Pre-school autism sleep personalized intervention system based on AI behavior analysis

Abstract

The invention belongs to the technical field of medical health monitoring and behavior intervention, and discloses a pre-school autism sleep personalized intervention system based on AI behavior analysis, which comprises a behavior acquisition unit, an environment monitoring unit, a living habit management unit, a fusion analysis unit, an intervention execution unit, a self-adaptive optimization unit and a family interaction unit, according to the system, body movement, heart rate, skin electricity and sleeping posture information of children and environment parameters such as illumination, noise and temperature and humidity of a bedroom are synchronously collected, diet and exercise balance indexes are combined for fusion analysis, and behavior risk indexes are generated; a personalized intervention scheme is automatically generated according to the risk level, and intervention is implemented through light, sound and visual guidance modes. The system dynamically adjusts analysis parameters according to an intervention result, realizes closed-loop correlation analysis and continuous regulation and control of sleep behaviors, environment changes and parent feedback, and is suitable for individualized sleep management of children with autism.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical health monitoring and behavior intervention, and particularly relates to a pre-school autism sleep personalized intervention system based on AI behavior analysis. BACKGROUND

[0002] Children with autism spectrum disorder generally have problems such as sleep delay, frequent night awakening and insufficient deep sleep ratio in sleep behavior. Such sleep abnormalities are closely related to sensory sensitivity, environmental changes and emotional fluctuations. Research on this group of people shows that fluctuations in sleep quality not only affect daytime learning and social skills, but also exacerbate behavioral stereotypy and emotional instability.

[0003] Existing sleep monitoring devices mostly use single sensing equipment to collect physiological parameters such as body movement, heart rate, sleep posture, etc. to evaluate the overall sleep state. However, in the application of autistic children, due to the strong heterogeneity of their behavioral responses, relying on single source signals often cannot accurately reflect the real sleep situation. In addition, conventional systems mostly focus on result statistics, and lack of synchronous collection and correlation analysis of external influencing factors such as environmental light, noise, temperature and humidity.

[0004] In the family care scene, parents mainly rely on experience to adjust the bedroom environment or work and rest time, and lack of guidance based on objective data. At the same time, existing systems mostly adopt one-way information collection mode, and no effective interaction mechanism with the family end is established. The subjective feedback of parents and the emotional state of children cannot be timely included in the system analysis.

[0005] Therefore, the existing sleep monitoring and intervention technology generally has problems such as single data source, insufficient environmental correlation, lagging intervention feedback and weak human-computer interaction link in the individualized application of autistic children, and it is difficult to realize dynamic correlation analysis and continuous optimization of multi-dimensional information such as behavior, environment and emotion. SUMMARY

[0006] The purpose of the present application is to provide a pre-school autism sleep personalized intervention system based on AI behavior analysis to solve the problems raised in the background art.

[0007] In order to achieve the above purpose, the present application provides the following technical scheme: a pre-school autism sleep personalized intervention system based on AI behavior analysis, which comprises:

[0008] A behavior collection unit is used to acquire body movement, heart rate, skin electrical signal and sleep posture information of a pre-school autistic child, and to collect behavior data before falling asleep and during night awakening;

[0009] An environment monitoring unit is used to collect bedroom light, noise, temperature and humidity environmental parameters, and to perform synchronous sampling when a behavior change is detected;

[0010] a lifestyle management unit configured to record the child's diet and exercise information and calculate a diet and exercise balance index;

[0011] a fusion analysis unit configured to perform fusion analysis on the behavior data, environmental parameters, and the diet and exercise balance index based on a time convolutional neural network combined with an attention weighting algorithm, to identify key features affecting sleep onset and night waking, and to generate a behavior risk index;

[0012] an intervention execution unit configured to generate a personalized intervention plan based on the behavior risk index and to control visual guidance, environmental adjustment, and lifestyle adjustment interventions;

[0013] an adaptive optimization unit configured to evaluate the intervention effect and update the model weights in the fusion analysis unit based on sleep changes and emotional feedback information after intervention execution, to realize dynamic optimization of the intervention strategy;

[0014] a family interaction unit configured to display behavior, intervention, and sleep correlation information on a parent terminal and receive parent feedback to supplement model learning data.

[0015] Further, the behavior collection unit comprises:

[0016] a low-sensitivity wearable device worn on the body surface of a preschool autistic child, which is internally provided with three types of sensing components:

[0017] a body motion sensing component configured to detect the child's body displacement, turning over, and limb activity changes and generate a body motion amplitude signal;

[0018] a heart rate sensing component configured to obtain the child's pulse wave signal and calculate the instantaneous heart rate change;

[0019] a skin conductance sensing component configured to measure the skin conductance change on the child's skin surface;

[0020] The wearable device is further provided with a posture recognition module configured to determine the sleep posture state based on the body motion signal and store the sleep posture change data synchronously with the body motion signal.

[0021] The behavior collection unit is provided with a time labeling module configured to automatically label the time period information when the child falls asleep and when the behavior change is detected during night waking, and to transmit the body motion signal, heart rate signal, skin conductance signal, and sleep posture change data to the fusion analysis unit for feature analysis.

[0022] Further, the environmental monitoring unit comprises a light sensor, a noise sensor, and a temperature and humidity integrated detection module.

[0023] The light sensor is used to detect the illumination change of the bedroom space and output a light intensity signal; the noise sensor is used to collect the sound pressure level signal in the bedroom environment; the integrated temperature and humidity detection module is used to obtain the numerical parameters of air temperature and relative humidity; the environment monitoring unit receives the body movement signal or heart rate variation signal output by the behavior collection unit, and when it is detected that the signal amplitude exceeds the set threshold, automatically performs an environment parameter sampling; the light intensity signal, noise signal and temperature and humidity data obtained by sampling are transmitted to the fusion analysis unit after being attached with sampling time information, so as to be analyzed with the behavior data in the same time period.

[0024] Further, the lifestyle management unit includes a data entry module, a diet information analysis module and a sports information analysis module.

[0025] The data entry module is used to receive the daily diet records and sports records of children input by the parent terminal;

[0026] The diet information analysis module is used to extract food types, intake time and energy content information according to the diet records, and generate a diet data set;

[0027] The sports information analysis module is used to extract sports items, duration and intensity parameters according to the sports records, and generate a sports data set;

[0028] The lifestyle management unit calculates the ratio of diet energy intake value and sports energy consumption value according to the diet data set and the sports data set, and the ratio is used as a diet and sports balance index. After the calculation is completed, the balance index and the corresponding date information are stored and sent to the fusion analysis unit for joint analysis with the behavior data and environment data on the same day.

[0029] Further, the fusion analysis unit includes a data preprocessing module, a feature extraction module and a risk calculation module.

[0030] The data preprocessing module is used to perform time alignment and normalization processing on the multi-source data from the behavior collection unit, the environment monitoring unit and the lifestyle management unit, and generate a data input set on the same time sequence;

[0031] The feature extraction module adopts a time convolution calculation structure, and performs convolution operation on the data input set to obtain a feature sequence reflecting the time variation characteristics of various signals;

[0032] The feature extraction module further includes a weighting operation unit for performing weighting processing on the feature sequence according to the correlation coefficient of each signal and the sleep state or a preset weight value;

[0033] The risk calculation module receives the weighted feature sequence, performs weighted summation on different signal features according to a preset mathematical relationship, obtains a behavior risk index, and outputs the behavior risk index to an intervention execution unit for generating an intervention judgment basis.

[0034] Further, the intervention execution unit comprises an intervention scheme generation module, a control instruction module, and an execution device interface.

[0035] The intervention scheme generation module is configured to receive the behavior risk index output by the fusion analysis unit, and determine a corresponding intervention type according to a preset risk grading rule, wherein the risk grading rule comprises at least three risk intervals.

[0036] The control instruction module is configured to generate a control instruction according to the intervention type, wherein the control instruction comprises a visual guidance control instruction, an environment adjustment control instruction, and a lifestyle habit prompting instruction.

[0037] The visual guidance control instruction is configured to call a preset interactive content output by a display or voice playing device to guide the child to complete an attention diversion activity.

[0038] The environment adjustment control instruction is configured to control a light, audio, and air conditioning device connected to the system to adjust parameters such as brightness, color temperature, audio frequency, and ventilation state, and the lifestyle habit prompting instruction is configured to send diet or exercise adjustment information to a parent terminal.

[0039] The intervention execution unit records execution time and instruction type while sending the control instruction, and feeds back the record information to the adaptive optimization unit for subsequent correlation analysis of intervention effect.

[0040] Further, the adaptive optimization unit comprises a feedback collection module, an effect evaluation module, and a parameter updating module.

[0041] The feedback collection module is configured to receive the execution record information output by the intervention execution unit, and obtain sleep monitoring data and emotional score data after intervention, wherein the emotional score data is input by a parent terminal or collected by an expression recognition device.

[0042] The effect evaluation module is configured to compare sleep duration, night waking frequency, and emotional score changes before and after intervention.

[0043] The parameter updating module is configured to adjust feature weight parameters in the fusion analysis unit according to the intervention effect value.

[0044] When the intervention effect value is lower than a preset threshold, the adjustment coefficient of the corresponding feature weight is increased; when the intervention effect value is higher than the preset threshold, the adjustment coefficient is decreased; and the feature weight parameters in the fusion analysis unit are periodically updated in this way to correct subsequent intervention strategies.

[0045] Further, the home interaction unit comprises an information display module and a feedback collection module.

[0046] The information display module is used to display the behavior monitoring data output by the fusion analysis unit, the execution record information of the intervention execution unit and the sleep change result fed back by the adaptive optimization unit on the parent terminal interface, so as to form a corresponding relationship curve of behavior, intervention and sleep.

[0047] The feedback collection module is used to receive the child daily behavior notes, bedtime environment description and subjective emotion evaluation information input by the parents, and store the feedback information in the form of structured data in the system database.

[0048] The feedback information in the database is used to supplement the input data of the fusion analysis unit to improve the data sample in the subsequent parameter updating process.

[0049] The beneficial effects of the present application are as follows:

[0050] 1. The present application can simultaneously obtain the body movement, heart rate, skin electricity and sleep posture signals of preschool children with autism by setting the behavior collection unit and the environment monitoring unit, and synchronously sample the environmental parameters such as bedroom illumination, noise, temperature and humidity, the multi-source fusion structure realizes the time alignment of sleep behavior data and environmental state, avoids the defect that a single sensor is easily disturbed by the outside world, significantly improves the continuity and reliability of the data, and through the key collection before falling asleep and the night waking stage, the physiological response and environmental adaptation of children in different sleep stages can be truly reflected, and stable basic data is provided for subsequent correlation analysis.

[0051] 2. The present application establishes a unified analysis framework among behavior, environment and daily routine through the life habit management unit and the fusion analysis unit in the system, quantifies the diet and exercise records into balance indexes, and comprehensively analyzes the behavior monitoring data, so as to identify the key factors affecting falling asleep and night waking, the structure makes the sleep state no longer depend on physiological signal judgment, but models from multiple angles of life rules, environmental conditions and behavior characteristics, realizes the identification and risk grading of individual differences of children with autism, and provides accurate basis for subsequent intervention.

[0052] 3. This invention constructs an intervention closed-loop structure through an intervention execution unit, an adaptive optimization unit, and a family interaction unit. The system generates personalized intervention plans based on the behavioral risk index and achieves real-time response through lighting adjustment, sound guidance, and parent feedback. The adaptive optimization unit automatically adjusts the analysis parameters according to the intervention effect, so that the intervention strategy is dynamically matched with the child's response. The family interaction unit incorporates parent feedback into the sample update to achieve human-machine collaborative optimization. This closed-loop design ensures the continuous effectiveness of intervention measures, improves the shortcomings of traditional systems such as delayed response and lack of feedback correction, and significantly enhances the stability and scalability of individualized intervention. Attached Figure Description

[0053] Fig. 1 This is a flowchart of the multi-source data acquisition and fusion analysis process of this invention;

[0054] Fig. 2 This is a flowchart illustrating the personalized intervention execution and feedback process of this invention;

[0055] Fig. 3 This is a flowchart illustrating the adaptive optimization process of the intervention strategy of this invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] like Figs. 1 to 3 As shown, this embodiment of the invention provides a personalized sleep intervention system for preschool children with autism based on AI behavioral analysis. The system includes:

[0058] The behavior data collection unit is used to acquire information on body movement, heart rate, skin conductance signals and sleeping posture of preschool children with autism, and to collect behavioral data before falling asleep and during nighttime awakenings.

[0059] The environmental monitoring unit is used to collect environmental parameters such as bedroom light, noise, temperature and humidity, and to perform synchronous sampling when behavioral changes are detected.

[0060] The lifestyle habits management unit is used to record children's diet and exercise information and calculate the diet and exercise balance index;

[0061] The fusion analysis unit is used to perform fusion analysis on behavioral data, environmental parameters, and diet and exercise balance index based on temporal convolutional neural network and attention weighting algorithm, identify key features that affect falling asleep and nighttime awakening, and generate a behavioral risk index.

[0062] An intervention execution unit is configured to generate a personalized intervention plan according to the behavior risk index and control visual guidance, environmental adjustment and lifestyle adjustment intervention measures;

[0063] An adaptive optimization unit is configured to evaluate the intervention effect and update the model weight in the fusion analysis unit according to the sleep change and emotional feedback information after the intervention is performed, so as to realize dynamic optimization of the intervention strategy.

[0064] A family interaction unit is configured to display the behavior, intervention and sleep association information on the parent terminal and receive the parent feedback to supplement the model learning data.

[0065] The behavior collection unit is configured to acquire the body movement, heart rate, skin conductance and sleep posture information of the preschool autistic children during the sleep process and perform time labeling before falling asleep and during the night waking stage, so as to realize the synchronous collection and management of the multi-source behavior data.

[0066] The behavior collection unit includes a low-sensitivity wearable device and a time labeling module. The wearable device is worn on the surface of the child's body (such as the wrist or the chest) and adopts a medical silicone shell to reduce the tactile stimulation and is suitable for long-term wearing by the autistic children.

[0067] The wearable device internally integrates three types of sensing components:

[0068] The body movement sensing component is composed of a three-axis acceleration sensor and a gyroscope and is configured to detect the body displacement, turning over and limb activity change of the children. The sampling frequency is adjustable between 20 Hz and 100 Hz. The system sets a threshold value according to the amplitude change of the body movement signal to distinguish between micro body movement and large displacement.

[0069] The heart rate sensing component adopts a photoelectric plethysmogram sensor (PPG module) to measure the pulse wave curve through the infrared and green light dual-wavelength light source emission and reflection signals and remove the ambient light noise by a filter circuit. The instantaneous heart rate HR is calculated by the formula:

[0070]

[0071] In the formula, T p is the time interval (seconds) between adjacent pulse peaks.

[0072] The skin conductance sensing component is composed of a double-electrode sensing sheet and a constant-voltage detection circuit and is configured to measure the skin conductance change. The calculation formula of the conductance G s is as follows:

[0073]

[0074] In the formula, I s is the detected skin current (ampere) and U0 is the detection voltage (volt).

[0075] In addition, the wearable device further comprises a posture recognition module, which recognizes four postures of supine, prone, left lateral recumbency and right lateral recumbency by analyzing three-axis acceleration data using a threshold interval matching method, and outputs sleep posture change data, which is stored synchronously with the body movement signal;

[0076] The time labeling module is used to automatically start timing within a preset period before the child falls asleep, and insert a time label in the corresponding data frame when a night waking event is detected, and the module is built-in with a real-time clock (RTC), and the timing accuracy is preferably 0.1 seconds, and has a power failure retention function;

[0077] The time-labeled body movement, heart rate, skin conductance and sleep posture data are transmitted to the fusion analysis unit through a wireless communication module (BLE or Wi-Fi), and the data transmission adopts a CRC redundancy check and packet number verification mechanism to ensure data integrity;

[0078] Through multi-dimensional signal synchronous acquisition and time labeling mechanism, high-precision recording of sleep behavior of autistic children is realized, which provides a reliable basis for subsequent fusion analysis; compared with the prior art, the unit has significant innovation and practicality in data acquisition integrity, low sensitivity wearing and time labeling mechanism.

[0079] The environment monitoring unit is used to monitor the light, noise and temperature and humidity parameters in the sleep environment of preschool autistic children, and automatically trigger the sampling of environmental parameters when the body movement or heart rate signal output by the behavior acquisition unit exceeds the threshold value, to realize time-synchronous analysis of environment and behavior data;

[0080] The environment monitoring unit includes a light sensor, a noise sensor and a temperature and humidity integrated detection module, and each sensor is connected to the micro-processing module through a signal acquisition bus, and the acquisition results are transmitted to the fusion analysis unit after time labeling;

[0081] The light sensor adopts a high-sensitivity photodiode array structure, with a response waveband of 380nm-780nm and a sampling frequency of 1Hz-10Hz, and the calculation formula of the illumination value L is:

[0082] L=K L ×V L

[0083] In the formula, L is the illumination value (lux), V L is the output voltage of the photodiode (V), and K L is the calibration coefficient (lux / V);

[0084] When the detected illumination is lower than 5Iux, the automatic gain adjustment module will amplify the detection signal to maintain the measurement resolution;

[0085] Noise sensor: adopts electret microphone array structure, sampling frequency is 8kHz-16kHz, its output sound pressure level signal SPL calculation formula is:

[0086]

[0087] In the formula, SPL is sound pressure level (dB), P is instantaneous sound pressure value (Pa), P0 is reference sound pressure value (20μPa); the sensor can detect indoor burst sound source and output digital signal;

[0088] Temperature and humidity integrated detection module: including digital temperature sensor and capacitive humidity detection unit, temperature measurement range is 0℃-50℃, humidity measurement range is 0%RH-100%RH, module shell is provided with anti-condensation breathable film to prevent water vapor interference;

[0089] Trigger sampling logic: the environmental monitoring unit continuously receives the body motion signal and heart rate signal of the behavior collection unit, when detecting that the body motion amplitude exceeds 0.5g or the heart rate change exceeds 10bpm, automatically executing once environmental sampling, collecting current light, noise and temperature and humidity data, each sampling is generated time label (accuracy is not less than 0.1 second) by time labeling module, and is stored synchronously with sampling data;

[0090] Data transmission and synchronization: after CRC check, the sampling result is sent to the fusion analysis unit through the wireless communication module, realizing the time sequence correspondence of multi-modal data, the system main clock uniformly controls all modules, and ensures that the time error of behavior and environment data is not more than 0.1 second;

[0091] Through the behavior-driven environmental trigger sampling mechanism, the precise capture of the environmental state of the autism children night wake-up process is realized, which is different from the fixed period sampling mode in the prior art, and can identify the corresponding relationship between light, noise, temperature and humidity change and night wake-up behavior, the design enhances the analysis ability of the system to complex sleep environment, and has remarkable creativity and practicality.

[0092] Among them, the life habit management unit is used for collecting, sorting and quantifying the diet and exercise information of the preschool autistic children, forming a daily "diet and exercise balance index", which is used for joint analysis with behavior data and environment data in the fusion analysis unit;

[0093] The unit includes data entry module, diet information analysis module and exercise information analysis module, each module is connected through system bus and controlled by unified processor;

[0094] Data entry module

[0095] The data entry module is configured to receive daily diet records and exercise records of the child input by the parent terminal, wherein the diet records include meal times, food names, intake times and portions, and the exercise records include exercise items, durations and intensity levels.

[0096] The system is preset with a food database and an exercise database, and the parent can input data in a standardized manner by matching options. After the input is completed, the data is stored in a structured format, which facilitates subsequent analysis and calling.

[0097] The diet information analysis module

[0098] The diet information analysis module automatically matches the energy parameters corresponding to the food in the food database according to the diet records input by the parent. The energy of each food is obtained by multiplying the food portion by the unit energy value, and the system automatically accumulates to generate the total energy intake value of the daily diet, which is attached with date labels and meal information to form a "diet data set" for corresponding with the exercise data.

[0099] The exercise information analysis module

[0100] The exercise information analysis module calculates the exercise energy consumption of the child according to the exercise items, the weight of the child, the exercise time and the intensity level, in combination with the metabolic equivalent (MET) values in the database. The system sets a metabolic correction coefficient for children of different ages to take into account the difference in basal metabolic rate. After the daily exercise data is sorted, a "exercise data set" is formed, which is attached with a date index.

[0101] Calculation of diet and exercise balance index

[0102] The life habit management unit calculates the diet and exercise balance index according to the daily diet energy intake value and the exercise energy consumption value. The index is the ratio of the two, which is used to represent the balance degree of the daily energy intake and consumption of the child. When the ratio is close to 1, it means that the energy intake and consumption are basically balanced. When the ratio is higher than 1.2, it means that the intake is too much. When the ratio is lower than 0.8, it means that the consumption is too much. The index is stored together with the date information and sent to the fusion analysis unit for joint analysis with the behavior data and environment data of the same day.

[0103] The behavior collection unit provides body movement and heart rate data, which can indirectly verify the authenticity of the exercise records. The environment monitoring unit provides temperature, humidity and light data, which are used to correct the exercise energy consumption evaluation. The fusion analysis unit receives the balance index and aligns it with the behavior and environment data in time for analysis, which is used to identify the potential impact of life habits on sleep quality.

[0104] The synergistic relationship forms a data closed loop from "life habits → behavior state → sleep performance", which makes the analysis results traceable.

[0105] The fusion analysis unit is used for synchronous fusion and feature recognition of multi-source data from the behavior acquisition unit, the environment monitoring unit and the habit management unit, so as to determine the key behavior mode causing sleep delay or frequent night wake-up, and generate a behavior risk index as an input basis for the subsequent intervention execution unit;

[0106] The unit is composed of a data preprocessing module, a feature extraction module and a risk calculation module, and the modules sequentially transmit data and jointly complete the whole process from data alignment to risk quantification;

[0107] The data preprocessing module

[0108] The main function of the data preprocessing module is to perform time alignment and numerical standardization on asynchronous data from different sources, to ensure that the input data is comparable in time and scale;

[0109] Time alignment processing:

[0110] The body movement signal, heart rate change and sleep posture state output by the behavior acquisition unit are marked with millisecond-level time stamps; the light, noise and temperature and humidity data output by the environment monitoring unit are sampled at a second level; the diet and exercise balance index output by the habit management unit is updated in units of days; the data preprocessing module maps the multi-source signals to a fixed step time sequence through interpolation and time window slicing algorithm, for example, with 1 minute as the alignment period, to generate the average value or weighted average value in the corresponding time slice, so as to obtain a multi-source data matrix that is time continuous and dimension unified;

[0111] Normalization processing:

[0112] Different signal types have different dimensions and ranges of change (such as heart rate in bpm, temperature in ℃, and noise in dB), and if directly input into the model, it will cause bias, the module adopts linear interval scaling or Z-score standardization method to map all input features to [0, 1] or standard normal interval, to eliminate the instability of the model caused by the difference of physical quantities, and the processed result forms a "unified time series data input set" and is transmitted to the feature extraction module;

[0113] Feature extraction module

[0114] The feature extraction module is used for extracting feature patterns with time sequence correlation from the normalized time series, to reflect the dynamic change characteristics of different signals over time;

[0115] Time convolution calculation structure: The module uses a sliding window method to perform local convolution operation on the input sequence to extract characteristic parameters such as time change rate, peak fluctuation frequency and signal synchronization. Unlike traditional recurrent neural networks, this structure does not rely on deep recursion, but captures short-term dependencies through convolution kernel sliding, has the characteristics of low delay and high calculation efficiency, and is more suitable for real-time monitoring scenarios;

[0116] Weighted operation unit: Since different signals have different effects on sleep state, the module sets a weighted operation unit to realize feature importance adjustment. The weight value can be imported through a pre-set correlation coefficient table or offline training results. For example, heart rate variation has a higher weight in the 30 minutes before falling asleep, and noise weight increases during night waking. The module performs weighted aggregation on the feature sequence generated by convolution to form a unified dimension weighted feature vector. The weighted feature sequence output by this module is used as the input of the risk calculation module;

[0117] Risk calculation module: The risk calculation module is used to integrate the multi-source information reflected in the weighted feature sequence to generate a behavior risk index for evaluating the current sleep risk state of the individual. The module uses a linear weighted superposition mathematical relationship to combine and calculate each signal feature. The behavior risk index is defined as the weighted sum of each type of weighted feature. Its output is a normalized numerical value, which is used to quantify the sleep risk level of the individual on the same day or night. This result is transmitted in real time to the intervention execution unit as the basis for triggering personalized intervention strategies;

[0118] For example, when the risk index is higher than a pre-set threshold (such as 0.7), the system automatically sends a trigger signal to the intervention execution unit to start visual guidance or environmental adjustment measures;

[0119] Data flow and collaborative logic between modules

[0120] Input end collaboration: The data from the behavior collection, environment monitoring and lifestyle management units are all attached with a unified timestamp before entering this unit. The data preprocessing module can directly align and format the data;

[0121] Intermediate layer collaboration: The feature extraction module will feed back the feature fluctuation range to the preprocessing module in real time during the calculation process for adaptive adjustment of the normalization parameter;

[0122] Output end collaboration: The risk index output by the risk calculation module is read in real time by the intervention execution unit, combined with the current state of the child and historical intervention records, and determines whether to perform a specific intervention action;

[0123] This feedback coupling between modules forms a closed-loop response mechanism, ensuring data integrity and improving the real-time response of the system.

[0124] The intervention execution unit is configured to generate a personalized intervention scheme and execute a control action according to the risk index output by the fusion analysis unit, and the unit comprises an intervention scheme generation module, a control instruction module and an execution device interface.

[0125] The intervention scheme generation module receives the risk index output by the fusion analysis unit, determines the intervention type according to the risk grading rule, and the risk rule is divided into three grades: when the index is less than 0.3, it is low risk, and a life habit prompt is executed; when the index is between 0.3 and 0.7, it is medium risk, and an environment adjustment is executed; and when the index is greater than 0.7, it is high risk, and a visual guide is executed. The module generates an intervention scheme according to the risk level, which includes the intervention category, the target device and the execution parameter.

[0126] The control instruction module generates specific control instructions according to the intervention scheme. The visual guide control instruction calls the display or voice device to play soothing content; the environment adjustment instruction controls the light brightness (10-100 lux), color temperature (2700-4000K), audio frequency (100-1000Hz) or ventilation gear; and the life habit prompt instruction pushes diet and exercise adjustment information to the parent terminal.

[0127] The execution device interface is configured to convert the control instruction into a device recognizable protocol, supports Wi-Fi, Bluetooth and ZigBee communication, records the execution time, device feedback state and instruction number when the instruction is issued, and uploads the execution result to the adaptive optimization unit for subsequent intervention effect analysis.

[0128] The unit realizes the whole process from risk assessment, intervention generation, control execution to feedback transmission. Compared with the existing sleep assistance system which is limited to single device control, the present application improves the precision and response speed of the intervention through multi-dimensional linkage intervention and real-time feedback mechanism, and is suitable for individualized sleep intervention scene of autistic children.

[0129] The adaptive optimization unit is configured to receive the execution record information and the monitoring data after the intervention fed back by the intervention execution unit, evaluate the intervention effect, and dynamically correct the feature weight parameters in the fusion analysis unit, so as to realize the periodic optimization of the intervention strategy.

[0130] The unit comprises a feedback collection module, an effect evaluation module and a parameter updating module, which are connected in sequence to form an optimization cycle of execution result acquisition-effect evaluation-parameter correction.

[0131] The feedback collection module

[0132] Input content and source

[0133] The module receives execution record information from the intervention execution unit, including intervention type, execution time, duration, target device, and feedback status (success / failure); at the same time, the module obtains sleep monitoring data after intervention from the system monitoring channel, including sleep duration (minutes), number of night awakenings (times / night), and deep sleep duration proportion (%); in addition, the module also synchronously obtains emotional score data, which is obtained by two ways: subjective emotional rating input by the parent terminal (0-5 points); and objective emotional characteristics collected by the expression recognition device, which generates a score value after system standardization;

[0134] Data integration and time matching

[0135] The module indexes the execution record, sleep monitoring data, and emotional score data by intervention execution time for one-to-one correspondence;

[0136] If there is time overlap or missing data, the module automatically adopts the nearest time matching principle for pairing to ensure that the data within the same intervention period can be compared;

[0137] Finally, a structured "intervention feedback dataset" is generated and transmitted to the effect evaluation module;

[0138] Effect evaluation module

[0139] Intervention effect calculation logic

[0140] The module compares the change trend of the same type of data before and after the intervention, quantifies the intervention effect value, and calculates the comprehensive result according to the differences in sleep duration, number of night awakenings, and emotional score;

[0141] For example:

[0142] If the average sleep duration after intervention is prolonged and the number of night awakenings is reduced, the intervention is determined to be effective; if the emotional score is improved, it is considered that the psychological relaxation effect is significant;

[0143] Effect value normalization processing

[0144] The module maps the change value of each index to the 0-1 interval after linear normalization, forming a unified intervention effect value E f ;

[0145] Where:

[0146] E f Close to 1 indicates ideal effect;

[0147] E f Close to 0 indicates poor effect;

[0148] The module transmits this value and the corresponding intervention type to the parameter update module;

[0149] Parameter update module

[0150] Update logic and decision condition

[0151] The module is used to update the feature weight parameter in the fusion analysis unit according to the intervention effect value E f When E is lower than a preset threshold (for example, 0.4), the system considers that the current intervention strategy is insufficient to improve the sleep state; when E f When E is higher than the threshold (for example, 0.8), it indicates that the current weight distribution has excessively emphasized a certain feature, and the weight needs to be appropriately weakened. f

[0152] Parameter adjustment mechanism

[0153] The module sets an adjustment coefficient ki for each feature dimension;

[0154] If the intervention effect is not ideal, the weight adjustment coefficient of the corresponding feature is increased; if the effect is significant, the coefficient is decreased, and the updated feature weight parameter is written back to the fusion analysis unit to affect the subsequent risk calculation process. The parameter update period can be set to be performed once a day or once after each intervention ends;

[0155] Data security and history record management

[0156] The module records the version of each parameter update operation, including the update time, adjustment direction and change amplitude, so as to enable the system to trace back and manually review. If the intervention effect does not improve significantly after three consecutive updates, the system triggers a manual review prompt to prevent parameter drift.

[0157] Among them, the family interaction unit is used to establish an information communication channel between the parents and the system, and to realize the visual display of the monitoring results and the effective collection of the parents' subjective information;

[0158] The unit includes an information display module and a feedback collection module, which cooperatively form a circulation path of data output-manual feedback-sample supplement, thereby enhancing the individualized data support capability of the system;

[0159] Information display module: used to display the behavior monitoring data output by the fusion analysis unit, the execution record information of the intervention execution unit, and the sleep change result fed back by the adaptive optimization unit on the parent terminal interface;

[0160] The data display is a multi-dimensional curve graph with time as the horizontal coordinate and body movement times, night wake-up times and sleep quality scores as the vertical coordinates, and the occurrence time points of intervention events are marked on the curve to reflect the corresponding change trend of behavior, intervention and sleep result. This module supports daily, weekly and monthly period switching display, and allows the parents to export statistical reports;

[0161] ​The feedback collection module is used for receiving the child daily behavior notes, bedtime environment description and subjective emotion score information input by the parents. The input content is converted into a structured data table after field processing. The fields include time, category, parameter value and note information.

[0162] The module is provided with an input verification mechanism to prompt abnormal or missing data. All feedback data is stored in the system database with a unified time index and participates in the next round of data processing as a supplementary sample in the fusion analysis unit.

[0163] The family interaction unit establishes a visual information display and artificial feedback collection channel, so that the system has an individualized data updating mechanism with human-computer cooperation based on automatic collection. This module ensures the dynamic correspondence between intervention effect, behavior data and subjective feedback, and provides data support for the continuous optimization of sleep intervention for children with autism.

[0164] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0165] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A personalized sleep intervention system for preschool children with autism based on AI behavioral analysis, characterized in that: The system includes: The behavior data collection unit is used to acquire information on body movement, heart rate, skin conductance signals and sleeping posture of preschool children with autism, and to collect behavioral data before falling asleep and during nighttime awakenings. The environmental monitoring unit is used to collect environmental parameters such as bedroom light, noise, temperature and humidity, and to perform synchronous sampling when behavioral changes are detected. The lifestyle habits management unit is used to record children's diet and exercise information and calculate the diet and exercise balance index; The fusion analysis unit is used to perform fusion analysis on behavioral data, environmental parameters, and diet and exercise balance index based on temporal convolutional neural network and attention weighting algorithm, identify key features that affect falling asleep and nighttime awakening, and generate a behavioral risk index. The intervention execution unit is used to generate personalized intervention plans based on the behavioral risk index and control intervention measures such as visual guidance, environmental adjustment, and lifestyle habit adjustment. The adaptive optimization unit is used to evaluate the intervention effect and update the model weights in the fusion analysis unit based on sleep changes and emotional feedback information after the intervention is implemented, so as to achieve dynamic optimization of the intervention strategy. The family interaction unit is used to display information related to behavior, intervention, and sleep on the parent terminal, and to receive parent feedback to supplement the model's learning data.

2. The personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 1, characterized in that: The behavior acquisition unit includes: A low-sensitivity wearable device worn on the surface of the body of a preschool child with autism, which incorporates three types of sensing components: The motion sensing component is used to detect changes in a child's body displacement, rolling over, and limb movements, and to generate motion amplitude signals. Heart rate sensing components are used to acquire the child's pulse wave signal and calculate instantaneous heart rate changes; Skin conductivity sensing components are used to measure changes in electrical conductivity on the surface of a child's skin. The wearable device is also equipped with a posture recognition module, which is used to determine the sleeping posture based on body movement signals and store the sleeping posture change data and body movement signals synchronously. The behavior acquisition unit is equipped with a time labeling module, which is used to automatically label time period information when a preset period of time is set before falling asleep and when behavioral changes are detected during nighttime awakenings. The body movement signal, heart rate signal, skin conductance signal and sleep posture change data are transmitted to the fusion analysis unit for feature analysis.

3. The personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 2, characterized in that: The environmental monitoring unit includes: a light sensor, a noise sensor, and an integrated temperature and humidity detection module; The light sensor is used to detect changes in illuminance in the bedroom space and output a light intensity signal; the noise sensor is used to collect sound pressure level signals in the bedroom environment; the integrated temperature and humidity detection module is used to acquire numerical parameters of air temperature and relative humidity; the environmental monitoring unit receives body movement signals or heart rate change signals output by the behavior acquisition unit, and automatically performs an environmental parameter sampling when the amplitude of the signal exceeds a set threshold; the sampled light intensity signal, noise signal, and temperature and humidity data, along with sampling time information, are transmitted to the fusion analysis unit.

4. The personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 3, characterized in that: The lifestyle management unit includes: a data entry module, a diet information analysis module, and an exercise information analysis module; The data entry module is used to receive children's daily diet and exercise records input by parents on the terminal; The dietary information analysis module is used to extract information on food types, intake time, and energy content from the dietary records, and generate a dietary dataset. The sports information analysis module is used to extract sports events, duration and intensity parameters from the sports records and generate sports datasets; The lifestyle management unit calculates the ratio of dietary energy intake to exercise energy expenditure based on the diet dataset and exercise dataset. This ratio serves as the diet-exercise balance index. Once the calculation is complete, the balance index and the corresponding date information are stored and sent to the fusion analysis unit for joint analysis with behavioral and environmental data from the same day.

5. A personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 4, characterized in that: The fusion analysis unit includes: a data preprocessing module, a feature extraction module, and a risk calculation module; Among them, the data preprocessing module is used to perform time alignment and normalization processing on multi-source data from the behavior collection unit, environmental monitoring unit and lifestyle management unit to generate a data input set on the same time series; The feature extraction module employs a temporal convolution computation structure to perform convolution operations on the data input set in order to obtain feature sequences that reflect the temporal variation characteristics of various signals. The feature extraction module further includes a weighted operation unit, which is used to perform weighted processing on the feature sequence based on the correlation coefficient or preset weight value between each signal and the sleep state. The risk calculation module receives the weighted feature sequence, performs weighted summation on different signal features according to a preset mathematical relationship to obtain a behavioral risk index, and outputs the behavioral risk index to the intervention execution unit to generate intervention judgment criteria.

6. The personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 5, characterized in that: The intervention execution unit includes: an intervention plan generation module, a control command module, and an execution device interface; The intervention plan generation module is used to receive the behavioral risk index output by the fusion analysis unit and determine the corresponding intervention type according to the preset risk grading rules, which include at least three risk ranges. The control instruction module is used to generate control instructions based on the intervention type, including visual guidance control instructions, environmental adjustment control instructions, and lifestyle habit prompting instructions. Visual guidance control commands are used to invoke display or voice playback devices to output preset interactive content in order to guide children to complete attention-shifting activities; The environmental control command is used to control the lighting, audio and air conditioning devices connected to the system, and to adjust parameters such as lighting brightness, color temperature, audio frequency and ventilation status. The lifestyle habit prompt command is used to send dietary or exercise adjustment information to the parent terminal. The intervention execution unit records the execution time and instruction type while sending control instructions, and feeds back the recorded information to the adaptive optimization unit for subsequent correlation analysis of intervention effects.

7. A personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 6, characterized in that: The adaptive optimization unit includes: a feedback acquisition module, an effect evaluation module, and a parameter update module; The feedback acquisition module is used to receive the execution record information output by the intervention execution unit, and to acquire the sleep monitoring data and emotion score data after the intervention execution. The emotion score data is input by the parent terminal or collected by the expression recognition device. The effectiveness assessment module is used to compare changes in sleep duration, number of nighttime awakenings, and mood scores before and after the intervention; The parameter update module is used to adjust the feature weight parameters in the fusion analysis unit according to the intervention effect value; When the intervention effect value is lower than a preset threshold, the adjustment coefficient of the corresponding feature weight is increased; when the intervention effect value is higher than the preset threshold, the adjustment coefficient is decreased; in this way, the feature weight parameters in the fusion analysis unit are updated periodically to correct subsequent intervention strategies.

8. A personalized sleep intervention system for preschool autism based on AI behavior analysis according to claim 7, characterized in that: The home interaction unit includes: an information display module and a feedback collection module; The information display module is used to display the behavior monitoring data output by the fusion analysis unit, the execution record information of the intervention execution unit, and the sleep change results fed back by the adaptive optimization unit on the parent terminal interface, so as to form a curve showing the correspondence between behavior, intervention and sleep. The feedback collection module is used to receive notes on children's daily behavior, descriptions of the bedtime environment, and subjective emotional evaluation information input by parents, and stores the feedback information in the system database in the form of structured data. The feedback information in the database is used to supplement the input data of the fusion analysis unit.

Images