An adaptive dynamic parameter adjustment method and system for insulin infusion

Abstract

The present application relates to the technical field of parameter adjustment, in particular to an insulin infusion method and system with adaptive dynamic parameter adjustment, comprising the following steps: obtaining patient device data and unifying time axis, extracting key physiological indicators and classifying current state, predicting blood glucose trend to form adjustment expectation, adjusting injection speed to generate parameter group and uploading to cloud, updating synchronization strategy according to frequency threshold, writing parameter instructions based on doctor's advice, and completing remote update of injection parameter set. The present application classifies the physiological state of the current period by means of a deep neural network structure, compares the consistency of the trend direction and the current injection rate, and combines the motion intensity and heat intake information of the specific period to perform dynamic adjustment of the injection rate, improve the response accuracy of insulin infusion, combine the feedback parameters and the continuous period at the remote doctor end to complete the update, and improve the stability and safety of the overall blood glucose regulation.

Description

Technical Field

[0001] This invention relates to the field of parameter adjustment technology, and in particular to an adaptive dynamic parameter adjustment method and system for insulin infusion. Background Technology

[0002] The field of parameter tuning technology involves real-time or periodic adjustments to key control parameters in complex systems based on dynamic inputs or system states to maintain system stability, improve response speed, and enhance control accuracy. This field is commonly applied in automatic control, intelligent decision-making, biomedical devices, and robotic systems, and its typical characteristics include data-driven adaptive mechanisms, multivariate input processing capabilities, and closed-loop control strategies. In healthcare scenarios, particularly in chronic disease management, parameter tuning technology integrates patient physiological data (such as blood glucose, body temperature, and exercise intensity) with treatment responses to construct dynamic tuning models. This enables personalized, context-aware intervention plans, improving the safety and effectiveness of treatment.

[0003] Among them, the adaptive dynamic parameter adjustment insulin infusion method is an intelligent treatment solution for diabetic patients. It aims to predict and dynamically adjust insulin infusion parameters in real time through multi-source data modeling and intelligent control. It can automatically adapt to changes in diet, exercise, emotions, and physiology in the patient's daily life, achieving precise, efficient, and individualized insulin management, thereby reducing the risk of blood glucose fluctuations and improving the safety and long-term stability of blood glucose control.

[0004] Traditional infusion methods often misclassify or delay response when recognizing state transitions such as exercise or eating. During parameter adjustment, they typically rely on the current blood glucose level and fixed adjustment rules, failing to consider the dynamic trend of blood glucose. This can lead to deviations between injection adjustments and actual needs. Furthermore, the lack of an adaptive frequency control mechanism in data uploading and remote interaction means that frequent uploading or delayed synchronization affects the real-time nature of physician intervention. For example, if a patient's blood glucose drops rapidly after strenuous exercise but parameter updates are delayed, it can trigger the risk of hypoglycemia and limit the accuracy of individualized management. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose an adaptive dynamic parameter adjustment method and system for insulin infusion.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an insulin infusion method with adaptive dynamic parameter adjustment, comprising the following steps:

[0007] S1: Acquire real-time detection data from the patient's device, extract and record meal time and calorie information from the device, arrange the data in parallel according to a unified Unix timestamp format and correct time axis errors to generate a fused physiological input set;

[0008] S2: Based on the fused physiological input set, a deep neural network model trained with the patient's historical cycle data is invoked to perform classification operations on the current physiological state using a multilayer perceptron structure, and time series regression prediction is performed to obtain the trend of blood glucose concentration changes in future continuous cycles, thereby obtaining the expected trend set of insulin regulation.

[0009] S3: Based on the expected trend set of insulin regulation, call the currently set basal infusion rate in the insulin pump, and adjust the infusion rate of the insulin pump according to the exercise intensity and meal calorie value in the corresponding time period to generate a dynamically adjusted infusion parameter set;

[0010] S4: Call the dynamically adjusted infusion parameter group and the fused physiological input set to construct the data block before and after adjustment, upload it to the cloud patient database, and update the current data synchronization frequency according to the time interval value of three consecutive uploads to generate an adaptive frequency adjustment record package.

[0011] The present invention is improved in that the fused physiological input set includes a blood glucose concentration curve, a cadence time series, and a skin temperature distribution; the insulin regulation expected trend set includes physiological state classification labels, blood glucose trend interval descriptions, and insulin regulation response instructions; the dynamically adjusted infusion parameter set includes adjustment amplitude values, target infusion cycle indexes, and rate adjustment direction labels; and the adaptive frequency adjustment record includes data upload interval change instructions, frequency setting parameter sets, and historical synchronization delay values.

[0012] The present invention is improved in that the step of obtaining the fused physiological input set is specifically as follows:

[0013] S111: Acquire real-time detection data from the patient's device, including blood glucose concentration value output by the blood glucose monitoring device, step frequency value output by the step sensor, and skin temperature value recorded by the thermistor probe. Index and extract the three types of data according to the built-in timestamp, align the data with a unified sampling period as the standard, remove data segments with time errors exceeding the set synchronization offset threshold, implement multi-source timestamp accuracy unification, and obtain three types of physiological synchronization data frames.

[0014] S112: Based on the three types of physiological synchronization data frames, extract the meal timestamp and corresponding calorie information value recorded in the recording device, call the timestamp as the primary key index to insert the meal data into the corresponding position in the data frame, and perform data frame structure expansion to expand the meal record content into data field columns to obtain an embedded extended meal data frame.

[0015] S113: Based on the field columns in the embedded meal extended data frame, construct a time series axis according to a unified Unix timestamp format, perform parallel arrangement processing of column dimensions on the data items, and call the calibrated timestamp field to perform error comparison and position reorganization, establish a full physiological parameter structure, and generate a fused physiological input set.

[0016] The present invention is improved in that the step of obtaining the expected trend set of insulin regulation is specifically as follows:

[0017] S211: Based on the fused physiological input set, extract the blood glucose concentration change sequence, step frequency value, skin temperature value and meal time point in the current cycle, construct a multi-dimensional state feature indexed by timestamp, and normalize it according to the unit standard of each feature. Establish a unified input structure through tensor reconstruction to obtain the physiological feature input array.

[0018] S212: Call the physiological feature input array and input it into the multilayer perceptron neural structure trained with historical period data. Based on the maximum probability index position corresponding to the weight layer output result, determine the physiological state label corresponding to the time period. Combine the probability distribution score and the label set to perform label assignment and obtain the current period state label set.

[0019] S213: Based on the hidden state values ​​of the neural network output layer corresponding to the current periodic state label set, connect the output sequence to the last layer of the regression structure, transform the regression output through the activation function, construct a set of concentration change curve trends within the prediction period based on the sign trend and fluctuation direction of the output values, and generate an expected trend set of insulin regulation.

[0020] The present invention is improved in that the step of obtaining the dynamically adjusted infusion parameter group is specifically as follows:

[0021] S311: Based on the blood glucose change direction markers presented by the expected trend of insulin regulation, extract the record segments marked as upward or downward trends in the current cycle, and call the baseline infusion rate value in the current cycle for direction matching judgment. If the trend direction is inconsistent with the current infusion rate regulation direction, record the inconsistent cycle index and generate a trend deviation index sequence.

[0022] S312: Based on the index position of the corresponding period in the trend deviation index sequence, extract the step frequency value and meal calorie value under the period, calculate and obtain the injection rate adjustment coefficient for each period, adjust the injection rate of the current period, and obtain the updated injection rate sequence.

[0023] S313: Call the updated injection rate sequence, retrieve the continuous injection time field value from the injection configuration parameter set according to the cycle position corresponding to each adjustment range, set the corresponding duration standard according to the speed adjustment level, adjust the duration and perform joint mapping with the speed adjustment value, establish an update association table between the cycle and the parameter value, and generate a dynamically adjusted infusion parameter group.

[0024] The present invention is improved in that the formula for calculating the injection rate adjustment coefficient for each cycle is specifically as follows:

[0025] ;

[0026] in, This represents the normalized value of meal calories consumed within the current period. This represents the normalized value of the heat generated during the current cycle. This represents the normalized reference value for meal calories. This represents the injection rate adjustment factor.

[0027] The present invention is improved in that the step of obtaining the adaptive frequency adjustment recording packet is specifically as follows:

[0028] S411: Call the record of each parameter change in the dynamically adjusted infusion parameter group and the attached time period label, combine it with the original data recorded in the fused physiological input set, perform periodic time index alignment, construct a complete data structure before and after each parameter change cycle, and obtain the adjustment data block sequence;

[0029] S412: According to the adjusted data block sequence, upload the data blocks to the cloud patient database, extract the time of each data upload, calculate the time span of three consecutive upload intervals, call the synchronization load threshold set by the cloud, calculate and obtain the upload interval deviation value. If the upload interval deviation value exceeds the allowable deviation range, it is regarded as a synchronization abnormal cycle, and the synchronization abnormal cycle is marked to generate a synchronization offset cycle index set.

[0030] S413: Call the synchronization abnormal cycle position in the synchronization offset cycle index set, update the current data synchronization frequency according to the maximum allowed synchronization frequency and the minimum limit interval, reset the synchronization trigger condition parameters under the cycle, update the synchronization scheduling configuration table in the cloud database, and generate an adaptive frequency adjustment record package.

[0031] The present invention has an improvement, wherein the method further includes the following steps:

[0032] S5: Based on the adaptive frequency adjustment recording packet, send the blood glucose concentration value and exercise intensity value to the doctor's remote terminal, extract the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, call the two parameter contents provided by the doctor, write the update command to the injection unit port in the patient's insulin pump, and generate the remotely updated injection parameter set;

[0033] The injection parameter set that has been remotely updated includes the rate setting value, duration configuration value, and pump response status code provided by the doctor.

[0034] The present invention is improved in that the step of obtaining the injection parameter set after remote update is specifically as follows:

[0035] S511: Based on the adaptive frequency adjustment recording package, the recorded blood glucose concentration fluctuation amplitude, the number of changes in the basic infusion rate and the physiological state classification label are adjusted. The blood glucose concentration value and step frequency value corresponding to the current cycle are extracted, and the time index and classification label are paired. The two types of indicators are sent to the doctor's remote terminal interface in the form of structured data frames to obtain the set of physiological indicators for remote access.

[0036] S512: Based on each record of data in the remotely invoked physiological indicator set, extract the recommended infusion rate value and adjustment duration period value parameters fed back by the doctor in the doctor's remote management terminal, and write the parameters into the injection parameter queue cache area to generate doctor configuration injection parameter information;

[0037] S513: Call the infusion rate value and adjustment cycle value in the injection parameter information configured by the doctor, match the injection instruction queue under the insulin pump port with the time index field, write the corresponding parameter instruction into the dynamic scheduling buffer of the injection unit port, implement pump end parameter update and record the synchronization status, and generate a remotely updated injection parameter set.

[0038] An adaptive dynamic parameter-adjustable insulin infusion system, the adaptive dynamic parameter-adjustable insulin infusion system being used to implement the aforementioned adaptive dynamic parameter-adjustable insulin infusion method, the system comprising:

[0039] The physiological parameter fusion module acquires real-time detection data from the patient's device and extracts the meal time and calorie information recorded by the device. The data is then arranged in parallel according to a unified Unix timestamp format, and time axis errors are corrected to generate a fused physiological input set.

[0040] The blood glucose trend analysis module, based on the fused physiological input set, calls a deep neural network model trained with the patient's historical period data, performs classification operations on the current physiological state using a multilayer perceptron structure, and performs time series regression prediction to obtain the trend of blood glucose concentration changes in future continuous periods, thus obtaining the expected trend set of insulin regulation.

[0041] The pancreatic parameter adjustment module, based on the expected trend set of insulin regulation, calls the currently set basal infusion rate in the insulin pump, and adjusts the infusion rate of the insulin pump according to the exercise intensity and meal calorie value in the corresponding time period, generating a dynamically adjusted infusion parameter set;

[0042] The parameter integration and upload module calls the dynamically adjusted infusion parameter group and the fused physiological input set to construct data blocks before and after adjustment, uploads them to the cloud patient database, and updates the current data synchronization frequency based on the time interval value of three consecutive uploads to generate an adaptive frequency adjustment record package.

[0043] The remote update and adjustment module sends the blood glucose concentration value and exercise intensity value to the doctor's remote terminal according to the adaptive frequency adjustment recording package, extracts the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, calls the two parameter contents provided by the doctor, writes the update command to the injection unit port of the patient's insulin pump, and generates the remotely updated injection parameter set.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0045] In this invention, multiple physiological data are arranged in parallel and time axis errors are corrected to achieve high synchronization of data fusion, enhancing the timeliness and specificity of blood glucose prediction. A deep neural network structure is used to clearly classify the physiological state of the current cycle, and based on this classification, continuous cycle blood glucose trend regression prediction is achieved. This allows infusion parameters to be predictive of trends rather than relying on fixed rules. Based on a consistency comparison between the trend direction and the current injection rate, combined with exercise intensity and calorie intake information for specific time periods, the injection rate is dynamically adjusted to improve the responsiveness of insulin infusion. By aligning the data blocks before and after adjustment along the timeline and combining them with cloud load thresholds for adaptive adjustment of synchronization frequency, upload redundancy and synchronization lag are reduced. Feedback parameters and continuous cycle updates are completed at the remote doctor's end, achieving high adaptability and closed-loop control of insulin management under individual dynamic conditions, improving the overall stability and safety of blood glucose regulation. Attached Figure Description

[0046] Figure 1 This is a flowchart of the method of the present invention;

[0047] Figure 2 This is a flowchart illustrating the process of obtaining a fused physiological input set according to the present invention;

[0048] Figure 3 This is a flowchart illustrating the process of obtaining the expected trend set of insulin regulation according to the present invention;

[0049] Figure 4 This is a flowchart for obtaining a dynamically adjusted infusion parameter group according to the present invention;

[0050] Figure 5 This is a flowchart illustrating the process of obtaining the adaptive frequency adjustment recording packet according to the present invention;

[0051] Figure 6 This is a flowchart illustrating how the invention obtains the remotely updated set of injection parameters. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0054] Please see Figure 1 This invention provides a technical solution: an insulin infusion method with adaptive dynamic parameter adjustment, comprising the following steps:

[0055] S1: Acquire real-time detection data from the patient's device, including blood glucose concentration values ​​output by the blood glucose monitoring device, step frequency values ​​output by the step sensor, and skin temperature values ​​recorded by the thermistor probe. Extract meal time and calorie information recorded by the recording device, arrange the four types of data in parallel according to a unified Unix timestamp format, and correct time axis errors to generate a fused physiological input set.

[0056] S2: Based on the fusion of physiological input sets, extract the blood glucose concentration change sequence, step frequency value, skin temperature value and meal time point of the current cycle, call the deep neural network model trained by the patient's historical cycle data, and perform classification operation on the current physiological state using a multilayer perceptron structure to identify it as post-meal, exercise or resting state. On the basis of classification, perform time series regression prediction to obtain the description of the rising or falling trend of blood glucose concentration changes in future continuous cycles under the current state, and obtain the expected trend set of insulin regulation.

[0057] S3: Based on the expected trend of blood glucose increase or decrease presented by insulin regulation, the current set basal infusion rate in the insulin pump is called to compare the trend direction consistency. If there is a situation where the predicted direction is inconsistent with the current infusion rate adjustment direction, the infusion rate of the insulin pump is adjusted according to the exercise intensity and meal calorie value in the corresponding time period to generate a dynamic adjustment infusion parameter group.

[0058] S4: Call the changes to each parameter in the dynamically adjusted infusion parameter group and the corresponding time period label, align it with the execution timeline of the raw data collected from the fusion physiological input set, construct the data blocks before and after adjustment, upload them to the cloud patient database, and match them with the cloud data synchronization load threshold based on the time interval value of three consecutive uploads to determine whether there is redundancy or delay in the synchronization configuration, update the current data synchronization frequency, and generate an adaptive frequency adjustment record package.

[0059] S5: Based on the adaptive frequency adjustment of the blood glucose concentration fluctuation range, the number of changes in the basic infusion rate and the physiological state classification label recorded in the recording package, send the blood glucose concentration value and exercise intensity value to the doctor's remote terminal, extract the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, call the two parameter contents provided by the doctor, write the update command to the injection unit port in the patient's insulin pump and complete the parameter update, and generate the injection parameter set that has been remotely updated.

[0060] The fusion of physiological input sets includes blood glucose concentration curves, cadence time series, and skin temperature distribution; the expected insulin regulation trend set includes physiological state classification labels, blood glucose trend interval descriptions, and insulin regulation response instructions; the dynamically adjusted infusion parameter set includes regulation amplitude values, target infusion cycle indexes, and rate adjustment direction labels; the adaptive frequency adjustment record includes data upload interval change instructions, frequency setting parameter sets, and historical synchronization delay values; and the remotely updated injection parameter set includes rate setting values, duration configuration values, and pump response status codes provided by the physician.

[0061] Please see Figure 2 The specific steps for obtaining the fused physiological input set are as follows:

[0062] S111: Acquire real-time detection data from the patient's device, including blood glucose concentration value output by the blood glucose monitoring device, step frequency value output by the step sensor, and skin temperature value recorded by the thermistor probe. Index and extract the three types of data according to the built-in timestamp, align the data with a unified sampling period as the standard, remove data segments with time errors exceeding the set synchronization offset threshold, implement multi-source timestamp accuracy unification, and obtain three types of physiological synchronization data frames.

[0063] The blood glucose concentration value output by the blood glucose monitoring device at time point 1678886100 Output at time 1678886400 The step count sensor outputs the step frequency value at time point 1678886100.100. Output at time 1678886400.150 The thermistor probe recorded the skin temperature value at time point 1678886100.300. ℃, recorded at time point 1678886400.900 ℃, these three types of data are indexed and extracted according to the built-in Unix timestamp to unify the sampling period. Second( Data alignment is performed using minutes as the standard, and a synchronization offset threshold is set. milliseconds, this threshold is based on After statistical analysis of communication delays of different sensor devices in a group, the first group was selected. Percentile delay difference The timestamp for the blood glucose value is determined in milliseconds and rounded up, at time 1678886100. Step frequency value timestamp Skin temperature value timestamp The three relative to the reference time The maximum offset is milliseconds, less than Milliseconds, data retained, at time point 1678886400, blood glucose value timestamp. Step frequency value timestamp Skin temperature value timestamp Skin temperature timestamp offset milliseconds, greater than Millisecond threshold, this data segment ℃ was removed, and the previous cycle was used. After interpolating the ℃ value and unifying the multi-source timestamp precision, three types of physiological synchronization data frames are obtained, such as data frame 1: { , , , }, Data Frame 2: { , , , }

[0064] S112: Based on three types of physiological synchronization data frames, extract the meal timestamp and corresponding calorie information value recorded in the recording device, call the timestamp as the primary key index to insert the meal data into the corresponding position in the data frame, and perform data frame structure expansion to expand the meal record content into data field columns to obtain an embedded extended meal data frame.

[0065] Based on data frame 2: { , , , }, Extracting recording devices in The timestamp of the meal and the corresponding calorie information value entered at the time Call timestamp As the primary key index, it is linked to the sampling period. Matching data frames of seconds, because lie in and Between, and distance Closer (time difference) seconds, less than (seconds), so the meal data is inserted into the corresponding position in data frame 2, and the data frame structure is expanded, changing the original data frame structure {timestamp, blood glucose concentration, cadence, skin temperature} to {timestamp, blood glucose concentration, cadence, skin temperature, meal timestamp, meal calories}. If there is no meal record within the period, the corresponding field value is filled in. For example, data frame 1 is maintained as: { , , , , , }, Data frame 2 is updated to: { , , , , , }, thus obtaining the embedded dining extended data frame.

[0066] S113: Based on the field columns in the embedded meal extended data frame, construct a time series axis according to a unified Unix timestamp format, perform parallel arrangement processing of the column dimensions on the data items, and call the calibrated timestamp field to perform error comparison and position reorganization, establish a full physiological parameter structure, and generate a fused physiological input set;

[0067] According to data frame 1: { , , , , , } and data frame 2: { , , , , , The time series axis is constructed according to a unified Unix timestamp format. This involves extracting the first field of each frame [1678886100, 1678886400, ...], performing parallel column-dimensional permutation on the data items, generating blood glucose concentration sequences [8.0, 8.2, ...], step frequency sequences [90, 0, ...], skin temperature sequences [34.5, 34.5, ...], and meal calorie sequences [0, 500, ...]. The corrected timestamp field [1678886100, 1678886400] is then used for error comparison and position reassembly. For example, upon receiving a new data frame 3: { , , , , , After that, confirm its timestamp. Greater than Append it to the end of the sequence if out-of-order data frame 0 is received: { , , , , , }, then by comparison Less than The data is then recombined and inserted into the beginning of the sequence to establish a complete physiological parameter structure and generate a fused physiological input set.

[0068] Please see Figure 3 The specific steps for obtaining the expected trend set of insulin regulation are as follows:

[0069] S211: Based on the fusion of physiological input sets, extract the blood glucose concentration change sequence, step frequency value, skin temperature value and meal time point in the current cycle, construct a multi-dimensional state feature indexed by timestamp, and normalize it according to the unit standard of each feature. Establish a unified input structure through tensor reconstruction to obtain the physiological feature input array.

[0070] Based on the current period (timestamp) The sequence of blood glucose concentration changes within (e.g., taking the most recent 3 periods: [ , , ]) Current step frequency value Current skin temperature value With the most recent mealtime Build with timestamps The index is a multidimensional state feature, and normalization is performed according to the unit standard of each feature, blood glucose concentration (physiological range) to ) Normalization Step frequency (range) to ) Normalization Skin temperature (range) to ) Normalization Meal time Convert to distance from current Time difference Seconds, press Second( (hours) range normalized to Blood glucose sequence , , Normalization to [ , , A unified input structure is established through tensor recombination, for example, by concatenating them into a one-dimensional vector. , , , , , , ], thus obtaining the physiological feature input array.

[0071] S212: Call the physiological feature input array and input it into the multilayer perceptron neural structure trained with historical period data. Based on the maximum probability index position corresponding to the weight layer output result, determine the physiological state label corresponding to the time period. Combine the probability distribution score and the label set to perform label assignment and obtain the current period state label set.

[0072] Call [0.361, 0.05, 0.44, 0.111, 0.333, 0.344, 0.361], inputting historical periodic data (e.g.) In a multilayer perceptron neural structure trained with 10,000 sets of physiological data labeled as "stable," "rising," and "falling," the structure contains 3 hidden layers and 1 output layer. The output layer has 3 nodes, corresponding to the three labels "stable," "rising," and "falling," respectively. The node is determined by the maximum probability index position corresponding to the output of the weight layer (e.g., the Softmax layer outputs [0.15, 0.75, 0.10]). Corresponding index Determine the physiological state label (index) corresponding to the time period. ="stable", index =“rising”, index =“decline”, which means “rise”, combined with the probability distribution score Perform label assignment with the label set and set a confidence threshold. This threshold is based on historical data. The ROC curve analysis of the second prediction ensures that the recall rate is not lower than [percentage missing]. The specific point determined at that time Greater than After confirming the assignment is valid, obtain the current cycle status label set, namely {Timestamp: 1678886700, Label: 'Rising', Confidence: 0.75}.

[0073] S213: Based on the hidden state values ​​of the neural network output layer corresponding to the current cycle state label set, connect the output sequence to the last layer of the regression structure, transform the regression output through the activation function, construct the concentration change curve trend set within the prediction cycle based on the sign trend and fluctuation direction of the output value, and generate the expected trend set of insulin regulation.

[0074] Based on the hidden state value of the previous layer of the neural network output layer corresponding to the current periodic state label set {Label: 'rising'}, for example, a A vector of dimension [0.92, -0.15, ..., 0.88] is concatenated to the output sequence into a vector of dimension [0.92, -0.15, ..., 0.88]. The final layer of the regression structure with output nodes, which is used to predict the next 3 cycles ( , , The amount of change in blood glucose, through an activation function (e.g. The regression output is transformed to obtain the original output [+0.4, +0.2, +0.05] as [0.380, 0.197, 0.050]. Based on the sign trend (all positive) and fluctuation direction (decreasing value) of the output values, a set of concentration change curve trends within the prediction period is constructed. This trend is marked as "slowing rate of increase", generating the expected trend set of insulin regulation.

[0075] Please see Figure 4 The specific steps for obtaining dynamically adjusted infusion parameter sets are as follows:

[0076] S311: Based on the blood glucose change direction markers that are concentrated in the expected trend of insulin regulation, extract the record segments marked as upward or downward trends in the current cycle, and call the baseline infusion rate value in the current cycle for direction matching judgment. If the trend direction is inconsistent with the current infusion rate regulation direction, record the inconsistent cycle index and generate a trend deviation index sequence.

[0077] Based on the presented blood glucose change direction marker, i.e., "slowing rate of increase", the current period (timestamp) is extracted. The record segment marked as having an upward trend is retrieved, and the baseline infusion rate value for the current period is called. and its adjustment direction markings (e.g.) Minutes ago, to address the risk of hypoglycemia, the infusion adjustment direction was set to "decrease". A direction matching judgment was performed; if the blood glucose trend "increasing" did not align with the infusion adjustment direction "decrease", the inconsistent cycle index was recorded. The trend deviation index sequence is generated as [1678886700].

[0078] S312: Based on the index position of the corresponding period in the trend deviation index sequence, extract the step frequency value and meal calorie value for that period using the following formula:

[0079] ;

[0080] The injection rate adjustment coefficient for each cycle is obtained through calculation, and the injection rate for the current cycle is adjusted to obtain the updated injection rate sequence.

[0081] in, This represents the normalized value of meal calorie intake within the current period, derived from dimensionless proportional data obtained by dividing the recorded meal calorie intake by the maximum calorie statistical value. This represents the normalized value of the heat generated during the current cycle. This represents the normalized reference value for meal calories. This represents the injection rate adjustment factor;

[0082] Based on index position Extract the step frequency value under this period. (Corresponding to normalized kinetic energy) ) and meal calorie value (Corresponding to normalized meal calories) ), using the formula:

[0083] ;

[0084] in, This represents the injection rate adjustment factor. This represents the normalized value of meal calorie intake within the current period, derived from recorded meal calorie intake values. Divide by the maximum calorie count (based on the patient's recent...) The largest single meal in a day The dimensionless proportional data obtained by setting (the parameters) is acquired as follows: , The normalized value representing the amount of energy generated during the current cycle is derived from the step frequency value. (continued (minutes) convert to calories burned , divided by The maximum calorie count within a minute period (e.g.) The process of obtaining it is as follows: , This represents a normalized reference value for meal calories; the coefficient is based on... name Type 2 diabetes patients Mean normalized peak blood glucose response after a standard meal of glucose Therefore, The calculation logic of this formula is as follows: first calculate The normalized value of net calorie intake is obtained, i.e. Then divide by ( The net calorie value is compared with the standard meal response, and the absolute value is taken and the square root is calculated. To smooth the response amplitude, while calculate This is used to non-linearly scale the food portion size, and finally multiply the two ( The advantage of this formula lies in the introduction of exercise expenditure. Calories from meals The difference allows for an accurate assessment of the net heat load, and utilizes reference values. Standardize and combine The nonlinear amplification effect of large meal portions makes the adjustment coefficient... It can simultaneously reflect the counteracting effect of exercise and the actual impact force of food, and calculate the injection rate adjustment coefficient for each cycle:

[0085] ;

[0086] The result This indicates that the base infusion rate for the current cycle should be increased. For the current cycle infusion rate Adjustments were made, and the updated speed is: The updated injection rate sequence is obtained as {Timestamp: 1678886700, NewRate: 1.3023}.

[0087] S313: Call the updated injection rate sequence, retrieve the continuous injection time field value from the injection configuration parameter set according to the cycle position corresponding to each adjustment range, set the corresponding duration standard according to the speed adjustment level, adjust the duration and perform joint mapping with the speed adjustment value, establish an update association table between the cycle and the parameter value, and generate a dynamically adjusted infusion parameter group.

[0088] Call the updated injection rate sequence {Timestamp: 1678886700, NewRate: 1.3023}, and adjust the magnitude of each item accordingly. The corresponding period position Retrieve the value of the continuous injection time field from the injection configuration parameter set, which is based on clinical data ( The rate-effect relationship was established based on the statistics of the adjustment (secondary adjustment), for example, the speed adjustment range was divided into levels: Level 1 ( - Corresponding duration Minutes, Level 2 ( - )correspond Minutes, Level 3 ( - )correspond Minutes, current amplitude Belongs to level 3, the corresponding duration standard is set as follows Minutes, after adjusting the duration and speed value Perform joint mapping to establish an update association table between the period and parameter values, namely {Timestamp: 1678886700, Rate: 1.3023, Duration: 40}, and generate a dynamically adjusted infusion parameter group.

[0089] Please see Figure 5 The specific steps for obtaining the adaptive frequency adjustment recording packet are as follows:

[0090] S411: Call the record of each parameter change in the dynamically adjusted infusion parameter group and the attached time period label, combine it with the raw data recorded in the fused physiological input set, perform periodic time index alignment, construct the complete data structure before and after each parameter change cycle, and obtain the adjustment data block sequence;

[0091] Retrieve the change record for each parameter {Timestamp: 1678886700, Rate: 1.3023, Duration: 40} along with the accompanying time period label (period index). ), combined with raw data recorded in the fusion physiological input set (e.g. Periodic blood glucose Step frequency ,meal ) and the rate before adjustment Perform periodic time index alignment and construct a complete data structure before and after each parameter change period, for example: {Index: 1678886700, Glucose: 8.5, Steps: 10, Meal: 500, OldRate: 1.0, NewRate: 1.3023, Duration: 40}, to obtain the adjusted data block sequence.

[0092] S412: Based on the adjusted data block sequence, upload the data blocks to the cloud-based patient database, extract the upload time for each data upload, calculate the time span between three consecutive uploads, and call the synchronization load threshold set by the cloud using the formula:

[0093] ;

[0094] The upload interval deviation value is calculated. If the upload interval deviation value exceeds the allowable deviation range, it is regarded as a synchronization abnormal period, and the synchronization abnormal period is marked to generate a synchronization offset period index set.

[0095] in, This indicates the upload interval deviation value. For the first The time interval for each upload is derived from the difference between the timestamps of two consecutive data uploads. To observe the maximum value among the three upload interval values ​​within the window, The cloud synchronization load threshold is set based on the average upload interval deviation value calculated from the historical upload records collected over a long period of time and three times the standard deviation limit. It is used to define the boundary conditions between normal synchronization state and abnormal fluctuation state.

[0096] Upload the data block {Index: 1678886700, ...} to the cloud-based patient database, extract the upload time for each data upload, and calculate the time span between three consecutive uploads. For example, the timestamps of the first three data block uploads are as follows: , , (This time), calculation Second, Seconds, (to make up 3 intervals, there are still 3 seconds left) ), Seconds, call the synchronization load threshold set in the cloud. ,Should It is based on the cloud server. During the concurrent write test, the CPU utilization remained at [percentage missing]. The following average safe interval The second setting, that is Seconds, using the formula:

[0097] ;

[0098] in, To upload the interval deviation value, For the first The time interval between uploads, i.e. , , , The maximum value among the three upload intervals is obtained as follows: Second, Cloud synchronization load threshold The formula's operational logic lies in calculating each interval. With threshold The relative deviation, and through Normalization is performed. and The relative positions of the current interval and the threshold within the maximum observation window were calculated separately. The normalized bias was obtained by subtracting the two, and then divided by... The ratio of this deviation to the normalization threshold is obtained, the absolute value is taken as the square root to smooth the fluctuations, and finally the average of three values ​​is calculated. The advantage of this formula is that it allows for... Normalization and A relative comparison, The value not only reflects the absolute magnitude of the deviation, but also the relative severity of the deviation, making the assessment independent of a fixed time scale and adaptable to the fluctuations in upload frequency at different times, and calculating the upload interval deviation value.

[0099] :

[0100] , , ;

[0101] :

[0102] , ;

[0103] :

[0104] , ;

[0105] ;

[0106] The allowable deviation range is set based on the average upload interval deviation value calculated from historical upload records collected over a long period of time. and three standard deviations jointly determined, that is The result Exceeding the allowable deviation range If the upload interval deviation value exceeds the allowable deviation range, it is considered a synchronization anomaly period and the synchronization anomaly period is marked. ( (corresponding to the end point), generate the synchronization offset cycle index set, i.e. [1678886700].

[0107] S413: Call the synchronization abnormal cycle position in the synchronization offset cycle index set, update the current data synchronization frequency according to the maximum allowed synchronization frequency and the minimum limit interval, reset the synchronization trigger condition parameters under the cycle, update the synchronization scheduling configuration table in the cloud database, and generate an adaptive frequency adjustment record package.

[0108] Calling the synchronization exception cycle position According to the maximum permissible synchronization frequency (e.g.) Second-rate / (seconds) and minimum limit interval (e.g.) Second-rate / (seconds), update the current data synchronization frequency, because ( (seconds) indicates that the interval is too large, causing deviation. The system will increase the synchronization frequency (shorten the interval) and adjust the current threshold. seconds, updated to Seconds, new interval Seconds in Seconds and Between seconds, the synchronization trigger condition parameter under the reset period is: The time is incremented and updated to the synchronous scheduling configuration table in the cloud database, generating an adaptive frequency adjustment record package, namely {PatientID: P001, NewTheta: 2287.0, TriggerIndex: 1678886700}.

[0109] Please see Figure 6 The specific steps for obtaining the injection parameter set after remote update are as follows:

[0110] S511: Based on the adaptive frequency adjustment of the blood glucose concentration fluctuation amplitude, the number of changes in the basic infusion rate and the physiological state classification label recorded in the recording package, extract the blood glucose concentration value and step frequency value corresponding to the current cycle, pair them with the time index and classification label, and send the two types of indicators to the doctor's remote terminal interface in the form of structured data frames to obtain the set of physiological indicators for remote access.

[0111] Based on the magnitude of blood glucose concentration fluctuations (e.g.) ), number of changes in baseline infusion rate (e.g.) (Time) and physiological state classification label ("rising"), extract the current cycle. Corresponding blood glucose concentration value With step frequency value Indexed by time The data is paired with the category label "rising" to form a data frame {Index: 1678886700, Label: 'rising', Glucose: 8.5, Steps: 10}. The two types of indicators are sent to the doctor's remote interface in a structured data frame format (e.g., JSON format: {"Index": 1678886700, "Label": "rising", "Glucose": 8.5, "Steps": 10}) to obtain the set of physiological indicators for remote access.

[0112] S512: Based on each record of data in the remote physiological indicator set, extract the recommended infusion rate value and adjustment duration period value parameters fed back by the doctor in the doctor's remote management terminal, write the parameters into the injection parameter queue cache area, and generate doctor configuration injection parameter information;

[0113] Based on the recorded data {Index: 1678886700, Label: 'Rising', Glucose: 8.5, Steps: 10}, the doctor's actions after viewing this record were retrieved within the doctor's remote management terminal. Recommended infusion rate value based on real-time feedback With adjustment of duration value minutes (the doctor will) Fine-tuning ,Will Minutes adjusted to (minutes), and write the parameters {Rate: 1.25, Duration: 45, Source: 'Dr. Zhang'} into the injection parameter queue cache to generate doctor configuration injection parameter information.

[0114] S513: Call the infusion rate value and adjustment cycle value in the doctor's configured injection parameter information, match the injection instruction queue under the insulin pump port with the time index field, write the corresponding parameter instruction into the dynamic scheduling buffer of the injection unit port, implement pump end parameter update and record synchronization status, and generate a remotely updated injection parameter set.

[0115] Call infusion rate value With adjustment cycle value Minutes, indexed by time field The injection command queue at the insulin pump port is matched, and the command to be executed generated by S313 {Timestamp: 1678886700, Rate: 1.3023, Duration: 40} is found. The doctor-configured parameter command {Rate: 1.25, Duration: 45} is written into the dynamic scheduling buffer of the injection unit port, overwriting the original command, and the pump-side parameters are updated. The pump hardware then begins execution. The inflow, continued Minutes are recorded, and the synchronization status is set to "Updated - Doctor," generating a set of injection parameters that has been remotely updated.

[0116] An adaptive dynamic parameter-adjustable insulin infusion system is provided to implement the aforementioned adaptive dynamic parameter-adjustable insulin infusion method. The system includes:

[0117] The physiological parameter fusion module acquires real-time detection data from the patient's device and extracts the meal time and calorie information recorded by the device. The data is then arranged in parallel according to a unified Unix timestamp format, and time axis errors are corrected to generate a fused physiological input set.

[0118] The blood glucose trend analysis module, based on the fusion of physiological input sets, calls a deep neural network model trained with the patient's historical period data, performs classification operations on the current physiological state using a multilayer perceptron structure, and performs time series regression prediction to obtain the trend of blood glucose concentration changes in future continuous periods, thus obtaining the expected trend set of insulin regulation.

[0119] The pancreatic parameter adjustment module, based on the expected trend set of insulin regulation, calls the currently set basal infusion rate in the insulin pump, and adjusts the infusion rate of the insulin pump according to the exercise intensity and meal calorie value in the corresponding time period, generating a dynamically adjusted infusion parameter set;

[0120] The parameter integration and upload module calls the dynamically adjusted infusion parameter group and the fused physiological input set to construct data blocks before and after adjustment, uploads them to the cloud patient database, and updates the current data synchronization frequency based on the time interval of three consecutive uploads to generate an adaptive frequency adjustment record package.

[0121] The remote update and adjustment module adjusts the recording packet according to the adaptive frequency, sends the blood glucose concentration value and exercise intensity value to the doctor's remote terminal, extracts the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, calls the two parameter contents provided by the doctor, writes the update command to the injection unit port of the patient's insulin pump, and generates the remotely updated injection parameter set.

[0122] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An insulin infusion method with adaptive dynamic parameter adjustment, characterized in that, Includes the following steps: S1: Acquire real-time detection data from the patient's device, extract and record meal time and calorie information from the device, arrange the data in parallel according to a unified Unix timestamp format and correct time axis errors to generate a fused physiological input set; The specific steps for obtaining the fused physiological input set are as follows: S111: Acquire real-time detection data from the patient's device, including blood glucose concentration value output by the blood glucose monitoring device, step frequency value output by the step sensor, and skin temperature value recorded by the thermistor probe. Index and extract the three types of data according to the built-in timestamp, align the data with a unified sampling period as the standard, remove data segments with time errors exceeding the set synchronization offset threshold, implement multi-source timestamp accuracy unification, and obtain three types of physiological synchronization data frames. S112: Based on the three types of physiological synchronization data frames, extract the meal timestamp and corresponding calorie information value recorded in the recording device, call the timestamp as the primary key index to insert the meal data into the corresponding position in the data frame, and perform data frame structure expansion to expand the meal record content into data field columns to obtain an embedded extended meal data frame. S113: Based on the field columns in the embedded meal extended data frame, construct a time series axis according to a unified Unix timestamp format, perform parallel arrangement processing of column dimensions on the data items, and call the calibrated timestamp field to perform error comparison and position reorganization, establish a full physiological parameter structure, and generate a fused physiological input set; S2: Based on the fused physiological input set, a deep neural network model trained with the patient's historical cycle data is invoked to perform classification on the current physiological state using a multilayer perceptron structure, and time series regression prediction is performed to obtain the trend of blood glucose concentration changes in future continuous cycles, thereby obtaining the expected trend set of insulin regulation. S3: Based on the expected trend set of insulin regulation, call the currently set basal infusion rate in the insulin pump, and adjust the infusion rate of the insulin pump according to the exercise intensity and meal calorie value in the corresponding time period to generate a dynamically adjusted infusion parameter set; S4: Call the dynamically adjusted infusion parameter group and the fused physiological input set to construct the data block before and after adjustment, upload it to the cloud patient database, and update the current data synchronization frequency according to the time interval value of three consecutive uploads to generate an adaptive frequency adjustment record package; The fused physiological input set includes blood glucose concentration curves, cadence time series, and skin temperature distribution; the insulin regulation expected trend set includes physiological state classification labels, blood glucose trend interval descriptions, and insulin regulation response instructions; the dynamically adjusted infusion parameter set includes regulation amplitude values, target infusion cycle indexes, and rate adjustment direction labels; and the adaptive frequency adjustment record package includes data upload interval change instructions, frequency setting parameter sets, and historical synchronization delay values.

2. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 1, characterized in that, The specific steps for obtaining the expected trend set of insulin regulation are as follows: S211: Based on the fused physiological input set, extract the blood glucose concentration change sequence, step frequency value, skin temperature value and meal time point in the current cycle, construct a multi-dimensional state feature indexed by timestamp, and normalize it according to the unit standard of each feature. Establish a unified input structure through tensor reconstruction to obtain the physiological feature input array. S212: Call the physiological feature input array and input it into the multilayer perceptron neural structure trained with historical period data. Based on the maximum probability index position corresponding to the weight layer output result, determine the physiological state label corresponding to the time period. Combine the probability distribution score and the label set to perform label assignment and obtain the current period state label set. S213: Based on the hidden state values ​​of the neural network output layer corresponding to the current periodic state label set, connect the output sequence to the last layer of the regression structure, transform the regression output through the activation function, construct a set of concentration change curve trends within the prediction period based on the sign trend and fluctuation direction of the output values, and generate an expected trend set of insulin regulation.

3. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 2, characterized in that, The specific steps for obtaining the dynamically adjusted infusion parameter set are as follows: S311: Based on the blood glucose change direction markers presented by the expected trend of insulin regulation, extract the record segments marked as upward or downward trends in the current cycle, and call the baseline infusion rate value in the current cycle for direction matching judgment. If the trend direction is inconsistent with the current infusion rate regulation direction, record the inconsistent cycle index and generate a trend deviation index sequence. S312: Based on the index position of the corresponding period in the trend deviation index sequence, extract the step frequency value and meal calorie value under the period, calculate and obtain the injection rate adjustment coefficient for each period, adjust the injection rate of the current period, and obtain the updated injection rate sequence. S313: Call the updated injection rate sequence, retrieve the continuous injection time field value from the injection configuration parameter set according to the cycle position corresponding to each adjustment range, set the corresponding duration standard according to the speed adjustment level, adjust the duration and perform joint mapping with the speed adjustment value, establish an update association table between the cycle and the parameter value, and generate a dynamically adjusted infusion parameter group.

4. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 3, characterized in that, The formula for obtaining the injection rate adjustment coefficient for each cycle is as follows: ； in, This represents the normalized value of meal calories consumed within the current period. This represents the normalized value of the heat generated during the current cycle. This represents the normalized reference value for meal calories. This represents the injection rate adjustment factor.

5. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 4, characterized in that, The specific steps for obtaining the adaptive frequency adjustment recording packet are as follows: S411: Call the record of each parameter change in the dynamically adjusted infusion parameter group and the attached time period label, combine it with the original data recorded in the fused physiological input set, perform periodic time index alignment, construct a complete data structure before and after each parameter change cycle, and obtain the adjustment data block sequence; S412: According to the adjusted data block sequence, upload the data blocks to the cloud patient database, extract the time of each data upload, calculate the time span of three consecutive upload intervals, call the synchronization load threshold set by the cloud, calculate and obtain the upload interval deviation value. If the upload interval deviation value exceeds the allowable deviation range, it is regarded as a synchronization abnormal cycle, and the synchronization abnormal cycle is marked to generate a synchronization offset cycle index set. S413: Call the synchronization abnormal cycle position in the synchronization offset cycle index set, update the current data synchronization frequency according to the maximum allowed synchronization frequency and the minimum limit interval, reset the synchronization trigger condition parameters under the cycle, update the synchronization scheduling configuration table in the cloud database, and generate an adaptive frequency adjustment record package.

6. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 5, characterized in that, The method further includes the following steps: S5: Based on the adaptive frequency adjustment recording packet, send the blood glucose concentration value and exercise intensity value to the doctor's remote terminal, extract the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, call the two parameter contents provided by the doctor, write the update command to the injection unit port in the patient's insulin pump, and generate the remotely updated injection parameter set; The injection parameter set that has been remotely updated includes the rate setting value, duration configuration value, and pump response status code provided by the doctor.

7. The insulin infusion method with adaptive dynamic parameter adjustment according to claim 6, characterized in that, The specific steps for obtaining the injection parameter set after the remote update are as follows: S511: Based on the adaptive frequency adjustment recording package, the recorded blood glucose concentration fluctuation amplitude, the number of changes in the basic infusion rate and the physiological state classification label are adjusted. The blood glucose concentration value and step frequency value corresponding to the current cycle are extracted, and the time index and classification label are paired. The two types of indicators are sent to the doctor's remote terminal interface in the form of structured data frames to obtain the set of physiological indicators for remote access. S512: Based on each record of data in the remotely invoked physiological indicator set, extract the recommended infusion rate value and adjustment duration period value parameters fed back by the doctor in the doctor's remote management terminal, and write the parameters into the injection parameter queue cache area to generate doctor configuration injection parameter information; S513: Call the infusion rate value and adjustment cycle value in the injection parameter information configured by the doctor, match the injection instruction queue under the insulin pump port with the time index field, write the corresponding parameter instruction into the dynamic scheduling buffer of the injection unit port, implement pump end parameter update and record the synchronization status, and generate a remotely updated injection parameter set.

8. An adaptive dynamic parameter adjustment insulin infusion system, characterized in that, The system is used to implement the adaptive dynamic parameter adjustment insulin infusion method according to any one of claims 1-7, the system comprising: The physiological parameter fusion module acquires real-time detection data from the patient's device and extracts the meal time and calorie information recorded by the device. The data is then arranged in parallel according to a unified Unix timestamp format, and time axis errors are corrected to generate a fused physiological input set. The blood glucose trend analysis module, based on the fused physiological input set, calls a deep neural network model trained with the patient's historical period data, performs classification on the current physiological state using a multilayer perceptron structure, and performs time series regression prediction to obtain the trend of blood glucose concentration changes in future continuous periods, thus obtaining the expected trend set of insulin regulation. The pancreatic parameter adjustment module, based on the expected trend set of insulin regulation, calls the currently set basal infusion rate in the insulin pump, and adjusts the infusion rate of the insulin pump according to the exercise intensity and meal calorie value in the corresponding time period, generating a dynamically adjusted infusion parameter set; The parameter integration and upload module calls the dynamically adjusted infusion parameter group and the fused physiological input set to construct data blocks before and after adjustment, uploads them to the cloud patient database, and updates the current data synchronization frequency based on the time interval value of three consecutive uploads to generate an adaptive frequency adjustment record package. The remote update and adjustment module sends the blood glucose concentration value and exercise intensity value to the doctor's remote terminal according to the adaptive frequency adjustment recording package, extracts the doctor's data to formulate recommended infusion parameter values ​​and adjustment duration period, calls the two parameter contents provided by the doctor, writes the update command to the injection unit port of the patient's insulin pump, and generates the remotely updated injection parameter set.

Images