An intelligent medical micro-injection pump management system

Abstract

The present invention discloses an intelligent medical micro-injection pump management system, and the method includes a core device module, a sensor module, a data processing unit, a precision injection control module, an intelligent analysis module and an alarm module. The present invention belongs to the field of medical information technology, and specifically uses a hierarchical clustering collaboration model based on a feedback mechanism in the precision injection control module to control the injection pump, and through the multi-granularity data analysis capability of the hierarchical clustering collaboration model, the subtle changes in the key parameters of the liquid medicine are captured more finely; the DBSCAN algorithm is used in constructing the initial architecture of the hierarchical clustering collaboration model, and the noise data and outliers are effectively identified and filtered through the density threshold to ensure that the clustering results are more accurate; the meta-heuristic algorithm is used in the intelligent analysis module in combination with reinforcement learning for in-depth analysis, and while exploring new solutions, the existing optimal solution is effectively used to find a better global optimal solution, thereby improving the overall reliability of the system.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an intelligent medical micro-injection pump management system. Background Art

[0002] Medical micro-bolus refers to the process of accurately and continuously delivering drugs or nutrient liquids to the patient's body at an extremely low rate. This technology is widely used in various medical fields. Modern micro-bolus equipment has been highly automated and combined with intelligent technology for personalized management. Traditional control methods have problems such as low accuracy, delayed response, and drug waste. They rely on a single dimension or simple rules to control the bolus process and cannot capture subtle changes in the key parameters of liquid drugs, resulting in insufficient bolus accuracy and possible deviation of drug dosage from the target value. At the same time, they cannot respond and adjust in time during the bolus process, cannot correct deviations in real time, and cannot dynamically adjust bolus parameters according to individual differences of patients and drug absorption, resulting in excessive or insufficient use of drugs, resulting in drug waste or poor treatment effect. Ordinary data analysis has problems such as low analysis accuracy, low computational efficiency, and high false alarm rate. During the analysis process, parameters cannot be dynamically adjusted according to actual conditions, resulting in insufficient adaptability to data characteristics, low accuracy of analysis results, and the algorithm wastes a lot of computing resources in invalid search areas, reducing overall computing efficiency. It may perform poorly in identifying abnormal data, increasing false alarm rate and missed alarm rate, and reducing system reliability. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent medical micro-bolus pump management system. In view of the problems of low precision, response delay and drug waste in traditional control methods, the present invention uses a hierarchical clustering collaboration model based on a feedback mechanism in the precise injection control module to control the injection pump. Through the multi-granularity data analysis capability of the hierarchical clustering collaboration model, the subtle changes in the key parameters of the liquid drug can be captured more finely, thereby achieving high-precision control of the injection process and reducing drug dosage errors. Based on the real-time feedback mechanism, adaptive adjustments are made according to actual conditions during the injection process, thereby optimizing drug utilization. At the same time, the DBSCAN algorithm is used in constructing the initial architecture of the hierarchical clustering collaboration model to effectively identify and filter noise data and anomalies through density thresholds. value, ensuring that the clustering results are more accurate. By correcting the clustering at a finer granularity level, the boundary points and sparse area data that may be ignored in the clustering process are reduced. In view of the problems of low analysis accuracy, low computational efficiency and high false alarm rate in ordinary data analysis, the present invention uses a metaheuristic algorithm combined with reinforcement learning in an intelligent analysis module for in-depth analysis. The combined use can effectively utilize the existing optimal solution while exploring new solutions, so as to find a better global optimal solution. Reinforcement learning adjusts the parameters of the metaheuristic algorithm according to real-time feedback, enhances the adaptability of the model to data characteristics, and improves the analysis accuracy. The global search capability of the metaheuristic algorithm combined with the dynamic optimization capability of reinforcement learning can more accurately identify abnormal points in the data, reduce the false alarm rate and missed alarm rate, and improve the overall reliability of the system.

[0004] The technical solution adopted by the present invention is as follows: an intelligent medical micro-bolus pump management system provided by the present invention includes a core equipment module, a sensor module, a data processing unit, a precision bolus control module, an intelligent analysis module and an alarm module;

[0005] The core equipment module includes a push pump body and a flow regulating device, the push pump body delivers liquid medicine into the patient's body, and the flow regulating device is used to control the push speed and dosage of the liquid medicine;

[0006] The sensor module installs sensors inside and around the injection pump to monitor key parameters of the liquid drug in real time, including drug concentration, drug viscosity, drug pH value, drug temperature sensitivity, flow rate, pressure and temperature, and transmits the key parameters of the liquid drug to the data processing unit through a feedback mechanism;

[0007] The data processing unit is arranged inside the push pump, and pre-processes the key parameters of the liquid medicine and transmits them to the precise push control module and the intelligent analysis module;

[0008] The precise injection control module controls the injection pump through the key parameters of the liquid drug using a hierarchical clustering collaborative model based on a feedback mechanism. The control parameters include starting, stopping, adjusting flow, injection speed and injection dosage.

[0009] The intelligent analysis module receives the key parameters and control parameters of the pre-processed liquid medicine, uses a meta-heuristic algorithm combined with reinforcement learning to perform in-depth analysis, obtains an analysis report and transmits it to the alarm module;

[0010] The alarm module receives the analysis report and performs abnormality detection, and issues an alarm when an abnormal situation occurs.

[0011] Furthermore, in the precise injection control module, a hierarchical clustering collaborative model based on feedback mechanism is used to control the injection pump, which specifically includes the following steps:

[0012] Step S1: data preparation, receiving the key parameters of the pre-processed liquid medicine and performing standardization;

[0013] Step S2: Initialize the model and build the initial architecture of the hierarchical clustering collaborative model;

[0014] Step S3: real-time feedback, real-time monitoring of key parameters of the liquid drug during the injection process to obtain feedback information;

[0015] Step S4: Model evaluation, evaluating the hierarchical clustering collaboration model to obtain evaluation results;

[0016] Step S5: Model update, updating the initial architecture of the hierarchical clustering collaborative model according to the evaluation results and feedback information;

[0017] Further, in step S2, the initialization model is processed based on the key parameters of the liquid medicine after the standardization processing in step S1, specifically including the following steps:

[0018] Step S21: hierarchical structure setting, dividing the key parameters of liquid medicine into four levels, namely, ultra-fine granularity layer, fine granularity layer, medium granularity layer and coarse granularity layer, to capture multi-granularity information;

[0019] Step S22: Preliminary clustering: Use the DBSCAN algorithm to perform preliminary clustering on each level. In the clustering process, a downward masking strategy is adopted to filter false negatives by combining multi-granularity information. In each layer of clustering, data points located at the center of the cluster are retained to generate preliminary clustering results. A potential partition is provided for each level, and each potential partition represents a set of similar data points at that level.

[0020] Step S23: Cluster optimization, using the density peak cluster optimization algorithm to optimize the preliminary clustering results, while introducing an upward masking strategy to remove outliers in the cluster from each potential partition;

[0021] Step S24: prototype initialization, selecting the data point located at the cluster center from the clustering results of each level as the clustering prototype;

[0022] Step S25: Prototype optimization, using the sharp DBSCAN algorithm to adjust the position of the cluster prototype;

[0023] Furthermore, in step S21, the key parameters of the liquid medicine are divided into four levels, specifically including the following:

[0024] The ultra-fine-grained layer and the fine-grained layer capture detailed information, while the medium-grained layer and the coarse-grained layer capture the overall trend;

[0025] Ultra-fine granularity layer: captures tiny fluctuations in flow or pressure every millisecond;

[0026] Fine-grained layer: captures flow fluctuations and instantaneous pressure changes every second;

[0027] Medium granularity layer: captures the average flow, average pressure, and average temperature in the range of 1 to 5 minutes, reflecting the short-term trend of drug bolus;

[0028] Coarse-grained layer: total flow, cumulative bolus dose, average pressure, and average temperature over a 15-minute to 1-hour analysis range;

[0029] Further, step S25 specifically includes the following steps:

[0030] Step S251: input data and take each cluster prototype as the initial point;

[0031] Step S252: constructing a grid, dividing the grid units, and constructing a grid map to store the grid units;

[0032] Step S253: Neighboring grid calculation: calculate the neighborhood points for each initial point and calculate the density center of the points in the neighborhood. The formula used is as follows:

[0033] ;

[0034] ;

[0035] in, is the starting point, is the set of neighborhood points, is the center point of other grid cells, is the distance threshold, is the density center of the points in the neighborhood, is a point in the neighborhood point set;

[0036] Step S254: Distance calculation, calculate the distance between the initial point and the neighboring point. And adjust the position of the initial point. The formula used is as follows:

[0037] ;

[0038] in, is the learning rate;

[0039] Step S255: Merge grid cells. The formula used is as follows:

[0040] ;

[0041] Among them, this formula is for adjacent grid cells, if the distance between them is less than or equal to , then merge;

[0042] ;

[0043] in, is the center position of the merged grid unit, set as the new initial point;

[0044] Step S256: repeat steps S251 to S255 until the initial point position is stable and unchanged;

[0045] Step S257: Update the clustering results, apply the stable initial point to reallocate the clusters, and output the updated clustering results.

[0046] Furthermore, in the intelligent analysis module, a meta-heuristic algorithm combined with reinforcement learning is used for in-depth analysis, which specifically includes the following steps:

[0047] Step Q1: Feature extraction, using principal component analysis technology to extract key features from the key parameters and control parameters of the pre-treated liquid drug;

[0048] Step Q2: Initial analysis of the metaheuristic algorithm, using genetic algorithm to preliminarily optimize the key parameters and control parameters of liquid drugs, and perform multi-objective optimization at the same time to find the parameter combination and set it as the initial solution;

[0049] Step Q3: Training of reinforcement learning model: Initialize the reinforcement learning model, define the state space and action space, the state space includes the current push state and the characteristics of the liquid drug, the action space includes the parameter combination and the push strategy, simulate the push process, train the reinforcement learning model, use the initial solution of the metaheuristic algorithm as the starting point, and use the reinforcement learning model to select and optimize the push strategy;

[0050] Step Q4: Balance between exploration and exploitation, using the ε-greedy strategy, including the exploration phase, the exploitation phase, and adaptive learning;

[0051] Step Q5: Analysis result generation, using meta-heuristic algorithm to cluster and classify the data during the bolus injection process, identify the drug response pattern under the bolus injection strategy, and obtain the analysis results;

[0052] Furthermore, the three stages in step Q4 specifically include the following:

[0053] Exploration phase: finding potential optimization space through mutation operations of metaheuristic algorithms;

[0054] Utilization stage: using the historical optimal strategy to perform push injection operations;

[0055] Adaptive learning: Combined with the initial solution of the metaheuristic algorithm, a reinforcement learning model is used to adjust the ratio of the exploration phase to the exploitation phase.

[0056] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0057] (1) In view of the problems of low precision, response delay and drug waste in traditional control methods, the present invention uses a hierarchical clustering collaborative model based on a feedback mechanism in the precise push injection control module to control the push injection pump. Through the multi-granularity data analysis capability of the hierarchical clustering collaborative model, the subtle changes in the key parameters of the liquid drug can be captured more finely, thereby achieving high-precision control of the push injection process and reducing drug dosage errors. Based on the real-time feedback mechanism, adaptive adjustments are made according to actual conditions during the push injection process, thereby optimizing drug utilization.

[0058] (2) At the same time, the DBSCAN algorithm is used in the initial architecture of the hierarchical clustering collaborative model to effectively identify and filter noise data and outliers through density thresholds to ensure more accurate clustering results. By correcting the clustering at a finer level, the boundary points and sparse area data that may be ignored in the clustering process are reduced;

[0059] (3) In view of the problems of low analysis accuracy, low computational efficiency and high false alarm rate in ordinary data analysis, the present invention uses a metaheuristic algorithm combined with reinforcement learning in the intelligent analysis module for in-depth analysis. The combined use can effectively utilize the existing optimal solution while exploring new solutions, thereby finding a better global optimal solution. Reinforcement learning adjusts the parameters of the metaheuristic algorithm according to real-time feedback, enhances the adaptability of the model to data characteristics, and improves the analysis accuracy. The global search capability of the metaheuristic algorithm combined with the dynamic optimization capability of reinforcement learning can more accurately identify anomalies in the data, reduce the false alarm rate and missed alarm rate, and improve the overall reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a module of an intelligent medical micro-injection pump management system provided by the present invention;

[0061] Figure 2 A schematic diagram of the process flow of the precision injection control module provided by the present invention;

[0062] Figure 3 is a schematic flow chart of step S2;

[0063] Figure 4 A schematic diagram of the flow of the intelligent analysis module provided by the present invention.

[0064] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0066] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0067] Example 1, see Figure 1 , the present invention provides an intelligent medical micro-bolus pump management system, including a core equipment module, a sensor module, a data processing unit, a precision bolus control module, an intelligent analysis module and an alarm module;

[0068] The core equipment module includes a push pump body and a flow regulating device, the push pump body delivers liquid medicine into the patient's body, and the flow regulating device is used to control the push speed and dosage of the liquid medicine;

[0069] The sensor module installs sensors inside and around the push pump to monitor key parameters of the liquid drug in real time, including drug concentration, drug viscosity, drug pH value, drug temperature sensitivity, flow rate, pressure and temperature, and transmits the key parameters of the liquid drug to the data processing unit through a feedback mechanism;

[0070] The data processing unit is arranged inside the push pump, and pre-processes the key parameters of the liquid medicine and transmits them to the precise push control module and the intelligent analysis module;

[0071] The precise injection control module controls the injection pump through the key parameters of the liquid drug using a hierarchical clustering collaborative model based on a feedback mechanism. The control parameters include starting, stopping, adjusting flow, injection speed and injection dosage.

[0072] The intelligent analysis module receives the key parameters and control parameters of the pre-processed liquid medicine, uses a meta-heuristic algorithm combined with reinforcement learning to perform in-depth analysis, obtains an analysis report and transmits it to the alarm module;

[0073] The alarm module receives the analysis report and performs abnormality detection, and issues an alarm when an abnormal situation occurs.

[0074] Example 2, see Figure 2 This embodiment is based on the above embodiment, and uses a hierarchical clustering collaboration model based on a feedback mechanism in a precise push injection control module to control the push injection pump, which specifically includes the following steps:

[0075] Step S1: data preparation, receiving the key parameters of the pre-processed liquid medicine and standardizing them using the Z-score standardization method;

[0076] Step S2: Initialize the model and build the initial architecture of the hierarchical clustering collaborative model;

[0077] Step S3: Real-time feedback, real-time monitoring of key parameters of liquid medicine during the injection process, and feedback to the control center via the Zigbee low-speed short-distance transmission wireless network protocol;

[0078] Step S4: Model evaluation, evaluating the clustering effect of the hierarchical clustering collaborative model to obtain an evaluation result;

[0079] Step S5: Model update: According to the evaluation results and feedback information, the initial architecture of the hierarchical clustering collaborative model is updated by merging clusters, dispersing clusters and adjusting model parameters.

[0080] Example 3, see Figure 3 This embodiment is based on the above embodiment. In step S2, the initialization model is processed based on the key parameters of the liquid medicine after the standardization process in step S1, and specifically includes the following steps:

[0081] Step S21: hierarchical structure setting, dividing the key parameters of liquid medicine into four levels, namely, ultra-fine granularity layer, fine granularity layer, medium granularity layer and coarse granularity layer, to capture multi-granularity information;

[0082] Step S22: Preliminary clustering: Use the DBSCAN algorithm to perform preliminary clustering on each level. In the clustering process, a downward masking strategy is adopted to filter false negatives by combining multi-granularity information. In each layer of clustering, data points located at the center of the cluster are retained to generate preliminary clustering results. A potential partition is provided for each level, and each potential partition represents a set of similar data points at that level.

[0083] Step S23: Cluster optimization, using the density peak cluster optimization algorithm to optimize the preliminary clustering results, while introducing an upward masking strategy to remove outliers in the cluster from each potential partition;

[0084] Step S24: prototype initialization, selecting the data point located at the cluster center from the clustering results of each level as the clustering prototype;

[0085] Step S25: Prototype optimization, using the sharp DBSCAN algorithm to adjust the position of the cluster prototype;

[0086] In this embodiment, in step S22, in the preliminary clustering of each layer, a downward masking strategy is adopted, and multi-granularity information is combined to retain points that appear normal on the surface but show abnormalities in the ultra-fine granularity layer and the fine granularity layer as candidate abnormalities. By capturing abnormal details in the fine granularity layer and the ultra-fine granularity layer, the anomalies ignored in the medium granularity layer and the coarse granularity layer are further filtered to identify false negative data that are misjudged as normal. For example, a point is judged to be normal in the coarse granularity layer, but the pressure fluctuation of the point is found to be abnormal in the ultra-fine granularity layer, thereby identifying the point as a false negative.

[0087] In step S23, an upward masking strategy is introduced to remove outliers in the cluster from each potential partition. After finding the outliers in the clusters of the ultra-fine-grained layer and the fine-grained layer, the outlier information is used to optimize the clustering of the medium-grained layer and the coarse-grained layer. If some cluster centers are found in the ultra-fine-grained layer and the fine-grained layer that are inconsistent with the medium-grained layer and the coarse-grained layer, the cluster contains false negative points.

[0088] In view of the problems of low precision, response delay and drug waste in traditional control methods, the present invention uses a hierarchical clustering collaboration model based on a feedback mechanism in the precise push injection control module to control the push pump. Through the multi-granularity data analysis capability of the hierarchical clustering collaboration model, subtle changes in key parameters of liquid drugs can be captured more finely, thereby achieving high-precision control of the push injection process and reducing drug dosage errors. Based on the real-time feedback mechanism, adaptive adjustments are made according to actual conditions during the push injection process, thereby optimizing drug utilization.

[0089] Embodiment 4: This embodiment is based on the above embodiment. In step S21, the key parameters of the liquid medicine are divided into four levels, which specifically include the following contents:

[0090] The ultra-fine-grained layer and the fine-grained layer capture detailed information, while the medium-grained layer and the coarse-grained layer capture the overall trend;

[0091] Ultra-fine granularity layer: captures tiny fluctuations in flow or pressure every millisecond;

[0092] Fine-grained layer: captures flow fluctuations and instantaneous pressure changes every second;

[0093] Medium particle size layer: captures the average flow rate, average pressure and average temperature within 3 minutes, and obtains the effect of gradual absorption of drugs in the body;

[0094] Coarse-grained layer: Calculate total flow, cumulative bolus dose, average pressure, and average temperature over 30 minutes.

[0095] Example 5, see Figure 3 This embodiment is based on the above embodiment, and step S25 specifically includes the following steps:

[0096] Step S251: input data and take each cluster prototype as the initial point;

[0097] Step S252: constructing a grid, dividing the grid units, and constructing a grid map to store the grid units;

[0098] Step S253: Neighboring grid calculation: calculate the neighborhood points for each initial point and calculate the density center of the points in the neighborhood. The formula used is as follows:

[0099] ;

[0100] ;

[0101] in, is the starting point, is the set of neighborhood points, is the center point of other grid cells, is the distance threshold, is the density center of the points in the neighborhood, is a point in the neighborhood point set;

[0102] Step S254: Distance calculation, calculate the distance between the initial point and the neighboring point. And adjust the position of the initial point. The formula used is as follows:

[0103] ;

[0104] in, is the learning rate;

[0105] Step S255: Merge grid cells. The formula used is as follows:

[0106] ;

[0107] Among them, this formula is for adjacent grid cells, if the distance between them is less than or equal to , then merge;

[0108] ;

[0109] in, is the center position of the merged grid unit, set as the new initial point;

[0110] Step S256: repeat steps S251 to S255 until the initial point position is stable and unchanged;

[0111] Step S257: Update the clustering results, apply the stable initial point to reallocate the clusters, and output the updated clustering results.

[0112] In this embodiment, it is assumed that there are 5 initial points, and the set is recorded as P={p1,p2,p3,p4,p5}, where p1={2,3},p2={4,5}p3={6,7}p4={8,9}p5={10,11};

[0113] Divide the space into multiple grid cells with a side length of 1, construct a grid graph G to store the grid cells, set the side length of each grid cell to 1, and record the set of grid center points as C={(0.5,0.5),(1.5,1.5),…,(9.5,9.5)};

[0114] For each initial point , calculate the set of its neighborhood points , and calculate the density center of the points in the neighborhood , assuming that the neighborhood point set , then the density center for ;

[0115] set up is 0.1, ;

[0116] set up is 1, merge. ,Will Set as a new initial point;

[0117] Repeat steps S251 to S255 until .

[0118] The DBSCAN algorithm is used in the initial architecture of constructing the hierarchical clustering collaborative model. The density threshold is used to effectively identify and filter noise data and outliers to ensure more accurate clustering results. By correcting the clustering at a finer level, the boundary points and sparse area data that may be ignored in the clustering process are reduced.

[0119] Example 6, see Figure 4 Based on the above embodiment, this embodiment uses a meta-heuristic algorithm combined with reinforcement learning in the intelligent analysis module for in-depth analysis, which specifically includes the following steps:

[0120] Step Q1: Feature extraction, using principal component analysis technology to extract key features from the key parameters and control parameters of the pre-treated liquid drug;

[0121] Step Q2: Initial analysis of the metaheuristic algorithm, using genetic algorithm to preliminarily optimize the key parameters and control parameters of liquid drugs, and perform multi-objective optimization at the same time to find the parameter combination and set it as the initial solution;

[0122] Step Q3: Training of reinforcement learning model: Initialize the reinforcement learning model, define the state space and action space, the state space includes the current push state and the characteristics of the liquid drug, the action space includes the parameter combination and the push strategy, simulate the push process, train the reinforcement learning model, use the initial solution of the metaheuristic algorithm as the starting point, and use the reinforcement learning model to select and optimize the push strategy;

[0123] Step Q4: Balance between exploration and exploitation, using the ε-greedy strategy, including the exploration phase, the exploitation phase, and adaptive learning;

[0124] Step Q5: Analysis result generation, using meta-heuristic algorithm to cluster and classify the data during the bolus injection process, identify the drug response pattern under the bolus injection strategy, and obtain the analysis results;

[0125] In this embodiment, after principal component analysis is performed on the key parameters and control parameters of the pretreated liquid drug, the first principal component is mainly determined by the drug concentration and flow rate, while the second principal component is mainly determined by the injection speed and pressure, and the key features are drug concentration, flow rate, injection speed and pressure.

[0126] Embodiment 7: This embodiment is based on the above embodiment, and the three stages in step Q4 specifically include the following contents:

[0127] Exploration phase: finding potential optimization space through mutation operations of metaheuristic algorithms;

[0128] Utilization stage: using the historical optimal strategy to perform push injection operations;

[0129] Adaptive learning: Combined with the initial solution of the metaheuristic algorithm, the reinforcement learning model is used to adjust the ratio of the exploration phase to the exploitation phase;

[0130] In this embodiment, the exploration phase includes the following contents:

[0131] Population initialization: Early in the bolus process, a random parameter population is generated, with each individual representing a set of bolus parameters, such as bolus rate, pressure, and drug flow rate;

[0132] Fitness function: A fitness function is defined for each individual. The fitness function evaluates the effect of the group of injection parameters on drug absorption and patient response. By monitoring the patient's physiological response, the injection effect and drug absorption rate are taken as the main optimization goals.

[0133] Mutation operation: Through mutation operation, small random adjustments are made to individuals in the population to generate new parameter combinations and expand the exploration space;

[0134] Adaptive learning includes the following:

[0135] Initial settings: In the initial stage of bolus injection, a higher ε value is set according to the initial solution provided by the metaheuristic algorithm. In this embodiment, it is set to 0.9, which means that 90% of the time is spent on exploration and 10% of the time is spent on selecting the historical optimal strategy;

[0136] Dynamic adjustment: As the injection process progresses, the ε value is adjusted. If it is detected that the effect of the mutation operation in the exploration phase is gradually getting worse, the ε value is gradually reduced, the exploration is reduced, and the time for selecting the historical optimal strategy is increased.

[0137] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0138] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0139] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent medical micro-injection pump management system, characterized by: It includes core equipment module, sensor module, data processing unit, precision injection control module, intelligent analysis module and alarm module; The core equipment module includes a push pump body and a flow regulating device, the push pump body delivers liquid medicine into the patient's body, and the flow regulating device is used to control the push speed and dosage of the liquid medicine; The sensor module installs sensors inside and around the injection pump to monitor key parameters of the liquid drug in real time, including drug concentration, drug viscosity, drug pH value, drug temperature sensitivity, flow rate, pressure and temperature, and transmits the key parameters of the liquid drug to the data processing unit through a feedback mechanism; The data processing unit is arranged inside the push pump, and pre-processes the key parameters of the liquid medicine and transmits them to the precise push control module and the intelligent analysis module; The precise injection control module controls the injection pump through the key parameters of the liquid drug using a hierarchical clustering collaborative model based on a feedback mechanism. The control parameters include starting, stopping, adjusting flow, injection speed and injection dosage. The intelligent analysis module receives the key parameters and control parameters of the pre-processed liquid medicine, uses a meta-heuristic algorithm combined with reinforcement learning to perform in-depth analysis, obtains an analysis report and transmits it to the alarm module; The alarm module receives the analysis report and performs abnormality detection, and issues an alarm when an abnormal situation occurs; In the precise injection control module, a hierarchical clustering collaborative model based on feedback mechanism is used to control the injection pump, which specifically includes the following steps: Step S1: data preparation, receiving the key parameters of the pre-processed liquid medicine and performing standardization; Step S2: Initialize the model and build the initial architecture of the hierarchical clustering collaborative model; Step S3: real-time feedback, real-time monitoring of key parameters of the liquid drug during the injection process to obtain feedback information; Step S4: Model evaluation, evaluating the hierarchical clustering collaboration model to obtain evaluation results; Step S5: Model update, updating the initial architecture of the hierarchical clustering collaborative model according to the evaluation results and feedback information; In step S2, the initialization model is processed based on the key parameters of the liquid medicine after the standardization process in step S1, specifically including the following steps: Step S21: hierarchical structure setting, dividing the key parameters of liquid medicine into four levels, namely, ultra-fine granularity layer, fine granularity layer, medium granularity layer and coarse granularity layer, to capture multi-granularity information; Step S22: Preliminary clustering: Use the DBSCAN algorithm to perform preliminary clustering on each level. In the clustering process, a downward masking strategy is adopted to filter false negatives by combining multi-granularity information. In each layer of clustering, data points located at the center of the cluster are retained to generate preliminary clustering results. A potential partition is provided for each level, and each potential partition represents a set of similar data points at that level. Step S23: Cluster optimization, using the density peak cluster optimization algorithm to optimize the preliminary clustering results, while introducing an upward masking strategy to remove outliers in the cluster from each potential partition; Step S24: prototype initialization, selecting the data point located at the cluster center from the clustering results of each level as the clustering prototype; Step S25: Prototype optimization, using the sharp DBSCAN algorithm to adjust the position of the cluster prototype.

2. The intelligent medical micro-injection pump management system according to claim 1, characterized in that: Step S25 specifically includes the following steps: Step S251: input data and take each cluster prototype as the initial point; Step S252: constructing a grid, dividing the grid units, and constructing a grid map to store the grid units; Step S253: Neighboring grid calculation: calculate the neighborhood points for each initial point and calculate the density center of the points in the neighborhood. The formula used is as follows: ； ； in, is the starting point, is the set of neighborhood points, is the center point of other grid cells, is the distance threshold, is the density center of the points in the neighborhood, is a point in the neighborhood point set; Step S254: Distance calculation, calculating the distance between the initial point and the neighboring points and adjusting the position of the initial point. The formula used is as follows: ； in, is the learning rate; Step S255: Merge grid cells. The formula used is as follows: ； Among them, this formula is for adjacent grid cells, if the distance between them is less than or equal to , then merge; ； in, is the center position of the merged grid unit, set as the new initial point; Step S256: repeat steps S251 to S255 until the initial point position is stable and unchanged; Step S257: Update the clustering results, apply the stable initial point to reallocate the clusters, and output the updated clustering results.

3. The intelligent medical micro-injection pump management system according to claim 1, characterized in that: In the intelligent analysis module, a meta-heuristic algorithm is used in combination with reinforcement learning for in-depth analysis, which includes the following steps: Step Q1: Feature extraction, using principal component analysis technology to extract key features from the key parameters and control parameters of the pre-treated liquid drug; Step Q2: Initial analysis of the metaheuristic algorithm, using genetic algorithm to preliminarily optimize the key parameters and control parameters of liquid drugs, and perform multi-objective optimization at the same time to find the parameter combination and set it as the initial solution; Step Q3: Training of reinforcement learning model: Initialize the reinforcement learning model, define the state space and action space, the state space includes the current push state and the characteristics of the liquid drug, the action space includes the parameter combination and the push strategy, simulate the push process, train the reinforcement learning model, use the initial solution of the metaheuristic algorithm as the starting point, and use the reinforcement learning model to select and optimize the push strategy; Step Q4: Balance between exploration and exploitation, using the ε-greedy strategy, including the exploration phase, the exploitation phase, and adaptive learning; Step Q5: Analysis result generation, using meta-heuristic algorithm to cluster and classify the data during the push injection process, identify the drug response pattern under the push injection strategy, and obtain the analysis results.

4. The intelligent medical micro-injection pump management system according to claim 1, characterized in that: In step S21, the key parameters of the liquid medicine are divided into four levels, specifically including the following: The ultra-fine-grained layer and the fine-grained layer capture detailed information, while the medium-grained layer and the coarse-grained layer capture the overall trend; Ultra-fine granularity layer: captures tiny fluctuations in flow or pressure every millisecond; Fine-grained layer: captures flow fluctuations and instantaneous pressure changes every second; Medium granularity layer: captures the average flow, average pressure, and average temperature in the range of 1 to 5 minutes, reflecting the short-term trend of drug bolus; Coarse-grained layer: total flow, cumulative bolus dose, mean pressure, and mean temperature over a range of 15 minutes to 1 hour were analyzed.

5. The intelligent medical micro-injection pump management system according to claim 3, characterized in that: The three stages in step Q4 specifically include the following: Exploration phase: finding potential optimization space through mutation operations of metaheuristic algorithms; Utilization stage: using the historical optimal strategy to perform push injection operations; Adaptive learning: Combined with the initial solution of the metaheuristic algorithm, a reinforcement learning model is used to adjust the ratio of the exploration phase to the exploitation phase.

Images