood glucose fluctuations caused by cycle coupling deviations, ultimately achieving precise matching between infusion parameters and actual metabolic needs, and enhancing the blood glucose control effectiveness and long-term safety of the insulin pump control system.

[0020] like Figure 2The flowchart illustrating the general overview of the insulin pump control system infusion parameter generation method based on time-series patterns shows the following steps: Adaptive selection of the time-series dual-mode basal rate is performed. It checks if basal rate data from the previous day exists. If so, the first-day basal rate initialization mode is executed; otherwise, the first total basal rate value is obtained. Based on the first total basal rate value, the next-day basal rate iteration mode is executed, obtaining the second total basal rate value and the initial basal rate. Basal rate correction is performed based on the initial basal rate to obtain the time-series basal rate curve. Dawn phenomenon forced compensation is then performed to obtain the basal rate compensation curve. The adaptive selection of the time-series dual-mode basal rate is then completed. Multi-period coupling analysis of blood glucose time-series signals is performed, along with temporal mode decomposition and Hilbert transform, ultimately yielding a blood glucose time-series coupling feature vector. Based on this feature vector, multi-period coupling discrimination of the blood glucose time-series signal is conducted, determining whether the Euclidean distance of the time-series coupling satisfies a condition greater than a preset distance threshold. If so, blood glucose time-series coupling correction is performed, and the corrected blood glucose prediction value is used as the final blood glucose prediction value. Infusion parameters are then generated based on the blood glucose prediction value. Otherwise, a linear extrapolation prediction value is directly obtained and used as the blood glucose prediction value, and infusion parameters are generated based on the blood glucose prediction value.

[0021] Furthermore, the specific process of adaptive basal rate selection in the time-series dual-mode is as follows: Read the historical basal rate records stored locally, determine if there is basal rate data from the previous day, including the total basal rate value of the previous day and the corresponding 24-hour segmented basal rate array; if there is no basal rate data from the previous day, execute the first-day basal rate initialization mode; if there is basal rate data from the previous day, obtain the total basal rate value of the previous day, record it as the first total basal rate value, and execute the next-day basal rate iteration mode based on the first total basal rate value; the first-day basal rate initialization mode is used to achieve a smooth transition for first-time use, and the specific process is as follows: Obtain the first fasting blood glucose value, and based on the preset blood glucose range in which the first fasting blood glucose value is located, select from a preset number of preset time-series basal rates... A corresponding total basal rate value and basal rate array are matched in the rate profile. Multi-period coupled analysis of blood glucose time-series signals is performed based on the total basal rate value and basal rate array. The preset blood glucose intervals and preset quantities are pre-set by preset personnel. The first fasting blood glucose value is obtained by a continuous glucose monitor at a preset morning fasting sampling time point in the first-day basal rate initialization mode. The preset morning fasting sampling time point is also pre-set by preset personnel. The preset time-series basal rate profile refers to a pre-stored set of standardized basal rates covering different fasting blood glucose levels. Each basal rate profile corresponds to a total basal rate value used to set the total daily insulin infusion volume and a set of hourly segmented basal rate arrays indicating the infusion rate for each hour, with units of U / h. The basal rate array contains basal rate values ​​for 24 time periods, and the basal rate distribution for each time period simulates the diurnal fluctuation characteristics of the physiological rhythm. The basal rate array of the selected basal rate profile for 24 hours is used as the basal rate output for the day and stored as a historical basal rate record. In an optional embodiment, the preset time-series basal rate profile obtained based on the first fasting blood glucose value can be set as follows: 1. First fasting blood glucose value < 7.0 mmol / L: total basal rate 6.1U, corresponding to a 24-hour basal rate array of [0.2, 0.2, 0.2, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.3, 0.3, 0.3, 0.3, 0.2, 0.2, 0.2, 0.2, 0.3 ... 1. 3, 0.3, 0.2, 0.2, 0.2]; 2. First fasting blood glucose value 7.0~8.5mmol / L: total basal rate 8.4U, corresponding to the 24-hour basal rate array [0.3, 0.3, 0.3, 0.4, 0.4, 0.4, 0.4, 0.4, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.4, 0.4, 0.4, 0.4, 0.3, 0.3, 0.3]; 3. First fasting blood glucose value 8.5~9.0mmol / L: total basal rate 10.0U, corresponding to the 24-hour basal rate array [0.4, 0.4, 0.4, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.4, 0.5].4,0.4,0.4,0.4,0.4,0.4,0.5,0.5,0.5,0.5,0.5,0.5,0.4,0.4,0.4]; 4. First fasting blood glucose value 9.0~12.0mmol / L: total basal rate 13.5U, corresponding to the 24-hour basal rate array [0.5,0.5,0.5,0.6,0.6,0.6,0.6,0.6,0.6,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.6,0.6,0.7,0.7,0.6,0.6,0.5,0.5]; 5. First fasting blood glucose value 12.0~14.0mmol / L: total basal rate 16.0U, corresponding to the 24-hour basal rate array [0.5,0.5,0.5,0.6,0.6,0.5,0.5]; The basal rate array is [0.6,0.6,0.6,0.7,0.7,0.7,0.7,0.7,0.7,0.7,0.6,0.6,0.6,0.6,0.6,0.6,0.6,0.7,0.7,0.8,0.8,0.7,0.7,0.6,0.6]; 6. First fasting blood glucose value > 14.0 mmol / L: total basal rate 18.3U, corresponding to the 24-hour basal rate array [0.7,0.7,0.7,0.8,0.8,0.8,0.8,0.8,0.8,0.7,0.7,0.7,0.7,0.7,0.7,0.7,0.7,0.8,0.8,0.9,0.9,0.8,0.8,0.7,0.7].

[0022] Furthermore, the specific process of the next-day basal rate iteration mode is as follows: Obtain the second fasting blood glucose value; the second fasting blood glucose value is obtained by a continuous glucose monitor at a preset morning fasting sampling time point in the next-day basal rate iteration mode; obtain the second total basal rate value based on the first total basal rate value, and apply safety boundary constraints during the acquisition of the second total basal rate value. The specific acquisition process is as follows: if the second fasting blood glucose value is less than the preset first low basal rate value, then the first total basal rate value is compared with the preset first high basal rate compensation value using a deviation calculation, and the result of the deviation calculation is used as the candidate second total basal rate value; if the second fasting blood glucose value is not less than the preset first low basal rate value and not greater than the preset second low basal rate value, then the first total basal rate value is adjusted accordingly. The first total basal rate is used as a candidate second total basal rate value after a deviation calculation is performed between the first total basal rate value and a preset first low basal rate compensation value. If the second fasting blood glucose value is greater than the preset second low basal rate value but less than the preset basal rate value, then the first total basal rate value is used as a candidate second total basal rate value. If the second fasting blood glucose value is not less than the preset basal rate value and not greater than the preset second high basal rate value, then the first total basal rate value is summed with the preset low basal rate compensation value, and the result is used as a candidate second total basal rate value. If the second fasting blood glucose value is not less than the preset second high basal rate value and not greater than the preset high basal rate value, then the first total basal rate value is summed with the preset first low basal rate compensation value, and the result is used as the candidate second total basal rate value. Candidate second total basal rate value; if the second fasting blood glucose value is greater than the preset high basal rate value, then the first total basal rate value and the preset first high basal rate compensation value are summed, and the result of the summation is used as the candidate second total basal rate value; the preset first high basal rate compensation value is greater than the preset first low basal rate compensation value, and is greater than the preset low basal rate compensation value, and all are preset by preset personnel; the preset high basal rate value is greater than the preset second high basal rate value, greater than the preset basal rate value, greater than the preset second low basal rate value, and greater than the preset first low basal rate value, and all are preset by preset personnel; in a specific embodiment, the process of obtaining the second total basal rate value is as follows: 1. Second fasting blood glucose value < 5.1 mmol / L: candidate second total basal rate value 1. Second total basal rate value = first total basal rate value - 4.0U; 2. Second fasting blood glucose value 5.1~6.3mmol / L: candidate second total basal rate value = first total basal rate value - 2.0U; 3. Second fasting blood glucose value 6.3~7.6mmol / L: candidate second total basal rate value = first total basal rate value; 4. Second fasting blood glucose value 7.6~8.3mmol / L: candidate second total basal rate value = first total basal rate value + 0.4U; 5. Second fasting blood glucose value 8.3~10.5mmol / L: candidate second total basal rate value = first total basal rate value + 2.0U; 6. Second fasting blood glucose value > 10.5mmol / L: candidate second total basal rate value = first total basal rate value + 4.0U.

[0023] Specifically, the process for safety boundary constraints is as follows: If the candidate second total basal rate value is less than the preset lower limit protection value, then the second total basal rate value is set to the preset lower limit protection value; if the candidate second total basal rate value is greater than the preset upper limit protection value, then the second total basal rate value is set to the preset upper limit protection value; if the candidate second total basal rate value is neither less than the preset lower limit protection value nor greater than the preset upper limit protection value, then the corresponding candidate second total basal rate value is set to the second total basal rate value. The preset lower limit protection value is used to prevent the problem of elevated blood glucose monitoring values ​​due to an excessively low basal rate, and the preset upper limit protection value is used to prevent the problem of decreased blood glucose monitoring values ​​due to an excessively high basal rate. Both the preset lower limit protection value and the preset upper limit protection value are preset by designated personnel.

[0024] The time-series basal rate curve is obtained based on the second total basal rate value. The specific acquisition process is as follows: Divide the second total basal rate value (after safety boundary constraints) by 24 to obtain the average hourly basal rate; multiply the average hourly basal rate by the corresponding preset time-period rhythm coefficient to obtain the initial basal rate for each hour; the preset time-period rhythm coefficient represents a set of pre-set rhythm values ​​that correspond one-to-one with the hour sequence number. This value is lower than a preset low rhythm threshold in the early morning period to simulate diurnal insulin sensitivity variations. In a specific embodiment, the preset time-period rhythm coefficient can be set to [0.85, 0.85, 0.85, 0.85, 0.90, 1.00, 1.10, 1.15, 1.10, 1.00, 0.9]. [5,0.95,0.95,0.95,0.95,0.95,0.95,1.00,1.05,1.10,1.15,1.10,1.00,0.90,0.85]; Basal rate correction is performed based on the initial basal rate, specifically as follows: Obtain the second fasting blood glucose value; Determine whether the second fasting blood glucose value is higher than the preset hyperglycemia threshold. If so, multiply the initial basal rate by a preset blood glucose correction coefficient greater than 1 to obtain the time-series basal rate correction value; otherwise, compare the second fasting blood glucose value with the preset hypoglycemia threshold. Both the preset hyperglycemia threshold and the preset hypoglycemia threshold are pre-set by preset personnel; If the second fasting blood glucose value is lower than the preset hypoglycemia threshold, then the initial basal rate is adjusted accordingly. The basal rate is multiplied by a preset blood glucose correction factor less than 1 to obtain the time-series basal rate correction value. If the second fasting blood glucose value is not higher than the preset hyperglycemia threshold and not lower than the preset hypoglycemia threshold, the initial basal rate remains unchanged, and the corresponding initial basal rate is used as the time-series basal rate correction value. The time-series basal rate correction value is multiplied by 10 and then rounded. The result of the rounding operation is divided by 10 to obtain the time-series basal rate rounded value. The 24-hour time-series basal rate rounded values ​​are combined in hourly order, for example, from 0:00 to 23:00, to form a time-series basal rate curve containing 24 data points. Since there is a regular upward trend in the blood glucose monitoring data during the early morning period, approximately 3:00 to 6:00, if... The time-series basal rate curve generated by the above steps alone may not be sufficient to offset the signal rise during this period. Therefore, it is necessary to perform forced compensation for the dawn phenomenon in the preset specific time period in the early morning to obtain the basal rate compensation curve. Based on the second total basal rate value and the basal rate compensation curve, multi-cycle coupling analysis of blood glucose time-series signals is performed. The forced compensation for the dawn phenomenon means that in each hour corresponding to the preset specific time period in the early morning, such as the 3rd, 4th and 5th hours, corresponding to the time-series basal rate curve from 3:00 to 6:00, a preset fixed compensation amount, such as 0.1 U / h, is added to the rounded value of the time-series basal rate. This preset fixed compensation amount is a fixed value independent of all correction coefficients and is used to directionally increase the infusion rate during this period. It is set in advance by the preset personnel.

[0025] In this embodiment, the time-series dual-mode adaptive basal rate selection helps to quickly invoke the first-day basal rate initialization mode when there is no historical data. It matches a safe and standardized basal rate profile based on the current fasting blood glucose value, reducing the blind infusion caused by missing parameters. When historical data is available, it switches to the next-day basal rate iteration mode, forming a history-driven, daily-updated adaptive closed loop. This allows the basal rate parameter to continuously approach the optimal value as blood glucose drifts slowly over time. The next-day basal rate iteration mode uses a fine division of six levels, which helps to achieve graded response to different degrees of deviation. This makes the basal rate iteration optimization process clearly interpretable, improving the adaptability of the insulin pump control system and the feasibility of its promotion at the grassroots level.

[0026] Furthermore, the specific process of multi-period coupling analysis of blood glucose time-series signals is as follows: Step 1, acquire the blood glucose monitoring sequence for the data acquisition period; the blood glucose monitoring sequence represents a sequence composed of a series of blood glucose concentration values ​​arranged in chronological order; Step 2, perform time-series signal mode decomposition on the blood glucose monitoring sequence based on the online empirical mode decomposition algorithm to obtain each intrinsic mode function component, including time-series high-frequency components, diurnal rhythm components, exercise aftereffect components, and long-period components; the time-series high-frequency components represent the rapidly changing blood glucose signal fluctuation components within several hours. For example, the time-series high-frequency components can be represented as [h1, h2, h3, ..., h n ], where n is the total number of sampling points within the sliding window, h_k represents the instantaneous amplitude of the component at the k-th sampling time, and k represents the sampling point number, unit: mmol / L; the circadian rhythm component represents the time-related fluctuation component of the basic blood glucose signal with a period of approximately 24 hours within a day; the exercise post-effect component represents the blood glucose shift caused by changes in insulin sensitivity lasting for several hours to tens of hours after exercise; the long-period component represents the slow trend change spanning several weeks or even longer timescales; Step 3, based on the online Hilbert transform algorithm, analyze each intrinsic mode function component The analytic signal is obtained by performing a Hilbert transform; the magnitude of the analytic signal is taken as the instantaneous amplitude of the blood glucose time series; the principal argument of the analytic signal is taken as the instantaneous phase of the blood glucose time series; the instantaneous amplitude of the blood glucose time series represents the energy intensity of the current intrinsic mode function component at the current moment. For example, at a certain sampling moment, the instantaneous amplitude of the diurnal rhythm component is 2.3 mmol / L, indicating that the contribution intensity of this component to the original blood glucose signal at that moment is 2.3 mmol / L; the instantaneous phase of the blood glucose time series represents the oscillation position of the current intrinsic mode function component at the current moment, ranging from 0 to 2π.

[0027] Specifically, the process of temporal signal mode decomposition is as follows: Obtain all local extrema of the blood glucose monitoring data during the data acquisition period, including local maxima and local minima; based on the cubic spline interpolation function, connect all local maxima to form an upper envelope, and connect all local minima to form a lower envelope, ensuring that all data points of the current signal to be decomposed lie between these two envelopes; the current signal to be decomposed refers to the signal sequence that has not yet been decomposed and is currently in the screening stage during the current iteration of intrinsic mode function component extraction. Initially, the current signal to be decomposed is the blood glucose monitoring sequence, i.e., the original blood glucose time-series signal; for each acquisition... At each sampling time, the blood glucose concentration values ​​of the upper and lower envelopes at that time are added together and divided by 2 to obtain the mean blood glucose concentration value at that time. The mean blood glucose concentration values ​​at all times are then connected sequentially to form a mean curve, which is used as the local average trend estimation curve for the current signal to be decomposed. The mean blood glucose concentration value of the current signal to be decomposed at each sampling time is subtracted from the mean blood glucose concentration value of the mean curve at the same time to obtain the first-order candidate component. It is then determined whether the current first-order candidate component satisfies the intrinsic mode function (IMF) criterion. If it does not, the first-order candidate component is used to replace the signal to be decomposed, and the time-series signal mode decomposition is repeated. If the number of repetitions is greater than [a certain number], the process continues. If a maximum number of attempts is preset, a mode decomposition failure warning for the timing signal is sent. If the condition is met, the current first-order candidate component is the first intrinsic mode function (IMF) component extracted, and IMF components are extracted sequentially based on the first IMF component. The IMF determination condition indicates that during the data acquisition period, the number of all local extrema and the number of zero-crossings must be equal or differ by at most one, and at any time point, the local mean of the upper and lower envelopes is zero. The specific process of sequentially extracting IMF components is as follows: the first extracted IMF component is subtracted from the current signal to be decomposed to obtain the remaining component; the remaining component is used as the new signal to be decomposed. The signal undergoes repetitive temporal signal mode decomposition, sequentially extracting the next intrinsic mode function component. When the remaining components meet the decomposition stopping condition, the temporal signal mode decomposition process terminates. At this point, multiple intrinsic mode function components arranged from high to low frequency have been obtained. The decomposition stopping condition includes the number of extreme points of the remaining components being less than a preset extreme point number threshold, or the signal energy of the remaining components being lower than a preset signal energy threshold. The preset extreme point number threshold is set in advance by a preset person, and the preset signal energy threshold is represented by the average value of the remaining component signal energy over a historical time period. The signal energy of the remaining components is obtained by summing the squares of the blood glucose concentration values ​​at each sampling point in the remaining component sequence.

[0028] Step 4: Pair each intrinsic mode function (IMF) component in pairs according to frequency from high to low, and obtain the instantaneous phase difference and instantaneous amplitude ratio of each pair of IMF components. The instantaneous phase difference is represented by the absolute value of the subtraction of the instantaneous phase values ​​of each pair of IMF components. The instantaneous amplitude ratio is represented by the ratio of the instantaneous amplitude of each pair of IMF components. The instantaneous amplitude, instantaneous phase, instantaneous phase difference, and instantaneous amplitude ratio of each IMF component are sorted together in the order of these parameters to form a blood glucose time-series coupling feature vector. Based on this feature vector, multi-cycle coupling discrimination of the blood glucose time-series signal is performed. The specific process is as follows: The Euclidean distance between the blood glucose time-series coupling feature vector and the preset time-series coupling feature vector is obtained using the Euclidean distance algorithm, and this distance is marked as the time-series coupling feature vector. The Euclidean distance is used, where the preset temporal coupling feature vector is pre-set by preset personnel. If the temporal coupling Euclidean distance is greater than the preset distance threshold, it indicates that there is a significant multi-cycle nonlinear coupling deviation in the current time period, and blood glucose temporal coupling correction needs to be performed. The preset distance threshold is represented by the average value of the temporal coupling Euclidean distance of the historical time period. Otherwise, it indicates that the components of each cycle in the current time period approximately satisfy a linear superposition relationship. The linear extrapolation prediction value of each intrinsic mode function component at the next sampling time is obtained. The linear extrapolation prediction values ​​of all intrinsic mode function components at the same time are summed to obtain the linear superposition prediction value of blood glucose. The linear superposition prediction value of blood glucose is used as the blood glucose prediction value, and the infusion parameters are generated based on the blood glucose prediction value. The specific method of obtaining the linear extrapolation prediction value is as follows: for the current intrinsic mode function component, the intrinsic mode function component value at the current time, i.e., the previous preset time, is taken and input into the preset autoregressive model to obtain the linear extrapolation prediction value of the current intrinsic mode function component at the next sampling time.

[0029] It should be noted that the autoregressive model preset in this embodiment is specifically a second-order autoregressive model. In another optional embodiment, a first-order autoregressive model, a third-order autoregressive model, or a state-space model based on exponential smoothing can also be used. The specific parameter estimation and training process of the second-order autoregressive model is as follows: First, collect the continuous sampled value sequence of the intrinsic mode function component within the historical time window to form a training sample set. Each training sample consists of values ​​from three consecutive sampling times: the values ​​from the first two times are used as the model input, and the value from the last time is used as the prediction target. Then, the least squares method is used. To estimate the unknown parameters of the model, specifically, for each sample consisting of values ​​from the previous two time points and the next time point, a linear equation is established containing a constant term, two autoregressive coefficients, and random errors. By solving for the parameter combination that minimizes the sum of squared errors for all samples, the optimal estimates of the constant term and the two autoregressive coefficients are obtained. The above parameter estimation can be performed offline once every preset period, or updated online using the recursive least squares method. After training, the constant term and the two autoregressive coefficients are fixed and can then be used for real-time linear extrapolation prediction. The preset period is set in advance by preset personnel, for example, 24 hours.

[0030] Furthermore, the specific process of blood glucose temporal coupling correction is as follows: The blood glucose temporal coupling feature vector is input into a preset nonlinear regression model, such as a random forest or a third-order polynomial regressor, which outputs blood glucose nonlinear coupling correction coefficients. These coefficients represent the correction strength to be superimposed under the current coupling state; a positive sign indicates upward correction, and a negative sign indicates downward correction. In actual operation, each time a new blood glucose temporal coupling feature vector is obtained, the model immediately outputs the corresponding blood glucose nonlinear coupling correction coefficient. The blood glucose nonlinear coupling correction coefficient is multiplied by the corresponding preset baseline correction amplitude to obtain the blood glucose coupling correction term. This term represents the additional blood glucose fluctuation offset caused by the superposition of multiple periods of nonlinearity. The preset baseline correction amplitude is represented by the average instantaneous amplitude of each intrinsic mode function component over a historical time period. The blood glucose linear superposition prediction value is added to the blood glucose coupling correction term to obtain the corrected blood glucose prediction value.

[0031] It should be noted that the preset nonlinear regression model in this application embodiment is specifically a third-order polynomial regressor. In an optional embodiment, a random forest regressor or a support vector regressor can also be used. Specifically, the training process of the preset nonlinear regression model is as follows: First, collect blood glucose monitoring sequences within a historical sliding window to obtain the blood glucose temporal coupling feature vector at each sampling time. Simultaneously, calculate the difference between the actual blood glucose value at the next sampling point and the linearly superimposed predicted value, using this difference as a supervision label. Use the blood glucose temporal coupling feature vector as the model input and the supervision label as the model output target. Divide all samples into training and validation sets according to time order. Use regularized least squares to estimate parameters, including the third-order polynomial regression model. The constant term, the coefficients of the first, second, and third terms of each input feature, and the coefficients of the product terms of each pair of features are used as the basis for the mean squared error between the predicted label and the true supervised label on the training set. A regularization coefficient multiplied by the sum of squared parameters is added to this loss function, and the optimal parameter vector is obtained by minimizing this loss function. Then, the optimal regularization coefficient is selected based on five-fold cross-validation, choosing the regularization coefficient with the smallest average validation error. Finally, the model is retrained using the entire training set and the selected optimal regularization coefficient to obtain the final multinomial regression parameters. After training, for each new blood glucose time-series coupled feature vector obtained, the model directly calculates the output blood glucose nonlinear coupling correction coefficient without retraining.

[0032] In this embodiment, by performing multi-period coupling analysis and correction of blood glucose time-series signals, it is helpful to automatically separate periodic components of multiple time scales, such as circadian rhythm, short-term response, and post-exercise effect, from the raw blood glucose monitoring data and quantify their nonlinear interaction intensity. This reduces the systematic bias in blood glucose prediction caused by ignoring multi-period coupling, reduces the probability of misjudgment by single-period or linear superposition models in high-dynamic scenarios, and improves the adaptability of infusion parameters to the individual's true metabolic rhythm. In this way, it enables the insulin pump control system to achieve accurate and safe all-weather regulation in complex modes.

[0033] like Figure 3The logic diagram of the infusion parameter generation method for the insulin pump control system based on time-series patterns shows that: infusion parameters are generated, and the blood glucose prediction deviation is obtained. It is determined whether the absolute value of the blood glucose prediction deviation is less than a preset allowable fluctuation threshold. If so, the system continues to wait for the next blood glucose sampling point to arrive and repeats the above steps to form a continuous adaptive control closed loop. Otherwise, the infusion parameters are dynamically adjusted. If the blood glucose prediction deviation is positive, it is further determined whether the absolute value of the blood glucose prediction deviation exceeds a preset hyperglycemia risk threshold. If so, it is determined that the infusion rate needs to be increased and the insulin adjustment dose needs to be adjusted. Otherwise, only a slight upward adjustment is needed. If the blood glucose prediction deviation is negative, it is determined whether the absolute value of the blood glucose prediction deviation exceeds a preset hypoglycemia risk threshold. If so, it is determined that the infusion rate needs to be decreased and the insulin adjustment dose needs to be adjusted. Otherwise, only a slight downward adjustment is needed. After the dynamic adjustment of the infusion parameters is completed, the actual pump infusion dose is obtained, and the infusion parameters are modified based on the actual pump infusion dose.

[0034] Furthermore, the specific process for generating infusion parameters is as follows: The predicted blood glucose value is compared with the target blood glucose value to obtain a blood glucose prediction deviation used to quantify the degree to which future blood glucose deviates from the ideal state. If the absolute value of the blood glucose prediction deviation is less than a preset allowable fluctuation threshold, it is determined that the predicted blood glucose is within the ideal range. At this time, the current infusion parameters remain unchanged, including the current basal rate value and the existing pre-meal high-dose instruction, and the system continues to wait for the next blood glucose sampling point to arrive, repeating the above steps to form a continuous adaptive control closed loop. The preset allowable fluctuation threshold is represented by the average of the absolute values ​​of blood glucose prediction deviations over historical time periods. If the absolute value of the blood glucose prediction deviation is not less than the preset allowable fluctuation threshold, the infusion parameters are dynamically adjusted. The specific process for dynamically adjusting the infusion parameters is as follows: If the blood glucose prediction deviation is positive, it is further determined whether the absolute value of the blood glucose prediction deviation exceeds a preset high blood glucose risk threshold. If a high blood glucose risk threshold is exceeded, the infusion rate is increased and the insulin adjustment dose is adjusted. Otherwise, only a slight upward adjustment is needed. The preset high blood glucose risk threshold is pre-set by a pre-defined personnel. A slight upward adjustment involves multiplying the insulin adjustment dose by a preset fine-tuning coefficient greater than one to obtain the actual pump infusion dose, and then modifying the infusion parameters based on the actual pump infusion dose. For example, the preset fine-tuning coefficient is set to 0.5 to suppress over-adjustment. The preset fine-tuning coefficient is pre-set by a pre-defined personnel. If the blood glucose prediction deviation is negative, it is determined whether the absolute value of the blood glucose prediction deviation exceeds the preset low blood glucose risk threshold. If it does, the infusion rate is reduced and the insulin adjustment dose is adjusted. Otherwise, only a slight downward adjustment is needed. A slight downward adjustment involves multiplying the insulin adjustment dose by a preset fine-tuning coefficient less than one to obtain the actual pump infusion dose, and then modifying the infusion parameters based on the actual pump infusion dose.

[0035] Specifically, the process for adjusting the insulin dosage is as follows: Step 1: Calculate the deviation between the predicted blood glucose level and the preset allowable fluctuation threshold. Divide the result of the deviation calculation with the preset dose response coefficient. Use the result of the division calculation as the pump infusion dosage adjustment value. The preset dose response coefficient represents the reduction in blood glucose per unit of insulin, which is preset by the operator. For cases requiring an increase in infusion rate, the pump infusion dosage adjustment value is positive, representing the dosage to be increased. For cases requiring a decrease in infusion rate, the absolute value of the pump infusion dosage adjustment value represents the dosage to be decreased. Step 2: Determine if the absolute value of the pump infusion dosage adjustment value is greater than the preset maximum adjustment dose. If so, adjust the corresponding pump infusion dosage. The dose adjustment value is set to the preset maximum adjustment dose, which is then marked as the actual pump infusion dose, keeping the original sign unchanged. Conversely, if the actual dose is less than the preset maximum adjustment dose, the corresponding pump infusion dose adjustment value is directly marked as the actual pump infusion dose. The preset maximum adjustment dose is set in advance by a preset user. The specific process for modifying the infusion parameters is as follows: The sampling time corresponding to the predicted blood glucose value is marked as the prediction time, and the basal rate time period to which the prediction time belongs is determined. The basal rate time period refers to each hourly segment in the 24-hour segmented basal rate compensation curve or basal rate array. For example, 00:00 to 01:00 is the first time period, and so on, with each time period corresponding to a basal rate value. If the actual pump infusion dose is less than the preset small dose threshold, the basal rate segment value of the prediction time is adjusted. The specific process is as follows: Modify step 1 to obtain the basal rate value corresponding to the predicted time in the basal rate array or the basal rate compensation curve corresponding to the predicted time. The preset small dose threshold is pre-set by a preset person. Modify step 2 to sum the predicted basal rate value and the actual pump infusion dose if the actual pump infusion dose is positive, and use the sum as the actual basal rate value corresponding to the predicted time. Modify step 3 to calculate the deviation between the predicted basal rate value and the absolute value of the actual pump infusion dose, and use the result as the actual basal rate value corresponding to the predicted time. Modify step 4 to adjust the infusion parameters if the actual pump infusion dose is not less than the preset small dose threshold. The control center then sends a pre-meal bolus adjustment instruction. This instruction is a one-time infusion dose instruction based on the feedback relationship between pre-meal and post-meal blood glucose monitoring values ​​for the same meal in history, such as the previous day's breakfast, lunch, or dinner. The instruction is based on the relative position of the post-meal blood glucose monitoring value in historical data to the preset target blood glucose range: if the post-meal blood glucose monitoring value of the corresponding meal the previous day is higher than the upper limit of the preset target blood glucose range, the actual pump infusion dose for the corresponding meal on the current day is increased; if it is lower than the lower limit of the preset target blood glucose range, the actual pump infusion dose for the corresponding meal on the current day is decreased. This allows for a one-time infusion of the adjusted dose before the nutritional intake event occurs, thus achieving closed-loop correction of post-meal blood glucose fluctuations.This pre-meal high-dose adjustment instruction administers the actual pump infusion dose in a single infusion within a predetermined time window before the predicted time, such as 15 minutes before the predicted time. It should be noted that if the actual pump infusion dose is negative, the infusion volume will be zero, i.e., no infusion will be administered.

[0036] Specifically, multi-level safety checks are performed on the adjusted basal rate value or pre-meal bolus dose: the corresponding basal rate value for each hour is not lower than the preset minimum basal rate limit, nor higher than the preset maximum basal rate limit; the pre-meal bolus dose is not lower than zero, nor higher than the preset maximum single bolus dose limit; at the same time, the total basal rate for the whole day after the dynamic adjustment of infusion parameters does not exceed the preset total basal rate upper limit. Among them, the preset minimum basal rate limit, the preset maximum basal rate limit, the preset maximum single bolus dose limit, and the preset total basal rate upper limit are all set in advance by preset personnel; if the adjusted basal rate does not meet the multi-level safety checks, it is forcibly set to the closest boundary value, and the corresponding basal rate value or pre-meal bolus dose is uploaded to the infusion parameter generation control center; if the adjusted basal rate meets the multi-level safety checks, the adjusted basal rate value or pre-meal bolus dose instruction is uploaded to the control register of the insulin pump, and the pump's drive actuator implements the infusion at the corresponding time. At the same time, the adjustment record, including the predicted blood glucose value, the actual pump infusion dose, and the basal rate, is stored in the infusion parameter generation control as a historical basis for the next round of iterative optimization.

[0037] In this embodiment, by generating infusion parameters, it is helpful to convert the multi-cycle coupling information and basal rate adaptive adjustment results in the blood glucose time series signal into precise pump control commands, continuously and dynamically adjust and gradually converge to the optimal value, effectively reduce excessive fluctuations in blood glucose monitoring values ​​caused by the inability of a single fixed algorithm to accurately match diverse dynamic needs, enable the system to maintain stable control performance in long-term operation, reduce abnormal fluctuations in blood glucose monitoring values ​​caused by prediction bias and dose lag, and improve the accuracy and response speed of infusion parameter matching of the insulin pump control system in complex life scenarios.

[0038] Example 2, as an alternative to multi-period coupling analysis of blood glucose time series signals, addresses the issue of strong high-frequency noise interference in blood glucose monitoring signals or irregular fluctuations in sampling intervals due to equipment limitations. Since online empirical mode decomposition (EMD) is sensitive to noise and its boundary effects easily lead to mode aliasing between different frequency components, the decomposed intrinsic mode function components may lose their clear physical periodic meaning, resulting in a significant decrease in the accuracy of the blood glucose time series coupling feature vector. Therefore, multi-period coupling analysis of blood glucose time series signals is necessary. The specific process is as follows: Step A: Using the blood glucose monitoring sequence as input to a preset multi-scale time series feature extraction network, and outputting four multi-scale time series feature maps representing local change patterns at different time scales, labeled as the first, second, third, and fourth multi-scale time series feature maps. The first multi-scale time series feature map corresponds to the output of a convolutional layer with a kernel length of approximately 15 minutes, used to capture high-frequency rapid fluctuation components in the blood glucose signal with a period of approximately 15 minutes to one hour. The second multi-scale time series feature map corresponds to the output of a convolutional layer with a kernel length of approximately three hours. The first multi-scale temporal feature map is used to capture fast response patterns with a period of approximately three to five hours; the second multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of approximately twelve hours, used to capture post-exercise effects or diurnal metabolic state changes with a period of approximately six to twelve hours; the third multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of twenty-four hours, used to capture the basic diurnal rhythm with a period of approximately twenty-four hours; Step B, the four multi-scale temporal feature maps are concatenated along the channel dimension to form a multi-scale feature tensor; the multi-scale feature tensor is input into a preset deep learning model, such as Transformer, and the output is an attention-enhanced multi-scale feature vector; this process automatically learns the correlation weights between features at each scale, thereby quantifying the nonlinear coupling strength between different periodic components without explicit signal decomposition; Step C, the multi-scale feature vector is input into a preset feedforward neural network, and the output is the blood glucose prediction value at the next sampling time; the blood glucose prediction value has implicitly incorporated the nonlinear interaction between multi-scale features during the calculation process, which is equivalent to applying a coupling correction term dynamically generated by the attention mechanism to the linear superposition prediction value; infusion parameters are generated based on the blood glucose prediction value.

[0039] It should be noted that the preset multi-scale temporal feature extraction network used in this embodiment is specifically a one-dimensional convolutional neural network with four parallel branches. Each branch contains a convolutional layer and a non-linear activation function. The kernel lengths of the four convolutional layers correspond to 15 minutes, 3 hours, 12 hours, and 24 hours, respectively. In another optional embodiment, a multi-scale feature extraction network based on wavelet packet decomposition can also be used. This network decomposes signals at different time scales using pre-designed wavelet basis functions without requiring training. The specific training process of the preset multi-scale temporal feature extraction network is as follows: First, a large number of historical blood glucose monitoring sequence samples are collected, and the corresponding real blood glucose value at the next sampling time is labeled for each sample. All samples are then divided into training and validation sets. The training set is then fed into a multi-scale temporal feature extraction network. Four branches extract feature maps at different time scales, which are then concatenated along the channel dimension. This concatenation is followed by a global average pooling layer and a fully connected layer to output the predicted blood glucose value for each sample. The mean squared error between the predicted and actual blood glucose values ​​is calculated as the loss function. A backpropagation algorithm and an adaptive moment estimation optimizer are used to iteratively update the convolutional kernel weights and bias terms in the network at a preset small batch size (e.g., 32 samples). After each complete training cycle, the prediction error of the current network is calculated using a validation set. If the validation error does not decrease for several consecutive cycles, training is terminated early, and the network parameters at this point are saved as the final model parameters.

[0040] It should also be noted that the preset deep learning model (such as Transformer) used in this embodiment, specifically the preset deep learning model used in the output attention-enhanced multi-scale feature vector, is a basic Transformer encoder containing four attention heads and two encoding layers. Each encoding layer includes a multi-head self-attention sub-layer, a feedforward sub-layer, and residual connections and layer normalization. In another optional embodiment, a network combining a long short-term memory network and an attention mechanism can also be used. This network extracts time dependencies through gated recurrent units and then superimposes attention layers. The specific training process of the preset deep learning model in this embodiment is as follows: First, prepare the training sample set in the same way as the aforementioned multi-scale feature extraction network. Each sample is a 48-hour blood glucose monitoring sequence, labeled with the actual blood glucose value at the next moment. Encode the input blood glucose monitoring sequence by time position, mapping the absolute time position of each sampling point to a fixed-dimensional vector, adding it to the blood glucose value, and using it as the encoder's vector. Input; next, the blood glucose monitoring sequence with position encoding is input into the multi-head self-attention sublayer of the first encoding layer: the input is transformed into a query matrix, a key matrix, and a value matrix respectively through three trainable weight matrices. Each attention head independently calculates its attention weight, and then the outputs of the four heads are concatenated and linearly transformed; after residual connection and layer normalization, it is input into the feedforward sublayer, and after residual connection and layer normalization again, the output of the first encoding layer is completed; the above process is repeated to obtain the output of the second encoding layer, and then the output vector of the last time step of the blood glucose monitoring sequence is taken, and then passed through a linear output layer to obtain the blood glucose prediction value; then, the mean absolute error between the blood glucose prediction value and the actual blood glucose value is calculated as the loss function, and the backpropagation algorithm and the stochastic gradient descent optimizer with momentum are used for iterative training with a preset learning rate and batch size; in each iteration, the prediction error is evaluated on the validation set. If the prediction error is lower than the preset error threshold, the current model parameters are saved and training is terminated. The preset error threshold is represented by the average value of the prediction error over a historical time period; It should also be noted that the preset feedforward neural network used in this embodiment is specifically a multilayer perceptron containing two fully connected hidden layers. The first hidden layer has 64 nodes, the second hidden layer has 32 nodes, the number of input layer nodes depends on the dimension of the coupled feature vector, and the number of output layer nodes is one. In another optional embodiment, a radial basis function network can also be used, which performs nonlinear mapping on the input features through a Gaussian kernel function. The specific training process of the preset feedforward neural network is as follows: First, obtain the blood glucose time-series coupled feature vector for historical time periods, and record the residual between the actual blood glucose value and the linearly superimposed predicted value at the corresponding next sampling time. Use the blood glucose time-series coupled feature vector as the network input, and the residual as the supervision label. Divide the dataset into a training set and a validation set. Then, input the blood glucose time-series coupled feature vector of the training set into the multilayer perceptron layer in sequence, and calculate layer by layer. The output of the first hidden layer = nonlinear activation function (input × weight matrix 1 + bias vector 1), and the output of the second hidden layer = The nonlinear activation function (output of the first hidden layer × weight matrix 2 + bias vector 2) is used, and the final output = output of the second hidden layer × weight matrix 3 + bias vector 3. The mean squared error between the output value and the label residual is calculated as the loss function. The backpropagation algorithm and adaptive moment estimation optimizer are used for training with a preset mini-batch size and initial learning rate. After each training cycle, the learning rate is multiplied by a preset decay coefficient. When the loss function value on the validation set no longer decreases for a preset number of consecutive cycles, training is stopped, and the network parameters corresponding to the minimum validation error are saved as the final model. The preset mini-batch size, initial learning rate, and preset decay coefficient are all set in advance by a preset team.

[0041] In this embodiment, the multi-cycle coupling analysis of blood glucose time-series signals in Example 2 helps to directly capture the nonlinear interaction between different cycle components by using multi-scale convolutional feature extraction and multi-head self-attention mechanism when there is strong noise or irregular sampling interval in the blood glucose monitoring signal. This reduces the dependence on signal quality and sampling consistency, reduces the risk of feature distortion caused by mode mixing or boundary effects, and improves the prediction robustness and infusion parameter generation accuracy of the system under abnormal conditions such as complex electromagnetic environment or poor sensor contact. This enables end-to-end high-disturbance insulin pump control closed-loop regulation.

[0042] like Figure 4The schematic diagram of the time-series-based insulin pump control system infusion parameter generation device shows that it includes: a basal rate mode adaptive selection module, a multi-cycle coupling monitoring module, a multi-cycle coupling correction module, and an infusion parameter generation module. Specifically, the basal rate mode adaptive selection module initiates a time-series dual-mode basal rate adaptive selection during the insulin pump control system infusion parameter generation process, selecting between the first-day basal rate initialization mode and the second-day basal rate iteration mode. The multi-cycle coupling monitoring module, after completing the aforementioned time-series dual-mode basal rate adaptive selection, performs multi-cycle coupling analysis of blood glucose time-series signals by extracting periodic components from different time scales of the blood glucose monitoring signal, followed by multi-cycle coupling discrimination of the blood glucose time-series signal. The multi-cycle coupling correction module determines whether to initiate blood glucose time-series coupling correction based on the discrimination result. The infusion parameter generation module, if correction is required, first performs blood glucose time-series coupling correction, then generates infusion parameters based on the corrected result; if correction is not required, it directly generates infusion parameters.

[0043] In this embodiment, the basal rate mode adaptive selection module, multi-cycle coupling monitoring module, multi-cycle coupling correction module, and infusion parameter generation module help improve the insulin pump control system's ability to deeply analyze blood glucose time-series patterns and its response speed, reducing blood glucose fluctuations caused by parameter mismatch or coupling deviation. Specifically, the basal rate mode adaptive selection module provides a basal rate benchmark that fits individual fasting blood glucose changes for the first day of startup or daily iteration, ensuring that subsequent coupling analysis is based on a reasonable infusion baseline. After the basal rate benchmark is determined, the multi-cycle coupling monitoring module decomposes the periodic components at different time scales from the real-time blood glucose signal and quantifies their nonlinear interactions, providing a reliable basis for the correction module. The multi-cycle coupling correction module selectively compensates for prediction deviations based on the judgment results. The infusion parameter generation module maps the corrected blood glucose prediction value to basal rate fine-tuning or pre-meal high-dose instructions. The four modules are interconnected and progressively enhance each other.

[0044] The above description is only an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating infusion parameters for an insulin pump control system based on time-series patterns, characterized in that, The method includes: During the generation of infusion parameters for the insulin pump control system, a time-sequential dual-mode adaptive basal rate selection is initiated to select between the first-day basal rate initialization mode and the second-day basal rate iteration mode. After completing the above-mentioned adaptive selection of basal rate in dual-mode time series, multi-cycle coupling analysis of blood glucose time series signals with periodic components of different time scales extracted from blood glucose monitoring signals is carried out, followed by multi-cycle coupling discrimination of blood glucose time series signals. The decision on whether to initiate blood glucose time-series coupling correction is based on the discrimination results; If correction is determined to be necessary, blood glucose timing coupling correction is performed first, and then infusion parameters are generated based on the corrected results. If correction is determined not to be necessary, infusion parameters are generated directly.

2. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 1, characterized in that, The specific process of adaptive selection of the time-series dual-mode base rate is as follows: Read the historical base rate records stored locally to determine if there is base rate data for the previous day, including the total base rate value for the previous day and the corresponding 24-hour segment base rate array; If the base rate data for the previous day is not available, then execute the base rate initialization mode for the first day. If the base rate data for the previous day exists, then obtain the total base rate value for the previous day, record it as the first total base rate value, and execute the base rate iteration mode for the next day based on the first total base rate value; The specific process of the first-day base rate initialization mode is as follows: Obtain the first fasting blood glucose value. Based on the first fasting blood glucose value and the preset blood glucose range where the body weight is located, match a corresponding total basal rate value and basal rate array from a preset number of preset time-series basal rate profiles. Perform multi-cycle coupling analysis of blood glucose time-series signals based on the total basal rate value and basal rate array. The 24-hour basal rate array in the selected basal rate profile is used as the basal rate output for the day and stored as a historical basal rate record.

3. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 2, characterized in that, The specific process of the next-day base rate iteration mode is as follows: Obtain the second fasting blood glucose value; The second fasting blood glucose value is obtained by a continuous glucose monitor at a preset morning fasting sampling time point in the next day's basal rate iteration mode; The second total base rate value is obtained based on the first total base rate value, and safety boundary constraints are applied during the process of obtaining the second total base rate value. The specific process is as follows: If the second fasting blood glucose value is less than the preset first low basal rate value, then the first total basal rate value and the preset first high basal rate compensation value are used to calculate the deviation, and the result of the deviation calculation is used as the candidate second total basal rate value. If the second fasting blood glucose value is not less than the preset first low basal rate value and not greater than the preset second low basal rate value, then the first total basal rate value and the preset first low basal rate compensation value are used to perform a deviation calculation, and the result of the deviation calculation is used as the candidate second total basal rate value. If the second fasting blood glucose value is greater than the preset second low basal rate value but less than the preset basal rate value, then the first total basal rate value will be used as the candidate second total basal rate value. If the second fasting blood glucose value is not less than the preset basal rate value and not greater than the preset second high basal rate value, then the first total basal rate value and the preset low basal rate compensation value are summed, and the result of the summation is used as the candidate second total basal rate value. If the second fasting blood glucose value is not less than the preset second high basal rate value and not greater than the preset high basal rate value, then the first total basal rate value and the preset first low basal rate compensation value are summed, and the result of the summation is used as the candidate second total basal rate value. If the second fasting blood glucose value is greater than the preset high basal rate value, the first total basal rate value and the preset first high basal rate compensation value are summed, and the result of the summation is used as the candidate second total basal rate value. The preset first high base rate compensation value is greater than the preset first low base rate compensation value and is greater than the preset low base rate compensation value. The preset high base rate value is greater than the preset second high base rate value, greater than the preset base rate value, greater than the preset second low base rate value, and greater than the preset first low base rate value. The specific process for establishing the safety boundary constraints is as follows: If the candidate second total base rate value is less than the preset base rate lower limit protection value, then the second total base rate value is set to the preset base rate lower limit protection value; If the candidate second total base rate value is greater than the preset base rate upper limit protection value, then the second total base rate value is set to the preset base rate upper limit protection value; If the candidate second total base rate value is not less than the preset base rate lower limit protection value and not greater than the preset base rate upper limit protection value, then the corresponding candidate second total base rate value is set as the second total base rate value. The time series base rate curve is obtained based on the second total base rate value.

4. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 3, characterized in that, The specific process for obtaining the time-series base rate curve is as follows: Divide the second total base rate value, which has been constrained by the safety boundary, by 24 to obtain the average hourly base rate; Multiply the average hourly base rate by the corresponding preset time period rhythm coefficient to obtain the initial base rate for each hour; The base rate is adjusted based on the initial base rate, and the specific process is as follows: Obtain the second fasting blood glucose value; Determine whether the second fasting blood glucose value is higher than the preset hyperglycemia threshold. If so, multiply the initial basal rate by a preset blood glucose correction coefficient greater than 1 to obtain the time-series basal rate correction value. Otherwise, compare the second fasting blood glucose value with the preset hypoglycemia threshold. If the second fasting blood glucose value is lower than the preset hypoglycemia threshold, the initial basal rate is multiplied by a preset blood glucose correction coefficient less than 1 to obtain the time-series basal rate correction value. If the second fasting blood glucose value is not higher than the preset hyperglycemia threshold and not lower than the preset hypoglycemia threshold, the initial basal rate remains unchanged, and the corresponding initial basal rate is used as the time-series basal rate correction value. Multiply the timing base rate correction value by 10 and then round it down. Divide the result of the rounding operation by 10 to obtain the rounded value of the timing base rate. The 24-hour time series base rate values ​​are rounded down and arranged in hourly order, for example, from 0:00 to 23:00, to form a time series base rate curve containing 24 data points.

5. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 4, characterized in that, The specific process of the multi-period coupling analysis of the blood glucose time series signal is as follows: Obtain the blood glucose monitoring sequence for the data collection period; Based on the online empirical mode decomposition algorithm, the blood glucose monitoring sequence is decomposed into time-series signal modes to obtain the intrinsic mode function components, including time-series high-frequency components, diurnal rhythm components, exercise aftereffect components, and long-period components. The analytical signals of each intrinsic mode function component are obtained by performing Hilbert transform on the online Hilbert transform algorithm. The magnitude of the analytical signal is used as the instantaneous amplitude of the blood glucose time series. The principal argument value of the analytical signal is used as the instantaneous phase of the blood glucose time series; The instantaneous amplitude of the blood glucose time series represents the energy intensity of the current intrinsic mode function component at the current moment; Each intrinsic mode function component is paired up in pairs according to frequency from high to low, and the instantaneous phase difference and instantaneous amplitude ratio of each pair of intrinsic mode function components are obtained. The instantaneous amplitude and phase of the blood glucose time series of each intrinsic mode function component, as well as the instantaneous phase difference and instantaneous amplitude ratio of each pair of intrinsic mode function components, are sorted together into a blood glucose time series coupled feature vector according to the order of the instantaneous amplitude, phase, and instantaneous phase difference and instantaneous amplitude ratio of each pair of intrinsic mode function components. The process of multi-period coupling discrimination of blood glucose time series signals based on blood glucose time series coupling feature vectors is as follows: The Euclidean distance between the blood glucose temporal coupling feature vector and the preset temporal coupling feature vector is obtained based on the Euclidean distance algorithm, and it is marked as the temporal coupling Euclidean distance. If the temporal coupling Euclidean distance is greater than a preset distance threshold, perform blood glucose temporal coupling correction. Otherwise, obtain the linear extrapolation prediction value of each intrinsic mode function component at the next sampling time, sum the linear extrapolation prediction values ​​of all intrinsic mode function components at the same time to obtain the linear superposition prediction value of blood glucose, use the linear superposition prediction value of blood glucose as the blood glucose prediction value, and generate infusion parameters based on the blood glucose prediction value.

6. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 5, characterized in that, The specific process of the time-series signal mode decomposition is as follows: Obtain all local extreme points of blood glucose monitoring data during the data collection period, including local maxima and local minima; Connect all the local maxima to form the upper envelope, and connect all the local minima to form the lower envelope; For each sampling time, the blood glucose concentration values ​​of the upper envelope and the lower envelope at that time are added together and then divided by 2 to obtain the mean blood glucose concentration value at that time. The mean blood glucose concentration values ​​of all times are connected in chronological order to form a mean curve. This mean curve is used as the local average trend estimation curve of the current signal to be decomposed. The first-order candidate component is obtained by subtracting the mean blood glucose concentration value of the mean curve at the same time from the blood glucose concentration value of the current signal to be decomposed at each sampling time. Determine whether the current first-order candidate component satisfies the intrinsic mode function determination condition. If it does not, replace the signal to be decomposed with the first-order candidate component and repeat the time-series signal mode decomposition. If the condition is met, the current first-order candidate component is the first intrinsic mode function component decomposed, and the intrinsic mode function components are extracted successively based on the first intrinsic mode function component. The specific process of successive extraction of the intrinsic mode function components is as follows: The first extracted intrinsic mode function component is subtracted from the current signal to be decomposed to obtain the remaining components; The remaining components are used as new signals to be decomposed. The time-series signal mode decomposition is repeated, and the next intrinsic mode function component is extracted in sequence. The time-series signal mode decomposition process terminates when the remaining components meet the decomposition stopping condition. The decomposition stopping conditions include the number of extreme points of the remaining components being less than a preset threshold for the number of extreme points, or the signal energy of the remaining components being lower than a preset threshold for signal energy.

7. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 5, characterized in that, The specific process of the blood glucose time-series coupling correction is as follows: Input the blood glucose time-series coupled feature vector into a preset nonlinear regression model, and output the blood glucose nonlinear coupling correction coefficient; The blood glucose nonlinear coupling correction coefficient is multiplied by the corresponding preset benchmark correction amplitude to obtain the blood glucose coupling correction term. The corrected blood glucose prediction value is obtained by adding the linear superposition prediction value of blood glucose to the blood glucose coupling correction term. Infusion parameters are generated based on the corrected blood glucose prediction values.

8. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 7, characterized in that, The specific process for generating the infusion parameters is as follows: The deviation between the predicted blood glucose value and the target blood glucose value is calculated to obtain the blood glucose prediction deviation; If the absolute value of the blood glucose prediction deviation is less than the preset allowable fluctuation threshold, then keep the current infusion parameters unchanged and continue to wait for the next blood glucose sampling point to arrive, and repeat the above steps. If the absolute value of the blood glucose prediction deviation is not less than the preset allowable fluctuation threshold, then the infusion parameters are dynamically adjusted. The specific process for dynamically adjusting the infusion parameters is as follows: If the blood glucose prediction deviation is positive, further determine whether the absolute value of the blood glucose prediction deviation exceeds the preset hyperglycemia risk threshold. If it does, it is determined that the infusion rate needs to be increased and the insulin dose needs to be adjusted. Otherwise, only a slight increase is needed. The micro-adjustment refers to multiplying the insulin adjustment dose by a preset micro-adjustment coefficient greater than one to obtain the actual pump infusion dose, and then modifying the infusion parameters based on the actual pump infusion dose; If the blood glucose prediction deviation is negative, it is determined whether the absolute value of the blood glucose prediction deviation exceeds the preset hypoglycemia risk threshold. If it does, it is determined that the infusion rate needs to be reduced and the insulin dose needs to be adjusted. Otherwise, it is determined that only a slight reduction is needed. The micro-adjustment refers to multiplying the insulin adjustment dose by a preset micro-adjustment coefficient less than one to obtain the actual pump infusion dose, and then modifying the infusion parameters based on the actual pump infusion dose; The specific process for adjusting the insulin dosage is as follows: The deviation of the blood glucose prediction is calculated by comparing it with the preset allowable fluctuation threshold. The result of the deviation calculation is divided by the preset dose response coefficient. The result of the division is used as the pump infusion dose adjustment value. Determine whether the absolute value of the pump infusion dose adjustment value is greater than the preset maximum adjustment dose. If it is, set the corresponding pump infusion dose adjustment value to the preset maximum adjustment dose, mark it as the actual pump infusion dose, and keep the original sign unchanged. Otherwise, mark the corresponding pump infusion dose adjustment value directly as the actual pump infusion dose. The specific process for modifying the infusion parameters is as follows: Mark the sampling time corresponding to the predicted blood glucose value as the prediction time and determine the basal rate period to which the prediction time belongs; If the actual pump infusion dose is less than the preset low dose threshold, the basal rate segment value at the predicted time will be adjusted, as follows: Obtain the base rate value corresponding to the prediction time in the base rate array or the base rate value corresponding to the prediction time in the base rate compensation curve; If the actual pump infusion dose is positive, the basal rate value corresponding to the predicted time is summed with the actual pump infusion dose, and the result of the summation is used as the actual basal rate value corresponding to the predicted time. If the actual pump infusion dose is negative, the deviation between the basal rate value corresponding to the predicted time and the absolute value of the actual pump infusion dose will be calculated, and the result of the deviation calculation will be used as the actual basal rate value corresponding to the predicted time. If the actual pump infusion dose is not less than the preset small dose threshold, the infusion parameter generation and control center will generate a pre-meal large dose instruction. Perform multi-level safety checks on the adjusted basal rate values ​​or pre-meal bolus doses: If the adjusted basal rate does not meet the multi-level safety verification, it will be forcibly set to the closest boundary value, and the corresponding basal rate value or pre-meal high dose will be uploaded to the infusion parameter generation and control center. If the adjusted basal rate meets the multi-level safety check, the adjusted basal rate value or pre-meal bolus command is uploaded to the insulin pump's control register, and the pump's drive actuator performs the infusion at the corresponding time, while recording this adjustment.

9. The method for generating infusion parameters for an insulin pump control system based on time-series patterns as described in claim 4, characterized in that, The multi-period coupling analysis of the blood glucose time-series signal also includes: Using blood glucose monitoring sequences as input to a pre-defined multi-scale temporal feature extraction network, and multi-scale temporal feature maps as output, these are labeled as the first multi-scale temporal feature map, the second multi-scale temporal feature map, the third multi-scale temporal feature map, and the fourth multi-scale temporal feature map. The first multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of approximately fifteen minutes; The second multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of approximately three hours; The third multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of approximately twelve hours. The fourth multi-scale temporal feature map corresponds to the output of a convolutional layer with a kernel length of 24 hours. The four multi-scale temporal feature maps are concatenated along the channel dimension to form a multi-scale feature tensor; Input the multi-scale feature tensor into the preset deep learning model and output the attention-enhanced multi-scale feature vector. The multi-scale feature vector is input into a preset feedforward neural network, which outputs the predicted blood glucose value at the next sampling time. Infusion parameters are generated based on predicted blood glucose levels.

10. A device for generating infusion parameters for an insulin pump control system based on time-series patterns, characterized in that, The device includes: a basal rate mode adaptive selection module, a multi-cycle coupling monitoring module, a multi-cycle coupling correction module, and an infusion parameter generation module. The basal rate mode adaptive selection module is used to initiate a time-sequential dual-mode basal rate adaptive selection during the generation of insulin pump control system infusion parameters, which is used to select the first day's basal rate initialization mode and the second day's basal rate iteration mode. The multi-cycle coupling monitoring module is used to perform multi-cycle coupling analysis of blood glucose time-series signals by extracting periodic components of different time scales from blood glucose monitoring signals after completing the above-mentioned adaptive selection of the time-series dual-mode basal rate, and then perform multi-cycle coupling discrimination of blood glucose time-series signals. The multi-cycle coupling correction module is used to decide whether to initiate blood glucose time-series coupling correction based on the judgment result; The infusion parameter generation module is used to perform blood glucose time-series coupling correction first if it is determined that correction is needed, and then generate infusion parameters based on the correction result; if it is determined that correction is not needed, it directly generates infusion parameters.

Images