Plant extract transdermal absorption dynamic regulation and control method based on artificial intelligence

Abstract

The invention discloses a plant extract transdermal absorption dynamic regulation and control method based on artificial intelligence. The method comprises the following steps: S1, forming a structured data sequence; s2, generating preliminary prediction data; s3, forming an optimized GRU prediction model; s4, performing real-time prediction on a future transdermal absorption state by utilizing the optimized GRU prediction model, and generating prediction data of an absorption rate, a drug concentration in skin and an accumulated absorption amount; and S5, according to the prediction data and a preset drug release target, carrying out global search and optimization on drug release control parameters through the firefly population optimization algorithm, generating a corresponding dynamic regulation and control instruction, and adjusting the release rate and carrier diffusion characteristics of a drug release device. The release rhythm can be actively adjusted according to the predicted data, so that the concentration stability and the curative effect persistence of the plant active ingredients in the target cortex are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of plant extracts, and in particular to a method for dynamically regulating the transdermal absorption of plant extracts based on artificial intelligence. Background Art

[0002] With the continuous deepening of natural medicine research and development, plant extracts are increasingly widely used in transdermal drug delivery systems due to their multiple biological activities, high safety and few side effects. Various plant components such as essential oils, plant water solvents and plant elixirs are widely used to relieve skin inflammation, analgesia, anti-oxidation and immunomodulation. However, the transdermal absorption process of plant extracts is extremely complex and is affected by multiple factors such as skin barrier status, environmental temperature and humidity, individual physiological differences, and the concentration of the extract's own components. This leads to significant fluctuations in its absorption rate, intensity of action and duration, making it difficult to achieve effective, controllable and stable drug release.

[0003] At present, most transdermal drug formulation systems rely mainly on passive diffusion for release control. Transdermal performance has been improved to a certain extent only by changing the carrier structure or adding penetration enhancers to increase penetration efficiency. However, there is still a lack of real-time prediction and feedback regulation capabilities for the dynamic process of drug absorption in the skin. This is especially true when faced with multi-component, multi-target, and non-constant diffusion systems such as plant extracts. Traditional methods are difficult to deal with effectively. In addition, some studies have begun to try to apply artificial intelligence models to drug absorption prediction, but most of them remain in the static modeling stage and are unable to perceive and adjust drug release parameters in real time. Conventional neural network models lack a deep understanding of dynamic changes in time series and are prone to misjudgment when data fluctuates violently or is under abnormal conditions.

[0004] What is more prominent is that there is currently no closed-loop intelligent optimization method for transdermal absorption that can integrate physiological parameter perception, adaptive modeling and real-time regulation. In the current context of strong volatility of plant extract ingredients, variable absorption pathways, and dynamic changes in skin conditions, how to establish a dynamic regulation mechanism that integrates real-time response, physiological perception, and absorption prediction feedback through the collaboration of artificial intelligence and optimization algorithms is a key technical bottleneck in the current field of transdermal absorption.

[0005] In summary, the existing technology has the following prominent problems in the control of transdermal absorption of plant extracts: first, the ability to perceive the dynamic absorption process is weak, and it is unable to accurately characterize the penetration process under the influence of multiple factors; second, the artificial intelligence prediction model lacks the optimization capability for the complex transdermal mechanism; third, there is a lack of real-time linkage mechanism with the drug release device, and it is impossible to achieve personalized and dynamic regulation, which seriously restricts the intelligent and precise development of the transdermal application of plant extracts. An innovative method that integrates data perception, intelligent modeling and global optimization is urgently needed to solve it. Summary of the Invention

[0006] One purpose of the present invention is to propose a method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence. The present invention can actively adjust the release rhythm according to predicted data, thereby ensuring the stability of the plant active ingredients in the target cortical concentration and the continuity of therapeutic efficacy.

[0007] According to an embodiment of the present invention, a method for dynamically regulating transdermal absorption of plant extracts based on artificial intelligence comprises the following steps:

[0008] S1. Real-time collection of multi-dimensional time series data on skin condition, physiological environment, and plant extract concentration to form a raw data set, and data preprocessing of the raw data set to form a structured data sequence;

[0009] S2. Construct a gated recurrent unit network prediction model based on the structured data sequence. Use the gated recurrent unit network prediction model to dynamically model and predict the transdermal absorption state of plant extracts at different time points to generate preliminary prediction data.

[0010] S3. Using the preliminary prediction data and the corresponding prediction error as evaluation indicators, the firefly swarm optimization algorithm is used to globally optimize the weights, biases, and learning rate parameters of the gated recurrent unit network prediction model to form an optimized GRU prediction model;

[0011] S4. Use the optimized GRU prediction model to make real-time predictions of future transdermal absorption status, generating predicted data on absorption rate, drug concentration in the skin, and cumulative absorption amount;

[0012] S5. Based on the predicted data and the pre-set drug release target, the firefly swarm optimization algorithm is used to globally search and optimize the drug release control parameters, generate corresponding dynamic control instructions, and adjust the release rate of the drug release device and the diffusion characteristics of the carrier.

[0013] Optionally, the S1 includes the following steps:

[0014] S11. Set up a transdermal absorption experimental environment parameter collection plan, dynamically monitor the transdermal absorption process of plant extracts within the pre-qualified time window, collect skin status, physiological environment and plant extract concentration data, and construct the original data set D raw :

[0015]

[0016] Among them, d i Indicates that at time point t i A data record collected, T i Indicates the skin surface temperature, H i Represents local humidity, R irepresents skin resistance, C i represents the drug concentration of the plant extract in the skin surface, and N is the total number of sampling points within the set time window;

[0017] S12. Unify the collection frequency and time interval of the original data set, use equal interval sampling to ensure the consistency of the time series, and establish a data index sequence to keep the time synchronization of each dimension data. i ,R i}, physiological environment data {H i} and plant extract drug concentration data {C i} Perform feature integrity analysis, remove data points whose missing rate exceeds the set threshold, and convert the complete original data set D raw Input into the structured processing module, normalize the data features at each time point, and obtain the normalized original data set D norm , and output the normalized original data set as a structured data sequence.

[0018] Optionally, the S2 includes the following steps:

[0019] S21. Input the normalized raw data set into the gated recurrent unit network, set the time series input window length τ, and construct the input matrix X and the target output sequence Y:

[0020]

[0021] Among them, x j Indicates the time point t j As the starting point, the data segment of τ consecutive time steps, is the skin temperature subsequence, which is used to capture the effect of temperature on drug diffusion rate. is the local humidity subsequence, representing the changing process of local humidity regulating skin permeability. is the skin resistance subsequence, reflecting the dynamic changes of skin barrier function. is the plant extract concentration subsequence, indicating the gradual accumulation of drugs in the skin surface, y j =C j+τ Indicates that at time point t j+τ Target concentration at 5°C;

[0022] S22. Introduce an absorption dynamic gating mechanism into the gated recurrent unit network, control the dynamic behavior of the update gate by introducing a skin state modulation factor, and construct the update gate vector z t :

[0023] z t =σ(W z ·xt +U z ·h t-1 +γ T ·T t +γ R ·R t +γ H ·H t +b z );

[0024] Among them, z t is the update gate vector of the tth time step, x t is the current input feature vector, including skin surface temperature, skin resistance and local humidity, h t-1 is the hidden state of the previous time step, γ T , γ R , γ H are the modulation coefficients of skin surface temperature, skin resistance, and local humidity, respectively, reflecting their regulatory strength on drug penetration under different physiological conditions, and σ is the Sigmoid function;

[0025] S23. Introducing the transdermal absorption trend perception weight function α t , used to perceive the temporal fluctuation trend of drug concentration and dynamically assign attention weights to the gated recurrent unit network:

[0026]

[0027] Among them, α t C is the response intensity of the current time step to the absorption change trend. The larger the value, the more intense the absorption fluctuation at that moment. t 、C t-1 are the drug concentrations of the plant extract in the skin surface at the current and previous time points, respectively; δ is the length of the local trend analysis window; and ∈ is a stabilization factor to prevent the denominator from being zero.

[0028] S24. Based on the absorption dynamic gating mechanism and the transdermal absorption trend perception weight function, the update of the hidden state sequence is jointly controlled to construct an improved state transfer function:

[0029]

[0030] Among them, h t is the updated hidden state sequence, is the candidate hidden state sequence of the current time step, which represents the potential absorption trend under the current skin state and drug characteristics, and ⊙ is the Hadamard product operation;

[0031] S25. Using the updated hidden state sequence h t Generate preliminary forecast data.

[0032] Optionally, the S25 includes the following steps:

[0033] S251. Input the updated hidden state sequence into the prediction output layer, and use the linear mapping function to transform the hidden state to obtain the prediction output vector

[0034] S252. Predict the output vector Split into three subcomponents:

[0035]

[0036] in, represents the predicted concentration of the plant extract in the skin surface at time point t, It indicates the predicted value of transdermal absorption rate per unit time, which is used to evaluate the penetration efficiency of drugs in the current skin state. It represents the predicted cumulative absorbed dose, reflecting the accumulation of the drug in the body over time.

[0037] S253. Based on the predicted value of transdermal absorption rate per unit time The cumulative absorbed dose is calculated at time step intervals to obtain the predicted value of the cumulative drug absorption

[0038] S254. Predicting the concentration of plant extracts in the skin surface Predicted value of transdermal absorption rate per unit time and the predicted value of cumulative drug absorption Composing a preliminary prediction dataset Where T represents the total number of steps in the forecast time series.

[0039] Optionally, S3 includes the following steps:

[0040] S31. Through the preliminary prediction data set and the actual monitoring data set The prediction error between them is used to construct the objective function L for the firefly swarm optimization algorithm. GRU :

[0041]

[0042] Among them, ω1, ω2, and ω3 are weighted coefficients, which respectively control the proportion of predicted concentration, transdermal absorption rate, and cumulative absorption amount in the total error;

[0043] S32. Initialize the firefly population and set the number of individuals to N f , each individual F iRepresents a set of GRU network parameters to be optimized, including the input weights, state weights, and biases of the update gate, and the input weights, state weights, and biases of the candidate hidden states;

[0044] S33. For each individual F i Calculate the brightness function I according to its corresponding prediction error i :

[0045]

[0046] in, represents the objective function obtained by running the GRU model with the i-th individual parameter combination;

[0047] S34. Traverse all individuals F in the population i , for each pair of individuals F i With F j , if the brightness function I j >I i , then update individual F i Parameter location:

[0048]

[0049] Among them, β0 is the initial attraction intensity, γ1 is the light intensity attenuation coefficient, is the Euclidean distance between two individuals in the parameter space, α is the perturbation coefficient, ∈ t is a zero-mean normal random vector used to enhance the search space exploration capability;

[0050] S35. Repeat the iterative update until the maximum number of cycles is reached or the overall brightness value fluctuation of the population is less than the set convergence threshold, and select the individual F with the largest brightness function value in the last generation population. i *The corresponding parameter group is used as the optimal parameter group of the GRU prediction model.

[0051] Optionally, corresponding dynamic control instructions are generated based on the drug release control parameters to adjust the release rate and diffusion behavior of the drug release device. The dynamic control instruction rules are as follows:

[0052] When the predicted concentration And predict the absorption rate When the control instruction: increase the release rate r rate and the diffusion coefficient r diff

[0053] When the predicted concentration And the cumulative absorbed dose When the control instruction: reduce the release rate r rate , keeping the diffusion coefficient unchanged;

[0054] When predicting absorption rate When the concentration is close to the target range, the control instruction is to moderately reduce the diffusion coefficient r diff , limiting the diffusion depth;

[0055] If the predicted concentration and Control instructions: Maintain the current release parameters stable to avoid adjustment fluctuations.

[0056] Optionally, the dynamic control instructions are used to drive the drug release device in real time, automatically adjusting the release behavior according to the deviation between the predicted state and the target state of the transdermal absorption process of the plant extract, thereby achieving control over the release concentration, rate and dosage.

[0057] The beneficial effects of the present invention are:

[0058] (1) The present invention proposes an absorption dynamic gating mechanism and a trend-aware weighting function, which jointly drives the state update behavior of the gating unit by using skin state variables and absorption concentration change trends. This enables the model to maintain a high responsiveness to key points in the time series when facing drastic fluctuations in physiological state or sudden changes in drug concentration, thereby improving the model's accuracy in depicting transdermal absorption changes under dynamic conditions.

[0059] (2) The present invention introduces the firefly swarm optimization algorithm into the GRU network parameter optimization process, constructs a brightness function based on multidimensional prediction error as the fitness evaluation criterion, and adopts the attraction-mutation mechanism to collaboratively search and optimize the core parameters of weight, bias and learning rate, which significantly improves the convergence efficiency and optimal solution quality of the model, and overcomes the weakness of the traditional single-point gradient descent method in multi-peak function search.

[0060] (3) The present invention dynamically generates controllable release rate and diffusion coefficient control instructions through real-time deviation analysis between the predicted output and the set target, thereby realizing adaptive closed-loop control of the drug release device. In particular, in the context of different physiological states or the use of essential oil-water solvent compound dosage forms, the release rhythm can be actively adjusted according to the predicted data, thereby ensuring the stability of the plant active ingredients at the target cortical concentration and the continuity of the therapeutic effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0062] Figure 1 This is a flow chart of a method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence proposed in the present invention. DETAILED DESCRIPTION

[0063] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0064] refer to Figure 1 , a method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence, comprising the following steps:

[0065] S1. Real-time collection of multi-dimensional time series data on skin condition, physiological environment, and plant extract concentration to form a raw data set, and data preprocessing of the raw data set to form a structured data sequence;

[0066] S2. Construct a gated recurrent unit network prediction model based on the structured data sequence. Use the gated recurrent unit network prediction model to dynamically model and predict the transdermal absorption state of plant extracts at different time points to generate preliminary prediction data.

[0067] S3. Using the preliminary prediction data and the corresponding prediction error as evaluation indicators, the firefly swarm optimization algorithm is used to globally optimize the weights, biases, and learning rate parameters of the gated recurrent unit network prediction model to form an optimized GRU prediction model.

[0068] S4. Use the optimized GRU prediction model to make real-time predictions of future transdermal absorption status, generating predicted data on absorption rate, drug concentration in the skin, and cumulative absorption amount;

[0069] S5. Based on the predicted data and the pre-set drug release target, the firefly swarm optimization algorithm is used to perform a global search and optimization of the drug release control parameters, generate corresponding dynamic control instructions, and adjust the release rate of the drug release device and the diffusion characteristics of the carrier.

[0070] In this embodiment, S1 includes the following steps:

[0071] S11. Set up a transdermal absorption experimental environment parameter collection plan, dynamically monitor the transdermal absorption process of plant extracts within the pre-qualified time window, collect skin status, physiological environment and plant extract concentration data, and construct the original data set D raw :

[0072]

[0073] Among them, d i Indicates that at time point t i A data record collected, T i Indicates the skin surface temperature, H i Represents local humidity, R i represents skin resistance, C irepresents the drug concentration of the plant extract in the skin surface, and N is the total number of sampling points within the set time window;

[0074] S12. Unify the collection frequency and time interval of the original data set, use equal interval sampling to ensure the consistency of the time series, and establish a data index sequence to keep the time synchronization of each dimension data. i ,R i}, physiological environment data {H i} and plant extract drug concentration data {C i} Perform feature integrity analysis, remove data points whose missing rate exceeds the set threshold, and convert the complete original data set D raw Input into the structured processing module, normalize the data features at each time point, and obtain the normalized original data set D norm , and output the normalized original data set as a structured data sequence.

[0075] In this embodiment, S2 includes the following steps:

[0076] S21. Input the normalized raw data set into the gated recurrent unit network, set the time series input window length τ, and construct the input matrix X and the target output sequence Y:

[0077]

[0078] Among them, x j Indicates the time point t j As the starting point, the data segment of τ consecutive time steps, is the skin temperature subsequence, which is used to capture the effect of temperature on drug diffusion rate. is the local humidity subsequence, representing the changing process of local humidity regulating skin permeability. is the skin resistance subsequence, reflecting the dynamic changes of skin barrier function. is the plant extract concentration subsequence, indicating the gradual accumulation of drugs in the skin surface, y j =C j+τ Indicates that at time point t j+τ Target concentration at 5°C;

[0079] S22. Introduce an absorption dynamic gating mechanism into the gated recurrent unit network, control the dynamic behavior of the update gate by introducing a skin state modulation factor, and construct the update gate vector z t :

[0080] z t =σ(W z ·x t +U z·h t-1 +γ T ·T t +γ R ·R t +γ H ·H t +b z );

[0081] Among them, z t is the update gate vector of the tth time step, x t is the current input feature vector, including skin surface temperature, skin resistance and local humidity, h t-1 is the hidden state of the previous time step, γ T , γ R , γ H are the modulation coefficients of skin surface temperature, skin resistance, and local humidity, respectively, reflecting their regulatory strength on drug penetration under different physiological conditions, and σ is the Sigmoid function;

[0082] S23. Introducing the transdermal absorption trend perception weight function α t , used to perceive the temporal fluctuation trend of drug concentration and dynamically assign attention weights to the gated recurrent unit network:

[0083]

[0084] Among them, α t C is the response intensity of the current time step to the absorption change trend. The larger the value, the more intense the absorption fluctuation at that moment. t 、C t-1 are the drug concentrations of the plant extract in the skin surface at the current and previous time points, respectively; δ is the length of the local trend analysis window; and ∈ is a stabilization factor to prevent the denominator from being zero.

[0085] S24. Based on the absorption dynamic gating mechanism and the transdermal absorption trend perception weight function, the update of the hidden state sequence is jointly controlled to construct an improved state transfer function:

[0086]

[0087] Among them, h t is the updated hidden state sequence, is the candidate hidden state sequence of the current time step, which represents the potential absorption trend under the current skin state and drug characteristics, and ⊙ is the Hadamard product operation;

[0088] S25. Using the updated hidden state sequence h t Generate preliminary forecast data.

[0089] In this embodiment, S25 includes the following steps:

[0090] S251. Input the updated hidden state sequence into the prediction output layer, and use the linear mapping function to transform the hidden state to obtain the prediction output vector

[0091] S252. Predict the output vector Split into three subcomponents:

[0092]

[0093] in, represents the predicted concentration of the plant extract in the skin surface at time point t, It indicates the predicted value of transdermal absorption rate per unit time, which is used to evaluate the penetration efficiency of drugs in the current skin state. It represents the predicted cumulative absorbed dose, reflecting the accumulation of the drug in the body over time.

[0094] S253. Based on the predicted value of transdermal absorption rate per unit time The cumulative absorbed dose is calculated at time step intervals to obtain the predicted value of the cumulative drug absorption

[0095] S254. Predicting the concentration of plant extracts in the skin surface Predicted value of transdermal absorption rate per unit time and the predicted value of cumulative drug absorption Composing a preliminary prediction dataset Where T represents the total number of steps in the forecast time series.

[0096] In this embodiment, S3 includes the following steps:

[0097] S31. Through the preliminary prediction data set and the actual monitoring data set The prediction error between them is used to construct the objective function L for the firefly swarm optimization algorithm. GRU :

[0098]

[0099] Among them, ω1, ω2, and ω3 are weighted coefficients, which respectively control the proportion of predicted concentration, transdermal absorption rate, and cumulative absorption amount in the total error;

[0100] S32. Initialize the firefly population and set the number of individuals to N f , each individual F i Represents a set of GRU network parameters to be optimized, including the input weights, state weights, and biases of the update gate, and the input weights, state weights, and biases of the candidate hidden states;

[0101] S33. For each individual F i Calculate the brightness function I according to its corresponding prediction error i :

[0102]

[0103] in, represents the objective function obtained by running the GRU model with the i-th individual parameter combination;

[0104] S34. Traverse all individuals F in the population i , for each pair of individuals F i With F j , if the brightness function I j >I i , then update individual F i Parameter location:

[0105]

[0106] Among them, β0 is the initial attraction intensity, γ1 is the light intensity attenuation coefficient, is the Euclidean distance between two individuals in the parameter space, α is the perturbation coefficient, ∈ t is a zero-mean normal random vector used to enhance the search space exploration capability;

[0107] S35. Repeat the iterative update until the maximum number of cycles is reached or the overall brightness value fluctuation of the population is less than the set convergence threshold, and select the individual F with the largest brightness function value in the last generation population. i *The corresponding parameter group is used as the optimal parameter group of the GRU prediction model.

[0108] In this embodiment, the corresponding dynamic control instructions are generated by the drug release control parameters to adjust the release rate and diffusion behavior of the drug release device. The dynamic control instruction rules are as follows:

[0109] When the predicted concentration And predict the absorption rate When the control instruction: increase the release rate r rate and the diffusion coefficient r diff

[0110] When the predicted concentration And the cumulative absorbed dose When the control instruction: reduce the release rate r rate , keeping the diffusion coefficient unchanged;

[0111] When predicting absorption rate When the concentration is close to the target range, the control instruction is to moderately reduce the diffusion coefficient rdiff , limiting the diffusion depth;

[0112] If the predicted concentration and Control instructions: Maintain the current release parameters stable to avoid adjustment fluctuations.

[0113] In this embodiment, the dynamic control instructions are used to drive the drug release device in real time, and automatically adjust the release behavior according to the deviation between the predicted state and the target state of the transdermal absorption process of the plant extract to achieve control of the release concentration, rate and dosage.

[0114] Example 1:

[0115] On November 3, 2024, in the second clinic of the Dermatology Department of the Third Hospital of a certain university, a doctor was performing local application treatment for chronic pruritus on a 34-year-old female patient. Liang Ying had been suffering from the disease for more than half a year. She had tried various treatment methods including Chinese herbal lotions, Western ointments, and packing, but the efficacy was unstable. On that day, the doctor decided to try the latest essential oil-water-soluble composite plant extract transdermal formulation, and combined it with a set of "intelligent dynamic absorption regulation system" that had just been put into the trial stage for treatment, in order to explore ways to improve the problems of uneven local drug efficacy and discontinuous treatment.

[0116] At 9:20, nurse Wang Yue attached the transdermal patch to the inside of Liang Ying's left forearm in accordance with the operating procedures. The patch contains: 10% tea tree oil, 2% rosemary essential oil, 5% licorice water extract and a compound plant agent. A thin film sensor array is integrated between the patch and the skin to collect the patient's local skin temperature, resistance and humidity information in real time.

[0117] At this time, the system background displays:

[0118] Initial skin temperature: 34.2°C; epidermal resistance: 2.8 kΩ;

[0119] Humidity: 62.7% RH; preset target absorption concentration: C t * = 0.84 μg / cm 2 ; Initial predicted absorption rate: v t =0.019μg / cm 2 / min.

[0120] At 9:31, the system detected that the predicted absorption rate values for three consecutive time periods were lower than the set threshold, indicating an abnormal absorption trend. The changes are as follows:

[0121]

[0122] The system automatically triggers an anomaly flag, calls the absorbing dynamic gating mechanism to reweight the state weights, and restarts the Firefly Swarm Optimization algorithm to quickly readjust the GRU model weights in the background. This optimization uses 20 individuals for parallel evolution, converges after 3 rounds of iterations, takes 1.52 seconds, and obtains a new optimal prediction model.

[0123] The skin barrier state is detected to be high. The increase in resistance is expected to be related to a skin reaction to midday medication. The current absorption efficiency is low, and it is recommended to increase the release rate by 12.5% and the diffusion coefficient by 8.1%. The system generates a control instruction package and sends it to the patch's microcontroller unit to adjust the release valve opening interval and the diffusion interface temperature control parameters.

[0124] At 10:05, the system responded:

[0125] Predicted concentration C t =0.86μg / cm 2 ; Actual detection concentration = 0.84 μg / cm 2 ; Deviation is less than ±3%, entering the target range; System status: stable.

[0126] Within two hours thereafter, the itching sensation did not reappear, and the patient complained of a "fast onset and long-lasting" effect. However, the control group (using a traditional patch on the right arm without a control function) still experienced fluctuations at 10:40, with an absorption concentration as low as 0.61 μg / cm 2 , patients responded that “sometimes it relieves the itching, and sometimes it has no effect.”

[0127] Even more convincing is that the system also recorded a sudden change in absorption efficiency throughout the afternoon:

[0128] At 2:52 pm, Liang Ying wiped the patch area with warm water, and the skin surface temperature rose to 36.1°C, the local humidity increased to 78.3%, and the skin resistance dropped to 1.9kΩ.

[0129] The system detects "heat-induced penetration enhancement events" in real time and predicts that the absorption rate will increase sharply in the next 5 minutes, posing a risk of exceeding the safe dose.

[0130] The system responded quickly, adjusting the release rate to decrease by 23.5%, reducing the diffusion coefficient to 0.74 times the initial value, and successfully avoiding concentration exceeding the limit.

[0131] Estimated cumulative absorption in the next 10 minutes before regulation The actual measured value after adjustment was 8.7 μg, and the control deviation was within the standard range (±10%).

[0132] At the end of treatment at 18:00 on the same day, the complete time series data exported by the system is as follows:

[0133]

[0134] The statistics of the right arm traditional group at 18 points are:

[0135] The cumulative absorption amount is only 9.8μg; the target concentration range is maintained for only 2.1 hours; the user's subjective evaluation is "occasionally relieves itching, but the duration is not enough."

[0136] The system used 3,280 pre-trained data sets, and the incremental learning phase incorporated 486 new data sets collected that day. Using a GRU structure with a window size of τ = 6 and 64 hidden nodes, and global optimization using the Firefly algorithm, the model achieved a 58.7% reduction in prediction rate (vt) error compared to an unoptimized GRU. When responding to thermal stimulation and sudden nonlinear resistance changes, the system achieved an average response time of 2.12 minutes, significantly outperforming traditional fixed-release schemes.

[0137] The diagnosis and treatment process not only allows patients to truly feel the significant improvement in the stability of transdermal drug efficacy, but Example 1 also fully demonstrates that the present invention can be effectively operated in an actual clinical environment, and achieves the goal of intelligent dynamic regulation of transdermal absorption of plant extracts through data-driven and algorithm-coordinated optimization.

[0138] The present invention proposes an absorption dynamic gating mechanism and a trend-aware weighting function. By jointly driving the state update behavior of the gating unit through skin state variables and absorption concentration change trends, the model can maintain a high responsiveness to key points in the time series when facing drastic fluctuations in physiological state or sudden changes in drug concentration, thereby improving the model's accuracy in depicting transdermal absorption changes under dynamic conditions.

[0139] The present invention introduces the firefly swarm optimization algorithm into the GRU network parameter optimization process, constructs a brightness function based on multidimensional prediction error as the fitness evaluation criterion, and adopts the attraction-mutation mechanism to collaboratively search and optimize the core parameters of weight, bias and learning rate, which significantly improves the convergence efficiency and optimal solution quality of the model, and overcomes the weakness of the traditional single-point gradient descent method in multi-peak function search.

[0140] The present invention dynamically generates adjustable release rate and diffusion coefficient control instructions through real-time deviation analysis between predicted output and set targets, thereby achieving adaptive closed-loop control of the drug release device. In particular, in situations where different physiological states are being addressed or essential oil-water solvent compound formulations are being used, the release rhythm can be actively adjusted according to the predicted data, thereby ensuring the stability of the target cortical concentration of the plant active ingredients and the persistence of their therapeutic effects.

[0141] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence, characterized in that: The steps include: S1. Real-time collection of multi-dimensional time series data on skin condition, physiological environment, and plant extract concentration to form a raw data set, and data preprocessing of the raw data set to form a structured data sequence; S2. Construct a gated recurrent unit network prediction model based on the structured data sequence. Use the gated recurrent unit network prediction model to dynamically model and predict the transdermal absorption state of plant extracts at different time points to generate preliminary prediction data. S3. Using the preliminary prediction data and the corresponding prediction error as evaluation indicators, the firefly swarm optimization algorithm is used to globally optimize the weights, biases, and learning rate parameters of the gated recurrent unit network prediction model to form an optimized GRU prediction model; S4. Use the optimized GRU prediction model to make real-time predictions of future transdermal absorption status, generating predicted data on absorption rate, drug concentration in the skin, and cumulative absorption amount; S5. Based on the predicted data and the pre-set drug release target, the firefly swarm optimization algorithm is used to globally search and optimize the drug release control parameters, generate corresponding dynamic control instructions, and adjust the release rate of the drug release device and the diffusion characteristics of the carrier.

2. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Set up a transdermal absorption experimental environment parameter collection plan, dynamically monitor the transdermal absorption process of plant extracts within the pre-qualified time window, collect skin status, physiological environment and plant extract concentration data, and construct the original data set D raw : Among them, d i Indicates that at time point t i A data record collected, T i Indicates the skin surface temperature, H i Represents local humidity, R i represents skin resistance, C i represents the drug concentration of the plant extract in the skin surface, and N is the total number of sampling points within the set time window; S12. Unify the collection frequency and time interval of the original data set, use equal interval sampling to ensure the consistency of the time series, and establish a data index sequence to keep the time synchronization of each dimension data. i ,R i }, physiological environment data {H i } and plant extract drug concentration data {C i } Perform feature integrity analysis, remove data points whose missing rate exceeds the set threshold, and convert the complete original data set D raw Input into the structured processing module, normalize the data features at each time point, and obtain the normalized original data set D norm , and output the normalized original data set as a structured data sequence.

3. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 2, characterized in that: The S2 comprises the following steps: S21. Input the normalized raw data set into the gated recurrent unit network, set the time series input window length τ, and construct the input matrix X and the target output sequence Y: Among them, x j Indicates the time point t j As the starting point, the data segment of τ consecutive time steps, is the skin temperature subsequence, which is used to capture the effect of temperature on drug diffusion rate. is the local humidity subsequence, representing the changing process of local humidity regulating skin permeability. is the skin resistance subsequence, reflecting the dynamic changes of skin barrier function. is the plant extract concentration subsequence, indicating the gradual accumulation of drugs in the skin surface, y j =C j+τ Indicates that at time point t j+τ Target concentration at 5°C; S22. Introduce an absorption dynamic gating mechanism into the gated recurrent unit network, control the dynamic behavior of the update gate by introducing a skin state modulation factor, and construct the update gate vector z t : z t =σ(W z ·x t +U z ·h t-1 +g T ·T t +g R ·R t +g H ·H t +b z ); Among them, z t is the update gate vector of the tth time step, x t is the current input feature vector, including skin surface temperature, skin resistance and local humidity, h t-1 is the hidden state of the previous time step, γ T , γ R , γ H are the modulation coefficients of skin surface temperature, skin resistance, and local humidity, respectively, reflecting their regulatory strength on drug penetration under different physiological conditions, and σ is the Sigmoid function; S23. Introducing the transdermal absorption trend perception weight function α t , used to perceive the temporal fluctuation trend of drug concentration and dynamically assign attention weights to the gated recurrent unit network: Among them, α t C is the response intensity of the current time step to the absorption change trend. The larger the value, the more intense the absorption fluctuation at that moment. t 、C t-1 are the drug concentrations of the plant extract in the skin surface at the current and previous time points, respectively; δ is the length of the local trend analysis window; ∈ is a stabilization factor to prevent the denominator from being zero; S24. Based on the absorption dynamic gating mechanism and the transdermal absorption trend perception weight function, the update of the hidden state sequence is jointly controlled to construct an improved state transfer function: Among them, h t is the updated hidden state sequence, is the candidate hidden state sequence of the current time step, which represents the potential absorption trend under the current skin state and drug characteristics, and ⊙ is the Hadamard product operation; S25. Using the updated hidden state sequence h t Generate preliminary forecast data.

4. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 3, characterized in that: The S25 includes the following steps: S251. Input the updated hidden state sequence into the prediction output layer, and use the linear mapping function to transform the hidden state to obtain the prediction output vector S252. Predict the output vector Split into three subcomponents: in, represents the predicted concentration of the plant extract in the skin surface at time point t, It indicates the predicted value of transdermal absorption rate per unit time, which is used to evaluate the penetration efficiency of drugs in the current skin state. It represents the predicted cumulative absorbed dose, reflecting the accumulation of the drug in the body over time. S253. Based on the predicted value of transdermal absorption rate per unit time The cumulative absorbed dose is calculated at time step intervals to obtain the predicted value of the cumulative drug absorption S254. Predicting the concentration of plant extracts in the skin surface Predicted value of transdermal absorption rate per unit time and the predicted value of cumulative drug absorption Composing a preliminary prediction dataset Where T represents the total number of steps in the forecast time series.

5. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 4, characterized in that: The S3 includes the following steps: S31. Through the preliminary prediction data set and the actual monitoring data set The prediction error between them is used to construct the objective function L for the firefly swarm optimization algorithm. GRU : Among them, ω1, ω2, and ω3 are weighted coefficients, which respectively control the proportion of predicted concentration, transdermal absorption rate, and cumulative absorption amount in the total error; S32. Initialize the firefly population and set the number of individuals to N f , each individual F i Represents a set of GRU network parameters to be optimized, including the input weights, state weights, and biases of the update gate, and the input weights, state weights, and biases of the candidate hidden states; S33. For each individual F i Calculate the brightness function I according to its corresponding prediction error i : in, represents the objective function obtained by running the GRU model with the i-th individual parameter combination; S34. Traverse all individuals F in the population i , for each pair of individuals F i With F j , if the brightness function I j >I i , then update individual F i Parameter location: Among them, β0 is the initial attraction intensity, γ1 is the light intensity attenuation coefficient, is the Euclidean distance between two individuals in the parameter space, α is the perturbation coefficient, ∈ t is a zero-mean normal random vector used to enhance the search space exploration capability; S35. Repeat the iterative update until the maximum number of cycles is reached or the overall brightness value fluctuation of the population is less than the set convergence threshold, and select the individual with the largest brightness function value in the last generation of the population The corresponding parameter group is used as the optimal parameter group of the GRU prediction model.

6. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 5, characterized in that: The corresponding dynamic control instructions are generated by the drug release control parameters to adjust the release rate and diffusion behavior of the drug release device. The dynamic control instruction rules are as follows: When the predicted concentration And predict the absorption rate When the control instruction: increase the release rate r rate and the diffusion coefficient r diff When the predicted concentration And the cumulative absorbed dose When the control instruction: reduce the release rate r rate , keeping the diffusion coefficient unchanged; When predicting absorption rate When the concentration is close to the target range, the control instruction is to moderately reduce the diffusion coefficient r diff , limiting the diffusion depth; If the predicted concentration and Control instructions: Maintain the current release parameters stable to avoid adjustment fluctuations.

7. The method for dynamic regulation of transdermal absorption of plant extracts based on artificial intelligence according to claim 6, characterized in that: The dynamic control instructions are used to drive the drug release device in real time, automatically adjust the release behavior according to the deviation between the predicted state and the target state of the plant extract transdermal absorption process, and realize the control of the release concentration, rate and dosage.

Images