Personalized medicine feeding intelligent regulation and control system based on machine learning

Abstract

The invention belongs to a personalized intelligent medicine feeding regulation and control system based on machine learning, and the system comprises the steps: collecting a skin multi-source bioelectricity signal, carrying out the case matching and feature extraction through employing an ant colony-mantis shrimp visual fusion algorithm, and generating a bioelectricity feature data set; the optimization direction of the bioelectricity characteristic data set is shared through decentralized lightweight federal learning, and a medicine feeding equipment interaction model with communication and energy supply capabilities is constructed; establishing a minimum block chain network, and recording equipment operation behaviors to generate a physical hash value; decrypting the encrypted data set and extracting key features, constructing a database according to user health information, and establishing a correlation model of medication parameters and health states; and predicting an optimal medication scheme of the user through the LSTM neural network, converting the scheme into a control instruction, and issuing the control instruction to the medicine feeding equipment interaction model to form a personalized medicine feeding regulation and control closed loop. According to the invention, multiple advantages are formed in physiological data medication regulation individuation and data safety protection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent drug administration control technology, specifically a personalized intelligent drug administration control system based on machine learning. Background Technology

[0002] Traditional intelligent medication administration systems suffer from rudimentary data acquisition methods, often relying on single sensors. This makes them susceptible to environmental interference leading to signal distortion, and basic processing methods alone are insufficient for standardizing multi-source signals. Feature analysis uses a single algorithm to extract surface information, and case matching relies on simple comparisons, prone to mismatches. Medication regimens lack scientific and physiological support. Control modes are rigid, often employing fixed protocols or simple dosage adjustments. They fail to integrate machine learning to establish a correlation between health status and medication parameters, making dynamic optimization based on real-time user feedback difficult and adaptable to individual differences.

[0003] Traditional systems suffer from fixed device interfaces, poor compatibility, low integration of communication modules, and susceptibility to communication interruptions. Furthermore, they lack scientifically defined adaptation requirements through distributed collaboration mechanisms. Data security relies solely on simple encryption, lacking hardware-level protection and tiered access control, resulting in high risks of sensitive data leakage and tampering. Simultaneously, the absence of complete device testing and optimization processes and emergency response mechanisms makes equipment prone to malfunctions and prevents the formation of a predictive-control-feedback-iterative closed loop. Consequently, the system's robustness falls far short of the high security and stability requirements of medical scenarios. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention proposes a personalized medication administration intelligent control system based on machine learning. This invention primarily addresses the problem that current intelligent medication administration systems cannot dynamically optimize based on real-time user feedback and are difficult to adapt to individual differences.

[0005] This invention provides a machine learning-based personalized medication administration intelligent control system, comprising: The electrical signal acquisition module is used to acquire multi-source bioelectrical signals from the skin through a three-dimensional microchannel, and then use a chaotic data normalization algorithm to map them to the same chaotic space to generate a standard bioelectrical signal dataset.

[0006] The feature extraction module is used to preprocess the standard bioelectric signal dataset. It uses an ant colony-mantis shrimp visual fusion algorithm for case matching and feature extraction to generate a bioelectric feature dataset.

[0007] The interactive system construction module is used to connect the mechanical interlock interface with the passive RFID communication module through decentralized lightweight federated learning to share bioelectric feature datasets, thereby constructing an interactive model for drug delivery equipment with communication and power supply capabilities.

[0008] The data encryption module is used to build a minimal blockchain network in the drug delivery device interaction model, calculate the physical hash value based on the device operation behavior, and obtain the encrypted dataset.

[0009] The medication modeling module is used to decrypt encrypted datasets, build a database based on user health information, use a random forest algorithm to filter key features of the database, and establish a medication data feature model of medication parameters and health status.

[0010] The medication administration module is used to predict the user's optimal medication plan based on the medication data feature model using an LSTM neural network, adjust the equipment parameters according to the user's optimal medication plan, and record the effect.

[0011] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes an electrical signal acquisition module comprising: The skin bioelectric signal acquisition unit is used to attach a three-dimensional microchannel structure to the skin surface, acquire bioelectric signals including electromyography and electrocardiography, and convert them into processable analog signals.

[0012] The bioelectric signal preprocessing unit preprocesses the analog signal to eliminate environmental interference and noise, and obtains multi-source electrical signals.

[0013] The chaotic data normalization processing unit maps different multi-source electrical signals to the same chaotic space through a chaotic data normalization algorithm, unifies the signal dimension and range, and generates a standard bioelectrical signal dataset.

[0014] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes a feature extraction module comprising: The data preprocessing unit is used to filter out effective signal segments for case diagnosis from the standard bioelectric signal dataset through signal segmentation and baseline correction.

[0015] The case matching unit, based on valid signal fragments, utilizes the global search capability of the ant colony algorithm and the multi-dimensional feature perception of the mantis shrimp's vision to match similar case features in the case database.

[0016] The feature extraction unit is used to analyze bioelectrical feature datasets of similar cases using a fusion algorithm, including signal time domain, frequency domain, and nonlinear feature parameters.

[0017] According to the machine learning-based personalized medication intelligent control system provided by the present invention, the specific steps for locating key signal features of similar cases in the case matching unit are as follows: Extract the basic feature parameters of effective signal segments, and construct a feature index library based on the basic feature parameters and historical case feature labels.

[0018] The ant colony algorithm is used to filter feature vectors in the case database, calculate the similarity threshold between effective signal segments and historical cases, and select a set of candidate similar cases.

[0019] Based on the candidate similar case set, the signal details and feature correlation of the candidate cases are deeply verified by the multi-dimensional feature perception of the mantis shrimp's vision, and false similar cases are eliminated to obtain effective signal features.

[0020] Based on the effective signal characteristics, they are sorted and integrated according to diagnostic priority to form a matching feature list.

[0021] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes an interaction system construction module comprising: The federated optimization unit is used to analyze the shared optimization direction of the bioelectric feature dataset using decentralized lightweight federated learning, determine the adaptation requirements of the drug delivery device interaction model, and output a list of data transmission requirements.

[0022] Design an interface unit to set up an adjustable meshing mechanical interlock interface for different drug feeding equipment ports based on the data transmission requirements list, and output a mechanical interlock interface structure table.

[0023] The passive communication unit is used to embed the passive RFID communication module into the core area of ​​the interface located according to the mechanical interlock interface structure table, thereby obtaining the mechanical interlock interface.

[0024] The test interface unit is used to connect the terminal and the drug feeding device through a mechanical interlock interface, test the smoothness of the passive RFID communication module, optimize the interface engagement structure and module position based on the test results, and output the optimized interface product.

[0025] The system construction unit is used to establish an interactive model for the drug feeding device based on the optimized interface product, terminal control function, and bioelectrical characteristic data sharing function under federated learning.

[0026] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes a data encryption module comprising: The network building unit is used to set up lightweight blockchain nodes and consensus mechanisms in the core data link of the drug delivery device interaction model, and build a minimal blockchain network.

[0027] The operation acquisition unit is used to collect data on the start-up and shutdown of the drug delivery device, dosage adjustment, and data transmission in real time according to the minimized blockchain network, and write the operation information into the blockchain block in chronological order to generate a chain of device operation behavior records.

[0028] The hash value generation unit is used to perform hash operations on the key fields of each operation record based on the device operation behavior record chain, and generate a unique corresponding physical hash value.

[0029] The data encryption unit is used to bind the physical hash value with the original device operation data, and to encrypt the bound data as a whole using the blockchain encryption algorithm, outputting an encrypted dataset.

[0030] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes a medication modeling module comprising: The data decryption unit decrypts the encrypted dataset, extracts medication operation records, bioelectrical characteristics, and equipment operation data, and forms the original data set.

[0031] The data integration unit is used to associate the original dataset with the user's static health record, remove invalid information, and build a comprehensive analysis database.

[0032] The key feature screening unit is used to evaluate the multidimensional features in the comprehensive analysis database, identify the core features that significantly affect the medication effect, and output a set of key features.

[0033] The association model building unit establishes a mapping association model between medication parameters and user health status based on the key feature set and clinical medication case patterns, optimizes parameter weights, and outputs a medication data feature model.

[0034] According to the machine learning-based personalized medication intelligent control system provided by the present invention, the specific steps for outputting the key feature set in the key feature screening unit are as follows: Based on the comprehensive analysis of the database, the range of multidimensional features to be evaluated is determined, and a list of features to be evaluated is formed.

[0035] Using a list of features to be evaluated, the Random Forest algorithm is used to assess the importance of the multidimensional features in the list, calculate the influence weight of each feature on the medication effect, and generate a feature weight ranking table.

[0036] The core features of the feature weight ranking table are filtered, the filtered core features are classified, and the key feature set is output.

[0037] According to the present invention, a personalized medication administration intelligent control system based on machine learning includes a medication administration execution module comprising: The medication prediction unit inputs user health-related parameters and historical medication records from the medication data feature model into the LSTM neural network. The LSTM neural network dynamically predicts the medication dosage and administration time to generate the optimal medication plan that suits the user's current state.

[0038] The control command conversion unit decomposes the medication parameters in the optimal medication plan into control commands that the device can recognize, according to the instruction protocol format of the medication feeding equipment interaction model, and outputs a standard control command set.

[0039] The equipment parameter control unit is used to send standard control instructions to the drug feeding equipment interaction model and adjust the equipment's operating parameters according to the instructions.

[0040] The regulation effect recording unit monitors the actual drug administration results and physiological feedback data after regulation, and integrates and records the regulation process, execution results and feedback information to form a regulation effect data archive.

[0041] The closed-loop control unit feeds back the control effect data archive to the medication data feature model to verify the effectiveness of the current medication regimen and realize a personalized medication control closed loop.

[0042] According to the machine learning-based personalized medication intelligent control system provided by the present invention, the specific steps for outputting a standard control instruction set in the equipment parameter control unit are as follows: Retrieve the instruction protocol document of the drug delivery device interaction model, determine the instruction types and data transmission formats supported by the device, output the correspondence between drug delivery parameters and device instructions, and establish a mapping table.

[0043] Based on the mapping table, the medication dosage in the optimal medication plan is broken down into motor rotation number commands that the equipment can recognize.

[0044] The disassembled instructions are formatted using an instruction verification tool, integrated into a standard control instruction set, and an instruction execution list is generated.

[0045] The present invention provides a personalized medication intelligent control system based on machine learning, which has multiple advantages in terms of the accuracy of physiological data collection, the scientific nature of feature analysis, the stability of device interaction, the personalization of medication control, and the protection of data security through modular design, cross-technology integration and closed-loop management of the whole process.

[0046] The beneficial effects of this invention are as follows: 1. This invention achieves high-precision capture of bioelectrical signals through a three-dimensional microchannel and layered preprocessing. The three-dimensional microchannel, with its flexible substrate conforming to the skin's curved surface and combined with an inner conductive coating and electrode array, can seamlessly capture multi-source raw signals such as electromyography and electrocardiography, avoiding signal loss due to poor contact in traditional acquisition methods. After layered preprocessing including low-frequency filtering, high-frequency denoising, and low-noise amplification, a chaotic data normalization algorithm maps different signals to a unified chaotic space, eliminating environmental interference and signal noise, and unifying signal dimensions. This provides a clean and standardized data source for subsequent feature analysis, solving the pain points of complex acquisition, difficult processing, and disordered formats of multi-source bioelectrical signals.

[0047] 2. This invention integrates the advantages of ant colony optimization (ACO) and mantis shrimp vision to improve the accuracy of feature extraction and the reliability of case matching. First, effective signal segments are screened through signal segmentation and baseline correction. Then, the global search capability of the ACO algorithm is used to traverse the case database to quickly locate candidate similar cases. Finally, the multi-dimensional perception advantage of mantis shrimp vision is used to deeply verify signal details and eliminate falsely similar cases. Simultaneously, core features are mined from the time domain, frequency domain, and nonlinear dimensions to form a structured bioelectrical feature dataset. Compared with traditional single algorithms, this significantly reduces feature omissions and the probability of case mismatch, providing scientific physiological feature support for medication regimen development.

[0048] 3. This invention achieves efficient adaptation and stable communication through decentralized federated learning and modular interface design. The federated learning framework allows each node to share only the optimization direction of the bioelectrical feature dataset, ensuring data privacy while collaboratively determining device adaptation requirements. The adjustable meshing structure of the mechanical interlocking interface adapts to different drug delivery device ports, and the integration of the passive RFID communication module with the interface, combined with an insulating isolation design, ensures stable communication. Further interface testing and optimization form a device interaction development process of design-integration-testing-optimization, avoiding problems such as poor device compatibility and communication interruptions, and improving the overall operational stability of the system.

[0049] 4. This invention achieves precise prediction, dynamic regulation, and iterative effect through personalized and closed-loop optimization of medication control, relying on machine learning models and a full-process feedback mechanism. The medication modeling module uses random forests to select key features and establish a correlation model between medication parameters and health status. The medication execution module uses an LSTM neural network to predict the optimal treatment plan based on user health parameters and historical medication records, then breaks down the plan into control commands that the device can recognize, adjusting device parameters in real time. The regulation effect recording unit collects device operation data and user physiological feedback, forming a data archive that is fed back to the model to drive dynamic model optimization. Compared to fixed medication plans, this system can adjust in real time according to changes in the user's health status, meeting personalized medication needs. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a block diagram of a personalized medication administration intelligent control system based on machine learning provided in an embodiment of the present invention; Figure 2 This is a flowchart of a personalized medication administration intelligent control system based on machine learning provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below according to specific embodiments.

[0053] like Figures 1 to 2 As shown in the figure, an embodiment of the present invention provides a personalized medication administration intelligent control system based on machine learning. The system includes: The electrical signal acquisition module collects multi-source bioelectrical signals from the skin through a three-dimensional microchannel, maps them to the same chaotic space using a chaotic data normalization algorithm, and generates a standard bioelectrical signal dataset.

[0054] The skin bioelectric signal acquisition unit is used to precisely capture multi-source bioelectric signals such as electromyography and electrocardiography by using a three-dimensional microchannel structure to fit the skin surface, and convert the raw electrical signals into processable analog signals.

[0055] First, the three-dimensional microchannel structure is aligned with the skin area to be sampled. A flexible substrate conforms to the skin's curvature, ensuring seamless contact between the microchannel and the skin surface. A conductive coating on the inner wall of the microchannel guides multi-source bioelectrical signals, such as electromyography (EMG) and electrocardiogram (ECG), into the channel. These raw electrical signals are then captured in real time by an internal electrode array. The captured raw electrical signals are transmitted to a signal conversion module, where they are converted into analog signals suitable for subsequent processing. Finally, they are temporarily stored in a buffer module, awaiting use by the preprocessing unit.

[0056] The bioelectric signal preprocessing unit filters, denoises, and amplifies the analog signal to eliminate environmental interference and signal noise, thus obtaining a pure multi-source bioelectric signal.

[0057] The analog signal is received from the buffer module of the acquisition unit and imported into the preprocessing flow through a dedicated interface. First, the low-frequency filtering module is activated to filter out low-frequency environmental interference such as power frequency. Then, the high-frequency denoising algorithm is called to eliminate high-frequency noise generated during signal transmission. Next, the low-noise amplification module amplifies the amplitude of the denoised signal to an appropriate range. The amplified signal is then subjected to integrity verification, and unqualified segments with disconnections or abnormal amplitudes are removed. Finally, a clean multi-source bioelectric signal is output to the normalization processing unit.

[0058] The chaotic data normalization processing unit maps signals of different types and amplitudes to the same chaotic space based on pure multi-source bioelectric signals using a chaotic data normalization algorithm, unifying the signal dimension and range, and generating a standard bioelectric signal dataset.

[0059] Based on pure multi-source bioelectrical signals, features such as amplitude range and frequency distribution of various signals are extracted to establish a feature database. A unified chaotic space with defined dimensions and boundary parameters is then established based on this feature database. A chaotic data normalization algorithm is then run to map different signals to this chaotic space. The mapped signals are verified to ensure no distortion, no overlap, and preservation of core features. These signals are then integrated to generate a standard bioelectrical signal dataset.

[0060] The feature extraction module is used to perform case matching and feature extraction based on a standard bioelectric signal dataset and an ant colony-mantis shrimp visual fusion algorithm to generate a bioelectric feature dataset.

[0061] The data preprocessing unit is used to process the standard bioelectric signal dataset through signal segmentation and baseline correction to screen out effective signal segments relevant to case diagnosis.

[0062] The system receives a standard bioelectrical signal dataset and first segments the signal according to a preset time interval or physiological cycle, breaking the continuous signal into independent signal segments. Baseline correction is performed on each segment by calculating and inversely compensating for baseline drift, eliminating baseline shifts caused by skin contact or device noise. Based on key diagnostic indicators such as signal amplitude range and frequency interval, a screening threshold is set to remove invalid segments exceeding the threshold, retaining only the valid signal segments relevant to the diagnosis.

[0063] The case matching unit, based on valid signal fragments, utilizes the global search capability of the ant colony algorithm and the multi-dimensional feature perception advantage of the mantis shrimp's vision to complete accurate matching in the case database and locate the key signal features of similar cases.

[0064] Extract the basic feature parameters of effective signal segments, and construct a feature index library by combining them with the feature labels of historical cases in the case database.

[0065] The system calls upon valid signal fragment features from the feature index library, utilizes the global optimization capability of the ant colony algorithm, traverses the feature vectors in the case database, calculates the similarity threshold between the signal features and historical cases, and filters out a set of candidate similar cases.

[0066] Based on the candidate similar case set, the signal details and feature correlation of the candidate cases are deeply verified by the multi-dimensional feature perception of the mantis shrimp's vision, and false similar cases are eliminated to obtain the key signal features for effective signal segment matching.

[0067] Based on the key signal features, they are sorted and integrated according to diagnostic priority to form a structured list of matching features.

[0068] The feature extraction unit is used to further mine the time domain, frequency domain and nonlinear feature parameters of the signal based on the key signal features matched and located in the case, and obtain the bioelectric feature dataset through the fusion algorithm.

[0069] The key signal features for case matching and localization are imported, and three feature analysis dimensions—time domain, frequency domain, and nonlinearity—are established to determine the specific parameters to be mined in each dimension. A fusion algorithm is run to calculate the statistical parameters of the signal in the time domain, obtain the frequency distribution characteristics through Fourier transform in the frequency domain, and analyze the signal complexity using chaos theory in the nonlinearity dimension. The feature parameters in the three dimensions are normalized to eliminate dimensional differences, and redundant parameters are removed through feature correlation analysis, retaining the core features with diagnostic value. The filtered feature parameters are then integrated by category to generate a structured bioelectrical feature dataset.

[0070] The interactive system construction module is used to build an interactive system for drug delivery equipment with communication and power supply capabilities by optimizing the shared bioelectric feature dataset through decentralized lightweight federated learning, and connecting the mechanical interlock interface with the passive RFID communication module.

[0071] The federated optimization unit is used to analyze the shared optimization direction of the bioelectrical feature dataset using decentralized lightweight federated learning, determine the adaptation requirements of the drug delivery device interaction system, and output a data transmission requirement list. A decentralized lightweight federated learning framework is set up, the bioelectrical feature dataset is imported, and efficiency bottlenecks and privacy protection pain points in the data sharing process are analyzed through distributed node collaboration to determine the optimization direction. Based on the optimization direction, the functional requirements of the drug delivery device interaction system are matched, key indicators such as data transmission rate, encryption level, and device compatibility are determined, and a data transmission requirement list is compiled.

[0072] The interface design unit is used to design an adjustable meshing mechanical interlock interface based on the differences in ports of different medication feeding devices according to the data transmission requirements list, and output a mechanical interlock interface structure table. It analyzes the equipment compatibility requirements in the data transmission requirements list, statistically analyzes the differences in size and interface type of different medication feeding device ports, establishes a port parameter database, designs an adjustable meshing mechanical interlock interface, determines the meshing stroke range, interface material and sealing method, and refines the interface design details, forming a mechanical interlock interface structure table including dimensional parameters, structural drawings, and compatible equipment models.

[0073] The passive communication unit is used to locate the core area of ​​the interface according to the mechanical interlock interface structure table. The passive RFID communication module is then embedded into this core area to obtain the mechanical interlock interface. The mechanical interlock interface structure table is read to locate the coordinates and spatial dimensions of the core area, ensuring a suitable installation space for the passive RFID communication module. The passive RFID communication module is then embedded into the core area. An insulating coating isolates the module antenna from the interface metal components to prevent signal interference. After module installation, the various components of the mechanical interlock interface are assembled to form a mechanical interlock interface with basic communication capabilities.

[0074] The test interface unit is used to connect the terminal and the medication delivery device via a mechanical interlock interface to test the smoothness of the RFID module. Based on the test results, the interface engagement structure and module position are optimized, and an optimized interface product is output. The mechanical interlock interface is used to connect the terminal to different models of medication delivery devices, and a communication test program is initiated to detect the smoothness and stability of data transmission from the RFID module, recording any abnormal data. For issues such as loose connections and signal interruptions encountered during testing, the adjustment precision of the interface engagement structure and the installation position of the RFID module are adjusted. The testing and adjustment are repeated until the interface performance meets the standards, outputting an optimized interface product with a stable structure and smooth communication.

[0075] The system construction unit is used to debug the collaboration based on the optimized interface product, terminal control function, and bioelectrical characteristic data sharing function under federated learning, and to establish the interaction model of the drug feeding device. It integrates the optimized interface product, terminal control function module, and federated learning bioelectrical characteristic data sharing module to build the hardware and software framework of the drug feeding device interaction system.

[0076] Initiate collaborative debugging to test the linkage effect of interface connection, terminal command issuance, and data sharing, fix functional connection vulnerabilities, solidify system functional logic after debugging, and establish a stable interactive model for drug feeding equipment.

[0077] The data encryption module is used in the drug delivery device interaction system. The main control chip integrates a silicon-based PUF chip and builds a minimal blockchain network to record the device operation behavior, generate physical hash values, and obtain encrypted datasets.

[0078] The network setup module is used to set up lightweight blockchain nodes and consensus mechanisms in the core data link of the drug delivery device interaction model, configure the communication protocols and data storage permissions of network nodes, and build a minimal blockchain network.

[0079] The operation acquisition unit is used to collect operational behaviors such as starting and stopping of the drug delivery device, adjustment of drug dosage, and data transmission in real time according to the minimized blockchain network, and write the operation information into the blockchain block in the order of timestamps to generate a chain of device operation behavior records.

[0080] The hash value generation unit is used to perform hash operations on the key fields of each operation record based on the device operation behavior record chain, and generate a unique corresponding physical hash value.

[0081] The data encryption unit is used to bind the physical hash value with the original device operation data, and to encrypt the bound data as a whole using the blockchain encryption algorithm, outputting an encrypted dataset.

[0082] The medication modeling module is used to decrypt encrypted datasets and extract key features, build a database based on user health information, use the random forest algorithm to filter key features, establish a correlation model between medication parameters and health status, and output a medication data feature model.

[0083] The data decryption unit decrypts the encrypted dataset according to a hierarchical access control mechanism, extracting medication operation records, bioelectrical characteristics, and equipment operation data to form the original data set. It acquires the encrypted dataset, invokes a preset hierarchical access control mechanism to verify the current operator's access level, and initiates the decryption process after confirming that the decryption permissions are met. After decryption, it extracts three core information categories from the data: medication operation records, bioelectrical characteristics, and equipment operation data. The extracted information undergoes preliminary formatting, removing obvious formatting errors to form the original data set.

[0084] The data integration unit is used to associate the raw dataset with users' static health records and real-time physiological feedback data, remove invalid and redundant information, and construct a comprehensive analytical database covering multi-dimensional information. Based on the raw dataset, it associates the corresponding user's static health record with the user's unique identifier, while simultaneously accessing real-time collected user physiological feedback data. The associated data is cleaned, identifying and removing invalid and redundant information, such as duplicate records and abnormal values. The cleaned multi-dimensional data is then classified and organized according to a unified standard to construct a comprehensive analytical database covering multi-dimensional information.

[0085] The key feature screening unit uses a random forest algorithm to evaluate and screen the multidimensional features in the comprehensive analysis database, identifying core features that significantly impact medication efficacy and outputting a set of key features. The unit reads data from the comprehensive analysis database, identifies all feature dimensions, determines the range of features to be evaluated, uses a random forest algorithm to assess the importance of each feature dimension, calculates the weight of each feature's impact on medication efficacy, and filters out the core features with significant impact based on the weight results. These features are then summarized and organized to output a set of key features.

[0086] The association model building unit establishes a mapping association model between medication parameters and user health status based on a key feature set and clinical medication case patterns. The model is trained to optimize parameter weights and outputs a medication data feature model. The process involves acquiring a key feature set and collecting clinical medication cases to extract medication patterns. These two sets serve as the foundational data for model building. Based on this foundational data, a mapping association model between medication parameters and user health status is established. Initial parameter weights are set, and the model is trained using historical medication data. The parameter weights are continuously adjusted and optimized to improve model accuracy, ultimately outputting a medication data feature model.

[0087] The medication execution module is used to predict the user's optimal medication plan based on the medication data feature model and through an LSTM neural network. The plan is then converted into control commands and sent to the medication feeding device interaction model to adjust the device parameters and record the effects, forming a personalized medication control closed loop.

[0088] The medication prediction unit inputs user health-related parameters and historical medication records from the medication data feature model into an LSTM neural network. The network dynamically predicts medication dosage and administration time, generating an optimal medication plan tailored to the user's current state. The user health-related parameters and historical medication records are extracted from the medication data feature model, and the data format is standardized before being imported into the LSTM neural network. The network dynamically adjusts its prediction logic based on the user's current health status, iteratively calculating medication dosage and administration time. Based on the calculation results, an optimal medication plan tailored to the user's current state is generated, ensuring a high degree of match between the plan and the user's physiological characteristics and medication history.

[0089] The control command conversion unit, based on the optimal medication plan and in accordance with the command protocol format of the medication feeding equipment interaction model, decomposes the medication parameters in the plan into control commands that the equipment can recognize, forming a standard control command set.

[0090] Using the optimal medication plan, an interface call tool is used to retrieve the instruction protocol document of the medication delivery device's interaction model. This determines the types of instructions supported by the device, the data transmission format, and the instruction verification rules. The correspondence between medication parameters and device instructions in the plan is then established, creating a complete mapping table. Based on this mapping table, the medication dosage in the optimal medication plan is broken down into motor rotation count instructions that the device can recognize, and the medication administration time is broken down into timer trigger instructions. A unique identifier and execution priority are added to each decomposed instruction to avoid conflicts between different instructions during execution. An instruction verification tool is used to verify the format of all decomposed instructions, checking whether the instruction parameters meet the device's hardware operating limits. After successful verification, the instructions are sorted in execution order and integrated into a standard control instruction set containing instruction identifiers, parameters, and priorities, generating a clear instruction execution list.

[0091] The equipment parameter control unit is used to send instructions to the drug feeding equipment interaction model according to the standard control instruction set, and adjust the operating parameters of the equipment according to the instruction requirements.

[0092] The system receives a standard control command set, reads the unique identifier and execution priority of each command through the command parsing module, and parses the command content line by line according to priority to determine the types of operating parameters that need to be adjusted and their corresponding target values. The types of operating parameters include the engagement depth of the mechanical interlock interface, the signal transmission frequency of the RFID communication module, etc.

[0093] A wireless communication module establishes an adaptive connection with the RFID module of the medication delivery device's interaction model. Parsed instructions are sent sequentially according to priority. After each instruction is sent, the system waits for a confirmation signal from the device before sending the next instruction, ensuring no loss or omission during transmission. During instruction execution, real-time data collection of the device's operating parameters, including motor speed and interface position sensor data, is performed. This data is compared in real-time with the target values ​​required by the instructions. If a deviation is found, a parameter correction instruction is immediately issued to adjust the device's operating status, ensuring accurate parameter compliance.

[0094] The regulation effect recording unit monitors the operating status and actual drug administration results after the equipment parameters are adjusted, collects physiological feedback data after the user takes the drug, and integrates and records the regulation process, execution results and feedback information to form a regulation effect data archive.

[0095] The device's built-in sensors monitor its operational status in real time after parameter adjustments. Dosage and time sensors record the actual number of motor rotations and timer trigger times, respectively. The number of motor rotations can be directly converted into the actual dosage. Simultaneously, a camera or weight sensor collects actual medication results to confirm accurate tablet placement in the medication dispenser, integrating this data into a device execution data log. At two key time points after medication administration, wrist-worn physiological monitors and other vital sign acquisition devices collect real-time physiological feedback data, including changes in blood pressure, heart rate, and blood glucose levels. A voice interaction module records the user's subjective feelings, identifying any discomfort such as dizziness or nausea, forming a complete user feedback dataset. The device execution data log, user feedback dataset, and standard control command set are linked, organized sequentially by command issuance time, device execution process, actual results, and user feedback timeline. Metadata such as data acquisition personnel and acquisition time is added, generating a control effect data archive containing text descriptions and data tables, which is ultimately stored in an encrypted database.

[0096] The closed-loop control unit transmits the control effect data archive back to the medication data feature model to verify the effectiveness of the current medication regimen and realize the closed-loop control of personalized medication administration.

[0097] Key information is extracted from the regulatory effect data archive, covering deviations in medication regimen execution, changes in user physiological feedback, and records of equipment malfunctions. Data statistical tools are used to calculate the regimen's effectiveness rate and determine the achievement of key indicators such as blood glucose targets. This key information and the regimen's effectiveness rate are fed back to the medication data feature model via a data interface. The model's update layer uses this information as new training samples, adjusting the weights of user health-related parameters within the model. If a user's blood pressure fluctuates significantly after medication, the weight of the blood pressure parameter in the model is increased to further optimize the prediction logic. Combining the updated model parameters, the actual effectiveness of the current medication regimen on the user is verified. If the regimen's effectiveness rate is high, the current regimen framework is maintained; if the effectiveness rate is low, a new round of medication prediction is triggered, and a new regimen is generated based on the updated model, forming a closed loop of personalized medication regulation encompassing prediction, regulation, recording, feedback, and updating.

[0098] In summary, this embodiment provides a personalized medication administration intelligent control system based on machine learning. Through modular design, cross-technology integration, and closed-loop management throughout the entire process, it achieves multiple advantages in terms of the accuracy of physiological data acquisition, the scientific nature of feature analysis, the stability of device interaction, the personalization of medication control, and data security. This invention achieves high-precision capture of bioelectrical signals through a three-dimensional microchannel and layered preprocessing. The three-dimensional microchannel, with its flexible substrate conforming to the skin's curved surface, combined with an inner conductive coating and electrode array, can seamlessly capture multi-source raw signals such as electromyography and electrocardiography, avoiding signal loss due to poor contact in traditional acquisition methods. Layered preprocessing, including low-frequency filtering, high-frequency denoising, and low-noise amplification, followed by a chaotic data normalization algorithm, maps different signals to a unified chaotic space. This eliminates environmental interference and signal noise, unifies signal dimensions, and provides a clean and standardized data source for subsequent feature analysis, solving the pain points of complex multi-source bioelectrical signal acquisition, difficult processing, and disordered formats.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized medication administration intelligent control system based on machine learning, comprising: The electrical signal acquisition module is used to acquire multi-source bioelectrical signals from the skin through a three-dimensional microchannel, and to map them to the same chaotic space using a chaotic data normalization algorithm to generate a standard bioelectrical signal dataset. The feature extraction module is used to preprocess the standard bioelectric signal dataset, and to perform case matching and feature extraction using an ant colony-mantis shrimp visual fusion algorithm to generate a bioelectric feature dataset. The interactive system construction module is used to share the bioelectric feature dataset through decentralized lightweight federated learning, connect the mechanical interlock interface with the passive RFID communication module, and build an interactive model of the drug feeding device with communication and power supply capabilities. The data encryption module is used to build a minimal blockchain network in the interaction model of the drug feeding device, calculate the physical hash value based on the device operation behavior, and obtain the encrypted dataset. The medication modeling module is used to decrypt the encrypted dataset, build a database based on user health information, use a random forest algorithm to filter key features of the database, and establish a medication data feature model of medication parameters and health status. The medication administration module is used to predict the user's optimal medication plan using an LSTM neural network based on the medication data feature model, adjust the equipment parameters according to the user's optimal medication plan, and record the effect.

2. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The electrical signal acquisition module includes: The skin bioelectric signal acquisition unit is used to attach a three-dimensional microchannel structure to the skin surface, acquire bioelectric signals including electromyography and electrocardiography, and convert them into processable analog signals. The bioelectric signal preprocessing unit preprocesses the analog signal to eliminate environmental interference and noise, and obtains multi-source electrical signals; The chaotic data normalization processing unit maps different multi-source electrical signals to the same chaotic space through a chaotic data normalization algorithm, unifies the signal dimension and range, and generates a standard bioelectrical signal dataset.

3. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The feature extraction module includes: The data preprocessing unit is used to filter out effective signal segments for case diagnosis from the standard bioelectric signal dataset through signal segmentation and baseline correction; The case matching unit, based on the effective signal fragments, utilizes the global search capability of the ant colony algorithm and the multi-dimensional feature perception of the mantis shrimp's vision to match similar case features in the case database; The feature extraction unit is used to analyze the bioelectrical feature dataset of the similar cases using a fusion algorithm, including signal time domain, frequency domain, and nonlinear feature parameters.

4. The personalized medication administration intelligent control system based on machine learning according to claim 3, characterized in that: In the case matching unit, the specific steps for locating key signal features of similar cases are as follows: Extract the basic feature parameters of the effective signal segments, and construct a feature index library based on the basic feature parameters and historical case feature labels; The ant colony algorithm is used to filter feature vectors in the case database, and the similarity threshold between the effective signal segment and historical cases is calculated to select a set of candidate similar cases. Based on the candidate similar case set, the signal details and feature correlations of the candidate cases are deeply verified through the multi-dimensional feature perception of the mantis shrimp's vision, and false similar cases are eliminated to obtain effective signal features; Based on the effective signal features, they are sorted and integrated according to diagnostic priority to form a matching feature list.

5. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The interactive system building modules include: The federated optimization unit is used to analyze the shared optimization direction of the bioelectric feature dataset using decentralized lightweight federated learning, determine the adaptation requirements of the drug delivery device interaction model, and output a data transmission requirement list. Design an interface unit to set up an adjustable meshing mechanical interlock interface for different drug feeding device ports according to the data transmission requirement list, and output a mechanical interlock interface structure table. A passive communication unit is used to embed a passive RFID communication module into the core area of ​​the interface located according to the mechanical interlock interface structure table, so as to obtain the mechanical interlock interface. The test interface unit is used to connect the terminal and the drug feeding device through the mechanical interlock interface, test the smoothness of the passive RFID communication module, optimize the interface engagement structure and module position based on the test results, and output the optimized interface product. The system construction unit is used to establish an interactive model for the drug feeding device based on the optimized interface product, terminal control function, and bioelectrical characteristic data sharing function under federated learning.

6. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The data encryption module includes: The network building unit is used to set up lightweight blockchain nodes and consensus mechanisms in the core data link of the drug feeding device interaction model, and build a minimal blockchain network. The operation acquisition unit is used to collect the start-up and shutdown of the drug delivery device, the adjustment of the drug dosage, and the data transmission in real time according to the minimized blockchain network, and write the operation information into the blockchain block in chronological order to generate a chain of device operation behavior records. The hash value generation unit is used to perform hash operations on the key fields of each operation record according to the device operation behavior record chain, and generate a unique corresponding physical hash value. The data encryption unit is used to bind the physical hash value with the original device operation data, and to perform overall encryption processing on the bound data using the blockchain encryption algorithm to output an encrypted dataset.

7. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The medication modeling module includes: The data decryption unit decrypts the encrypted dataset, extracts medication operation records, bioelectrical characteristics, and equipment operation data, and forms the original data set. The data integration unit is used to associate the original data set with the user's static health record, remove invalid information, and construct a comprehensive analysis database; The key feature screening unit is used to evaluate the multidimensional features in the comprehensive analysis database, identify the core features that significantly affect the medication effect, and output a set of key features; The association model construction unit establishes a mapping association model between medication parameters and user health status based on the key feature set and clinical medication case patterns, optimizes parameter weights, and outputs a medication data feature model.

8. The personalized medication administration intelligent control system based on machine learning according to claim 7, characterized in that: The specific steps for outputting the key feature set in the key feature filtering unit are as follows: Based on the comprehensive analysis database, the range of multidimensional features to be evaluated is determined, and a list of features to be evaluated is formed. Using the aforementioned list of features to be evaluated, the Random Forest algorithm is employed to assess the importance of the multidimensional features in the list, calculate the influence weight of each feature on the medication effect, and generate a feature weight ranking table. The core features of the feature weight ranking table are selected, the selected core features are classified, and a set of key features is output.

9. The personalized medication administration intelligent control system based on machine learning according to claim 1, characterized in that: The medication administration module includes: The medication prediction unit inputs user health-related parameters and historical medication records from the medication data feature model into the LSTM neural network. The LSTM neural network dynamically predicts the medication dosage and administration time to generate the optimal medication plan that is suitable for the user's current state. The control command conversion unit, according to the optimal medication plan and the command protocol format of the medication feeding device interaction model, decomposes the medication parameters in the plan into control commands that the device can recognize, and outputs a standard control command set. The equipment parameter control unit is used to send the standard control instruction set to the drug feeding equipment interaction model and adjust the operating parameters of the equipment according to the instruction requirements. The regulation effect recording unit monitors the actual drug administration results and physiological feedback data after regulation, and integrates and records the regulation process, execution results and feedback information to form a regulation effect data archive. The closed-loop control unit feeds back the control effect data archive to the medication data feature model to verify the effectiveness of the current medication regimen and realize a personalized medication control closed loop.

10. The personalized medication administration intelligent control system based on machine learning according to claim 9, characterized in that: The specific steps for outputting the standard control command set in the equipment parameter control unit are as follows: Retrieve the instruction protocol document of the drug delivery device interaction model, determine the instruction types and data transmission formats supported by the device, output the correspondence between drug administration parameters and device instructions, and establish a mapping table; Based on the mapping table, the medication dosage in the optimal medication plan is broken down into motor rotation number commands that the equipment can recognize; The disassembled instructions are formatted using an instruction verification tool, integrated into a standard control instruction set, and an instruction execution list is generated.

Images