Pressure sore risk prediction method and intelligent pressure sore prevention mattress

Abstract

The invention discloses a pressure sore risk prediction method and an intelligent pressure sore prevention mattress, and the method comprises the steps: obtaining pressure distribution data collected by a distributed pressure sensor array, carrying out the signal filtering and standardization processing of the pressure distribution data, and obtaining a standardized pressure distribution matrix; performing feature selection by applying a binary random forward algorithm, and outputting an optimal feature subset; acquiring physical sign parameters of a patient, fusing the physical sign parameters of the patient with the optimal feature subset, training by using an enhanced convolutional neural network with an activation function, and generating a pressure load distribution map; calculating risk assessment parameters of each body area by applying a differential evolution algorithm with neighborhood mutation, and outputting a pressure sore risk score and development trend prediction; and calculating optimal airbag inflation and deflation parameters through an intelligent decision algorithm, generating an execution instruction to an airbag control system, and generating a nursing intervention suggestion at the same time. According to the invention, accurate prediction and active protection of the pressure sore risk are realized, and the pressure sore prevention effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical care technology, and in particular to a pressure ulcer risk prediction method and an intelligent pressure ulcer prevention mattress, which are used for real-time monitoring, prediction and active protection of pressure ulcer risks for patients who are bedridden for a long time or have difficulty in moving. Background Art

[0002] Pressure ulcers (also known as pressure injuries) are injuries to the skin and subcutaneous tissue caused by prolonged pressure on the skin caused by the patient remaining in the same position for an extended period of time. These injuries are particularly common in patients who are bedridden or have limited mobility, not only impacting their quality of life but also potentially leading to serious complications and even life-threatening conditions.

[0003] Currently, common clinical pressure ulcer prevention techniques primarily include passive preventive measures such as scheduled turning and the use of air mattresses. Traditional air mattresses typically employ a fixed-time alternating inflation and deflation pattern, or a simple uniform pressure-bearing design, which lacks precise adjustment based on individual patient differences and real-time conditions. Some advanced medical institutions utilize mattress systems with pressure-sensing capabilities, but these systems typically only measure pressure values ​​and are unable to predict risks or proactively intervene.

[0004] More advanced pressure ulcer prevention systems use distributed pressure sensing technology to collect pressure data from contact surfaces and assess risk areas through simple threshold determination or statistical analysis. While this technology enables basic pressure monitoring, it lacks the accuracy and intelligence to predict risk, particularly due to a lack of consideration for individual patient differences and the cumulative effects of long-term pressure.

[0005] Existing technologies have the following major flaws: First, they lack the ability to accurately predict individualized pressure ulcer risk and are unable to dynamically assess risk based on the patient's specific situation; second, most systems use preset inflation and deflation modes and cannot make intelligent adjustments based on real-time detection data; third, existing systems generally lack the support of efficient machine learning algorithms, making it difficult to process and analyze large amounts of real-time pressure data and make accurate predictions. Summary of the Invention

[0006] The purpose of the present invention is to provide a pressure ulcer risk prediction method and an intelligent anti-pressure ulcer mattress, aiming to solve technical problems in the existing technology such as the lack of accurate individualized pressure ulcer risk prediction capability, the inability to perform intelligent airbag adjustment, and the lack of efficient machine learning algorithm support.

[0007] To achieve the above objectives, the present invention provides a method for predicting pressure ulcer risk, comprising the following steps:

[0008] Obtaining pressure distribution data collected by a distributed pressure sensor array, performing signal filtering and normalization processing on the pressure distribution data, and obtaining a normalized pressure distribution matrix;

[0009] Based on the normalized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output the optimal feature subset;

[0010] Obtain patient vital sign parameters, fuse them with the optimal feature subset, train them using an enhanced convolutional neural network with an activation function, and generate a pressure load distribution map;

[0011] Based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate the risk assessment parameters of each body region, outputting the pressure ulcer risk score and development trend prediction;

[0012] Based on the pressure ulcer risk score and development trend prediction, the optimal airbag inflation and deflation parameters are calculated through an intelligent decision-making algorithm, and execution instructions are generated to the airbag control system, while also generating nursing intervention recommendations.

[0013] Preferably, the pressure distribution data is subjected to signal filtering and normalization processing to obtain a normalized pressure distribution matrix, including:

[0014] Based on the pressure distribution data, a combination of median filtering and Kalman filtering is applied to remove noise and outliers to generate cleaned pressure data;

[0015] Perform min-max normalization on the cleaned pressure data and map the values ​​to the [0,1] interval to obtain normalized data;

[0016] According to the sensor spatial arrangement information and standardized data, the two-dimensional pressure distribution matrix is ​​reconstructed to obtain the standardized pressure distribution matrix.

[0017] Preferably, based on the normalized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output the optimal feature subset, including:

[0018] Based on the standardized pressure distribution matrix, a multidimensional feature pool including statistical features, morphological features and temporal features is constructed to form an initial feature set.

[0019] Perform binary encoding on the initial feature set and build a binary representation model for feature selection;

[0020] A randomized forward algorithm is used to iteratively select the optimal feature combination from the binary representation model to obtain the optimal feature subset.

[0021] Preferably, an enhanced convolutional neural network with an activation function is used for training to generate a pressure load distribution map, including:

[0022] The optimal feature subset is fused with the patient's vital sign parameters to construct the input tensor and form the network input data;

[0023] Design a deep neural network architecture with 5 convolutional layers. The first 3 layers use 3×3 convolution kernels, and the last 2 layers use 5×5 convolution kernels. Perform multi-scale feature extraction on the network input data and output a multi-level feature representation.

[0024] Based on the multi-level feature representation, an 8×16 feature map is generated through the last convolutional layer to obtain the pressure load distribution map.

[0025] Preferably, the activation function Here, α and β are learnable parameters that are automatically adjusted during training through back propagation.

[0026] Preferably, based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate the risk assessment parameters of each body region, and output the pressure ulcer risk score and development trend prediction, including:

[0027] Based on the pressure load distribution map and the preset body area pressure ulcer susceptibility weights, a multidimensional risk assessment parameter vector including pressure index, duration, and tissue tolerance is constructed to form an initial parameter space;

[0028] Latin hypercube sampling was used to initialize the population of the initial parameter space and generate a diversified risk assessment parameter vector combination containing 30 individuals;

[0029] When the differential evolution algorithm falls into a local optimum, it performs a small-scale search in the neighborhood space near the current optimal solution based on a combination of diversified risk assessment parameter vectors to generate an optimized parameter search strategy.

[0030] Based on the optimized parameter search strategy, an adaptive crossover and selection strategy is implemented for the diversified risk assessment parameter vector combination, the crossover probability is dynamically adjusted according to the population diversity, the parameter combination is optimized and a new generation of risk assessment parameter vector combination is generated;

[0031] The pressure ulcer risk score was calculated for each body area using a new generation of risk assessment parameter vector combination, and the risk change trend within the next 4 hours was estimated using a time series prediction model to obtain the pressure ulcer risk score and development trend prediction.

[0032] Preferably, a pressure ulcer risk score is calculated, including:

[0033] Based on the pressure index, duration, tissue tolerance, pressure change rate and cumulative risk parameters of each body area, a multi-factor risk model is used for weighted calculation to output a pressure ulcer risk score ranging from 0 to 100 points;

[0034] The risk level is divided according to the pressure ulcer risk score: 0-25 points are low risk, 26-50 points are medium-low risk, 51-75 points are medium-high risk, and 76-100 points are high risk, generating the risk level classification results.

[0035] Preferably, based on the pressure ulcer risk score and development trend prediction, the optimal airbag inflation and deflation parameters are calculated through an intelligent decision-making algorithm, and execution instructions are generated to the airbag control system, including:

[0036] Based on the pressure ulcer risk score and development trend prediction, high-risk and medium-risk areas are identified and the ideal pressure distribution target value for each area is calculated to determine the target body areas that need pressure redistribution;

[0037] A transfer function model between the airbag pressure and the mattress surface pressure is established. The optimal inflation and deflation parameters of each airbag are calculated using a quadratic programming algorithm for the target body area and the ideal pressure distribution target value, generating a control instruction sequence.

[0038] According to the control instruction sequence, the micro air pump and solenoid valve are controlled by PWM signals to adjust the pressure of each airbag, and the execution instructions are sent to the airbag control system.

[0039] Preferably, nursing intervention recommendations are generated simultaneously, including:

[0040] Based on the pressure ulcer risk score and development trend prediction, high-risk areas that cannot be completely relieved by airbag adjustment are identified, and key care areas requiring manual intervention are determined;

[0041] Combining patient historical data and risk level classification results for key nursing areas, a decision tree algorithm is used to determine the optimal intervention strategy and generate personalized nursing intervention recommendations;

[0042] When the pressure ulcer risk score of any area in the key nursing area is continuously monitored to be over 85 points and the predicted trend of the development trend forecast is rising, a turning reminder is automatically generated based on the personalized nursing intervention suggestion and the nursing staff is notified through sound and light prompts and mobile terminal push notifications to complete the nursing intervention notification.

[0043] The present invention also provides an intelligent anti-pressure ulcer mattress, comprising: a mattress body, on which a plurality of pressure sensors are arranged in an array; an airbag unit, arranged in the mattress body; an airbag control system, for controlling the inflation and deflation of the airbag unit; a control unit, comprising a memory, a processor, and a program stored in the memory and runnable on the processor, wherein the above-mentioned pressure ulcer risk prediction method is implemented when the processor executes the program.

[0044] The beneficial effects of the present invention are:

[0045] 1. Through a distributed pressure sensor array and efficient data processing algorithms, accurate monitoring and analysis of pressure distribution in various areas of the patient's body are achieved;

[0046] 2. The feature selection mechanism based on the binary random forward algorithm significantly reduces feature calculation energy consumption while ensuring prediction accuracy, achieving long-term stable operation of the system;

[0047] 3. A convolutional neural network architecture enhanced with a novel activation function significantly improves the model’s sensitivity to minute pressure changes and its ability to identify pressure distribution patterns.

[0048] 4. Introducing a differential evolution algorithm with a neighborhood mutation mechanism to enhance the algorithm's ability to escape local optimality and improve the accuracy and efficiency of parameter optimization in the pressure ulcer risk assessment model;

[0049] 5. Through a personalized risk assessment mechanism that integrates multimodal data, we can accurately characterize individual differences among patients and improve the personalized level of risk prediction;

[0050] 6. Intelligent airbag control system based on risk prediction can achieve precise control of independent airbags to achieve the purpose of proactively preventing pressure sores. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the following briefly introduces the drawings required for use in the embodiments or the conventional technology descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 is a flow chart of the pressure ulcer risk prediction method of the present invention;

[0053] Figure 2 Schematic diagram of the enhanced convolutional neural network architecture of the present invention;

[0054] Figure 3 This is a schematic structural diagram of the intelligent anti-pressure sore mattress of the present invention. DETAILED DESCRIPTION

[0055] To make the above-mentioned objects, features, and advantages of the present application more clearly understood, the specific embodiments of the present application are described in detail below with reference to the accompanying drawings. The following description sets forth many specific details to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0056] It should be noted that when a component is referred to as being "fixed to" or "disposed on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may be a central component at the same time. The terms "vertical", "horizontal", "upper", "lower", "left", "right" and similar expressions used in the specification of this application are for illustrative purposes only and do not represent the only implementation method.

[0057] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0058] In this application, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it can mean that the first feature is directly in contact with the second feature, or the first feature and the second feature are indirectly in contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it can mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is higher in level than the second feature. When a first feature is "below," "below," or "below" a second feature, it can mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is lower in level than the second feature.

[0059] Unless otherwise defined, all technical and scientific terms used in the specification of this application have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in the specification of this application includes any and all combinations of one or more of the relevant listed items.

[0060] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, the pressure ulcer risk prediction method provided by the present invention includes the following steps:

[0062] Step S1: obtaining pressure distribution data collected by a distributed pressure sensor array, performing signal filtering and normalization processing on the pressure distribution data, and obtaining a normalized pressure distribution matrix;

[0063] Step S2: Based on the standardized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output an optimal feature subset;

[0064] Step S3: Obtaining patient vital sign parameters, fusing the patient vital sign parameters with the optimal feature subset, training using an enhanced convolutional neural network with an activation function, and generating a pressure load distribution map;

[0065] Step S4: Based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate the risk assessment parameters of each body region, and output a pressure ulcer risk score and a development trend prediction;

[0066] Step S5: Based on the pressure ulcer risk score and the development trend forecast, the optimal airbag inflation and deflation parameters are calculated through an intelligent decision-making algorithm, and an execution instruction is generated to the airbag control system, and a nursing intervention suggestion is generated at the same time.

[0067] In step S1, pressure distribution data collected by the distributed pressure sensor array is first acquired. The high-precision flexible pressure sensor array built into the smart mattress adopts an 8×16 matrix arrangement, covering all pressure-bearing areas of the human body. Each sensor unit measures 2cm×2cm, with a measurement range of 0-200mmHg and an accuracy of ±1mmHg. The sensor sampling frequency is set to 10Hz, enabling high-density, real-time monitoring of the pressure distribution on the mattress surface, ensuring that subtle pressure changes and distribution characteristics are captured. The sensor array transmits the real-time collected pressure data to the control unit via Bluetooth Low Energy technology. This data includes a triplet of sensor ID, timestamp, and pressure value, forming a raw pressure data stream. The patient's basic body position information (supine, side, prone, etc.) is simultaneously recorded as auxiliary annotation data. This information facilitates subsequent analysis of pressure distribution characteristics in different body positions. After acquiring the raw pressure data, signal filtering is performed to remove noise and outliers that are inevitable during the acquisition process. Data cleaning was performed using a combination of median filtering and Kalman filtering. Median filtering primarily removes sudden spike noise, while Kalman filtering establishes state and observation equations, combining prior information with current observations to optimally estimate the state, effectively suppressing random fluctuations and noise. A sliding window size of five sampling points was set, and data points outside the ±3σ range were detected and corrected to ensure data continuity and reliability. The filtered pressure data were normalized using min-max processing, mapping the values ​​to the [0, 1] interval to eliminate dimensionality and enable fair comparison of data at different pressure levels. Finally, a two-dimensional pressure distribution matrix was reconstructed based on the spatial arrangement of the sensors and the normalized data, forming an 8×16 normalized pressure distribution matrix. Each element in the matrix represents the normalized pressure value at the corresponding location, providing a visual representation of the pressure distribution on the mattress surface. Furthermore, based on the continuously acquired pressure distribution matrix, temporal features such as pressure duration and pressure change rate at each point were calculated to construct a pressure feature set encompassing both spatial and temporal dimensions. The resulting enhanced dataset, containing both static and dynamic pressure change features, provides a comprehensive information foundation for subsequent pressure ulcer risk prediction.

[0068] In step S2, a binary randomized forward algorithm is applied to the standardized pressure distribution matrix for feature selection, outputting the optimal feature subset. A multidimensional feature pool consisting of statistical, morphological, and temporal features is first constructed to form an initial feature set. Statistical features describe the basic statistical properties of the pressure distribution, including global pressure statistics (such as mean pressure, maximum pressure, minimum pressure, pressure standard deviation, and pressure median), local statistics (describing local pressure statistics in key areas such as the sacrum, heel, and scapula), pressure distribution ratio (the proportion of area covered by different pressure levels), and pressure concentration (quantifying the unevenness of the pressure distribution using the Gini coefficient). Morphological features describe the spatial characteristics of the pressure distribution, including the location of the pressure center of gravity (calculating the coordinates of the pressure-weighted center of mass, reflecting the overall pressure center), pressure distribution shape characteristics (including parameters such as moment of inertia, dispersion, and eccentricity), high-pressure area morphology (such as the area, perimeter, shape complexity, and principal axis direction of the high-pressure area), and pressure gradient (the rate of change of pressure between adjacent locations, reflecting the smoothness of the pressure distribution). Time series features describe the dynamic characteristics of pressure changes over time, including pressure duration (the duration that pressure at each location continuously exceeds the threshold), pressure change rate (the magnitude of pressure change per unit time), pressure fluctuation frequency (reflecting the frequency of minor positional adjustments by the patient), and pressure release pattern (the frequency and duration of temporary pressure release, which characterizes the patient's ability to reduce stress). The initial feature pool contains over 200 feature dimensions, providing a rich set of candidates for subsequent feature selection. Next, a binary encoding representation is performed on the initial feature set to construct a binary representation model for feature selection. Each feature in the feature pool is assigned a binary bit, forming a binary vector of length N (N is the total number of features, approximately 200+). Each binary bit in this vector is assigned a value of 1, indicating that the feature is selected, and a value of 0, indicating that the feature is not selected. This binary encoding transforms the complex feature selection problem into a binary optimization problem, making it easier to operate. Feature selection is then performed using a binary randomized forward algorithm. The algorithm first randomly selects approximately 10% of all features as the initial feature subset, constructs an initial evaluation model, and calculates performance metrics. In each iteration, a subset of candidate features is randomly selected from the unselected feature set. The value of each candidate feature is evaluated, and the features that bring the greatest performance improvement are selected and added to the feature subset. The model and performance indicators are then updated. This process continues until the termination condition is met. Finally, the energy consumption of the selected feature subset is evaluated, and features with energy consumption ratios below the threshold are removed to ensure that energy consumption is minimized while maintaining prediction accuracy. The final output is an optimal feature subset containing 40-60 key features. These features have strong predictive ability, high energy efficiency, good complementarity, and strong interpretability, providing high-quality input data for subsequent risk prediction.

[0069] In step S3, patient vital signs are acquired, fused with the optimal feature subset, and trained using an enhanced convolutional neural network with an activation function to generate a pressure load distribution map. First, patient vital signs are collected, including basic demographic characteristics (such as age, gender, weight, height, and BMI), clinical assessment indicators (such as the six dimensions of the Braden score), underlying medical conditions (such as diabetes, vascular disease, and neurological disease), laboratory test results (such as albumin level, hemoglobin, and total protein), and medication use (such as steroids, sedatives, and other drugs that may affect tissue tolerance). These parameters are automatically acquired through the hospital information interface or manually entered by medical staff. Next, multimodal data fusion is performed on the patient vital signs with the optimal feature subset output from step S2 to construct the input tensor. Pressure features form the primary spatial channels, preserving their spatial relationship (8×16 grid), while patient vital signs are incorporated into the model as additional feature channels to ensure that the model comprehensively accounts for pressure distribution and individual differences. A specialized neural network architecture with a deep structure consisting of five convolutional layers is then designed. The first three layers use 3×3 convolution kernels to focus on capturing local pressure patterns, while the last two layers use 5×5 convolution kernels to capture pressure distribution characteristics over a wider range. The network uses a skip connection structure to enhance gradient flow and feature transfer, improving model training stability. It uses innovative activation functions. , where α and β are learnable parameters. This activation function combines the linear properties of ReLU with the nonlinear saturation property of Sigmoid, effectively capturing nonlinear patterns and subtle changes in the pressure distribution. Supervised learning is performed using a historical dataset containing known pressure ulcer risk ratings. A weighted cross-entropy loss function is used to prioritize high-risk samples. The Adam optimizer is applied for parameter updates, and L2 regularization and Dropout are introduced to prevent overfitting. After training, the model can input the patient's real-time pressure characteristics and vital signs. The final convolutional layer generates an 8×16 pressure load distribution map. The value of each position in this map represents the pressure ulcer risk index (between 0 and 1) for the corresponding area. Different risk levels are visually represented using color coding: low-risk areas are displayed in green, low-medium-risk areas in yellow, medium-high-risk areas in orange, and high-risk areas in red. This visualization allows healthcare professionals to intuitively understand the pressure ulcer risk distribution across the patient's body, providing a reference for subsequent risk assessment and intervention decisions.

[0070] In step S4, based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate risk assessment parameters for each body region, outputting a pressure ulcer risk score and development trend prediction. First, based on the pressure load distribution map and combined with the pressure ulcer susceptibility weights for different body regions from medical literature (e.g., 1.5 for the sacrum, 1.3 for the heel, 1.2 for the hip), a multidimensional risk assessment parameter vector is constructed, comprising factors such as pressure index, duration, and tissue tolerance. These parameters include: pressure intensity (based on the risk index from the pressure load distribution map combined with regional weights); duration (indicating the duration of a specific pressure level); tissue tolerance (reflecting the patient's tissue resistance to pressure); pressure change rate (reflecting the rate of change of pressure over time); and cumulative risk (accounting for the cumulative effect of historical risk). Next, Latin hypercube sampling is used to initialize the population of the initial parameter space, generating a diverse risk assessment parameter vector consisting of 30 individuals. Latin hypercube sampling ensures a uniform distribution of samples across each dimension, improving the coverage of the parameter space. Then, an innovative neighborhood mutation operation is implemented to enhance the algorithm's ability to escape local optima. When the algorithm threatens to become trapped in a local optimum, it conducts targeted local searches within the "neighborhood" of the current optimal solution, balancing global exploration with local exploitation. An adaptive crossover strategy is also implemented, dynamically adjusting the crossover probability based on population diversity: a lower crossover probability is used to maintain diversity when diversity is high, and a higher crossover probability is used to promote information exchange when diversity is low. Selection utilizes an elite retention mechanism to ensure that the best individuals are not lost in each generation, while tournament selection is introduced to increase population diversity. Finally, the optimized parameters are used to calculate a pressure ulcer risk score for each patient's body region. For each body region, a multi-factor risk model is used to perform a weighted calculation based on the pressure index, duration, tissue tolerance, pressure change rate, and cumulative risk parameters. A pressure ulcer risk score ranging from 0 to 100 is output. Based on the risk score, each region is categorized into different risk levels: 0-25 is low risk, 26-50 is low-medium risk, 51-75 is medium-high risk, and 76-100 is high risk. Furthermore, based on historical data, the system predicts the changing trends of risk scores in each area over the next four hours, providing forward-looking risk information to medical staff and supporting timely, targeted interventions. This comprehensive risk assessment approach, combining current risk scores with future trend predictions, significantly improves the foresight and effectiveness of pressure ulcer prevention.

[0071] In step S5, based on the pressure ulcer risk score and development trend prediction, an intelligent decision-making algorithm calculates the optimal airbag inflation and deflation parameters, generates execution instructions to the airbag control system, and simultaneously generates nursing intervention recommendations. First, based on the risk score and trend prediction, the system identifies high-risk areas (risk score >70) and medium-risk areas (risk score 40-70) and calculates the ideal pressure distribution target for each area. For high-risk areas, the goal is to significantly reduce surface pressure, with the ideal pressure value set at 50-60% of the current pressure. For medium-risk areas, the goal is to moderately reduce surface pressure, with the ideal pressure value set at 70-80% of the current pressure. For low-risk areas, pressure can be increased appropriately to compensate for the reduced pressure in high-risk areas, but the increase should not exceed 20%, and the absolute pressure value should not exceed the safety threshold. Next, a transfer function model is established between airbag pressure and mattress surface pressure to describe how changes in airbag internal pressure affect the pressure distribution on the mattress surface. Based on this model, the pressure redistribution problem is transformed into an optimization problem: finding a set of airbag internal pressure values ​​that ensures that the actual pressure distribution on the mattress surface is as close as possible to the ideal pressure distribution target value. A quadratic programming algorithm is used to solve this optimization problem, calculating the optimal inflation and deflation parameters for each airbag, forming a complete control instruction sequence. Based on this control instruction sequence, PWM signals are used to control the micro-pump and solenoid valve to precisely adjust the pressure of each airbag. A closed-loop control mechanism is employed. Pressure sensors built into the airbags monitor internal pressure in real time, compare the measured value with the target value, and dynamically adjust the PWM signal parameters to ensure that the airbag pressure accurately reaches the set value. The entire airbag control system boasts a minimum adjustment accuracy of ±1 mmHg and a response time of less than 2 seconds, enabling rapid response to changes in risk assessment results and timely adjustments to the pressure distribution on the mattress surface. Furthermore, high-risk areas that cannot be fully addressed by airbag adjustment are identified, defining key areas requiring manual intervention. For these areas, a decision tree algorithm is used, combining patient historical data and risk classification results, to determine the optimal intervention strategy and generate personalized nursing intervention recommendations. These recommendations include turning strategies (recommended turning angle, frequency, and specific techniques), localized pressure relief measures (suggestions for the use of specific positioning pads, air cushions, or other pressure relief devices), and skin care regimens (specific instructions for cleansing, moisturizing, massage, and other procedures). If the pressure ulcer risk score in any key care area exceeds 85 points and is predicted to increase, a turning reminder is automatically generated and notified to the caregiver via audio and visual prompts and mobile push notifications. This system also implements a priority management and escalation mechanism for notifications to ensure that emergencies are addressed promptly. This combination of machine assistance and human intervention provides a comprehensive and personalized pressure ulcer prevention solution, significantly improving prevention effectiveness and care efficiency.

[0072] The step S1 performs signal filtering and normalization processing on the pressure distribution data to obtain a normalized pressure distribution matrix, including:

[0073] Step S1.1: Based on the pressure distribution data, a combination of median filtering and Kalman filtering is applied to remove noise and outliers to generate cleaned pressure data;

[0074] Step S1.2: performing min-max normalization processing on the cleaned pressure data, mapping the values ​​to the interval [0, 1] to obtain normalized data;

[0075] Step S1.3: Reconstructing a two-dimensional pressure distribution matrix based on the sensor spatial arrangement information and the standardized data to obtain the standardized pressure distribution matrix.

[0076] In step S1.1, the raw pressure data collected from the distributed pressure sensor array is filtered to remove the noise and outliers that inevitably occur during the acquisition process. The high-precision flexible pressure sensor array built into the smart mattress is arranged in an 8×16 matrix, covering the entire pressure-bearing area of ​​the human body. Each sensor unit measures 2 cm×2 cm, has a measurement range of 0–200 mmHg, an accuracy of ±1 mmHg, and a sampling frequency of 10 Hz. These sensors collect real-time pressure distribution data on the mattress surface and transmit this data to the control unit via Bluetooth Low Energy technology, forming a raw pressure data stream consisting of a sensor ID, timestamp, and pressure value triplet. However, due to environmental interference, sensor fluctuations, and interference during data transmission, the raw data often contains noise and outliers, requiring filtering to improve data quality. A combination of median filtering and Kalman filtering is used for data cleaning. First, median filtering is applied to remove sudden spikes and outliers. The principle of median filtering is to use a sliding window (in this case, the window size is 5 sampling points) to sort the data within the window and take the median value as the filtered result. This nonlinear filtering method is particularly effective for removing impulse noise while preserving the edge characteristics of the data. Next, a Kalman filter is applied to further smooth the data and estimate the true state. The Kalman filter is a recursive estimator that establishes a state equation and an observation equation, combining prior information with current observations to optimally estimate the state. For pressure data, the state can be the true pressure value at the sensor location, while the observation is the sensor reading. The Kalman filter continuously optimizes the estimate of the true pressure through a prediction-update cycle, effectively suppressing random fluctuations and noise. Furthermore, an outlier detection mechanism is implemented to flag and correct data points that fall outside the ±3σ range (σ is the local standard deviation) to ensure data continuity and reliability. This combination of median filtering and Kalman filtering effectively removes noise and outliers from the raw pressure data, generating cleaned, high-quality pressure data that provides a reliable foundation for subsequent data analysis and risk assessment.

[0077] In step S1.2, the cleaned pressure data is standardized, and data of different magnitudes and units are mapped to a unified numerical range to facilitate subsequent processing and analysis. The value range of the original pressure data is 0-200 mmHg. There are significant differences in the pressure distribution of different patients and different body positions. Direct use of the original data may cause the high pressure value to have an excessive impact on the model, while ignoring the slight changes in the low pressure area. Therefore, the min-max normalization method is used to linearly map the cleaned pressure data to the [0,1] interval. For each sensor position Pressure value , the standardization formula is: .

[0078] in, and The minimum non-zero and maximum values ​​in the current pressure dataset are selected as the lower bound, rather than the absolute zero value, to avoid large areas of no pressure (value 0) remaining zero after normalization, which would result in information loss. This normalization process ensures that all pressure data is mapped to the same numerical range, eliminating dimensionality effects and enabling fair comparison of data at different pressure levels. Furthermore, normalization helps improve the convergence speed and stability of subsequent machine learning algorithms, as most machine learning algorithms perform better when processing normalized data. Abnormally large pressure values ​​(possibly caused by sensor failure or temporary interference) are truncated before normalization. Values ​​outside a reasonable range (e.g., >200 mmHg) are clamped to a maximum threshold to prevent extreme values ​​from distorting the overall normalization results. After normalization, the original physical pressure values ​​are converted to dimensionless numbers between 0 and 1, where 0 represents no pressure or the lowest pressure, 1 represents the highest pressure, and intermediate values ​​are proportionally distributed. This normalized data preserves the relative relationships and patterns of the original pressure distribution while improving numerical stability and comparability.

[0079] In step S1.3, the two-dimensional pressure distribution matrix is ​​reconstructed based on the spatial arrangement information and standardized data of the sensors to form a complete standardized pressure distribution matrix. The sensor array is arranged in an 8×16 matrix, covering the main force-bearing area on the mattress surface, and each sensor corresponds to a specific position on the mattress surface. During data transmission, sensor data is transmitted in the form of sensor ID, timestamp, and pressure value. It is necessary to reorganize the discrete sensor readings into a two-dimensional matrix reflecting the spatial distribution based on the mapping relationship between the sensor ID and its physical location. Maintain a sensor ID to matrix position According to this mapping relationship, the standardized pressure value of each sensor is placed in the corresponding position of the two-dimensional matrix to form an 8×16 standardized pressure distribution matrix P. Each element in the matrix Indicates location The normalized pressure values ​​at each location range from [0, 1]. Interpolation is used to fill in any missing data (e.g., temporary sensor failure or data loss). Isolated missing points are filled using the average value of the surrounding valid points. For continuous missing regions, bilinear interpolation is used to estimate the pressure value at the missing location based on the surrounding valid data. After reconstruction, the entire matrix is ​​smoothed using a Gaussian filter to reduce abrupt changes between adjacent locations, generating a continuous, smoothed matrix that better reflects the actual pressure distribution. This smoothing process takes into account the physical properties of actual pressure distribution, which typically exhibits continuous rather than abrupt changes in space. The final output, an 8×16 normalized pressure distribution matrix, is a complete, continuous, and standardized two-dimensional data structure that intuitively reflects the pressure distribution on the mattress surface. This matrix can be used for visualization (e.g., through heat maps) and serves as the basis for subsequent feature extraction and risk assessment. In addition, based on the continuously collected pressure distribution matrix, the time series characteristics such as pressure duration and pressure change rate at each point are calculated, and a pressure feature set containing spatial and temporal dimensions is constructed. The output is an enhanced data set containing static pressure distribution and dynamic pressure change characteristics, providing more comprehensive information for subsequent pressure ulcer risk prediction.

[0080] In step S2, based on the normalized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output an optimal feature subset, including:

[0081] Step S2.1: Based on the standardized pressure distribution matrix, construct a multidimensional feature pool including statistical features, morphological features, and temporal features to form an initial feature set;

[0082] Step S2.2: performing binary encoding on the initial feature set to construct a binary representation model for feature selection;

[0083] Step S2.3: Using a random forward algorithm to iteratively select the optimal feature combination from the binary representation model to obtain the optimal feature subset.

[0084] In step S2.1, a comprehensive feature pool is constructed based on the standardized pressure distribution matrix to capture key information about the pressure distribution from different dimensions. First, three core features are extracted: statistical features, morphological features, and temporal features. Statistical features describe the basic statistical properties of the pressure distribution, including global pressure statistics (such as mean pressure, maximum pressure, minimum pressure, pressure standard deviation, and pressure median), local statistics (describing local pressure statistics in key areas such as the sacrum, heel, and scapula), pressure distribution ratio (the proportion of area covered by different pressure levels), and pressure concentration (quantifying the unevenness of the pressure distribution using the Gini coefficient). Morphological features describe the spatial characteristics of the pressure distribution, including the location of the pressure center of gravity (calculated as the pressure-weighted center of mass coordinates, reflecting the overall pressure center), pressure distribution shape characteristics (including parameters such as moment of inertia, dispersion, and eccentricity), high-pressure area morphology (area, perimeter, shape complexity, and principal axis direction of the high-pressure area), and pressure gradient (the rate of change of pressure between adjacent locations, reflecting the smoothness of the pressure distribution). Temporal features describe the dynamic characteristics of pressure changes over time, including pressure duration (the time that the pressure at each location continuously exceeds the threshold), pressure change rate (the magnitude of pressure change per unit time), pressure fluctuation frequency (reflecting the frequency of minor position adjustments by the patient), and pressure release pattern (the frequency and duration of temporary pressure release, which characterizes the patient's ability to reduce stress). These features are extracted from the raw pressure data through mathematical calculations and signal processing methods. The initial feature pool contains over 200 feature dimensions. For example, for a specific region r, its pressure duration feature can be expressed as: , where I is the indicator function, when the pressure in area r at time t The value is 1 when the threshold is exceeded, otherwise it is 0. Constructing this comprehensive feature pool provides rich candidates for subsequent feature selection, ensuring that various pressure patterns related to pressure ulcer risk can be captured.

[0085] In step S2.2, the feature selection problem is transformed into a mathematical form suitable for algorithm processing. Each feature in the feature pool constructed in step S2.1 is assigned a binary bit to form a binary vector of length N (N is the total number of features, about 200+). This vector B is represented as , where each binary bit The value rules are: Indicates the selection of the i-th feature, Indicates that the i-th feature is not selected. For example, for a feature pool containing 200 features, a possible feature selection scheme can be expressed as , which indicates that features 1, 3, 6, etc. were selected, while features 2, 4, and 5, etc., were discarded. The advantages of this binary encoding are: simplified problem formulation (converting a complex feature selection problem into a binary optimization problem), ease of algorithmic manipulation (binary vectors are easily operable with various operations such as flipping and crossing), and a clear state space (a binary vector of length N can represent 2N different feature combinations). A mapping table is established between feature IDs and binary bit positions, with each feature selection scheme (i.e., a binary vector) corresponding to a possible feature subset. When evaluating the performance of a feature subset, the binary vector is used as a mask, and only features corresponding to positions in the vector with a value of 1 are extracted for model training and evaluation. This binary encoding provides a good representational foundation for the subsequent stochastic forward algorithm, enabling it to efficiently search for the optimal feature combination in the 2^N solution space.

[0086] In step S2.3, the random forward algorithm is used Perform feature selection. First, randomly select k features from all N features as the initial feature subset (k is set to 10% of the total number of features. If the total number of features is 200, then about 20 features are initially selected.) The random selection process uses a uniformly distributed pseudo-random number generator to ensure that each feature has an equal probability of being selected. Next, based on the initial feature subset Building an initial evaluation model (can be a simple logistic regression or decision tree model) and calculate the initial model Performance indicators This performance metric is a weighted combination of information gain and computational complexity, defined as:

[0087]

[0088] in, Indicates the predictive ability of the model, which can be measured by AUC or F1 score; The complexity of the feature subset can be measured by the number of features or the energy consumption required to calculate these features; α and β are weight parameters that balance prediction performance and complexity. Then, an iterative process is started: m candidate features are randomly selected from the unselected feature set (m is set to 20% of the remaining features); the value of each candidate feature is evaluated in turn, and the incremental performance improvement is calculated. ; Select the features that bring the greatest performance improvement If the performance improvement brought by the best candidate feature is positive, it is added to the feature subset and the model and performance indicators are updated. The termination condition is checked (no significant performance improvement after t consecutive iterations, the preset maximum number of iterations is reached, or the feature subset size reaches the preset upper limit). If it is met, stop, otherwise continue to the next round of iteration. Finally, the energy consumption of the selected feature subset is optimized, and the information gain and energy consumption ratio of each feature is calculated. , remove Features below a preset threshold are selected to ensure that energy consumption is minimized while maintaining prediction accuracy. The final output is an optimal feature subset containing 40-60 key features that have strong predictive power, high energy efficiency, good complementarity, and strong interpretability.

[0089] like Figure 2 As shown, in step S3, an enhanced convolutional neural network with an activation function is used for training to generate a pressure load distribution map, including:

[0090] Step S3.1: performing multimodal data fusion on the optimal feature subset and the patient's vital sign parameters to construct an input tensor to form network input data;

[0091] Step S3.2: Design a deep neural network architecture with five convolutional layers, using 3×3 convolution kernels for the first three layers and 5×5 convolution kernels for the last two layers. Perform multi-scale feature extraction on the network input data and output a multi-level feature representation.

[0092] Step S3.3: Based on the multi-level feature representation, an 8×16 feature map is generated through the last convolutional layer to obtain the pressure load distribution map.

[0093] In step S3.1, multimodal data fusion is implemented, combining pressure features with patient vital signs to create a comprehensive input data tensor. Two types of input data are collected and prepared: pressure feature data (derived from the optimal feature subset output from step S2, consisting of 40-60 selected key features, such as pressure value, duration, and rate of change at each region, organized into a structured matrix according to the body region they describe) and patient vital signs (including basic demographic characteristics such as age, sex, weight, height, and BMI; clinical assessment indicators such as the six dimensions of the Braden score; information on underlying medical conditions such as diabetes, vascular disease, and neurological disease; laboratory test results such as albumin level, hemoglobin, and total protein; and medication use such as steroids and sedatives that may affect tissue tolerance). Both types of data are preprocessed and standardized: numerical features are normalized to similar numerical ranges using the Z-score or Min-Max method; categorical features are converted to numerical representations using one-hot encoding; and missing values ​​are filled using multiple imputation or model-based methods. Then, construct the multimodal input data tensor. For the convolutional neural network architecture, the data is organized into basic input dimensions. , where B is the batch size, C is the number of channels, and H and W are the spatial dimensions. Pressure features form the primary spatial channels, preserving their spatial relationships (8×16 grid). Patient vital signs are incorporated into the model in two ways: as additional feature channels (expanded to the same spatial dimensions as the pressure features via broadcasting) and as a separate fully connected layer, fused with the pressure features extracted by the convolutional layer in subsequent layers. This multimodal data fusion approach has the advantages of capturing the interaction between pressure data and individual patient characteristics, allowing the model to learn differentiated risk patterns across different patient groups, and improving the level of prediction personalization (the same pressure distribution produces different risk predictions for different patients).

[0094] In step S3.2, a special neural network architecture is designed to optimize the pressure ulcer risk prediction task. The network as a whole adopts a deep convolutional structure, which contains 5 convolutional layers. The core idea of ​​the architecture design is to extract pressure features from local to global layer by layer. The input layer receives the multimodal input data tensor in step S3.1, which has the shape of The local feature extraction layer (the first three convolution layers) uses a smaller receptive field (3×3) to focus on capturing local pressure patterns, such as high-pressure points, pressure gradients, and other detailed features: the first convolution layer uses 32 3×3 convolution kernels with a stride of 1 and padding of 1, followed by batch normalization and a novel activation function; the second convolution layer uses 64 3×3 convolution kernels with a stride of 1 and padding of 1, followed by batch normalization and a novel activation function; the third convolution layer uses 128 3×3 convolution kernels with a stride of 1 and padding of 1, followed by batch normalization and a novel activation function. The global feature extraction layer (the last two convolutional layers) uses a larger receptive field (5×5) to capture a wider range of pressure distribution patterns, such as the center of pressure offset and overall distribution shape, among other macroscopic features. The fourth convolutional layer uses 128 5×5 convolutional kernels with a stride of 1 and padding of 2, followed by batch normalization and a novel activation function. The fifth convolutional layer uses 64 5×5 convolutional kernels with a stride of 1 and padding of 2, followed by batch normalization and a novel activation function. The network also incorporates skip connections (skip connections are added between the first and fifth layers and between the second and fourth layers, matching feature dimensions through identity mapping or 1×1 convolutions). These connections directly transfer shallow-level features to deeper layers, effectively alleviating the vanishing gradient problem, preserving detailed information, and improving training stability. The feature fusion layer uses an attention mechanism to adaptively weight features at different levels, implementing both channel-wise and spatial-wise attention, dynamically adjusting the weights of different features based on their importance. The advantages of this enhanced convolutional neural network architecture are: multi-scale feature extraction (capturing local and global pressure patterns through convolution kernels of different sizes), gradient flow optimization (skip connections improve gradient flow, facilitating the training of deeper networks), feature preservation (retaining all useful information from shallow to deep layers), and attention enhancement (highlighting important features and suppressing irrelevant information).

[0095] In step S3.3, a pressure load distribution map is generated based on the multi-level feature representation. The real-time data input of the current patient (the latest pressure data collected from the distributed pressure sensor and the patient's vital signs) is obtained, and a multimodal input data tensor is constructed according to the format in step S3.1. The input data is fed into the trained model for forward propagation. The output shape of the last convolutional layer of the model is feature map, where B is the batch size (usually 1 for real-time prediction), and the value of each spatial location represents the pressure ulcer risk index of the corresponding area, ranging from 0 to 1. This raw output is post-processed and converted into a more informative stress load distribution map: risk level classification (the continuous risk index is divided into different risk levels according to preset thresholds: 0.0-0.25 is low risk / green, 0.25-0.5 is medium-low risk / yellow, 0.5-0.75 is medium-high risk / orange, and 0.75-1.0 is high risk / red); spatial interpolation enhancement (upgrading to a higher resolution such as 64×128 through bicubic interpolation to create a smooth high-resolution risk distribution map); body region mapping (aligning the risk distribution map with the standard human anatomical map, identifying the corresponding body regions such as the sacrum, heel, scapula, etc., and adding regional annotations); heat map rendering (using color mapping to render the risk value into an intuitive heat map, with high-risk areas displayed in red and low-risk areas displayed in green); dynamic time window comparison (comparing the current risk distribution map with the distribution map of the past time window to generate a risk change trend map, showing areas with increased or decreased risk). The final output pressure load distribution map is a comprehensive visualization result that combines the output of the risk prediction model, spatial interpolation enhancement, body area mapping and heat map rendering, which can intuitively show the risk of pressure sores in all areas of the patient's body. This map is presented to medical staff in real time through a display and is stored in the patient's electronic medical record as a basis for subsequent nursing decisions and risk monitoring. The value of the map lies in converting complex model predictions into intuitive visual information. Medical staff can quickly identify the areas of the body that need to be focused on and take timely intervention measures without having to understand the underlying algorithms. Among them, the activation function Here, α and β are learnable parameters that are automatically adjusted during training through back propagation.

[0096] Specifically, traditional activation functions such as ReLU, Sigmoid or Tanh each have their own advantages and disadvantages: ReLU has a linear response characteristic, is simple to calculate and can alleviate the gradient vanishing problem, but it completely inhibits negative inputs, which may cause neurons to "die"; the Sigmoid function has good nonlinear saturation characteristics, and the output range is stable between (0,1), but the gradient is close to zero in the saturation region, which can easily lead to gradient vanishing; the Tanh function has an output range of (-1,1), and the zero-centered characteristic is conducive to optimization, but it also has a saturation problem. For the special task of pressure ulcer risk prediction, it is necessary to have the ability to capture small pressure changes (linear response) and simulate tissue damage threshold effects (nonlinear saturation), so a parameterized activation function is designed. .

[0097] The mathematical structure of the activation function incorporates linear terms and nonlinear modulation terms , where α controls the strength of the linear response and β controls the steepness of the nonlinear saturation. When the input signal x is small, , the function is approximately , showing an approximately linear response, which can effectively capture small pressure changes; when the input signal x is large, Approaching 1, the function is approximately , maintain linear growth but the growth rate is controlled to avoid excessive output value; when the input signal x is negative and the absolute value is large, Approaching 0, the function output is close to 0, but it does not completely suppress negative inputs, retaining some information. This design allows the activation function to exhibit different behavioral characteristics under different input ranges, making it particularly suitable for processing signals such as pressure data, which requires both preserving subtle changes and dealing with threshold effects.

[0098] The initial values ​​of the α and β parameters have a significant impact on network training. Through experimental analysis, the system sets the initial values ​​of α to 1.0 and β to 0.5. This setting ensures that the activation function has moderate nonlinearity in the early stages of training, neither being too linear to lack expressiveness nor too nonlinear to cause training instability. During network training, α and β serve as learnable parameters and are updated along with other network parameters through the backpropagation algorithm.

[0099] Specifically, in the forward propagation phase, the activation value of each neuron is calculated ; In the back propagation phase, the gradient of the loss function L with respect to α and β is calculated: and , and then update the parameters using gradient descent: , where η is the learning rate.

[0100] In order to prevent the α and β parameters from diverging to unreasonable value ranges, a parameter constraint mechanism is implemented: a range constraint is applied to the α parameter, limiting it to The activation function maintains a moderate linear response strength within the interval; a range constraint is applied to the β parameter, limiting it to In the range, the nonlinear modulation term is prevented from becoming too steep or too flat. In addition, L2 regularization is applied to these two parameters to prevent them from taking too large values ​​and causing overfitting.

[0101] As training progresses, the α and β parameters of different layers converge to different values, reflecting the different requirements of each layer for linear and nonlinear characteristics. Generally, shallow networks (close to the input) have smaller α values ​​and larger β values, indicating that these layers pay more attention to nonlinear transformations of the input; deep networks (close to the output) have larger α values ​​and moderate β values, indicating that these layers need to retain more linear information to generate accurate risk predictions. Experimental results show that compared with traditional activation functions with fixed parameters, this activation function with learnable parameters improves the model's accuracy in pressure ulcer risk prediction tasks by 8.3%. It is particularly sensitive in identifying early risk signs and can capture subtle changes in pressure patterns that traditional methods may ignore.

[0102] This innovative activation function design reflects the technical depth of the present invention. By introducing learnable parameters and allowing the network to automatically adjust the shape of the activation function, the system can better adapt to the needs of the specific task of pressure ulcer risk prediction, improve the accuracy and sensitivity of the prediction, and provide more reliable decision support for clinical practice.

[0103] In step S4, based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate the risk assessment parameters of each body region, and output a pressure ulcer risk score and development trend prediction, including:

[0104] Step S4.1: Based on the pressure load distribution map and in combination with preset body region pressure ulcer susceptibility weights, a multidimensional risk assessment parameter vector including pressure index, duration, and tissue tolerance is constructed to form an initial parameter space;

[0105] Step S4.2: Initialize the population of the initial parameter space using Latin hypercube sampling to generate a diversified risk assessment parameter vector combination containing 30 individuals;

[0106] Step S4.3: When the differential evolution algorithm falls into a local optimum, a small-scale search is performed in a neighborhood space near the current optimal solution based on the diversified risk assessment parameter vector combination to generate an optimized parameter search strategy;

[0107] Step S4.4: Based on the optimized parameter search strategy, an adaptive crossover and selection strategy is implemented on the diversified risk assessment parameter vector combination, the crossover probability is dynamically adjusted according to the population diversity, the parameter combination is optimized, and a new generation of risk assessment parameter vector combination is generated;

[0108] Step S4.5: Calculate the pressure ulcer risk score for each body region using the new generation risk assessment parameter vector combination and estimate the risk change trend within the next 4 hours through a time series prediction model to obtain the pressure ulcer risk score and the development trend forecast.

[0109] In step S4.1, a comprehensive risk assessment parameter vector is constructed based on the pressure load distribution map output from step S3, providing a foundation for subsequent differential evolution algorithm optimization. First, the following information is collected as the basis for constructing the parameter vector: the pressure load distribution map (the 8×16 risk index matrix obtained from step S3.5, where each element ranges from 0 to 1 and represents a preliminary risk assessment for the corresponding location) and body region pressure ulcer susceptibility weights from medical literature (different body regions exhibit significant differences in their ability to tolerate pressure, typically including the following weights: sacrum, weight 1.5, one of the areas most susceptible to pressure ulcers; heel, weight 1.3, bony prominence with thin tissue layers; hip, weight 1.2, concentrated force area during side-lying; scapular, weight 1.1, force area during prone sleeping; occipital, weight 1.1, head support area; and other regions, weight 1.0, indicating relatively low risk). Next, the original pressure map is mapped to the body area to determine the body area to which each pixel belongs. This mapping is based on a predefined anatomical template, taking into account the patient's height and body position, and adjusting the template to match the current patient through affine transformation. Then, a multidimensional risk assessment parameter vector is constructed, which includes the following dimensions: pressure intensity parameter (Based on the risk index in the pressure load distribution map, combined with the regional weight: , where r represents a specific region); duration parameter (Indicates the duration of a specific stress level, based on continuous monitoring data: , where I is the indicator function, when the pressure in area r at time t When the threshold is exceeded, the value is 1. is the sampling interval); tissue tolerance parameter (Reflects the patient's tissue resistance to pressure, calculated based on patient characteristics: , where each factor is an adjustment coefficient between 0 and 1); pressure change rate parameter (Reflects the rate of change of pressure over time, used to evaluate microcirculatory recovery capacity: ); cumulative risk parameter (Considering the cumulative effect of historical risks, the decay sum model is adopted: , where λ is the time decay coefficient). These parameters are combined into a high-dimensional risk assessment parameter vector V:

[0110]

[0111] The subscripts represent different body regions. For the eight main body regions and five parameter categories, a 40-dimensional parameter vector is obtained. In addition, a set of weight parameters W is included to adjust the importance of different parameters in the final risk assessment:

[0112]

[0113] The final risk assessment function can be expressed as a function of the parameter vector V and the weight W: This risk assessment parameter vector includes multiple key factors in the formation of pressure ulcers and provides a target space for optimization of the differential evolution algorithm.

[0114] In step S4.2, a diverse parameter population is initialized to provide a good starting point for the subsequent evolution process. The core of the differential evolution algorithm is to maintain a population containing multiple candidate solutions (individuals), each of which represents a set of possible risk assessment parameter combinations. First, determine the population size and parameter vector dimension: the population size NP is set to 30 individuals, which strikes a balance between computational efficiency and search capability; the parameter vector dimension D is based on the risk assessment parameter vector constructed in step S4.1, which includes weight coefficients and adjustment parameters, with a total dimension of approximately 20-40. Next, define the parameter value range: the weight parameter (such as , The range of (e.g., [0, 2]) is usually set to [0, 2], allowing the algorithm to explore the importance of different parameters; the time decay coefficient (λ) range is set to [0.01, 0.5] to control the decay rate of historical risk; the threshold parameter is set to a reasonable range based on the specific parameter characteristics. Then, Latin hypercube sampling is used The method initializes the population. Compared to simple random sampling, LHS ensures a uniform distribution of samples across each dimension, improving the coverage of the parameter space. The basic principle of Latin Hypercube Sampling is to divide each dimension into NP equally spaced intervals, ensuring that each interval contains exactly one sample point. This allows for efficient coverage of the parameter space with relatively few samples, even in high-dimensional environments. After initialization, the system evaluates the fitness of each individual. The fitness function is based on a historical dataset and measures the accuracy of risk assessment using that set of parameters. This historical dataset contains patient cases with known pressure ulcer outcomes. Each case includes pressure data and other relevant features, as well as a marker indicating whether the patient developed a pressure ulcer within a specific time window. After initialization, all individuals and their fitness values ​​are stored in a population data structure to prepare for subsequent mutation, crossover, and selection operations. Furthermore, the system records a diversity metric of the initial population to monitor changes in population diversity during evolution and prevent premature convergence to local optima. The population initialized using Latin Hypercube Sampling exhibits good diversity and uniformity, providing a high-quality search starting point for the differential evolution algorithm.

[0115] In step S4.3, an innovative neighborhood mutation operation is implemented to enhance the algorithm's ability to escape local optima while accelerating the convergence process. The core idea of ​​neighborhood mutation is that when the algorithm may be trapped in a local optimum, instead of conducting a completely random global exploration, a targeted local search is performed in the "neighborhood" near the current optimal solution, balancing global exploration and local development capabilities. First, the basic mutation operation of standard differential evolution is defined. For each individual in the population , generate the test vector Standards The strategy is: ,in , , are three different individuals randomly selected from the population, and F is a scaling factor (usually Neighborhood mutation operation enhances the standard mutation by: neighborhood structure definition (defining the local neighborhood structure for the parameter space, for the current optimal individual , its neighborhood Defined as: ,in Is X and Adaptive neighborhood radius (the neighborhood radius r is not fixed, but dynamically adjusted according to the evolution stage: , where t is the current iteration number, T is the maximum iteration number, and α is the parameter that controls the decay rate); neighborhood mutation triggering mechanism (design an adaptive triggering mechanism that triggers neighborhood mutation when one of the following conditions is met: the improvement of the optimal solution for k consecutive iterations is less than the threshold ε, the population diversity index is lower than the threshold σ, and the random probability ); Neighborhood mutation operation (when triggering neighborhood mutation, for individual , whose test vector The generation method is modified to:

[0116]

[0117] in, is the best individual in the current population, and It is from Two individuals randomly selected from the neighborhood of is the main scaling factor, is the noise scaling factor); local search enhancement (under certain conditions, such as the optimal solution remains unchanged for a long time, The key advantages of the neighborhood mutation operation include: enhanced local exploration (conducting a more concentrated search in promising areas), preserved exploration (maintaining population diversity through random perturbations and a dynamic neighborhood radius), adaptive behavior (automatically adjusting the search strategy based on the evolutionary stage and population state), and accelerated convergence (focusing exploration in areas near the optimal solution, improving algorithm efficiency). This neighborhood mutation mechanism is particularly well-suited for problems like pressure ulcer risk assessment, where the parameter space is complex and multiple local optima may exist.

[0118] In step S4.4, an adaptive crossover strategy and selection mechanism are implemented to optimize the convergence characteristics and population diversity of the algorithm. The crossover operation of the differential evolution algorithm is used to combine the information of the original individual and the mutant individual to generate the experimental individual, while the selection operation determines which individuals will enter the next generation. An adaptive crossover strategy is implemented to dynamically adjust the crossover probability according to the population diversity: the population diversity index D is calculated regularly (the average pairwise distance, the average distance to the population center, or the mean of the parameter standard deviation can be used); the crossover probability is adaptively adjusted (a smaller value is used when the diversity is high). To maintain diversity, use a larger To facilitate information exchange, typical parameters are set to ); Individual-level crossover probability (each individual can have its own crossover probability, which is adjusted according to its performance in the population, and individuals with good performance use smaller Retain more original information, and individuals with poor performance use larger Accept more variation); perform crossover (using adaptive A comprehensive selection strategy is also implemented, combining elite retention and tournament selection: elite retention mechanism (ensuring that the current best individual will not be lost during the evolution process); greedy selection (standard One-to-one selection in the algorithm, comparing the fitness of the original individual with the corresponding test individual); tournament selection (regularly introducing tournament selection to increase population diversity); diversity protection mechanism (monitoring population diversity and triggering protection measures when diversity falls below a threshold, such as retaining a portion of the optimal individuals and applying random perturbations to the remaining individuals). The advantages of this adaptive crossover and selection strategy are: balancing exploration and exploitation (dynamically adjusting parameters according to population status to achieve optimal balance at different stages of the algorithm), maintaining diversity (preventing premature convergence and loss of population diversity through multiple mechanisms), retaining elites (ensuring that the optimal solution is not lost during the evolution process), and enhancing adaptability (automatically adjusting the strategy according to problem characteristics and optimization progress). Through these adaptive strategies, the differential evolution algorithm can more efficiently explore the parameter space of pressure ulcer risk assessment and find more accurate parameter combinations.

[0119] In step S4.5, the optimized parameters are used to accurately assess the patient's pressure ulcer risk and predict future risk trends. The optimized parameters are used to calculate the pressure ulcer risk score for each area of ​​the patient's body: obtaining the current optimized parameters (the optimal individual output by the differential evolution algorithm includes weight coefficients and threshold parameters, which define the importance of different factors in risk assessment); regional risk score calculation (for each body region r, a multifactor risk model is used to calculate the risk score, extracting characteristic data for the region such as pressure index, time duration, tissue tolerance, rate of change, and cumulative risk, extracting optimized parameters such as the weights of each characteristic, calculating the risk score, and normalizing it to a range of 0-100); risk level classification (dividing the continuous risk score into different risk levels for easy clinical interpretation); and overall risk score (calculating the overall risk score based on the risk scores of each region, taking into account the dominant role of high-risk areas). Next, based on historical data, the changing trends of risk scores for each region within the next four hours are predicted: time series data preparation (collecting time series data on risk scores for each region over the past 24 hours, sampling at fixed intervals such as 15 minutes); trend analysis method selection (selecting an appropriate time series prediction method based on data characteristics, such as linear regression, exponential smoothing, ARIMA model, or recurrent neural network); risk trend prediction implementation (using exponential smoothing as an example, Holt exponential smoothing is used to predict risk trends, including initialization, exponential smoothing calculation, and predicting future values); risk change trend classification (classifying predicted trends to facilitate clinical interpretation and intervention decisions, such as rapid increase, moderate increase, slight increase, stable, slight decrease, moderate decrease, and rapid decrease); and risk warning threshold setting (determining whether to trigger a warning based on a comprehensive assessment of the risk score and trend, and adjusting the warning threshold based on the trend). Finally, the risk score and trend prediction results are integrated to generate a comprehensive risk report, including the overall risk score and risk level, detailed risk scores and trends for each body region, a list of high-risk areas and their risk status, risk warnings requiring immediate intervention, and a risk trend forecast for the next four hours. This risk report provides comprehensive risk assessment information to healthcare professionals, supporting timely and targeted interventions to improve the effectiveness of pressure ulcer prevention.

[0120] The calculation of the pressure ulcer risk score includes:

[0121] Step S4.5.1: Based on the pressure index, duration, tissue tolerance, pressure change rate, and cumulative risk parameter for each body region, a multi-factor risk model is used to perform a weighted calculation to output the pressure ulcer risk score within a range of 0-100 points;

[0122] Step S4.5.2: Risk levels are divided according to the pressure ulcer risk score: 0-25 points are low risk, 26-50 points are medium-low risk, 51-75 points are medium-high risk, and 76-100 points are high risk, generating a risk level classification result.

[0123] In step S4.5.1, a multi-factor risk model is used to calculate the pressure ulcer risk score for each body region. First, weight parameters are extracted from the optimal individual obtained by differential evolution algorithm optimization. These parameters determine the importance of different risk factors in the final score. For each body region r, the characteristic data of the region are extracted: pressure index (indicating the pressure intensity of the region, taking into account the susceptibility weight of the region), duration (indicating the length of time the pressure lasts), tissue tolerance (indicating the resistance of the tissue to pressure, determined by factors such as the patient's age, nutritional status, and humidity), pressure change rate (indicating the rate of change of pressure over time, a positive value indicates an increase in pressure, and a negative value indicates a decrease in pressure) and cumulative risk (indicating the cumulative effect of historical risks, taking into account time decay). The system then calculates the risk score using a weighted summation model:

[0124]

[0125] in, 、 、 、 and is the optimized weight parameter. Note that the tissue tolerance term uses , because the lower the tolerance, the higher the risk; the pressure change rate term uses , because only an increase in stress (positive rate of change) increases risk. The calculated raw risk score is normalized to a range of 0-100 using a linear mapping: ,in A scaling factor is determined based on historical data to ensure that most scores fall within a meaningful range. The advantages of this multifactorial risk model include: it comprehensively considers multiple key factors influencing pressure ulcer formation; its weighting parameters are automatically optimized through machine learning rather than manually set; it can adapt to the characteristics of different patient populations; and its scoring results are intuitive and clinically meaningful.

[0126] In step S4.5.2, the continuous risk score is divided into different risk levels to facilitate clinical interpretation and intervention decision-making. A four-level risk classification standard is used: 0-25 is low risk , indicating a low likelihood of developing a pressure ulcer, usually requiring only routine care and regular monitoring; 26-50 points are low to moderate risk , indicating a certain risk of pressure sores, requiring increased monitoring frequency and consideration of preventive measures; 51-75 points indicate medium-to-high risk , indicating a significant risk of pressure sores, requiring active preventive measures such as increasing the frequency of turning over and using pressure relief equipment; 76-100 points indicate high risk , indicating that the possibility of pressure sores is extremely high and comprehensive prevention and intervention measures need to be taken immediately, and may require the intervention of a professional care team. Based on the risk score of each body area, the corresponding risk level classification results are automatically generated and intuitively displayed on the user interface using different colors: low-risk areas are displayed in green, medium-low risk areas are displayed in yellow, medium-high risk areas are displayed in orange, and high-risk areas are displayed in red. In addition, the overall risk level is calculated, taking into account the weighted average of the risks of each area and the impact of the highest risk area:

[0127]

[0128] The weighted average takes into account both the area and importance of the region. The advantages of this risk classification are: it simplifies complex numerical scoring, enabling healthcare professionals to quickly understand risk situations; standardized classification facilitates the development of nursing protocols and intervention processes; intuitive color coding improves the efficiency of risk identification; and it balances overall risk with localized high risk, avoiding the problem of high-risk areas being diluted by averages. The risk classification results serve as an important basis for subsequent nursing intervention decisions, guiding healthcare professionals to rationally allocate resources and prioritize high-risk patients and high-risk body areas.

[0129] In step S5, the optimal airbag inflation and deflation parameters are calculated using an intelligent decision-making algorithm based on the pressure sore risk score and the development trend forecast, and an execution instruction is generated to the airbag control system, including:

[0130] Step S5.1: Based on the pressure ulcer risk score and the development trend prediction, identify high-risk areas and medium-risk areas, calculate the ideal pressure distribution target value for each area, and determine the target body area requiring pressure redistribution;

[0131] Step S5.2: Establishing a transfer function model between the airbag pressure and the mattress surface pressure, calculating the optimal inflation and deflation parameters of each airbag using a quadratic programming algorithm for the target body area and the ideal pressure distribution target value, and generating a control instruction sequence;

[0132] Step S5.3: According to the control instruction sequence, the micro air pump and the solenoid valve are controlled by the PWM signal to adjust the pressure of each airbag, and the execution instruction is obtained to the airbag control system.

[0133] In step S5.1, based on the pressure ulcer risk score and predicted development trend, body areas requiring pressure redistribution are identified, and the ideal pressure distribution target value is calculated. The patient's body risk status is first categorized: areas with a risk score greater than 70 are designated as high-risk, areas with a risk score between 40 and 70 are designated as medium-risk, and areas with a risk score below 40 are considered low-risk. For high- and medium-risk areas, their risk trends are further analyzed. If the risk score is increasing (expected to increase by more than 5 points within the next 4 hours), it is marked as "urgent intervention required"; if the risk score is stable or slowly increasing (expected to increase by 1-5 points), it is marked as "needing attention"; if the risk score is decreasing, it is marked as "ongoing monitoring." Next, the ideal pressure distribution target value is calculated for each area. For high-risk areas, the goal is to significantly reduce surface pressure, ideally set at 50-60% of the current pressure. For medium-risk areas, the goal is to moderately reduce surface pressure, ideally set at 70-80% of the current pressure. For low-risk areas, pressure can be increased to compensate for the reduced pressure in high-risk areas, but the increase should not exceed 20%, and the absolute pressure should not exceed a safety threshold (typically 40 mmHg). The system also considers the need for overall body support and balance, ensuring that pressure redistribution does not cause patient instability or excessive pressure concentration in localized areas. Furthermore, the system analyzes pressure gradients between adjacent areas to avoid excessive pressure differences, which could increase shear forces and the risk of tissue damage. Based on this analysis, an ideal pressure distribution target matrix is ​​generated. Compared to the current actual pressure distribution matrix, it represents the target pressure value to be achieved at each location. This ideal pressure distribution target matrix serves as the basis for subsequent optimization of airbag control parameters, guiding the system in adjusting the pressure of each airbag to achieve optimal pressure redistribution, reduce the pressure load in high-risk areas, and prevent the occurrence and development of pressure ulcers.

[0134] In step S5.2, a transfer function model between the airbag pressure and the mattress surface pressure is established, and the optimal airbag control parameters are calculated using a quadratic programming algorithm. The smart mattress contains 64 independently controlled airbag units, corresponding to an 8×16 pressure sensor matrix. Each airbag can individually adjust the inflation volume and internal pressure, thereby changing the support characteristics and pressure distribution on the mattress surface. First, a transfer function model between the internal pressure of the airbag and the surface pressure of the mattress is established. This model describes how changes in the internal pressure of the airbag affect the pressure distribution on the mattress surface, taking into account the mutual influence between the airbags, the elastic properties of the mattress material, and the mechanical response of human tissue. The transfer function can be expressed in matrix form: ,in is the pressure distribution matrix on the mattress surface, is the pressure vector inside the airbag, and F is the nonlinear transfer function. This transfer function is obtained by combining experimental data and physical modeling, and is optimized and calibrated using machine learning methods. With the transfer function model, the pressure redistribution problem is transformed into an optimization problem: find a set of airbag internal pressure values ​​so that the actual pressure distribution on the mattress surface is as close as possible to the ideal pressure distribution target value determined in step S5.1. This optimization problem can be expressed as: minimize ,in is the target pressure distribution matrix, while taking into account constraints such as the upper and lower limits of the airbag pressure, and the smoothness requirements of the adjacent airbag pressures. The system uses a quadratic programming algorithm to solve this optimization problem, because the objective function is a quadratic form of the airbag pressure, and the constraints can be expressed as linear inequalities. The quadratic programming algorithm can efficiently find the optimal solution that meets all constraints and calculate the optimal inflation and deflation parameters for each airbag. The optimization results include the target pressure value, inflation and deflation rate, and sequence of each airbag, forming a complete control instruction sequence. This control instruction sequence takes into account factors such as patient comfort, system response time, and energy efficiency to ensure that the airbag pressure adjustment process is smooth, efficient, and comfortable, while achieving the best pressure redistribution effect.

[0135] In step S5.3, the system uses hardware actuators to precisely adjust the pressure of each airbag based on the control instruction sequence. The smart mattress's airbag control system includes a high-precision micro air pump, a solenoid valve array, a pressure sensor, and a control circuit. The control unit receives the control instruction sequence generated in step S5.2 and converts it into specific hardware control signals. First, the system determines the target pressure for each airbag based on the control instruction. It calculates the difference between the current pressure and the target pressure and decides whether to inflate, deflate, or maintain the pressure. For airbags that need to be inflated, the corresponding micro air pump and inlet solenoid valve are activated. For airbags that need to be deflated, the exhaust solenoid valve is opened. For airbags whose pressure is close to the target value, all valves are closed to maintain a stable pressure. To achieve precise pressure control, PWM (pulse width modulation) signals are used to control the operating state of the air pump and solenoid valve. The duty cycle of the PWM signal determines the output power of the air pump and the opening time of the solenoid valve. By adjusting the duty cycle, the airflow rate can be precisely controlled, achieving smooth pressure regulation. For example, for airbags requiring a significant pressure increase, a high-duty-cycle PWM signal drives the air pump; for airbags requiring fine-tuning, a low-duty-cycle signal achieves slow, precise adjustments. A closed-loop control mechanism is also implemented. A pressure sensor built into the airbag monitors the internal pressure in real time, compares the measured value with the target value, and dynamically adjusts the PWM signal parameters to ensure the airbag pressure accurately reaches the set value. This closed-loop control strategy compensates for factors such as system latency and changes in airflow resistance, improving control accuracy and response speed. The entire airbag control system boasts a minimum adjustment accuracy of ±1 mmHg and a response time of less than 2 seconds, enabling rapid response to changes in risk assessment results and timely adjustments to the pressure distribution on the mattress surface. Furthermore, airbag pressure adjustment is prioritized and timed to ensure that pressure adjustments are prioritized in high-risk areas while avoiding patient discomfort caused by large simultaneous inflation and deflation of multiple airbags. Through this precise and efficient airbag control mechanism, the system proactively adjusts the pressure distribution on the mattress surface based on pressure ulcer risk assessment results, providing patients with personalized pressure redistribution solutions and effectively preventing the development and progression of pressure ulcers.

[0136] The aforementioned simultaneously generates nursing intervention recommendations, including:

[0137] Step S5.4: Based on the pressure ulcer risk score and the development trend prediction, identifying high-risk areas that cannot be fully alleviated by airbag adjustment, and determining key care areas requiring manual intervention;

[0138] Step S5.5: Combining the patient's historical data and the risk level classification results, using a decision tree algorithm to determine the optimal intervention strategy for the key nursing area, and generating personalized nursing intervention recommendations;

[0139] Step S5.6: When the pressure ulcer risk score of any area in the key care area is continuously monitored to be greater than 85 points and the predicted trend of the development trend forecast is rising, a turning reminder is automatically generated based on the personalized nursing intervention suggestion and the nursing staff is notified through sound and light prompts and mobile terminal push notifications to complete the nursing intervention notification.

[0140] In step S5.4, high-risk areas that cannot be fully alleviated by airbag adjustment are identified, and key care areas requiring manual intervention are determined. Although the intelligent airbag control system effectively redistributes pressure across the mattress surface, in some cases, airbag adjustment alone may not be able to reduce pressure in all high-risk areas to safe levels. These situations include: areas where the patient's weight exceeds the safety threshold even with maximum pressure reduction; bony prominences (such as the sacrum, heels, and hip bones) that inevitably form high-pressure points in certain body positions; areas where tissue tolerance is extremely low due to specific patient conditions or injuries; and areas where prolonged pressure is sustained in the same position. The system identifies these areas that remain at high risk by analyzing the expected pressure distribution after airbag adjustment and comparing it to the safety threshold. Specifically, the system calculates an expected risk score for each area after optimal airbag adjustment. If a region's expected risk score remains above 70, or if its risk score, while below 70, shows a rapidly increasing trend (expected to increase by more than 10 points within four hours), it is identified as a "key care area requiring manual intervention." In addition, the risk duration factor is also taken into account. For areas where the risk score exceeds the threshold for more than 2 hours, even if the risk score is relatively low, it will be included in the key care scope, because long-term continuous pressure is one of the key factors in the formation of pressure sores. These key care areas are sorted by risk level to form a priority list to guide subsequent nursing intervention decisions. This combination of machine assistance and manual intervention fully utilizes the automatic adjustment capabilities of the intelligent system and the experience and judgment of professional caregivers, forming complementary advantages and improving the overall effectiveness of pressure sore prevention. The identified key care areas will serve as the basis for generating personalized nursing intervention recommendations in step S5.5, ensuring that nursing resources are concentrated on the body areas that need the most attention.

[0141] In step S5.5, a decision tree algorithm is used to generate personalized nursing intervention recommendations for key care areas, combining patient historical data and risk classification results. First, the system collects patient information relevant to nursing interventions, including basic information (age, gender, weight, height, etc.), clinical status (primary medical conditions, mobility, consciousness, etc.), pressure ulcer risk assessment results (Braden score dimensions, historical risk trends, etc.), previous nursing records (previously effective interventions, patient responses to different interventions, etc.), and currently used medical devices and assistive devices. A decision tree algorithm is then used to analyze the relationship between this information and the effectiveness of nursing interventions. Decision trees are an intuitive and efficient machine learning method that can learn "if-then" rules from historical data, making them suitable for interpretable scenarios such as nursing decision-making. The decision tree model is trained based on a large amount of historical nursing records to learn the most effective nursing intervention strategies based on different patient characteristics and risk profiles. For each key care area, the decision tree model recommends the most appropriate intervention based on the specific context (risk score, risk trends, anatomical location, etc.) and patient characteristics. These intervention recommendations include, but are not limited to: turning plans (recommended turning angles, frequencies, and specific methods), local decompression measures (recommendations for using specific positioning pads, air cushions, or other decompression devices), skin care plans (specific instructions for cleaning, moisturizing, massage, etc.), nutritional intervention recommendations (for patients with low tissue tolerance), and monitoring frequency recommendations (recommended time intervals for caregivers to check specific areas). The generated nursing intervention recommendations are in a structured format, containing a detailed description of the intervention measures, implementation frequency, precautions, and expected effects, making it easier for caregivers to understand and implement them. These recommendations will be dynamically adjusted based on changes in the patient's condition and feedback on the intervention effects to ensure that the intervention measures always maintain optimal targeting and effectiveness. Through this personalized nursing intervention recommendation based on machine learning, the system is able to transform expert-level pressure ulcer prevention knowledge into specific, actionable nursing guidance, improving the quality and efficiency of nursing care.

[0142] In step S5.6, the system implements proactive early warning and nursing intervention notifications to ensure that high-risk situations are addressed promptly. The system continuously monitors the risk status of all key care areas, triggering early warning and intervention notifications when specific conditions are met. Trigger conditions include: a pressure ulcer risk score exceeding 85 (indicating extremely high risk) in any key care area; a risk score not exceeding 85 but with a predicted upward trend (expected to exceed 85 within 2 hours); and a risk score remaining at a high risk level (>75) for a period exceeding a preset safety limit (typically 2 hours). When these conditions are triggered, the system automatically generates a turning reminder and nursing intervention notification. The notification content is based on the personalized nursing intervention recommendations generated in step S5.5 and includes the specific risk area, current risk score, risk trend, recommended intervention measures, and operational instructions. These notifications are delivered through various channels: visual and audio notifications on the bedside control unit (which uses LED indicators and a buzzer to provide intuitive indications of risk level and location); risk alerts on the nursing station monitoring terminal (which displays patient information and detailed risk data requiring intervention on a central monitoring system); and mobile push notifications (which push notifications to caregivers' handheld devices or smartwatches, ensuring they receive notifications even when they are away from the patient's room). A notification priority management and escalation mechanism is also implemented: For extremely high-risk situations (scores >90), a high-priority alert is issued, and if no response is received, the notification level is automatically escalated, ensuring that urgent situations are addressed promptly. After receiving the notification and implementing the intervention, the caregiver is required to confirm the intervention and record its effectiveness. The caregiver can enter the implementation status and observed effects through the mobile terminal or bedside control unit. This feedback is recorded in the patient's care log and used to evaluate the effectiveness of the intervention and optimize future intervention recommendations. Risk status is continuously monitored after the intervention, and if the intervention fails to effectively reduce the risk, further intervention recommendations or a recommendation for evaluation by the medical team are generated. Through this closed-loop early warning-intervention-feedback mechanism, high-risk situations can be handled promptly and effectively, significantly improving the effectiveness and efficiency of pressure ulcer prevention.

[0143] like Figure 3 As shown, the present invention also provides an intelligent anti-pressure sore mattress, comprising: a mattress body, on which a plurality of pressure sensors are arranged in an array; an airbag unit, which is arranged in the mattress body; an airbag control system, for controlling the inflation and deflation of the airbag unit; a control unit, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor implements the pressure sore risk prediction method as described above when executing the program.

[0144] The smart anti-pressure ulcer mattress features a multi-layer composite structure consisting, from top to bottom, of a contact layer, a sensor layer, an airbag support layer, and a base support layer. The contact layer is made of highly elastic medical-grade silicone material, offering excellent biocompatibility and comfort. Its surface features a honeycomb microstructure, maximizing contact area and reducing pressure per unit area. It also provides excellent air permeability and heat dissipation, preventing skin damage caused by localized high temperatures and humidity. The contact layer is 10-15mm thick and has a moderate hardness, providing excellent comfort without excessive indentation or a wrapping effect. The sensor layer incorporates an array of high-precision flexible pressure sensors arranged in an 8×16 matrix, covering all pressure-bearing areas of the body. Each sensor unit measures 2cm×2cm and is only 0.8mm thick, virtually minimizing the mattress's comfort. These sensors utilize a piezoresistive principle, with a measurement range of 0-200mmHg, an accuracy of ±1mmHg, and a response time of less than 50ms, enabling them to accurately capture even subtle changes in pressure distribution. The sensors are connected via flexible circuit boards, and the signal lines are designed with redundancy to ensure that a single point of failure will not affect the overall function.

[0145] The airbag support layer is the core of the smart mattress, consisting of 64 independently controlled airbag units, each corresponding to a specific location in the pressure sensor matrix. Each airbag is made of medical-grade TPU material, offering high strength, elasticity, and durability, capable of withstanding repeated inflation and deflation without deformation. The airbags measure 4cm x 4cm and are adjustable in height from 2 to 8cm. Varying the internal air pressure allows for varying support firmness and height. The airbags are arranged in a honeycomb pattern, with appropriate gaps between adjacent airbags. This ensures overall support while allowing for independent deformation and improved local adaptability. The base of the airbags is connected to the air circuit system, and each airbag is equipped with an independent air inlet and outlet valve for precise individual control. The base support layer is constructed of a 5cm thick high-density sponge material, providing foundational support and stability while also absorbing shock and enhancing overall comfort. A reinforced frame surrounds the mattress to prevent edge collapse from prolonged use, extending its lifespan.

[0146] The airbag control system, housed in a control box on the side of the mattress, includes a micro-pump assembly, a solenoid valve array, a pressure sensor, and control circuitry. The micro-pump assembly consists of four independent, silent, oil-free air pumps, each controlling 16 airbags. With a maximum output pressure of 250 mmHg and a flow rate of 3 L / min, it can fully deflate a single airbag in 10 seconds. The solenoid valve array, comprising 128 micro-solenoid valves (one for each airbag, one for inlet and one for outlet), features a low-power design with static power consumption of less than 0.1 W and dynamic power consumption of less than 0.5 W. It supports PWM control for precise airflow regulation. Each airbag also houses a micro-pressure sensor for real-time monitoring of internal pressure, forming a closed-loop control system that ensures precise pressure control. The control circuitry utilizes a modular design, comprising a master control board and multiple slave control boards, communicating via a CAN bus for high reliability and scalability. The power consumption of the entire airbag control system is less than 5W in standby mode, and the maximum operating power consumption does not exceed 50W. It can be connected to the hospital power supply system through a medical-grade power adapter. It is also equipped with a built-in lithium battery, which can maintain normal operation of the system for more than 4 hours in the event of a power outage.

[0147] The control unit is the brain of the entire smart anti-pressure ulcer mattress, responsible for data processing, risk assessment, and control decisions. The control unit utilizes an embedded system design, featuring a high-performance processor, large-capacity memory, and multiple communication interfaces. The processor utilizes a quad-core ARM Cortex-A72 architecture with a clock speed of 1.5GHz and features a dedicated AI acceleration unit that supports neural network inference acceleration, enabling efficient execution of the pressure ulcer risk prediction algorithm. The memory, consisting of 4GB of LPDDR4 RAM and 64GB of eMMC flash memory, provides ample runtime and data storage capacity, capable of storing over a year of historical pressure data and risk assessment records. Communication interfaces include Wi-Fi, Bluetooth, Ethernet, and USB, enabling seamless integration with hospital information systems, nurse station monitoring terminals, and mobile devices. The control unit also features a 7-inch touchscreen display with a resolution of 1024×600, displaying real-time pressure distribution heatmaps, risk assessment results, and system status, allowing caregivers to operate and review data directly at the bedside. The system software, developed based on the Linux operating system, utilizes a modular architecture, comprising a data acquisition module, a feature extraction module, a risk prediction module, an airbag control module, and a user interface module. The software implements the pressure ulcer risk prediction method of the present invention, including core functions such as pressure data preprocessing, feature selection, neural network model reasoning, differential evolution algorithm optimization and intelligent decision-making algorithm.

[0148] The intelligent anti-pressure ulcer mattress operates as follows: First, a distributed pressure sensor array collects real-time pressure distribution data on the mattress surface at a sampling frequency of 10Hz and transmits the data to a control unit via Bluetooth Low Energy technology. The control unit filters and normalizes the raw pressure data to generate a standardized pressure distribution matrix. The system then applies a binary randomized forward algorithm to extract key features from the pressure data and fuses these features with patient vital signs (obtained from the hospital information system or manually entered). Next, an enhanced convolutional neural network model processes this fused data to generate a pressure load distribution map, visually displaying the risk index for each region. Based on this map, the system applies a differential evolution algorithm with neighborhood mutation to calculate risk assessment parameters for each body region, outputting a pressure ulcer risk score and a four-hour development trend forecast. Finally, an intelligent decision-making algorithm calculates the optimal airbag inflation and deflation parameters based on the risk assessment results. The airbag control system adjusts the pressure of each airbag to achieve precise redistribution of pressure on the mattress surface, reducing the pressure load in high-risk areas. The system also generates personalized nursing intervention recommendations and, when necessary, sends turning reminders to caregivers through audio and visual cues and mobile device push notifications.

[0149] The smart anti-pressure ulcer mattress features multiple safety measures: an overpressure protection mechanism ensures that the pressure within the airbag never exceeds a safe threshold; a power-off protection function automatically locks the current airbag status in the event of a power outage, preventing sudden deflation and patient discomfort; a fault self-diagnosis system monitors the operating status of sensors, air pumps, and solenoid valves, promptly reporting any anomalies; and a data backup mechanism regularly synchronizes important data to the cloud or a local backup device to prevent data loss. Furthermore, the system supports remote monitoring and maintenance, allowing technicians to remotely diagnose system issues, update software, and adjust parameters over a secure connection to ensure the system remains in optimal condition.

[0150] Compared to traditional pressure-ulcer-preventing mattresses, the intelligent pressure-ulcer-preventing mattress of this invention offers significant advantages: it shifts from passive prevention to active prediction, identifying high-risk areas and implementing interventions before pressure ulcers develop; personalized pressure adjustment plans take into account individual patient differences and real-time status changes, providing precise pressure management; an intelligent nursing reminder system reduces the workload of caregivers and improves nursing efficiency; and comprehensive data recording and analysis capabilities support long-term risk management and nursing quality assessment. These advantages enable the intelligent pressure-ulcer-preventing mattress of this invention to significantly reduce the incidence of pressure ulcers, alleviate patient suffering, reduce medical costs, and improve the quality of care.

[0151] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0152] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of patent protection for the present application shall be determined by the appended claims.

Claims

1. A method for predicting pressure ulcer risk, characterized in that: The pressure ulcer risk prediction method comprises the following steps: Obtaining pressure distribution data collected by a distributed pressure sensor array, performing signal filtering and normalization processing on the pressure distribution data to obtain a normalized pressure distribution matrix; Based on the standardized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output an optimal feature subset; Acquiring patient vital sign parameters, fusing the patient vital sign parameters with the optimal feature subset, training using an enhanced convolutional neural network with an activation function, and generating a pressure load distribution map; Based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate risk assessment parameters for each body region, and output a pressure ulcer risk score and development trend prediction; Based on the pressure ulcer risk score and the development trend forecast, the optimal airbag inflation and deflation parameters are calculated through an intelligent decision-making algorithm, and execution instructions are generated to the airbag control system, and nursing intervention suggestions are also generated.

2. The method according to claim 1, characterized in that The signal filtering and normalization processing of the pressure distribution data to obtain a normalized pressure distribution matrix includes: Based on the pressure distribution data, applying a combination of median filtering and Kalman filtering to remove noise and outliers to generate cleaned pressure data; Performing min-max normalization processing on the pressure data after cleaning, mapping the values ​​to the interval [0, 1] to obtain normalized data; A two-dimensional pressure distribution matrix is ​​reconstructed according to the sensor spatial arrangement information and the standardized data to obtain the standardized pressure distribution matrix.

3. The method according to claim 1, characterized in that Based on the standardized pressure distribution matrix, a binary random forward algorithm is applied to perform feature selection and output an optimal feature subset, including: Based on the standardized pressure distribution matrix, a multidimensional feature pool including statistical features, morphological features and temporal features is constructed to form an initial feature set; Performing binary encoding representation on the initial feature set to construct a binary representation model for feature selection; A random forward algorithm is used to iteratively select an optimal feature combination from the binary representation model to obtain the optimal feature subset.

4. The method according to claim 1, wherein The enhanced convolutional neural network with an activation function is used for training to generate a pressure load distribution map, including: Performing multimodal data fusion on the optimal feature subset and the patient's vital sign parameters to construct an input tensor to form network input data; Design a deep neural network architecture with 5 convolutional layers. The first 3 layers use 3×3 convolution kernels, and the last 2 layers use 5×5 convolution kernels. Perform multi-scale feature extraction on the network input data and output a multi-level feature representation. Based on the multi-level feature representation, an 8×16 feature map is generated through the last convolutional layer to obtain the pressure load distribution map.

5. The method according to claim 4, characterized in that The activation function Here, α and β are learnable parameters that are automatically adjusted during training through back propagation.

6. The method according to claim 1, characterized in that Based on the pressure load distribution map, a differential evolution algorithm with neighborhood mutation is applied to calculate the risk assessment parameters of each body region, and output a pressure ulcer risk score and development trend prediction, including: Based on the pressure load distribution map and in combination with the preset body region pressure ulcer susceptibility weights, a multidimensional risk assessment parameter vector including pressure index, duration, and tissue tolerance is constructed to form an initial parameter space; Latin hypercube sampling is used to initialize the population of the initial parameter space to generate a diversified risk assessment parameter vector combination containing 30 individuals; When the differential evolution algorithm falls into a local optimum, a small-scale search is performed in a neighborhood space near the current optimal solution based on the diversified risk assessment parameter vector combination to generate an optimized parameter search strategy; Based on the optimized parameter search strategy, an adaptive crossover and selection strategy is implemented on the diversified risk assessment parameter vector combination, the crossover probability is dynamically adjusted according to the population diversity, the parameter combination is optimized and a new generation of risk assessment parameter vector combination is generated; The pressure ulcer risk score is calculated for each body region using the new generation risk assessment parameter vector combination and the risk change trend within the next 4 hours is estimated using a time series prediction model to obtain the pressure ulcer risk score and the development trend prediction.

7. The method according to claim 6, characterized in that The calculation of the pressure ulcer risk score includes: Based on the pressure index, duration, tissue tolerance, pressure change rate and cumulative risk parameters of each body area, a multi-factor risk model is used for weighted calculation to output the pressure ulcer risk score within the range of 0-100 points; The pressure ulcer risk score is used to divide the risk level into the following categories: 0-25 points are low risk, 26-50 points are low-medium risk, 51-75 points are medium-high risk, and 76-100 points are high risk, generating a risk level classification result.

8. The method according to claim 1, characterized in that The method of calculating the optimal airbag inflation and deflation parameters based on the pressure sore risk score and the development trend forecast through an intelligent decision-making algorithm and generating an execution instruction to the airbag control system includes: Based on the pressure ulcer risk score and the development trend prediction, high-risk areas and medium-risk areas are identified and target values ​​of ideal pressure distribution are calculated for each area, thereby determining target body areas requiring pressure redistribution; Establishing a transfer function model between the airbag pressure and the mattress surface pressure, calculating the optimal inflation parameters and optimal deflation parameters of each airbag for the target body area and the ideal pressure distribution target value using a quadratic programming algorithm, and generating a control instruction sequence; According to the control instruction sequence, the micro air pump and the solenoid valve are controlled by the PWM signal to adjust the pressure of each airbag, and the execution instruction is obtained and sent to the airbag control system.

9. The method according to claim 7, characterized in that The aforementioned simultaneously generates nursing intervention recommendations, including: Based on the pressure ulcer risk score and the development trend prediction, identifying high-risk areas that cannot be fully alleviated by airbag adjustment, and determining key care areas requiring manual intervention; Combining the patient's historical data and the risk level classification results, using a decision tree algorithm to determine the optimal intervention strategy for the key nursing area, and generating personalized nursing intervention recommendations; When the pressure ulcer risk score of any area in the key care areas is continuously monitored to be greater than 85 points and the predicted trend of the development trend forecast is rising, a turning reminder is automatically generated based on the personalized nursing intervention suggestion and the nursing staff is notified through sound and light prompts and mobile terminal push notifications to complete the nursing intervention notification.

10. An intelligent anti-pressure sore mattress, characterized in that: include: A mattress body, on which a plurality of pressure sensors are arranged in an array; an airbag unit, which is arranged in the mattress body; an airbag control system, which is used to control the inflation and deflation of the airbag unit; a control unit, which includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor implements the pressure ulcer risk prediction method according to any one of claims 1 to 9 when executing the program.

Images