Method for evaluating damage of air pollution to human skin health based on PPO-LSTM fusion algorithm

Abstract

The invention belongs to the technical field of medical evaluation, and relates to a method for evaluating damage of air pollution to human skin health based on a PPO-LSTM fusion algorithm. The method comprises the following steps: acquiring multi-source data; preprocessing data, and constructing a multi-dimensional time sequence feature data set; establishing a PPO-LSTM fusion model, performing feature extraction and reinforcement learning optimization on the multi-dimensional time sequence features, and outputting skin health core indexes; performing prediction and quantitative evaluation on the functional states of the cuticle, the epidermal layer and the corium layer; generating a personalized nursing decision scheme for skin health maintenance; outputting corresponding early warning information and coping suggestions by judging the risk levels of the skin barrier function and the dermal metabolism function; and updating parameters of the skin health injury evaluation model. According to the method, multi-dimensional data are integrated in the same model at the same time, the evaluation dimension is more comprehensive, the accuracy of skin health injury evaluation is improved, and a complete closed loop from detection evaluation to nursing scheme generation and risk prompt is realized through decision making and early warning.

Description

Technical Field

[0001] This invention belongs to the field of medical assessment technology, specifically relating to a method for assessing the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm. Background Technology

[0002] With the development of dermatology, wearable sensors and artificial intelligence technologies, skin health management has evolved from simple beauty care to continuous, data-driven health monitoring and early intervention.

[0003] Some home or professional skin analysis instruments primarily detect single or limited indicators such as skin surface moisture, oil, pore size, and skin tone, often relying on capacitance, resistance, or optical reflectance measurements. These methods mainly describe the surface state and are insufficient to reflect the functional synergy and physiological changes between the stratum corneum, stratum spinosum, stratum basale, and dermis.

[0004] AI skin analysis products based on visible light / multispectral imaging use convolutional neural networks to identify and score pigmentation, fine lines, and enlarged pores in facial images. However, the data source is mainly single static images, lacking time-series dynamic information, making it impossible to accurately characterize the skin barrier function. It is also difficult to quantify the metabolic level of the dermis as time and environment change.

[0005] The existing technology mainly has the following technical problems: (1) The detection dimension is singular, making it impossible to construct a correlation model of the multi-layered structure of the skin; (2) Lack of adaptive evaluation mechanism for multi-source time series data; (3) The assessment results are difficult to automatically translate into personalized and actionable nursing decisions; (4) Lack of a systematic and trend-based barrier function risk warning module.

[0006] Therefore, it is necessary to provide an adaptive assessment and decision-making model that integrates skin bioelectric signals, temporal features, environmental behavior data, and reinforcement learning and deep learning algorithms to achieve closed-loop intelligent management of data collection, status assessment, risk warning, and nursing decisions. Summary of the Invention

[0007] To address the aforementioned technical problems, the present invention provides a solution. The present invention provides a method for assessing the damage of air pollution to human skin health based on a PPO-LSTM fusion algorithm, comprising: Multi-source data acquisition; multi-source data includes individual basic information, individual behavioral data, real-time bioelectrical and metabolic indicators, physical measurement indicators, environmental adaptation parameters and barrier function indicators; Data preprocessing to construct a multidimensional time-series feature dataset; A PPO-LSTM fusion model was established to extract features and optimize multidimensional temporal features through reinforcement learning, resulting in a skin health damage assessment model that outputs core skin health indicators. Based on core skin health indicators, the functional status of the stratum corneum, epidermis, and dermis is predicted and quantitatively assessed. Based on core skin health indicators and prediction results, a personalized skin health maintenance decision-making plan is generated. Based on the absolute values ​​and time trends of core skin health indicators, the risk levels of skin barrier function and dermal metabolic function are determined, and corresponding early warning information and coping suggestions are output. Train the skin health damage assessment model and update its parameters.

[0008] In some optional embodiments, basic individual information includes age, sex, height, weight, weight change level, skin type, family history, allergy history, and history of skin diseases; individual behavioral data includes skin care product ingredients, type of cleansing tools, and frequency of cleansing; real-time bioelectrical and metabolic indicators include dermal-stratum corneum resistance difference and dermal cell ATP concentration; physical measurement indicators include stratum corneum thickness, pore area ratio, epidermal temperature, skin tension, subcutaneous vascular flow velocity and pressure, skin moisture content, and skin breathability; environmental adaptation parameters include ambient temperature and humidity, ultraviolet radiation intensity, air particulate matter concentration, weather forecast air quality index, and sweat secretion rate; and barrier function indicators include transepidermal water loss rate and skin elasticity value.

[0009] In some optional embodiments, data preprocessing to construct a multidimensional time-series feature dataset includes: filtering, normalizing, environmental compensation, and time-series slicing of the collected data to obtain multidimensional time-series features. In some optional embodiments, a dual-band photoelectric anti-interference algorithm is used to filter the photoelectric signal, the skin surface tension and pressure are calculated based on the resistance change of the microbubble coating, and the temperature and humidity are corrected by an environmental compensation matrix.

[0010] In some optional embodiments, the PPO-LSTM fusion model includes an input layer, an LSTM feature extraction hidden layer, a dynamic feature cross-network, and a PPO adaptive optimization layer; The input layer concatenates the preprocessed multidimensional temporal feature data into a temporal input vector; The LSTM feature extraction hidden layer includes multiple memory modules. Each memory module processes the temporal input vector through an input gate, a forget gate, an output gate, and cell states, and outputs a hidden state vector to represent the dynamic changes in skin. The dynamic feature cross-network includes a 3D convolutional module and a gated recurrent unit module; the 3D convolutional module is used to extract dermal features. The space between the stratum corneum resistance difference and the stratum corneum thickness and pore structure Temporal coupling characteristics; the gated loop unit module is used to capture the temporal dependence of transepidermal water loss rate and sebum secretion-related indicators, and to calculate the environmental stress index; The PPO adaptive optimization layer consists of a policy network and a value network. The policy network takes the output of the dynamic feature cross-network and the output of the LSTM feature extraction hidden layer as input, and the output is used to adjust the evaluation correction coefficients of the core skin health indicators. The value network is used to evaluate the value of the current state and guide the policy update. In some optional embodiments, the reward function of the PPO adaptive optimization layer includes at least an error-based reward calculated based on the deviation between the current model evaluation value and the historical best benchmark value for the same age and skin type, as well as rewards for trend prediction accuracy, risk warning accuracy, and compliance with recommendations; the PPO adaptive optimization layer limits the policy update magnitude by introducing a proportional truncation function.

[0011] In some optional embodiments, key indicators of skin health include dermal turnover rate, stratum corneum aging rate, stratum corneum thickness, skin tension change rate, skin capillary flow velocity and pressure change rate, and skin permeability change rate. In some optional embodiments, the PPO adaptive optimization layer takes core skin health indicators as input and generates personalized care decision-making schemes for skin health maintenance through multi-objective optimization and strategy trees. Let the objective function be The time is The resistance difference between the dermis and the stratum corneum is The fluctuation of the dermal-stratum corneum resistance difference is The thickness of the stratum corneum is The growth rate of stratum corneum thickness is The area ratio of pores is Transdermal water loss rate , This is the weighting coefficient for the resistance difference fluctuation. This is a weighting coefficient for the growth rate of stratum corneum thickness. This is a weighting coefficient representing the proportion of pore area. Let be the weighting coefficient for transdermal water loss rate, then: ; Under the constraints of controlling the fluctuation of the dermal-stratum corneum resistance difference, increasing the growth rate of stratum corneum thickness, controlling the proportion of pore area and transepidermal water loss rate, a number of intervention programs were comprehensively selected, including microcurrent stimulation, moisturizing and hydrating, stratum corneum conditioning, barrier repair, sun protection and environmental avoidance.

[0012] In some optional embodiments, the risk level of skin barrier function and dermal metabolic function is determined based on the absolute value and time trend of core skin health indicators, and corresponding early warning information and coping suggestions are output. This includes: by performing trend analysis on the time series changes of core skin health indicators, setting risk judgment conditions for thinning of the stratum corneum barrier, damage to the hydrolipid film and slowing down of dermal metabolism, triggering an early warning when the corresponding conditions are met for multiple consecutive samplings, and outputting early warning information in the form of risk level, cause of formation and recommended care measures.

[0013] In some optional embodiments, when the average change in stratum corneum thickness in multiple consecutive samples is less than a preset negative threshold and the dermal-stratum corneum resistance difference is greater than a preset positive threshold, it is determined to be a risk of thinning of the stratum corneum barrier. When the skin's breathability exceeds the preset upper limit and the skin's surface tension is below the preset lower limit, it is considered to be at risk of hydrolipid film damage. When the ATP concentration in the dermis is below a preset lower limit and the dermal turnover rate is below a preset lower limit, it is considered a risk of slowed dermal metabolism.

[0014] The beneficial effects of this invention are: (1) Integrating multi-dimensional data in the same model simultaneously makes the evaluation dimensions more comprehensive; (2) Introducing the PPO-LSTM fusion architecture, using LSTM to capture temporal dynamics, and combining PPO reinforcement learning to adaptively optimize the evaluation strategy, solves the problem of insufficient adaptability of simple deep learning to individual differences, and improves the accuracy of skin health damage assessment. (3) Through decision-making and early warning, a complete closed loop is achieved from detection and assessment to nursing plan generation and risk warning, which greatly reduces the risk of misuse caused by users having to make their own judgments; (4) Organically link electrophysiological signals with the skin growth process to achieve quantitative assessment and regulation of skin growth and repair based on bioelectric signals; (5) Extending the traditional linear superposition of nursing decisions into a multi-dimensional strategy space that can be combined and weighed helps to obtain better combined nursing methods; (6) The early warning system is based on a trend judgment mechanism and can issue a warning in advance of the time window before the stratum corneum is obviously damaged, thereby improving the timeliness of care and repair intervention and reducing the probability of serious skin problems. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the method for assessing the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm provided in Embodiment 1 of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0017] Example 1 As an example, in order to solve the problems existing in the prior art, this example provides a method for assessing the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm.

[0018] The following is a detailed description of the implementation details of the method described in this embodiment. The following content is only for the convenience of understanding and is not necessary for implementing this solution.

[0019] The method for assessing the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm in this embodiment can be applied to electronic devices with communication, computing, and data storage capabilities. (See attached...) Figure 1 As shown, the method provided in this embodiment includes the following steps: Step 110: Multi-source data acquisition; Multi-source data includes individual basic information, individual behavioral data, real-time bioelectrical and metabolic indicators, physical measurement indicators, environmental adaptation parameters and barrier function indicators; In some optional embodiments, basic individual information includes age, sex, height, weight, weight change level, skin type, family history, allergy history, and history of skin diseases; individual behavioral data includes skin care product ingredients, type of cleansing tools, and frequency of cleansing; real-time bioelectrical and metabolic indicators include dermal-stratum corneum resistance difference and dermal cell ATP (adenosine triphosphate) concentration; physical measurement indicators include stratum corneum thickness, pore area ratio, epidermal temperature, skin tension, subcutaneous vascular flow velocity and pressure, skin moisture content, and skin breathability; environmental adaptation parameters include ambient temperature and humidity, ultraviolet radiation intensity, air particulate matter concentration, weather forecast air quality index, and sweat secretion rate; and barrier function indicators include transepidermal water loss rate and skin elasticity value.

[0020] In the user's daily life, data is collected at set time periods every day through a detection instrument that fits in contact with the skin, a microbubble tension measurement module, and an environmental data interface that connects to home WiFi.

[0021] Step 120: Data preprocessing to construct a multidimensional time-series feature dataset; In some optional embodiments, a dual-band photoelectric anti-interference algorithm is used to filter the photoelectric signal, the skin surface tension and pressure are calculated based on the resistance change of the microbubble coating, and the temperature and humidity are corrected by an environmental compensation matrix, as follows: Photoelectric signal anti-interference and normalization: Skin or finger tissue is irradiated with dual-band light sources of red light (about 630nm) and infrared light (about 950nm). The photoelectric receiver collects the reflected / transmitted signals, which are processed by a dual-channel anti-interference filtering algorithm to eliminate motion artifacts and ambient light interference. The collected electrical signals are then converted into standardized skin blood flow and oxygen content information.

[0022] Microbubble tension measurement: Microbubbles containing graphite oxide bamboo fiber are uniformly coated on the skin surface to form a microfilm. By applying a small electrical signal to monitor the resistance change, the pressure and tension parameters of the skin surface are calculated. A signal normalization model based on the skin's electrical properties is constructed to achieve cross-device data compatibility.

[0023] Environmental compensation and separation of internal and external factors: Daily weather and air quality data are obtained through network interfaces, and a microenvironment compensation matrix is ​​established to correct the effects of temperature, humidity, and air pressure on indicators such as resistivity and transdermal water loss rate; at the same time, an assessment matrix is ​​established to separate endogenous factors (such as metabolism, nutrition, and age) from exogenous factors (such as ultraviolet radiation, pollution, and friction) for subsequent model interpretation and intervention strategy classification.

[0024] Time series slicing and windowing: For continuous data from the past 7 days or longer, a time series segment of length L (e.g., 10 samples with intervals of 5 seconds or 30 minutes) is constructed in the form of a "sliding window" and used as the input sequence for LSTM (Long Short-Term Memory) / GRU (Gated Recurrent Unit) to preserve local trends and short-term fluctuations.

[0025] Step 130: Establish a PPO-LSTM fusion model, extract features and optimize multi-dimensional temporal features through reinforcement learning to obtain a skin health damage assessment model and output core skin health indicators. In some optional embodiments, the PPO-LSTM fusion model includes an input layer, an LSTM feature extraction hidden layer, a dynamic feature cross-network, and a PPO adaptive optimization layer; The input layer concatenates the preprocessed multidimensional time-series feature data into a time-series input vector; a multidimensional input layer structure is adopted, which includes six dimensions: individual basic attributes, bioelectrical indicators, physical structure indicators, environmental indicators, behavioral characteristics, and time-series features.

[0026] The LSTM feature extraction hidden layer includes multiple memory modules. Each memory module processes the temporal input vector through input gates, forget gates, output gates, and cell states, outputting a hidden state vector representing the dynamic changes in skin. Specifically, multiple memory modules (e.g., 15) are set, each corresponding to a set of time-series inputs; the cell dimension is set to 36 (corresponding to the mixed features of the six dimensions of data in the time dimension); input gates, forget gates, and output gates are used to control the flow of information. Through the Long Short-Term Memory (LSTM) network, patterns of change in skin resistance, electrophysiology, thickness, and moisture on short-term and multi-day scales can be captured, such as: a sudden drop in moisture content in a short period of time, increased transdermal water loss at night, and changes in dermal moisture in the morning. Dynamic characteristics such as increased resistance difference in the stratum corneum.

[0027] The dynamic feature cross-network includes a 3D convolutional module and a gated recurrent unit module; the 3D convolutional module is used to extract dermal features. The space between the stratum corneum resistance difference and the stratum corneum thickness and pore structure Temporal coupling characteristics; the gated loop unit module is used to capture the temporal dependence of transdermal water loss rate and sebum secretion-related indicators, and to calculate the environmental stress index.

[0028] The Dynamic Feature Cross-Network (DFCN) can uncover the spatial and temporal coupling relationships between different physical / physiological indicators. The DFCN includes a 3D convolution module, a GRU module, and an environmental perturbation simulator. The 3D convolution module performs 3D convolution on the joint temporal and spatial changes of dermal-stratum corneum resistance difference, stratum corneum thickness, and pore area ratio, extracting high-order coupling features between resistance, stratum corneum thickness, and pore deformation. The GRU module introduces gated recurrent units to capture the temporal dependence of transdermal water loss rate and sebum secretion-related indicators, obtaining a temporal description of barrier function changes. The environmental perturbation simulator constructs an approximate functional relationship between UV intensity and stratum corneum photodamage, predicts the nonlinear relationship between PM2.5 exposure and pore blockage rate, and outputs a skin stress index to quantify the real-time impact of environmental stress on barrier function. The DFCN output, along with the Long Short-Term Memory (LSTM) network output, is input into the PPO policy network for more accurate state representation.

[0029] The PPO adaptive optimization layer consists of a policy network and a value network. The policy network takes the output of the dynamic feature cross-network and the output of the LSTM feature extraction hidden layer as input, and the output is used to adjust the evaluation correction coefficients of the core skin health indicators. The value network is used to evaluate the value of the current state and guide the policy update.

[0030] In some optional embodiments, key indicators of skin health include dermal turnover rate, stratum corneum aging rate, stratum corneum thickness, skin tension change rate, skin capillary flow velocity and pressure change rate, and skin permeability change rate.

[0031] In some optional embodiments, the PPO layer is used to enhance the adaptability of the evaluation strategy to individual differences, and the specific structure includes: State space: a high-dimensional vector composed of the output of the LSTM hidden layer and the features fused by the dynamic feature cross-network; Behavioral space: For evaluating the correction coefficient vector, the adjustment weights correspond to indicators such as dermal renewal rate, stratum corneum aging rate, stratum corneum thickness, skin tension change rate, and skin permeability change rate, with values ​​ranging from [0.8, 1.2], for example. Policy network: It uses a two-layer fully connected network to map the state to the probability distribution of the action space; Value network: estimates the value of the current state, used to calculate the advantage function; Reward function: Based on the deviation between the current model's pre-evaluation value and the historical best baseline value for the same age and skin type; let the reward function be... The current model pre-evaluation value is The historical best benchmark value for the same skin type is ,but: The smaller the deviation, the higher the reward. Additionally, rewards can be added for trend matching and the accuracy of risk warnings.

[0032] In some optional embodiments, the reward function of the PPO adaptive optimization layer includes at least an error-based reward calculated based on the deviation between the current model evaluation value and the historical best benchmark value for the same age and skin type, as well as rewards for trend prediction accuracy, risk warning accuracy, and compliance with recommendations; the PPO adaptive optimization layer limits the policy update magnitude by introducing a proportional truncation function.

[0033] By repeatedly training the model with historical data through PPO, the evaluation strategy is optimized through continuous iteration, making it more robust to different populations and changing environments.

[0034] The output layer uses a linear activation function to ensure that the numerical range covers the physiologically reasonable range.

[0035] Step 140: Based on core skin health indicators, predict and quantify the functional status of the stratum corneum, epidermis, and dermis.

[0036] Using core skin health indicators as input, personalized intervention suggestions are generated through multi-objective optimization and strategy trees. For example, assuming the dermal cell turnover rate is... If the aging rate of the stratum corneum is Skin moisture content The thickness of the stratum corneum is The rate of change in skin breathability ,but: If the dermal cell turnover rate Less than 5% / day, generate a microcurrent stimulation program, for example, it is recommended to apply a 10μA / 5kHz microcurrent to gently stimulate intradermal microcirculation and ATP synthesis; If the aging rate of the stratum corneum Greater than 30% and skin moisture content If the concentration is less than 25%, a 5kDa hyaluronic acid infusion treatment is recommended after 6 hours to hydrate and repair the hydrolipid film. If the thickness of the stratum corneum If the value is less than 0.02mm, a decision will be made to disable products containing fruit acids or high concentrations of acids for the next 48 hours to avoid over-exfoliation; If the skin permeability changes With a rate greater than 15% / h and an increased transdermal water loss rate, it is recommended to use a ceramide-containing emulsion to lock in moisture, thereby reducing abnormal breathability and water loss by filling the lipid structure between the stratum corneum.

[0037] Step 150: Based on the core indicators of skin health and the prediction results, generate a personalized skin health maintenance decision plan.

[0038] In some optional embodiments, the PPO adaptive optimization layer takes core skin health indicators as input and generates personalized care decision-making schemes for skin health maintenance through multi-objective optimization and strategy trees. Let the objective function be The time is ,Genuine Leather The difference in stratum corneum resistance is ,Genuine Leather The fluctuation of the stratum corneum resistance difference is The thickness of the stratum corneum is The growth rate of stratum corneum thickness is The area ratio of pores is Transdermal water loss rate , Genuine leather The weighting coefficient for the fluctuation of stratum corneum resistivity difference. This is a weighting coefficient for the growth rate of stratum corneum thickness. This is a weighting coefficient representing the proportion of pore area. Let be the weighting coefficient for transdermal water loss rate, then: .

[0039] Under the constraints of controlling the fluctuation of the dermal-stratum corneum resistance difference, increasing the growth rate of stratum corneum thickness, controlling the proportion of pore area and transepidermal water loss rate, a number of intervention programs were comprehensively selected, including microcurrent stimulation, moisturizing and hydrating, stratum corneum conditioning, barrier repair, sun protection and environmental avoidance.

[0040] For example, intervention programs include: immediately adjusting indoor humidity, applying specific ingredients, and applying microcurrents; a 24-hour time-segmented care plan: such as using different skin care ingredients and physical methods in the morning, noon, and evening; environmental exposure avoidance tips for the next 3 to 5 days: such as reducing sun exposure during high UV periods and avoiding outdoor sports during periods of high pollution; and a list of contraindications: such as temporarily prohibiting the use of products containing fruit acids, salicylic acid, and exfoliating products.

[0041] Step 160: Based on the absolute values ​​and time trends of core skin health indicators, determine the risk level of skin barrier function and dermal metabolic function, and output corresponding warning information and coping suggestions.

[0042] In some optional embodiments, the risk level of skin barrier function and dermal metabolic function is determined based on the absolute value and time trend of core skin health indicators, and corresponding early warning information and coping suggestions are output. This includes: by performing trend analysis on the time series changes of core skin health indicators, setting risk judgment conditions for thinning of the stratum corneum barrier, damage to the hydrolipid film and slowing down of dermal metabolism, triggering an early warning when the corresponding conditions are met for multiple consecutive samplings, and outputting early warning information in the form of risk level, cause of formation and recommended care measures.

[0043] Risk categories include: risk of thinning of the stratum corneum barrier; risk of damage to the hydrolipid film; and risk of slowed dermal metabolism.

[0044] In some optional embodiments, when the average change in stratum corneum thickness in multiple consecutive samples is less than a preset negative threshold and the dermal-stratum corneum resistance difference is greater than a preset positive threshold, it is determined to be a risk of thinning of the stratum corneum barrier. When the skin's breathability exceeds the preset upper limit and the skin's surface tension is below the preset lower limit, it is considered to be at risk of hydrolipid film damage. When the ATP concentration in the dermis is below a preset lower limit and the dermal turnover rate is below a preset lower limit, it is considered a risk of slowed dermal metabolism.

[0045] Warning of risk of thinning stratum corneum barrier: If the stratum corneum thickness is low in 5 consecutive samples... The average change is less than 0.002mm, and genuine leather If the resistance difference of the stratum corneum is greater than 50Ω, it is determined that the stratum corneum is thin and dry, triggering a moderate or severe warning and pushing corresponding care contraindications and barrier repair solutions.

[0046] Hydrolipid film damage risk warning: If the skin breathability is greater than 20% / h and the skin surface tension is less than 0.03N / m, the skin surface lipid layer is determined to be damaged, the risk of barrier leakage increases, and the hydrolipid film damage warning is triggered.

[0047] Risk warning for slowed dermal metabolism: If the ATP concentration in dermal cells is lower than a preset threshold (e.g., 10...). -9 If the concentration of saturated cells is less than 1 mol / L and the dermal cell turnover rate is less than 3% / day, it indicates that the metabolic level of dermal cells is low, triggering a warning of slowed metabolism. It is recommended to combine microcurrent or photoelectric stimulation with nutritional supplementation.

[0048] The warning output includes the risk level, explanation of the cause, and response plan. For example: Moderate warning, risk of thinning of the stratum corneum barrier, explanation of the cause: continuous decrease in stratum corneum thickness and dermal... Increased resistance in the stratum corneum suggests: avoid using products containing fruit acids for 48 hours, use a barrier repair lotion containing ceramides and cholesterol, and add a moisturizing mask treatment every night.

[0049] Step 170: Train the skin health damage assessment model and update the parameters of the skin health damage assessment model.

[0050] Offline training phase: Initial model training is conducted using a large amount of historical data and manually labeled samples; Online learning phase: Data is collected in real time during actual use, and parameters are optimized through incremental learning or periodic retraining; Individualized fine-tuning: For a single user, while ensuring the stability of the global model, some parameters are individually fine-tuned to match their skin and behavioral characteristics; Genetic parameter iteration: The concept of genetic algorithm is introduced, and crossover and mutation operations are used on some hyperparameters (such as learning rate, weight initialization, and regularization coefficient) to automatically search for better combinations and improve the global optimal convergence performance of the model.

[0051] This invention integrates data on multi-layered skin electrical resistance, physical thickness, ATP metabolism, transepidermal water loss rate, elasticity, environmental exposure, and behavioral intervention to construct a holistic skin homeostasis model encompassing electrophysiology, physiology, metabolism, environment, and behavior, providing a more comprehensive assessment. By capturing temporal dynamics using LSTM and combining it with PPO reinforcement learning for adaptive optimization of the assessment strategy, it overcomes the limitations of simple deep learning in adapting to individual differences, enabling precise tracking of skin condition changes across different ages, skin types, and environments. Furthermore, it adds personalized care decision-making and early warning systems to traditional indicators, creating a complete closed loop from detection and assessment to care plan generation and risk alerts, significantly reducing the need for users to make their own judgments. Risks of misuse; Resistance difference-driven growth regulation mechanism: Organically linking electrophysiological signals with the skin growth process to achieve quantitative assessment and regulation of skin growth and repair based on bioelectric signals; Revealing the synergistic damage pattern of multiple external factors on barrier function, providing a basis for more reasonable daily care plans; By abstracting different intervention methods into superimposed "quantum states", the traditional linear superposition of care decisions is expanded into a combinable and weighable multidimensional strategy space, which helps to obtain better combined care methods; Early warning based on trend judgment mechanism can issue a prompt in advance of the time window before the stratum corneum is obviously damaged, improving the timeliness of care and repair intervention and reducing the probability of serious skin problems.

[0052] In some embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method for assessing the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm described in the above embodiments.

[0053] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

[0054] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the damage of air pollution to human skin health based on a PPO-LSTM fusion algorithm, characterized in that, The method comprises the following steps: Multi-source data collection; multi-source data includes individual basic information, individual behavior data, real-time bioelectricity and metabolic indicators, physical measurement indicators, environmental adaptation parameters and barrier function indicators; Data preprocessing, constructing a multi-dimensional time series feature dataset; Establishing a PPO-LSTM fusion model, performing feature extraction and reinforcement learning optimization on the multi-dimensional time series features to obtain a skin health damage assessment model, and outputting skin health core indicators; Based on the skin health core indicators, the functional state of the stratum corneum, epidermis and dermis is predicted and quantitatively evaluated; According to the skin health core indicators and the prediction results, an individualized nursing decision scheme for skin health maintenance is generated; According to the absolute value and time trend of the skin health core indicators, the risk level of skin barrier function and dermal metabolic function is determined, and corresponding warning information and response suggestions are outputted; Training the skin health damage assessment model to update the parameters of the skin health damage assessment model.

2. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to claim 1, characterized in that, The individual basic information includes age, gender, height, weight, body weight change level, skin type, family history, allergy history and skin disease history; the individual behavior data includes skin care product ingredients, cleaning tool types and cleansing frequency; the real-time bioelectricity and metabolic indicators include dermal-stratum corneum resistance difference and dermal cell ATP concentration; the physical measurement indicators include stratum corneum thickness, pore area ratio, epidermal temperature, skin tension, skin blood vessel flow rate pressure, skin water content and skin permeability efficiency; the environmental adaptation parameters include environmental temperature and humidity, ultraviolet intensity, air particulate matter concentration, weather forecast air quality index and sweat secretion rate; the barrier function indicators include transepidermal water loss rate and skin elasticity value. 3.The method of claim 1, wherein the method is characterized by, Data preprocessing, constructing a multi-dimensional time series feature dataset, including: filtering, normalizing, environment compensation and time series slicing processing on the collected data to obtain multi-dimensional time series features.

4. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to claim 3, characterized in that, The photoelectric signal is filtered by a dual-band photoelectric anti-interference algorithm, the skin surface tension and pressure are calculated based on the resistance change of the microbubble coating, and the temperature and humidity are corrected by an environment compensation matrix.

5. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to claim 1, characterized in that, The PPO-LSTM fusion model comprises an input layer, an LSTM feature extraction hidden layer, a dynamic feature cross network and a PPO adaptive optimization layer; The input layer splices the preprocessed multi-dimensional time series feature data into a time series input vector; The LSTM feature extraction hidden layer includes a plurality of memory modules, each memory module processes the time series input vector through an input gate, a forget gate, an output gate and a cell state, and outputs a hidden state vector representing the dynamic change characteristics of the skin; The dynamic feature cross network comprises a three-dimensional convolution module and a gated recurrent unit module; the three-dimensional convolution module is used for extracting dermis The difference in stratum corneum resistance and the space between stratum corneum thickness and pore structure Time coupling features; the gated recurrent unit module is used for capturing the time sequence dependent relationship between the transepidermal water loss rate and the sebum secretion related indexes, and calculating the environmental stress index; The PPO adaptive optimization layer includes a policy network and a value network; the policy network takes the output of the dynamic feature cross network and the output of the LSTM feature extraction hidden layer as input, and outputs an evaluation correction coefficient for adjusting the skin health core indicators; the value network is used to evaluate the value of the current state and guide the strategy update.

6. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to claim 5, characterized in that, The reward function of the PPO adaptive optimization layer includes at least an error reward calculated according to the deviation between the current evaluation value of the model and the historical optimal benchmark value of the same age and skin type, and reward items for trend prediction accuracy, risk warning accuracy and suggestion execution compliance; the PPO adaptive optimization layer limits the strategy update amplitude by introducing a proportional truncation function.

7. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to any one of claims 1 or 5, characterized in that, The skin health core indicators include dermal layer renewal rate, stratum corneum aging rate, stratum corneum thickness, skin tension change rate, skin capillary flow rate pressure change amount and skin air permeability change rate. 8.The method of claim 5, wherein the method is characterized by, The PPO adaptive optimization layer takes the skin health core indicators as input, and generates a personalized care decision scheme for skin health maintenance through multi-objective optimization and strategy tree. The objective function is , the time is , the resistance difference is , the resistance difference fluctuation degree is , the stratum corneum thickness is , the stratum corneum thickness growth rate is , the pore area ratio is , the transepidermal water loss rate is , is the weight coefficient of the resistance difference fluctuation degree, is the weight coefficient of the stratum corneum thickness growth rate, is the weight coefficient of the pore area ratio, is the weight coefficient of the transepidermal water loss rate. ； Under the constraints of controlling the fluctuation degree of dermal-stratum corneum resistance difference, improving the growth rate of stratum corneum thickness, controlling the pore area ratio and trans-epidermal water loss rate, a number of intervention schemes are selected, including micro-current stimulation, moisturizing and hydration, keratin conditioning, barrier repair, sun protection and environmental avoidance. 9.The method of claim 1, wherein the method is characterized by, According to the absolute value and time trend of the skin health core indicators, the risk level of skin barrier function and dermal metabolism function is determined, and the corresponding warning information and response suggestions are output, including: through trend analysis of the time series change of the skin health core indicators, the risk judgment conditions of stratum corneum barrier thinning, water-lipid membrane damage and dermal metabolism slowing down are set respectively, when the corresponding conditions are met for continuous multiple sampling, the warning is triggered, and the warning information is output in the form of risk level, formation reason and suggestion nursing measures.

10. The method for evaluating the damage of air pollution to human skin health based on the PPO-LSTM fusion algorithm according to claim 9, characterized in that, When the average change amount of stratum corneum thickness in continuous multiple sampling is less than a preset negative threshold and the dermal-stratum corneum resistance difference is greater than a preset positive threshold, it is determined that there is a risk of stratum corneum barrier thinning; When the skin air permeability is greater than a preset upper limit and the skin surface tension is lower than a preset lower limit, it is determined that there is a risk of water-lipid membrane damage; When the dermal ATP concentration is lower than a preset lower limit and the dermal layer renewal rate is lower than a preset lower limit, it is determined that there is a risk of dermal metabolism slowing down.

Images