Intelligent water purification system integrating safety monitoring and healthy drinking water management

Abstract

The invention relates to the technical field of smart home and health management, and discloses an intelligent water purification system integrating safety monitoring and healthy drinking water management, and the system comprises a data collection and preprocessing module which is used for collecting and preprocessing original data of a built-in sensor and external safety equipment; the water consumption behavior modeling module is used for identifying a water consumption event and analyzing a time distribution rule; the abnormal behavior recognition module is used for recognizing abnormal behaviors and fusing the abnormal behaviors with built-in sensor data for confirmation to obtain an abnormal behavior recognition result and a severity evaluation report; the healthy water drinking management module is used for calculating daily recommended water drinking amount, making a water drinking plan and carrying out achievement degree evaluation and trend analysis; the grading early warning execution module formulates a grading early warning strategy and executes the grading early warning strategy; the visual analysis module is used for carrying out visual presentation and trend analysis; the self-adaptive optimization module is used for carrying out system self-adaptive optimization and upgrading; according to the invention, non-perception safety monitoring and precise health management of the elderly living alone are realized.

Description

Technical Field

[0001] This invention relates to the field of smart home and health management technology, and more specifically, to a smart water purification system that integrates safety monitoring and healthy drinking water management. Background Technology

[0002] With the accelerating aging of my country's population, the number of elderly people living alone is constantly expanding, making their home safety and health management increasingly prominent issues. Due to the lack of family companionship, elderly people living alone face various safety risks in daily life, such as falls, sudden illnesses, and gas leaks, and often struggle to receive timely assistance. At the same time, many elderly people lack awareness of the importance of adequate hydration and do not drink enough water, while sufficient water intake is crucial for controlling chronic diseases such as hypertension and diabetes. How to effectively monitor the home safety and scientifically manage the health of the elderly without increasing their psychological burden has become an urgent social problem to be solved.

[0003] Existing safety monitoring solutions for the elderly mainly include wearable devices, video surveillance, and emergency call systems. Wearable devices require the elderly to actively wear and charge them, resulting in poor user compliance and easy forgetting or loss. While video surveillance can monitor the elderly's activities in real time, it raises privacy concerns, and the elderly generally resist installing cameras in their homes. Emergency call systems rely on the elderly actively pressing a button for help when in danger, but they are difficult to use if the elderly are unconscious or have limited mobility. All these solutions suffer from drawbacks such as high barriers to use, limited monitoring effectiveness, or insufficient privacy protection.

[0004] In terms of healthy drinking water management, existing smart water cups and drinking reminder apps mainly urge users to drink water through timed reminders. However, they lack accurate monitoring of users' actual drinking behavior and precise calculation of personalized needs. The reminder strategies are simple and mechanical, making it difficult to adapt to different users' lifestyles and health conditions. Furthermore, most existing solutions treat safety monitoring and health management as independent functions, failing to fully utilize the rich behavioral information contained in household water usage data, and failing to achieve an organic integration of safety monitoring and health management. Therefore, there is an urgent need for an intelligent system that can seamlessly collect user behavior data, accurately identify abnormal situations, and provide personalized health management services. Summary of the Invention

[0005] This invention provides an intelligent water purification system that integrates safety monitoring and healthy drinking water management, solving the technical problems of related technologies that lack accurate monitoring of users' actual drinking behavior and precise calculation of personalized needs, and whose reminder strategies are simple and mechanical.

[0006] This invention provides an intelligent water purification system integrating safety monitoring and healthy drinking water management, comprising: The data acquisition and preprocessing module is used to acquire raw data from built-in sensors and external safety devices and preprocess it to obtain a standard multi-source heterogeneous dataset. The water use behavior modeling module identifies water use events and analyzes their temporal distribution patterns based on standard multi-source heterogeneous datasets, and builds a personalized water use behavior baseline model. The abnormal behavior identification module, based on a personalized water use behavior baseline model, identifies abnormal behaviors and integrates standard multi-source heterogeneous datasets for anomaly confirmation, and obtains abnormal behavior identification results and severity assessment reports. The healthy drinking water management module, based on a personalized water behavior baseline model, calculates the recommended daily water intake and formulates a drinking water plan, conducts achievement assessment and trend analysis, and obtains a personalized drinking water solution. The tiered early warning execution module formulates and executes tiered early warning strategies based on the abnormal behavior identification results and severity assessment reports, and obtains the early warning execution results. The visualization analysis module, based on the results of abnormal behavior identification, personalized drinking water plans, and early warning execution results, provides visual presentation and trend analysis to generate a whole-house safety trend report; The adaptive optimization module performs adaptive optimization and upgrades of the system based on standard multi-source heterogeneous datasets and early warning execution results, resulting in an adaptive optimization scheme for the system.

[0007] In a preferred embodiment, the data acquisition and preprocessing module includes: Establish a unified data acquisition interface to encapsulate data from built-in sensors and external security devices into a unified data frame structure; The main controller is time-calibrated using the Network Time Protocol, and multi-source data is arranged by timestamp to form a time-series data stream; Establish multi-dimensional data quality assessment indicators, including completeness indicators, continuity indicators, reasonableness indicators, and consistency indicators, and use a weighted summation method to calculate the comprehensive quality score as the credibility weight; The health status of each sensor is evaluated using a sliding window statistical method, and dynamic weights are assigned to the sensors according to the health status grading standard.

[0008] In a preferred embodiment, the water use behavior modeling module includes: A threshold segmentation algorithm is used to identify independent water use events. When the water flow rate rises from below the water flow rate threshold to above the water flow rate threshold, it is determined that the water use event has started. When the water flow rate drops from above the water flow rate threshold to below the water flow rate threshold and continues for a preset time, it is determined that the water use event has ended. Feature parameters of each water use event are extracted. The sliding window method is used to calculate statistical characteristics at multiple time scales, with the window sliding forward daily to update the statistical indicators. Extract the time of first water use in the morning, the frequency of drinking water during the day, the water use at night, and the interval between water uses to construct a feature vector of water use behavior.

[0009] In a preferred embodiment, the water use behavior modeling module further includes: Calculate the statistical distribution parameters for each dimension of the water use behavior feature vector. For the normal distribution feature, calculate the mean and standard deviation. For the skewed distribution feature, calculate the median, quartiles and percentiles. Establish confidence intervals for features to define the boundaries of normal behavior; The correlation coefficient is used to calculate the degree of linear correlation between feature pairs. When the absolute value of the correlation coefficient is greater than the preset correlation threshold, it is determined to be a related feature. A conditional distribution model is then established for the related features. A sliding window strategy is used to periodically recalculate the parameters of the conditional distribution model. When updating the parameters, outlier data is excluded and normal data is used to update the model.

[0010] In a preferred embodiment, the abnormal behavior identification module includes: The standardized deviation measure method is used to calculate the difference between the current water use data and the corresponding feature in the personalized water use behavior baseline model, and divide the difference by the standard deviation to obtain the standardized deviation. The degree of deviation is determined based on the standardized deviation. Define abnormal modes such as prolonged inactivity, prolonged continuous water use, abnormal water use time patterns, and insufficient drinking water. Set corresponding judgment conditions and abnormal thresholds for each abnormal mode. An anomaly is confirmed by using a multi-sensor data fusion method. Data from human body sensors, door and window sensors, temperature sensors, pressure sensors, and humidity sensors are queried to assist in the judgment. The overall confidence level of the anomaly is calculated. When the confidence level exceeds a preset threshold, the anomaly is confirmed and an early warning is triggered.

[0011] In a preferred embodiment, the abnormal behavior identification module further includes: Define the assessment dimensions for the severity of anomalies, including the degree of deviation, duration, and historical frequency. A scoring mechanism is used to comprehensively determine the severity. Each dimension is scored based on its quantitative value, and the weighted sum is used to obtain the comprehensive score. Based on the comprehensive score, anomalies are divided into mild, moderate, and severe anomalies. Mild anomalies correspond to Level 1 alerts, which are local notifications on the device. Moderate anomalies correspond to Level 2 alerts, which send push messages to the user's mobile device. Severe anomalies correspond to Level 3 alerts, which trigger the highest level alarm and notify all emergency contacts.

[0012] In a preferred embodiment, the healthy drinking water management module includes: Establish user health records, calculate basic drinking water requirements using a weight-related water intake calculation method, and introduce age correction and disease correction factors; The system dynamically adjusts based on real-time environmental factors, collects outdoor temperature and indoor humidity data, and introduces temperature correction coefficients, humidity correction coefficients, and seasonal correction coefficients. The basic requirement is then multiplied by each correction coefficient to obtain the recommended total daily water intake. Develop personalized time-segmented water drinking plans, collect historical water consumption and frequency data for each time period to calculate the time period allocation ratio, set reminder points to check users' water consumption and trigger reminders; The system calculates the achievement rate by comparing the actual total water consumption with the recommended water consumption on a daily basis and establishes a trend analysis over several consecutive days. When the achievement rate is consistently lower than a preset threshold for more than a preset number of days, a serious water shortage warning report is generated.

[0013] In a preferred embodiment, the hierarchical early warning execution module includes: Build a structured early warning strategy knowledge base, where each rule includes triggering conditions, execution actions, notification targets, and escalation mechanisms; For minor anomalies, the water purifier plays a prompt voice through its built-in voice module and sets a delayed confirmation mechanism. After the reminder, it continues to monitor water usage behavior. If the user uses water within the first preset time, the anomaly mark will be automatically removed. For moderate anomalies, a push notification is sent to the user's mobile terminal. The push message includes a quick response button and a user response timeout judgment mechanism is set. If the user does not respond within a second preset time, it will automatically be upgraded to a severe warning. For severe anomalies, a multi-channel synchronous alarm mechanism is activated to send alarm notifications to users and all emergency contacts, and to perform device linkage control according to the anomaly type.

[0014] In a preferred embodiment, the hierarchical early warning execution module further includes: Collect complete data on historical early warning events; train an optimized early warning model using supervised learning methods, divide the historical early warning data into training and testing sets, construct feature vectors, and use labels to represent the authenticity of the early warning data; A classification algorithm is used to learn the correspondence between feature combinations and real anomalies. The output of the early warning optimization model is the probability that an anomaly is real under given feature conditions. The early warning judgment threshold is adjusted according to the prediction results of the early warning optimization model. Regularly retrain the early warning optimization model using the latest accumulated early warning data, update the parameters, convert the optimized early warning model parameters into adjustments to the early warning strategy rules, and push them to user devices.

[0015] In a preferred embodiment, the adaptive optimization module includes: Establish a sensor performance monitoring index system to monitor the stability of the output signal and response time of the water flow sensor; Establish a solenoid valve performance monitoring system to record the number of actuations and cumulative operating time. The filter life prediction algorithm is used to record the total flow of water, monitor the trend of water quality changes, calculate the filtration efficiency, use a weighted scoring method to comprehensively evaluate the life status, estimate the remaining days of use based on the average daily water consumption and the current status, and implement a phased reminder strategy. The system operation is modeled as a reinforcement learning problem, defining the state space, action space and reward function, and the decision model is trained using reinforcement learning algorithms; Establish a remote diagnostic and maintenance channel. When a device fault is detected, a fault report is automatically generated and uploaded to the cloud platform. The cloud platform's intelligent diagnostic system matches the fault cause and solution based on the fault knowledge base.

[0016] The beneficial effects of this invention are as follows: By fusing multi-source heterogeneous data and personalized behavior modeling, the system achieves seamless safety monitoring and precise health management for elderly people living alone. It accurately identifies various abnormal behavior patterns, such as prolonged periods without water use and continuous abnormal water usage, and effectively suppresses false alarms through cross-validation of multi-sensor data, ensuring the accuracy and reliability of early warnings. The tiered early warning mechanism adopts differentiated response strategies based on the severity of the anomaly, avoiding excessive disruption to users' daily lives while promptly triggering multi-channel alarms and device linkage control in emergencies, providing comprehensive safety protection for elderly people living alone.

[0017] In terms of healthy drinking water management, the system dynamically calculates personalized drinking water needs based on users' physiological characteristics, disease status, and environmental factors. It also develops time-segmented reminder plans based on users' historical drinking habits, effectively improving the elderly's compliance rate with drinking water requirements and their overall health. The visualization analysis module uses charts and trend analysis to help users and their families intuitively understand water safety and health conditions. Regularly generated weekly and monthly reports provide families with effective tools for remote care. The system's adaptive optimization capabilities enable it to continuously learn user behavior characteristics, constantly improving the accuracy of alerts and service quality, demonstrating significant practical value and social significance. Attached Figure Description

[0018] Figure 1 This is a block diagram of the present invention; Figure 2 This is a detailed flowchart of the present invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses an intelligent water purification system integrating safety monitoring and healthy drinking water management, such as Figures 1 to 2 As shown, it includes: The data acquisition and preprocessing module is used to acquire raw data from built-in sensors and external safety devices and preprocess it to obtain a standard multi-source heterogeneous dataset. Includes the following steps: Step 1.1: Establish a unified data acquisition interface and format conversion; Based on the hardware architecture of the water purification system, a modular sensor interface design is adopted to obtain a standardized data acquisition channel. Specifically, the water flow sensor uses the Hall effect principle, outputting a pulse signal by detecting the impeller speed; the pulse frequency is proportional to the water flow rate. The data acquisition module counts and measures the frequency of the pulse signal, converting it into a flow rate value. Water quality sensors include a turbidity sensor, a total dissolved solids sensor, and a pH sensor, which output analog voltage or current signals respectively. A high-precision analog-to-digital converter is used to convert the analog signals into digital signals, with the conversion accuracy set to 16 bits to ensure data accuracy. Pressure sensors are installed in the inlet pipe and before and after the filter cartridge, monitoring the pipe status and the degree of filter cartridge blockage by detecting changes in water pressure. A humidity sensor is installed at the bottom of the water purifier to detect abnormal humidity around the equipment, assisting in leak detection.

[0021] For external safety device integration, the system incorporates a built-in 433MHz wireless radio frequency receiver module. Smoke and gas alarms send specifically coded wireless signals upon detecting anomalies. The receiver module parses the device identification code and alarm type from the signal, converting them into standard safety event data. The human body sensor uses passive infrared technology, sending a trigger signal when human activity is detected. The emergency help button features a low-power design, sending a high-priority help signal when pressed.

[0022] To achieve unified management of different data sources, a standardized data format conversion protocol was established. All sensor data is encapsulated into a unified data frame structure. The data frame includes fields such as frame header identifier, data source type, device number, data value, timestamp, and checksum, enabling unified encapsulation and efficient parsing of different data sources. This standardized processing provides a unified data foundation for subsequent data fusion analysis, resulting in raw data in a standardized format.

[0023] Step 1.2: Use the Network Time Protocol (NTP) to achieve time synchronization of multi-source data; Based on standardized raw data, a time synchronization mechanism is employed to obtain a data stream with a unified time reference. Specifically, the main controller of the water purification system has a built-in real-time clock module, which automatically calibrates via a network time protocol after the system is connected to the network, ensuring time accuracy at the millisecond level. Data acquisition from all built-in sensors is uniformly scheduled by the main controller, and the acquisition time directly uses the main controller's clock time, naturally guaranteeing time consistency. The timestamps of external wireless devices are recorded using the signal reception time, and transmission delay is negligible.

[0024] A data caching queue mechanism is established to arrange received multi-source data in ascending order of timestamps, forming a time-series data stream. The queue adopts a circular buffer structure, with the buffer size set to store all data from the most recent 24 hours, satisfying both real-time analysis needs and supporting historical backtracking analysis. When new data arrives, it is inserted into the corresponding time position in the queue, maintaining the queue's temporal order. For out-of-order data, sorting and insertion are automatically completed by comparing timestamps, resulting in a time-aligned data stream.

[0025] Step 1.3: Construct a data quality assessment and outlier repair mechanism; Based on time-aligned data streams, a data quality detection algorithm is employed to obtain a reliable dataset with quality scores. Specifically, multi-dimensional data quality assessment metrics are established. The integrity metric is used to assess data loss; the integrity rate is obtained by calculating the ratio of the amount of data that should be received to the amount of data actually received per unit time. For periodically sampled sensors, such as a water flow sensor that samples 10 times per second, if only 7 data points are received in a second, the integrity rate for that second is 70%. The continuity metric is used to assess the temporal coherence of the data; the time interval between adjacent data points is calculated, and if the interval exceeds twice the normal sampling period, a data break is identified.

[0026] Reasonableness indicators are used to assess whether data values ​​are within the normal range. A preset valid data range is assigned to each sensor. Specifically, the valid range for a water flow sensor is 0 to the system's maximum design flow rate; for a turbidity sensor, it is 0 to 100 turbidity units; for a total dissolved solids sensor, it is 0 to 1000 mg / L; for a pH sensor, it is 0 to 14; and for a pressure sensor, it is 0 to the system's maximum operating pressure, such as 0.6 MPa. When a sensor output value exceeds its corresponding valid range, it is considered an abnormal value.

[0027] Consistency metrics are used to evaluate the logical relationships between related sensor data. For example, when a water flow sensor shows water flow, a pressure sensor should show pressure. If there is a contradiction where water flow is present but the pressure is zero, then the data is considered to have a problem.

[0028] A data quality score is calculated by integrating various indicators across multiple dimensions. A weighted summation method is used, where each indicator is normalized to a score between 0 and 100, and then summed according to its weights to obtain the overall quality score. The weights can be adjusted based on the application scenario. The quality score ranges from 0 to 100, with higher scores indicating better data quality. This quality score is then appended to each data record as a reliability weight during subsequent data fusion.

[0029] For detected outliers and missing values, an intelligent repair strategy is employed. For occasional outliers, median filtering is used to replace the outlier with the median of data from adjacent time points. For short-term data gaps (less than 5 seconds), linear interpolation is used to estimate the missing data based on data values ​​before and after the gap. For long-term data gaps, forward imputation is used to fill the missing period with the last valid data value before the gap. Simultaneously, this data is marked as repaired data in the data record, and its quality score is lowered to prevent the repaired data from having an excessive impact on the analysis results.

[0030] Step 1.4: Establish sensor health assessment and dynamic weight allocation; Based on the data quality assessment results, a sliding window statistical method was used to obtain a health assessment report for each sensor. Specifically, with a 24-hour assessment period, the average data quality score, outlier frequency, data loss rate, response latency, and other indicators for each sensor within that period were statistically analyzed. Sensor health was defined as a comprehensive evaluation of these indicators, expressed on a percentage scale.

[0031] Establish a sensor health grading standard. A health score of 90 or above is considered excellent, indicating that the sensor is working normally and the data is highly reliable. This sensor data is assigned a weight of 1.0 in data fusion, meaning it participates in subsequent analysis with its original confidence level. A health score between 70 and 90 is considered good, indicating that the sensor is basically normal but has minor issues. This sensor is assigned a weight of 0.8, meaning that its influence in multi-sensor fusion decisions is reduced to 80% of its normal state. A health score between 50 and 70 is considered average, indicating that the sensor has many problems and the data usability is reduced. This sensor is assigned a weight of 0.5. A health score below 50 is considered abnormal, indicating that the sensor may be faulty and the data is unreliable. This sensor data is assigned a weight of 0.2 or is excluded altogether.

[0032] Establish a sensor fault early warning mechanism. When the health score of a sensor is below 70 for three consecutive days, a sensor anomaly warning will be generated, and the user will be notified via mobile device that the sensor may require maintenance. When the health score is below 50, a sensor fault alarm will be generated, advising the user to check or replace the sensor as soon as possible to avoid affecting system functionality due to sensor failure.

[0033] This step outputs a standard multi-source heterogeneous dataset, including: timestamps, data source types, device IDs, data values, quality scores, and health weights, laying a high-quality data foundation for subsequent behavior analysis and anomaly detection.

[0034] The water use behavior modeling module identifies water use events and analyzes their temporal distribution patterns based on standard multi-source heterogeneous datasets, and builds a personalized water use behavior baseline model. Specifically, the following steps are included: Step 2.1: Use threshold segmentation and feature extraction to identify independent water use events; Based on the standard multi-source heterogeneous dataset output in step 1, data records whose data source type field is identified as water flow are selected and sorted by timestamp to form a water flow time-series data sequence. An event segmentation algorithm is then used to obtain independent water usage event sequences. Specifically, continuous water flow measurement data are arranged in chronological order, and a water flow threshold is set at 5% of the system's maximum flow rate. Flow below this value is considered no water usage, and flow above this value is considered water usage. When the water flow rises from below the threshold to above the threshold, it is determined as the start of a water usage event, and the start time is recorded. When the water flow falls from above the threshold back to below the threshold and remains below the threshold for a certain period, such as 5 seconds, it is determined as the end of a water usage event, and the end time is recorded.

[0035] For each identified water usage event, basic feature parameters are extracted. The start time feature records the exact moment the water usage occurred, accurate to the second, used to analyze the temporal distribution patterns of water usage. The duration feature calculates the time span from start to finish of the water usage event, reflecting the continuity of a single water usage event. The total flow feature is calculated by integrating the instantaneous flow rate within the event period, reflecting the total consumption of a single water usage event. The peak flow feature records the maximum instantaneous flow rate value occurring within the event period, reflecting the intensity of the water usage event. The average flow feature calculates the average flow rate within the event period, reflecting the stability of the water usage event.

[0036] Establish a water usage incident classification mechanism to categorize water usage incidents into different types based on duration and total flow, such as drinking water incidents, handwashing incidents, vegetable washing and cooking incidents, and dishwashing incidents. This incident classification allows for more refined analysis of users' lifestyle behavior patterns.

[0037] Step 2.2: Extract time distribution features using time period statistics and sliding window calculation; Based on the water usage event sequence obtained in step 2.1, time period segmentation and statistical analysis methods are used to obtain the statistical characteristics of the temporal distribution of water usage behavior. Specifically, the 24 hours of a day are divided into several time periods, which can be divided by fixed duration or according to daily routines, such as morning, forenoon, noon, afternoon, evening, night, and late night.

[0038] For each time period, the frequency of water use events is counted to reflect the level of water use activity during that period. The total flow rate during that period is also counted to reflect the total water consumption. The average duration and average flow rate of water use events during that period are calculated to reflect the typical characteristics of water use behavior during that period. The time interval between adjacent water use events during that period is calculated to reflect the intensity of water use behavior.

[0039] A sliding window method is used to calculate statistical characteristics across multiple time scales. Using a 7-day window, average water usage frequency and average water consumption are calculated for each time period within the most recent week, reflecting short-term water usage patterns. The window slides forward one day each day, meaning that data from the most recent 7 days is used for each calculation, automatically removing data older than 8 days and incorporating the latest data, thus dynamically updating the statistical indicators. A 14-day window is used to calculate statistical indicators for the most recent two weeks, reflecting medium-term water usage patterns. A 30-day window is used to calculate statistical indicators for the most recent month, reflecting long-term stable water usage patterns. This multi-time-scale statistical approach can account for both daily fluctuations and long-term trends in water usage behavior.

[0040] Extracting periodic characteristics, we analyze the differences in water usage between weekdays and weekends. We statistically analyze the average water usage patterns from Monday to Friday and on Saturday and Sunday, comparing the differences in time distribution, water consumption, and water usage frequency. For retired seniors, the difference between weekdays and weekends may not be significant, while for seniors still working or families living with their children, this difference may be more pronounced.

[0041] Step 2.3: Construct multidimensional water use behavior feature vectors and pattern representations; Based on the time distribution statistical characteristics obtained in step 2.2, a high-level water use behavior feature vector is obtained using feature engineering methods. Specifically, the feature of the first water use time in the morning is extracted. This feature reflects the user's wake-up time pattern and is of great significance for safety monitoring of elderly people living alone. By statistically analyzing the first water use time of each day over the past 30 days, the mean and standard deviation are calculated to obtain the user's typical wake-up time and its fluctuation range.

[0042] Multidimensional features were extracted, including: the time of first water use in the morning, the frequency of drinking water during the day, the distribution of single water consumption, the nighttime water use, and the interval between water uses. The mean and standard deviation were calculated by statistically analyzing the data of the most recent 30 days to establish a baseline.

[0043] The above features are organized into a water use behavior feature vector, with each dimension of the vector corresponding to a feature index. The feature vector comprehensively describes the user's behavioral characteristics in multiple aspects such as time distribution, water use frequency, water consumption, water use interval, and water use type, providing a rich data foundation for establishing a baseline model.

[0044] Step 2.4: Establish a personalized baseline model using statistical analysis and dynamic updates; Based on the water use behavior feature vector obtained in step 2.3, a personalized water use behavior baseline model is established using statistical modeling methods. Statistical distribution parameters are calculated for each dimension of the feature vector: for features with an approximately normal distribution, the mean and standard deviation are calculated; for features with a skewed distribution, the median, quartiles, and percentiles are calculated.

[0045] Establish confidence intervals for features to define the boundaries of normal behavior. For normally distributed features, use the mean ± 2 standard deviations as the 95% confidence interval; for non-normally distributed features, use the 5%–95th percentiles as the normal range. The setting of confidence intervals should balance sensitivity and specificity to avoid excessive false alarms or false negatives.

[0046] A feature association model is established, and the Pearson correlation coefficient is used to calculate the degree of linear correlation between feature pairs. When the absolute value of the correlation coefficient is greater than 0.3, 0.5, and 0.7, it is judged as a weak correlation, a moderate correlation, and a strong correlation, respectively. A conditional distribution model is established for the correlated features, and the changes of these features are considered together during anomaly detection.

[0047] A dynamic update mechanism for the baseline model is established to allow the model to adapt to slow changes in user behavior. A sliding window strategy is adopted, recalculating model parameters every 7 days based on the most recent 30 days of data. An exponentially weighted moving average method is used, with a decay coefficient set to 0.9, giving greater weight to recent data and achieving smooth model evolution.

[0048] Set model update constraints to prevent abnormal data from contaminating the baseline model. When updating parameters, use the current model to perform anomaly detection on the new data, exclude abnormal data, and only use normal data to update the model, ensuring that the baseline model always reflects the user's normal behavioral patterns.

[0049] To address the cold start problem for new users, a general template initialization strategy is adopted. Based on demographic characteristics such as age, gender, and living status, the most suitable general template (e.g., elderly men living alone, elderly couples, etc.) is selected from the template library as the initial baseline model. As user data accumulates, the template data is gradually replaced with actual data. When the accumulated data exceeds 30 days, the system completely switches to using the user's own data to build a personalized model.

[0050] This step establishes a personalized water use behavior baseline model, which includes statistical distribution parameters, confidence intervals, feature correlations, and temporal evolution trends for each feature, providing a reliable comparison benchmark for subsequent abnormal behavior identification.

[0051] The abnormal behavior identification module, based on a personalized water use behavior baseline model, identifies abnormal behaviors and integrates standard multi-source heterogeneous datasets for anomaly confirmation, and obtains abnormal behavior identification results and severity assessment reports. Specifically, the following steps are included: Step 3.1: Calculate behavioral anomaly indicators using standardized deviation metrics; Based on the personalized water use behavior baseline model output from step 2 and the current water use data, the deviation calculation method is used to obtain the abnormal indicators for each feature dimension. Specifically, for the feature of the first water use time in the morning, the actual first water use time of the day is obtained, the difference between this time and the mean of this feature in the baseline model is calculated, and then the difference is divided by the standard deviation for standardization to obtain the standardized deviation. The absolute value of the standardized deviation represents the degree of deviation; a value greater than 2 indicates a deviation of more than 2 standard deviations, which is considered a significant anomaly, and a value greater than 3 indicates a deviation of more than 3 standard deviations, which is considered a severe anomaly.

[0052] To determine the frequency of water consumption during the day, the number of water consumptions up to the current moment is counted and compared with the expected number of water consumptions for the same period in the baseline model. Since water consumption frequency accumulates over time, normalization is required for the comparison. Specifically, the proportion of the current moment in the day is calculated, and the expected number of water consumptions for the entire day in the baseline model is multiplied by this proportion to obtain the expected number of water consumptions for the current moment. The actual number of water consumptions is then compared with the expected value to determine if there is any abnormality in water intake.

[0053] For the water usage interval feature, the time interval between the current moment and the last water usage is calculated and compared with the maximum normal interval in the baseline model. If the current interval exceeds the 95th percentile interval in the baseline model, it indicates an abnormally long period of water inactivity. This feature is of significant value in detecting potential unexpected situations that elderly people may encounter.

[0054] To determine the frequency of nighttime water use, the number of times water was used the previous night was recorded and compared with the baseline model. If the number of nighttime water uses is significantly higher than normal, it may indicate that the elderly person has problems such as increased nocturia, sleep disorders, or physical discomfort.

[0055] For the total water consumption characteristics, the total water consumption up to the present day is calculated, normalized according to the time schedule, and compared with the expected value of the baseline model. A consistently low total water consumption may indicate reduced user activity or absence from home, while a consistently high total water consumption may indicate water leakage or abnormal water consumption behavior.

[0056] Step 3.2: Define various anomaly patterns and classification rules; Based on the deviations of each feature calculated in step 3.1, rule matching and pattern recognition methods are used to obtain the classification results of abnormal behavior types. Specifically, an abnormal pattern of prolonged inactivity is defined, and the judgment condition is that the time interval between the current moment and the last water usage exceeds the abnormal threshold of the baseline model, and the current time period belongs to the user's usually active time period. The abnormal threshold is defined as the 95th percentile of the water usage interval in the baseline model, that is, the probability of exceeding this threshold occurring under normal conditions is less than 5%.

[0057] Define an abnormal pattern of prolonged continuous water use, and determine it when the duration of a single water use event exceeds the normal duration limit for any water use type in the baseline model. For example, the baseline model shows that the longest water use event for a user is usually no more than 30 minutes, corresponding to washing vegetables and cooking. If a water use event with a duration exceeding 45 minutes is detected, a prolonged continuous water use anomaly is triggered, which may indicate that the user forgot to turn off the tap or that there is a leak in the pipes.

[0058] Abnormal prolonged water inactivity: Interval exceeding the 95th percentile threshold of the baseline. Severity grading: More than 1 hour is mild, 2 hours is moderate, and 4 hours is severe.

[0059] Anomalies in water usage patterns are defined as a systematic deviation from the baseline model when a user's water usage time distribution deviates from the baseline model. A systematic deviation is defined as a deviation in the same direction for more than three consecutive days, with the deviation exceeding one standard deviation.

[0060] An abnormal water intake pattern is defined, and the criteria for judgment are that the frequency and total amount of water consumed during the day are consistently lower than the healthy standard. This judgment not only refers to the user's historical baseline for assessment, but also triggers an abnormal water intake pattern when the actual water intake is lower than a certain percentage of the recommended value for several consecutive days, reminding the user to increase water intake.

[0061] Step 3.3: Integrate multi-sensor data for anomaly confirmation and false alarm suppression; Based on the suspected abnormal behaviors identified in step 3.2, a multi-sensor data fusion method is used to obtain a confirmed anomaly determination result. Specifically, when an anomaly of prolonged water inactivity is detected, an alert is not immediately triggered. Instead, data from other sensors in the same time period are first queried from the standard multi-source heterogeneous dataset output in step 1 for auxiliary judgment. The detection records of the human body sensor during the same time period are queried. If the human body sensor detects human activity indoors, it indicates that the user is at home but has not used water, and the anomaly may be normal; in this case, the anomaly confidence level is lowered. If the human body sensor has not detected activity for a long period, and this coincides with the period of water inactivity, the anomaly confidence level is increased, potentially indicating that the user has encountered an accident.

[0062] Query the data from the door and window sensors if they are installed to determine if the user has left the house. If the sensors show that the user recently opened the door and left without returning, then the prolonged lack of water usage is considered normal, and the abnormal warning is canceled. If the sensors show that the user has not left the house but shows no signs of activity for an extended period, then the abnormal confidence level is increased.

[0063] Check indoor temperature sensor data to determine if the environment is abnormal. If the room temperature is too low or too high, exceeding the comfort range, it may indicate that the user has failed to properly adjust the environment or that there is a malfunction in the heating or cooling system. Combined with prolonged inactivity, this increases the severity assessment of the abnormality.

[0064] When an abnormally prolonged period of continuous water use is detected, the pressure and humidity sensor data are queried to confirm a leak. If the pressure sensor shows an abnormal drop in pipeline pressure, and the humidity sensor detects a rapid increase in humidity around the water purifier, a leak is highly suspected, and a leak alarm is immediately triggered. If the pressure and humidity data are normal, it may be that the user is engaged in an activity requiring prolonged water use. In this case, a query message is first sent to the user to inquire whether they are currently using water, and the decision to cancel the alarm is based on the user's response.

[0065] A multi-sensor data fusion method is used to calculate the overall confidence level of anomalies. Data from each sensor is used as evidence to comprehensively assess the probability that an anomaly actually exists. An anomaly is confirmed and an alert is triggered only when the confidence level exceeds a set threshold; otherwise, suspected anomalies are logged without triggering an alert to avoid false alarms and user disruption.

[0066] Step 3.4: Establish an assessment model for the severity of anomalies; Based on the abnormal behaviors identified in step 3.3, a multi-factor comprehensive assessment method is used to obtain the severity level of the abnormality. Specifically, the assessment dimensions for the severity of the abnormality are defined as follows: Deviation dimension measures the extent of deviation of the abnormal behavior from the normal baseline; the greater the deviation, the higher the severity. Duration dimension measures the duration of the abnormal state; the longer the duration, the higher the severity. Historical frequency dimension measures the frequency of occurrence of this type of abnormality in history. If it is the first occurrence or a rare occurrence, the severity assessment needs to be more cautious; if it is a frequently occurring abnormality that has been confirmed as a false alarm, the severity can be appropriately reduced.

[0067] A three-tiered severity classification standard is established. A scoring mechanism is used to comprehensively determine severity, considering dimensions such as the degree of deviation, duration, and historical frequency. Each dimension is scored based on its quantified value, and the scores are weighted and summed to obtain a comprehensive score. Based on the comprehensive score, anomalies are categorized into three levels: mild (Level 1 alert), moderate (Level 2 alert), and severe (Level 3 alert). The handling strategy for mild anomalies is a local device notification, alerting the user via voice announcement or indicator light, without notifying family members to avoid excessive disturbance.

[0068] A moderate anomaly corresponds to a Level 2 alert, characterized by a significant deviation from the normal baseline, a prolonged duration, and potential health or safety risks. The handling strategy for a moderate anomaly is to send a push notification to the user's mobile device, inquiring about their current status and requesting a response. If the user responds promptly and confirms safety, the alert is lifted. If the user does not respond within a set time, such as 30 minutes, the alert is escalated to a severe anomaly.

[0069] Severe anomalies correspond to a Level 3 warning, characterized by a significant deviation from the normal baseline, a prolonged duration, or the receipt of a clear distress signal, indicating an imminent safety risk. The handling strategy for severe anomalies is to immediately trigger the highest-level alarm, simultaneously notify all emergency contacts, and implement coordinated control measures based on the anomaly type.

[0070] This step outputs the abnormal behavior identification results and severity assessment report, including: abnormality type, occurrence time, deviation value, severity level, overall confidence score, supporting evidence list, and recommended handling measures, providing accurate and reliable input for the subsequent generation of early warning strategies.

[0071] The healthy drinking water management module, based on a personalized water behavior baseline model, calculates the recommended daily water intake and formulates a drinking water plan, conducts achievement assessment and trend analysis, and obtains a personalized drinking water solution. Specifically, the following steps are included: Step 4.1: Calculate basic drinking water requirements using a weight-related model and disease correction. Based on the user's initial health information and the real-time environmental data collected in step 1, a user health profile is established, including information such as age, gender, weight, height, disease type, and medication status. A weight-related water intake calculation method is used, with the basic water requirement for adults being 30 to 40 ml per kilogram of body weight. For example, a user weighing 60 kg would have a basic requirement of 1800 to 2400 ml. For elderly individuals, due to decreased renal concentrating function, a correction factor of 10% to 15% is added to the basic requirement.

[0072] Introducing disease correction factors: 10% to 20% for patients with hypertension; 30% to 50% for patients with diabetes (especially when polyuria occurs due to poor glycemic control); 20% to 40% for patients with chronic kidney disease based on renal function stage; and fluid intake is restricted for patients with heart failure based on heart function classification.

[0073] Considering the effects of medication: Patients taking diuretics need to increase their water intake appropriately; when taking antihypertensive drugs, maintaining adequate water intake helps with drug metabolism and kidney protection; for elderly patients taking multiple drugs, increasing water intake helps with drug absorption and metabolism.

[0074] Establish a health record update mechanism: Users can update their weight, illnesses, medications, and other information at any time via mobile application, and the system will automatically recalculate the recommended dosage; for users who do not actively update, the system will send a reminder confirmation every month.

[0075] Step 4.2: Calculate dynamic drinking water adjustment amount using environmental perception and seasonal adaptation; Based on the fundamental requirements in step 4.1, dynamic adjustments are made in conjunction with real-time environmental factors. Outdoor temperature data is collected (via network access to weather information or outdoor temperature sensors), and a temperature-water consumption adjustment model is established to increase water consumption according to the temperature gradient during hot weather.

[0076] Collect indoor humidity data (via built-in humidity sensor or smart home devices): increase by 10% to 15% when relative humidity is below 40%; do not adjust between 40% and 60%; decrease by 5% when relative humidity is above 60%.

[0077] Seasonal adjustments: Increase by 20% to 30% in summer; increase by 10% in northern heating areas during winter; use the base value in spring and autumn.

[0078] Activity level inference: Activity level is indirectly assessed by water use behavior. If water use is frequent during the day, the recommended amount should be increased appropriately, while if there is no water use for a long period of time, the recommended amount should be reduced, but abnormal situations should be noted.

[0079] Comprehensive calculation: The basic requirement is multiplied by age, disease level, temperature, humidity, and seasonal correction factors in sequence to obtain the recommended total daily water intake. The system updates the recommended values ​​daily.

[0080] Step 4.3: Use behavioral preference analysis to develop a time-segmented water intake reminder plan; Based on the daily recommended total amount in step 4.2 and the water use behavior baseline model in step 2, a personalized time-segmented drinking water plan is developed. The day is divided into morning, forenoon, noon, afternoon, evening, and bedtime periods. The user's historical water consumption and frequency for each period are statistically analyzed, and the time-segment allocation ratio is calculated. A general template is used for new users: 20% in the morning, 15% in forenoon, 20% at noon, 25% in the afternoon, 15% in the evening, and 5% before bedtime.

[0081] Intelligent reminder mechanism: Reminders are set at the beginning and middle of a time period to check if the user has drunk water. If the requirement has been met, the reminder is skipped; otherwise, it is triggered. The reminder method is designed to be non-intrusive, using gentle voice or mobile push notifications to avoid forced interruptions.

[0082] Optimize reminder strategy: Record user response to reminders. If users respond frequently, maintain the strategy. If users ignore reminders frequently, adjust the reminder time or method (e.g., change from voice to light) to find a more acceptable approach for users.

[0083] Step 4.4: Generate drinking water health data using achievement assessment and trend analysis; Based on the drinking water plan established in step 4.3 and the actual drinking water data, a comparative analysis method is used to obtain drinking water achievement assessment data and health risk warnings. Specifically, the user's actual total water intake is counted daily; the actual water intake is compared with the recommended water intake to calculate the achievement rate. An achievement rate between 90% and 110% is considered ideal, and the system provides positive feedback, such as displaying a green achievement indicator; an achievement rate between 70% and 90% indicates slight deficiency, and the system displays a yellow warning indicator and sends a mild reminder; an achievement rate below 70% indicates significant deficiency, and the system displays a red warning indicator and sends a stronger reminder; an achievement rate between 110% and 130% indicates slight excess, and the system prompts attention but does not issue an alarm. An achievement rate exceeding 130% indicates significant excess, and the system sends an excessive water intake warning, suggesting consultation with a doctor.

[0084] Analyze the achievement of water intake time distribution. Assess whether the user's actual water intake in each time period meets the recommended amount for that time period, and identify which time periods are insufficient. If a user never drinks water during certain time periods, analyze the reasons, which may be that the user is out or engaged in specific activities during those times, and adjust the recommendation strategy for those time periods accordingly.

[0085] Establish trend analysis over multiple days. Calculate the average achievement rate over the past 7 days, 14 days, and 30 days, and observe the trend of the achievement rate. If the achievement rate shows a continuous downward trend, it may indicate that the user's awareness of drinking water has weakened, their health status has changed, or their lifestyle has changed, generating a health risk warning; if the achievement rate remains below 70% for more than 7 consecutive days, generate a serious water insufficiency warning report, push it to the user and emergency contacts via mobile device, and suggest paying attention to the user's health status.

[0086] Analyze the possible causes of insufficient water intake. Using a personalized water use behavior baseline model, if a user's overall activity level is found to be declining, such as reduced total water consumption or decreased water usage frequency, insufficient water intake may be a sign of decreased activity levels, suggesting a potential health problem. If other water-related behaviors are normal but water intake is significantly reduced, it may indicate insufficient attention to hydration or a drinking disorder, requiring further investigation.

[0087] The personalized drinking water plan output in this step includes: daily drinking water achievement rate, achievement status by time period, multi-day trend analysis results, and health risk warnings.

[0088] The tiered early warning execution module formulates and executes tiered early warning strategies based on the abnormal behavior identification results and severity assessment reports, and obtains the early warning execution results. Specifically, the following steps are included: Step 5.1: Establish an early warning strategy knowledge base and decision rule engine; A structured early warning strategy knowledge base is constructed using knowledge engineering methods. The rule base is stored using rule notation, and each rule includes elements such as triggering conditions, execution actions, notification targets, and escalation mechanisms, and records fields such as rule number, name, priority, activation status, and timestamp, supporting dynamic loading and hot updates.

[0089] The three-level alert for prolonged water inactivity includes: Mild: Water inactivity exceeds the baseline threshold by 1 hour and the confidence level is >70%, local voice alert is issued, and it escalates to moderate after 2 hours; Moderate: Water inactivity exceeds 2 hours or mild escalates, mobile push notification is sent to the user, and it escalates to severe if there is no response for 30 minutes or after 4 hours; Severe: Water inactivity exceeds 4 hours or moderate escalates or an emergency request for help is received, and an alarm is immediately triggered to notify all emergency contacts and dial the preset phone number.

[0090] The two-level warning system for abnormal continuous water usage includes: Moderate, where continuous water usage exceeds 1.5 times the baseline maximum value, triggering a push notification on the mobile device; if there is no response and the usage continues for more than 60 minutes, it is upgraded to severe; Severe, where continuous water usage exceeds 60 minutes or the pressure / humidity sensor malfunctions, sending a leak alarm to the user and emergency contacts, and triggering the shutdown of the solenoid valve to cut off the water supply.

[0091] The external safety device linkage rules include: smoke alarm: upon receiving a smoke signal, immediately alarm and notify all contacts, remotely close the solenoid valve according to the settings, and prompt to call 119; gas alarm: upon receiving a gas signal, immediately alarm and notify all contacts, automatically close the solenoid valve to prevent electric sparks, and prompt to open windows for ventilation and contact the gas company; emergency assistance: upon receiving an assistance signal, immediately alarm and notify all contacts (including user information, address, and time), and initiate continuous positioning until a response is received.

[0092] The decision rule engine monitors the system status in real time and automatically executes a sequence of actions when trigger conditions are met. The engine employs priority scheduling: high-priority rules (emergency assistance, gas leaks) are executed first; rules of the same priority are sorted by creation time; and conflicting rules adopt a conservative strategy (preferring to send alerts). The engine supports dynamic rule updates, allowing administrators or users to flexibly adjust policy parameters.

[0093] Step 5.2: Handle minor anomalies with gentle reminders and delayed confirmation; Based on the minor anomaly warning rules established in step 5.1 and the minor anomaly events identified in step 3, a non-intrusive reminder strategy is adopted to achieve localized user interaction. Specifically, when a minor anomaly of prolonged water inactivity is detected, the system plays a gentle prompt message through the water purifier's built-in voice module. The message content uses caring expressions rather than warnings to avoid causing psychological stress to the user.

[0094] Meanwhile, the water purifier's indicator light changes color or flashes at a different frequency to provide visual cues. The reminder is played only once and does not repeatedly disturb the user. If the water purifier has a touchscreen, the screen displays drinking water recommendations, and the user can manually cancel the reminder by clicking the "I'm fine" button on the touchscreen.

[0095] A delayed confirmation mechanism is implemented, eliminating the need for immediate user response. After issuing a local alert, the system continues to monitor water usage in real-time. Whenever new water usage data is received, it immediately determines whether the conditions for resolving the alert are met. If the user engages in any water usage within 30 minutes of the alert, the system automatically determines the user's status is normal, removes the anomaly flag, and takes no further action. This design philosophy is "reminding but not forcing," respecting user autonomy and avoiding excessive intervention.

[0096] If the user does not use water during the delayed confirmation period, the system assesses whether to escalate the alert level. If the human body sensor detects human activity signals indoors within the last 30 minutes of the delayed confirmation period, it determines that the user is at home but may be engaged in activities that do not require water use, and maintains the low alert status while extending the observation period. If the human body sensor does not detect activity for an extended period, the anomaly confidence level is increased, and preparations are made to escalate the alert to a medium alert.

[0097] Record the handling results of each minor alert, including the alert method, user response time, and whether an escalation was initiated. Evaluate the effectiveness of minor alerts through statistical analysis; if a particular alert method is found to have a significantly higher user response rate, that method should be prioritized.

[0098] Step 5.3: Use mobile interaction and user response to determine and handle moderate anomalies; Based on the moderate anomaly warning rules in step 5.1 and the moderate anomaly events identified in step 3, the judgment results of anomaly confirmation or resolution are obtained through mobile terminal push and user interaction.

[0099] When a moderate anomaly is detected, the system immediately sends a push notification to the user's linked mobile device. If the push fails, it will retry multiple times. If it continues to fail, it is determined that the user cannot receive notifications, and the alert is escalated to a severe warning, notifying emergency contacts. The notification title is concise and clear, such as "Safety Tip," and the content describes the anomaly, such as "The system detected that you have not used water this morning. Please confirm if it is safe." The priority is set to high.

[0100] The push notification includes quick response buttons: clicking "I'm fine" immediately cancels the alert, records false alarms, and adjusts the judgment threshold to reduce future false alarms; clicking "Need help" immediately escalates the alert to a severe level and initiates the emergency response process; clicking "I'm away" puts the system into away mode, suspends the detection of water usage anomalies, and asks for the estimated return time. By default, away mode lasts for 24 hours before automatically resuming monitoring, during which time it continues to monitor signals from external safety devices such as smoke detectors and gas alarms.

[0101] Set up a user response timeout mechanism: Start timing after push notification is sent. If the user does not respond within 30 minutes (neither clicking the button nor using water), the system automatically escalates to a severe alert. Push notifications will be sent every 10 minutes during the waiting period, up to a maximum of 3 times.

[0102] For moderate anomalies caused by prolonged and continuous water usage, a notification will be sent to the user asking if they are currently using water. Once the user confirms "using water," the system will pause the alert but continue to monitor the water usage duration. If the usage exceeds a reasonable timeframe, a reminder will be sent again to prevent the user from forgetting to turn off the tap.

[0103] The system records users' response patterns to moderate alerts (average response time, commonly used response options, frequency of going out, etc.) and optimizes alert strategies based on response habits.

[0104] Step 5.4: Use multi-channel alarm and linkage control to handle severe anomalies; Based on the severe anomaly warning rules in step 5.1 and the severe anomaly events identified in step 3 or the moderate anomalies upgraded in step 5.3, the highest priority emergency response strategy is adopted to achieve comprehensive alarm notification and equipment linkage control.

[0105] When a severe warning is triggered, the system immediately activates a multi-channel synchronous alarm mechanism: sends a mobile push with the highest priority to the user, with the title "Emergency Alert", and the content describes the nature of the emergency and recommended actions, and uses a strong reminder mode to ensure that an alarm sound will be emitted even if the phone is muted.

[0106] At the same time, send alarm notifications to all preset emergency contacts in order of priority. The first-priority contacts receive them first. If there is no response within 10 minutes, the second-priority contacts will be notified, and so on. If all contacts do not respond within 30 minutes, the system automatically dials the community service or emergency call according to the user's authorization. The alarm notification contains detailed abnormal information: abnormal type, occurrence time, user's basic information (age, living alone status, medical history), recommended actions. For the situation where the user actively presses the help button, it is clearly marked in the alarm and the highest level of urgency is prompted. The system also provides a function to directly dial the user's home phone or make a video call.

[0107] Execute device linkage control according to the abnormal type: when there is a gas leak, automatically close the water purification solenoid valve to cut off the water source to prevent electrical sparks; when there is continuous water use for a long time and there is a suspicion of water leakage, push a message to the user and emergency contacts to ask whether to authorize remote closing of the solenoid valve. If authorization is obtained within 5 minutes or there is no response and the risk of water leakage is extremely high, it will be automatically closed to stop the loss. The user and emergency contacts can remotely control the solenoid valve switch through the mobile application.

[0108] For severe abnormalities that have not been used for a long time and have been upgraded multiple times, the system automatically dials the community medical service, property management or 120 emergency call according to the user's preset authorization, and uses voice synthesis technology to broadcast the preset help information and the user's address.

[0109] Establish an alarm confirmation and cancellation mechanism: after a severe alarm is issued, if the user or emergency contact confirms it as a false alarm, the alarm can be cancelled through the mobile device to stop subsequent automatic responses; if the emergency contact confirms that the user has been contacted and the situation is safe, the alarm can be marked as processed and the warning status can be解除预警状态. All alarm events and processing results are completely recorded in the security event log for subsequent query and analysis.

[0110] Step 5.5, use machine learning methods to optimize the early warning strategy parameters; Based on the historical warning records and user feedback data executed in Steps 5.1 to 5.4, use supervised learning algorithms to obtain optimized early warning strategy parameters and rules. Specifically, collect the complete data of all warning events, including sensor data at the time of triggering, calculated feature values, determined abnormal types, severity levels, executed warning actions, user response situations, final confirmation results, etc.

[0111] Establish evaluation metrics for early warning effectiveness. Accuracy measures the proportion of truly abnormal events identified by the system; high accuracy indicates fewer false alarms. Recall measures the proportion of actual abnormal events successfully detected by the system; high recall indicates fewer missed alarms. Response time measures the time delay from the occurrence of an anomaly to the user or emergency contact receiving the notification; short response time indicates good real-time performance. User satisfaction is measured through user feedback, assessing users' overall evaluation of the early warning function.

[0112] A supervised learning method was used to train an optimized early warning model. Historical early warning data was divided into training and testing sets; the training set was used for model learning, and the testing set was used for performance evaluation. Feature vectors were constructed, including various sensor data, behavioral features, temporal features, and historical early warning frequency at the time of the early warning. Labels were assigned to indicate the authenticity of the early warning; genuine anomalies were marked as positive, and false alarms as negative.

[0113] A classification algorithm is used to learn which feature combinations are more likely to correspond to true anomalies and which are more likely to be false positives. The model output is the probability that an anomaly is true given the features. Based on the model's prediction results, the warning threshold is adjusted to achieve a balance between warning accuracy and recall, reducing false positives without increasing false negatives.

[0114] For different types of anomalies, specialized optimization models are trained separately, as the characteristic patterns and false alarm causes of different anomalies may vary. The optimization focus for anomalies caused by prolonged periods of inactivity is to reduce false alarms due to normal reasons such as users being away or sleeping. The optimization focus for anomalies caused by prolonged continuous water use is to distinguish between normal prolonged water usage activity and actual leaks or forgetting to turn off the water.

[0115] Regularly retrain the model, updating its parameters monthly or quarterly using the latest accumulated early warning data to enable it to adapt to long-term changes and seasonal fluctuations in user behavior. After updating the model, test it on a small group of users to verify whether the new model outperforms the old one. If the performance improvement is significant, roll it out to all users; if the performance declines, revert to the old model and analyze the reasons.

[0116] The optimized model parameters are then translated into adjustments to the early warning policy rules, including anomaly detection thresholds, delayed confirmation durations, and escalation trigger conditions. The updated rules are pushed to user devices via a cloud platform, enabling remote policy updates without manual user intervention.

[0117] This step outputs the early warning execution results, including various forms of emergency response actions such as local voice or light reminders, mobile push notifications, emergency contact alarms, device linkage control commands, and automatic dialing of emergency numbers, providing users with comprehensive safety protection.

[0118] The visualization analysis module, based on the results of abnormal behavior identification, personalized drinking water plans, and early warning execution results, provides visual presentation and trend analysis to generate a whole-house safety trend report; Specifically, the following steps are included: Step 6.1: Construct a data visualization module using charts and interactive interfaces; Based on the historical multi-source sensor data output in step 1, the abnormal behavior identification results in step 3, the drinking water health data in step 4, and the early warning execution records in step 5, the data dashboard adopts a real-time update mechanism. When new sensor data or analysis results are received, the relevant cards automatically refresh to display the latest information. For cards with high real-time requirements, such as water usage overview and drinking water health, the update frequency is set to once per minute; for cards with slower changes, such as device status, the update frequency is set to once per hour. Users can also manually pull down to refresh and obtain the latest data. Data visualization technology is used to obtain intuitive and easy-to-understand charts. Specifically, a data dashboard function is developed for the mobile application. The dashboard adopts a card layout, with each card displaying one type of data. The water usage overview card displays the total water consumption for the day and week, water consumption frequency, and a comparison with historical averages, using bar charts and line charts. The drinking water health card displays the daily water consumption, water consumption achievement rate, and weekly average achievement rate, using pie charts and trend curves.

[0119] The security incident card displays a list of abnormal events from the past week, including event type, occurrence time, warning level, warning method, and handling result. Abnormal events are marked with different colors: red for severe, yellow for moderate, and blue for mild. Users can click on an event to view detailed information, including sensor data at the time of triggering, system judgment criteria, and user response records, helping users understand the rationale behind the warning.

[0120] The device status card displays the operating status of the water purification system and each sensor, including the remaining life of the filter cartridge, sensor health, and the last maintenance date. The filter cartridge life is displayed using a progress bar; when the remaining life is below 20%, the progress bar turns red and a replacement prompt is displayed. Sensor health is displayed using a rating system; sensors with a health score below 70 are marked as abnormal, prompting the user to check them.

[0121] The water usage behavior timeline uses a heatmap format, with the horizontal axis representing 24 hours and the vertical axis representing the days of the week. The color intensity of each cell indicates the water usage activity during that period, with darker colors indicating frequent water usage and lighter colors indicating infrequent water usage. Through the heatmap, users can visually see their water usage patterns, and family members can observe changes in the heatmap to detect any abnormalities in the elderly person's behavior.

[0122] The water intake distribution map uses a radar chart format to show the distribution of water intake at different times of the day. Actual water intake is represented by a blue filled area, and recommended water intake is represented by a green dashed line. The degree of overlap between the two reflects the water intake compliance status, and time periods with significant differences indicate insufficient water intake during those periods and need to be improved.

[0123] The system uses a time-series database to store historical data, supporting efficient time-range queries. Data is partitioned and stored according to time. For older historical data, the system automatically compresses and archives the data, retaining key statistical indicators and deleting detailed raw data, thus saving storage space while ensuring the needs of long-term trend analysis are met.

[0124] All charts support time range selection, allowing users to view data for the most recent day, week, month, or a custom time period. Zooming and dragging are supported for easy viewing of details. Each chart is accompanied by brief text explanations to help users unfamiliar with charting understand the data.

[0125] Step 6.2: Use time series analysis to identify long-term behavioral trend changes; Based on the long-term historical data output from Step 1, the abnormal behavior identification results from Step 3, and the drinking water health data from Step 4, time series analysis is used to obtain the long-term evolution trend of user behavior patterns and potential health risk warnings. Specifically, long-term statistical indicators are extracted, and weekly and monthly data such as total water consumption, water consumption frequency, drinking water volume, and abnormal event frequency are calculated to form a time series. Trend decomposition is performed on the time series, and the moving average method is used to extract the long-term trend term. Specifically, a 7-day moving average is used to smooth short-term fluctuations, and a 30-day moving average is used to identify the long-term trend. The periodic difference method is used to extract the seasonal periodic term, and the average data for the same day each week is calculated as the seasonal benchmark for that day. The remaining part is treated as a random fluctuation term.

[0126] Analyze the long-term trend of total water consumption. Use linear regression or moving average methods to fit the trend of water consumption over time and calculate the trend slope. A significant trend change is defined as a slope exceeding a set threshold and the trend lasting for more than 30 days. If the slope is significantly negative, it indicates a continuous downward trend in water consumption lasting more than a month, generating a risk warning of decreased activity level, advising users to pay attention to their health or undergo a health check. If the slope is significantly positive, it indicates a continuous increase in water consumption, requiring investigation into possible leaks or changes in family members.

[0127] Analyze the long-term trend of the first water usage time in the morning. Calculate the first water usage time each day and fit its trend over time. If the first water usage time shows a gradually delayed trend, lasting for more than two weeks, it may indicate a change in the user's sleep schedule, or problems such as decreased sleep quality or difficulty waking up, generating a risk warning for abnormal sleep patterns.

[0128] Analyze the long-term trend of water intake compliance rate. Calculate the average weekly water intake compliance rate and observe its direction of change. If the compliance rate continues to decline, it indicates that users' awareness of water intake has weakened or their ability to drink water has decreased, generating a water intake health risk warning and suggesting strengthening water intake reminders or checking for health problems such as difficulty swallowing.

[0129] Analyze the long-term trend of abnormal event frequency. Statistically count the number of various abnormal events each week, observing whether there is an increase in the frequency of any particular type of abnormal event. If the frequency of abnormal events related to nighttime water use gradually increases, it may indicate that the user has a problem with increased nocturia, and it is recommended to pay attention to prostate health or blood sugar control. If the frequency of abnormal events related to prolonged periods without water use increases, it may indicate that the user's activity level has decreased or that they are going out more frequently.

[0130] Identifying seasonal patterns. Seasonal pattern identification requires the system to accumulate at least one year of historical data. For new users with less than a year of usage experience, the system uses a seasonal template from similar user groups in the same region as a reference. Once a year of user data has been accumulated, the system begins using the user's own historical data for seasonal analysis. Analyzing data from different years within the same season identifies seasonal patterns in water usage behavior. For example, summer water consumption and drinking water consumption are typically higher than winter consumption. If summer water consumption is significantly lower than the same period in previous years, it may indicate a change in the user's health condition, generating an anomaly alert.

[0131] The system employs a multi-verification mechanism to avoid false alarms, generating a health risk alert report only when the following conditions are met simultaneously: the duration of the trend change exceeds a set threshold, the magnitude of the trend change exceeds a specific multiple of the baseline standard deviation, the trend remains significant after excluding the influence of seasonal factors, and the user has not actively marked it as normal, such as planned outings or lifestyle adjustments.

[0132] For any significant trend changes detected, the system generates a health risk alert report, which is pushed to the user and their family via mobile device. The report includes the type of trend detected, its duration, possible cause analysis, and recommended actions. The report uses plain and easy-to-understand language, avoiding technical jargon, to ensure that elderly users and their families can comprehend it.

[0133] Step 6.3: Generate weekly and monthly reports to provide regular summaries of the security status; Based on the historical data from step 1, the abnormal event records from step 3, the drinking water achievement data from step 4, the visualization data from step 6.1, and the trend analysis results from step 6.2, structured weekly and monthly report documents are generated using report generation technology. Specifically, the weekly report is automatically generated and pushed to the user and authorized family members every Sunday evening.

[0134] The weekly report includes the following: A weekly overview summarizing total water consumption, number of water-using days, and average daily water consumption compared to last week, indicating the percentage increase or decrease and possible reasons. A drinking water health section summarizing average water consumption, achievement rate, and number of days meeting standards, listing the best and worst days, analyzing the main periods of insufficient water intake, and providing improvement suggestions. A safety incident section listing all abnormal events of the week, categorized by severity, analyzing the main types and timing of occurrence, and assessing the overall safety situation. If no abnormal events occur, it displays "Safety situation is good this week, no abnormal events occurred." An equipment status section reporting the operating status of the water purification system and sensors, indicating maintenance items, times, and operating methods. A health advice section providing personalized suggestions based on this week's data, such as "Low drinking water achievement rate, it is recommended to increase drinking frequency" or "Increased nighttime water usage, if discomfort occurs, it is recommended to seek medical attention." The weekly report uses a graphic format, with key data displayed in prominent numbers and charts, and trend changes indicated by arrows, resulting in a concise and clear layout for quick browsing.

[0135] The monthly report is automatically generated on the last day of each month, offering more comprehensive and in-depth content. In addition to summaries similar to weekly reports, it includes the following modules: Long-term trend analysis, displaying the change curves of key indicators such as water consumption, drinking water consumption, and frequency of abnormal events over the past three months to identify trend changes; Monthly health score, providing a health score from 0-100 based on dimensions such as water usage regularity, drinking water health, and safety status, and comparing it with the previous months, with higher scores indicating better health; Early warning system effectiveness evaluation, statistically analyzing the overall performance of the early warning system this month, including the total number of early warnings and month-on-month changes, early warning accuracy and trends, average user response time and improvements, early warning escalation rate and cause analysis, review of false alarms and improvement measures; Year-on-year analysis, comparing this month's data with the same month last year, eliminating the influence of seasonal factors and identifying the true changes between years; Family care suggestions, providing specific suggestions for families based on this month's data analysis, such as "suggest increasing visitation frequency and accompanying them on outdoor activities; reminding the elderly to take medication on time and drink water regularly; and suggesting arranging a health check-up."

[0136] Weekly and monthly reports can be exported and printed in PDF format, making them convenient for family members who do not use smartphones to view. Reports can also be sent to authorized family members via email for information sharing.

[0137] Step 6.4: Establish a group comparative analysis to provide a reference range for healthy behaviors; Based on the anonymized group data aggregated by the cloud platform, the user historical data from step 1, the abnormal behavior data from step 3, the drinking water health data from step 4, and the trend analysis results from step 6.2, a group statistical analysis method is used to obtain the user's relative position and health behavior reference range within the same group. Specifically, the cloud platform anonymizes and desensitizes all user data, removing personally identifiable information and retaining only classification tags such as age group, gender, region, and health status, as well as statistical data on water use behavior.

[0138] Based on user characteristics, the system matches similar user groups. For example, for elderly male patients aged 70 to 75 living alone with hypertension, the system retrieves all user data matching this characteristic from the cloud platform to form a comparison group. The matching rules employ a multi-level screening strategy, performing precise matching according to age group, gender, and living status. If the number of precisely matched users is less than 100, the age range is broadened by 5 years or the living status condition is relaxed to ensure a sufficient sample size for the comparison group. The system requires the comparison group to contain at least 50 users; if this is still not possible, the user is notified that the group comparison function is currently unavailable. The system calculates the statistical distribution of various indicators for the comparison group, including the mean and quantiles of total water consumption, the mean and quantiles of drinking water, and the statistical distribution of abnormal event frequencies.

[0139] The system compares a user's personal data with the distribution of a group to calculate the user's percentile ranking within that group. For example, if a user's average daily water intake is 1800 ml, and the median of the group is 2000 ml, the user would be at the 40th percentile. This means that 60% of users in the same category drink more water than the user, indicating that the user's water intake is below the group average and needs improvement.

[0140] Add a group comparison function to the data dashboard to visually compare users with the group. Use box plots, where the boxes represent the quartile range of the group, and the user's data points are marked on the plot to intuitively show the user's relative position. Accompanying text descriptions, such as "Your average daily water intake is 1800 ml, which is lower than the average of 2000 ml for your age group. We recommend increasing your water intake."

[0141] The group comparison function helps users understand whether their behavioral habits are healthy and provides a reference target for improvement. For users with excellent performance, such as those who consistently maintain a water intake compliance rate of over 95%, the system provides positive incentives, displaying "Your water intake habits are better than 85% of your peers, please continue to maintain this."

[0142] The use of group data strictly adheres to privacy protection principles. Users can choose whether to contribute their anonymized data to group statistics and can opt out of data sharing at any time. Group statistical results are for user reference only and are not used for any commercial purposes or shared with third parties.

[0143] This step outputs a whole-house safety trend report, including: a visual data dashboard, long-term trend analysis results, periodic summary reports, and group comparison analysis results.

[0144] The adaptive optimization module performs adaptive optimization and upgrades on the system based on standard multi-source heterogeneous datasets and early warning execution results, and obtains the system adaptive optimization scheme. Specifically, the following steps are included: Step 7.1: Establish a self-diagnostic mechanism for the equipment to monitor the health status of components in real time; Based on the equipment operating status data collected in step 1, a fault diagnosis algorithm is used to assess the health status of each component and provide fault warnings.

[0145] Establish a sensor performance monitoring index system: for water flow sensors, monitor the stability of the output signal (increased standard deviation indicates decreased accuracy) and response time (prolonged response time indicates aging); for water quality sensors, perform periodic self-calibration during periods of no water use, and indicate the need for calibration or replacement when the measured value deviates from the benchmark value beyond the allowable range; for pressure sensors, monitor whether the output value meets expectations under different operating conditions; for humidity sensors, monitor the rationality of the output value, and indicate faults if abnormal fluctuations occur.

[0146] Establish solenoid valve performance monitoring: record the number of actions and cumulative working time, and prompt preventive replacement when the rated life exceeds 80%; regularly check the response performance, verify the execution status through water flow and pressure sensors, and prompt possible jamming or failure when the command is inconsistent with the actual situation.

[0147] Monitor the performance of wireless communication modules: Statistically analyze the signal strength, transmission success rate, and packet loss rate of the 433MHz receiver module and Wi-Fi module, and generate alarm prompts to check the device location or network environment when an anomaly occurs.

[0148] Monitor the main controller's operating status: record processor load rate, memory usage rate, and storage space occupancy rate; automatically clean up expired data or prompt for hardware upgrades when resource usage is too high; record running time and number of restarts; frequent restarts prompt for further diagnosis.

[0149] All health issues are handled according to their severity: minor issues are logged and highlighted in weekly reports; medium issues generate maintenance suggestions and push them to users; serious issues (such as critical sensor failure or solenoid valve malfunction) trigger immediate alarms and recommend stopping use or contacting after-sales service.

[0150] Step 7.2: Use a filter life prediction model to provide intelligent replacement reminders; Based on the water quality and flow data collected in step 1, a filter cartridge life prediction algorithm is used to calculate the remaining service life and provide replacement recommendations.

[0151] The lifespan of a filter cartridge is affected by the flow rate and water quality. The system records the total flow rate and indicates the end of the lifespan when it approaches the rated treatment capacity. Simultaneously, it monitors water quality trends and calculates filtration efficiency: when the influent turbidity of a new filter cartridge is 10 NTU, the effluent should be below 1 NTU (efficiency > 90%). Replacement is indicated when the effluent turbidity exceeds the standard or the efficiency falls below 70%.

[0152] A comprehensive prediction model was established, employing a weighted scoring method: when the cumulative flow reaches 90% of the rated value or the filtration efficiency drops to 70%, the lifespan score falls to the warning threshold. The remaining usage days were estimated based on the average daily water consumption and the current status.

[0153] Implement a phased reminder strategy: when 30% remains, provide a preliminary reminder that "70% of the filter cartridge has been used and it is recommended to prepare for replacement"; when 10% remains, remind users weekly that "the filter cartridge is about to expire and should be replaced as soon as possible"; and when the filter cartridge is exhausted or its performance has seriously degraded, provide an emergency reminder that "the filter cartridge has expired and should be replaced immediately" and display an indicator on the screen.

[0154] The notification message includes the filter model, purchase link, and replacement video tutorial, and offers a one-click on-site service appointment function. It records the time of each replacement and the cumulative flow rate to verify and adjust model parameters to improve prediction accuracy.

[0155] Step 7.3: Optimize model parameters based on user feedback using reinforcement learning methods; Based on the early warning execution results and user feedback data output in step 5, including early warning trigger records, user response status, and early warning accuracy assessment, reinforcement learning algorithms are used to optimize the anomaly identification model parameters and early warning strategies.

[0156] The system operation is modeled as a reinforcement learning problem: the state space includes sensor data, user behavior characteristics, historical anomaly frequency, and historical warning execution records; the action space includes warning trigger decisions, level selection, method selection, and escalation timing selection; the reward function is defined based on the warning execution results and user feedback: accurate warnings with timely user responses receive high positive rewards, accurate warnings with delayed user responses receive medium positive rewards, false alarms receive negative rewards, the magnitude of the negative reward is related to the degree of disturbance, severe false alarms incur the largest penalty, and missed alarms receive large negative rewards, which are related to potential risks. By analyzing the warning execution history recorded in step 5, we can identify which state-action combinations yielded high rewards and which led to false alarms or user neglect, thereby optimizing the decision-making strategy.

[0157] The decision-making model is trained using Q-learning or deep Q-network algorithms. By trying different decision actions and observing user feedback, it learns the optimal strategy that maximizes warning accuracy while minimizing false alarm rate. The training process employs a balance between exploration and exploitation: a high exploration rate is used initially for rapid learning, and the exploration rate is reduced later to achieve stable operation.

[0158] Specialized models are trained for different types of anomalies, and a performance monitoring mechanism is established to regularly evaluate accuracy, recall, and user satisfaction. When performance meets targets, the learning rate is reduced to stabilize the parameters; when performance deteriorates, learning is restarted to adapt to new situations. Continuous updates are achieved through online learning, realizing a truly adaptive intelligent system.

[0159] Step 7.4: Establish a remote diagnostic and maintenance channel to support after-sales service; Based on the equipment health issues detected in step 7.1 and the maintenance recommendations generated in step 7.2, efficient fault diagnosis and maintenance support are achieved using IoT remote service technology. Specifically, when the system detects equipment faults or performance anomalies, it automatically generates a fault report and uploads it to the cloud platform. The fault report includes detailed information such as a description of the fault phenomenon, the time of occurrence, relevant sensor data, and system log records.

[0160] The cloud platform's intelligent diagnostic system analyzes fault reports and matches possible causes and solutions based on a fault knowledge base. For common faults such as loose sensor connections or clogged filters, the diagnostic system provides specific troubleshooting steps and solutions, which are pushed to users via a mobile application to guide them in resolving the issues themselves. The push notifications include text or video tutorials, along with a diagram of the sensor's location and a video demonstrating how to reconnect the connector.

[0161] For complex faults that users cannot resolve themselves, the diagnostic system automatically generates an after-sales service order, including user information, equipment model, fault diagnosis results, and suggested repair solutions, and dispatches it to the nearest after-sales service center. After-sales personnel can view detailed fault information and historical maintenance records through the service system, prepare the necessary spare parts in advance, and improve the efficiency of on-site repairs.

[0162] A remote assistance function is established, allowing after-sales technicians to remotely view real-time operating data of user equipment through a cloud platform and perform remote diagnostics. With user authorization, technicians can remotely adjust equipment parameters, execute diagnostic programs, and view detailed logs to quickly locate the root cause of problems. For certain software issues, technicians can repair them remotely without on-site service, saving users time.

[0163] Establish equipment maintenance records, documenting the installation date, maintenance history, fault records, and replacement parts list for each piece of equipment. These records help after-sales personnel understand the equipment's usage status and develop targeted maintenance plans. For equipment that frequently experiences problems, special attention should be paid and dedicated technical support provided.

[0164] Step 7.5: Regularly push out model updates and feature upgrades to maintain the system's advanced nature; Based on the massive user data accumulated on the cloud platform and the improved model trained in step 7.3, remote upgrade technology is used to update the device-side algorithms and functions.

[0165] The cloud platform aggregates all user data to form a large-scale dataset, which the R&D team uses to train and improve models, optimize algorithms, and develop new features. After the new version passes small-scale testing, an upgrade notification explaining the improvements is pushed through the cloud platform. Users can choose to upgrade immediately or postpone it; the system automatically executes the upgrade late at night by default.

[0166] The upgrade uses incremental updates to reduce data transfer and time, maintaining basic functionality throughout the process and automatically restarting upon completion. A version rollback mechanism is established, allowing manual or automatic revert to a stable version if issues arise with the new version.

[0167] Regular upgrades include algorithm optimization and the addition of new features. Algorithm optimization is based on the analysis of early warning execution effectiveness, such as optimizing the anomaly detection threshold based on early warning accuracy data from a large number of users, optimizing the timing of early warning upgrades based on user response patterns, and optimizing the multi-sensor fusion weights based on false alarm types. New features include voice interaction, smart home integration, and support for new types of sensors. Through continuous remote upgrades, the system constantly evolves to maintain its technological advancement, allowing users to enjoy new features without replacing hardware and extending the lifespan of their devices.

[0168] This step outputs the system's adaptive optimization plan, including: equipment health status assessment report, filter replacement reminder, optimized model parameters, remote maintenance instructions, and system upgrade package.

[0169] In one embodiment of the present invention, a specific example is provided: Mr. Zhang, 72 years old, lives alone in a two-bedroom apartment in a certain city. He suffers from hypertension and mild diabetes, and his children work in other cities. In October 2025, the intelligent water purification system of this invention was installed in Mr. Zhang's home. His children can be linked as emergency contacts through a mobile application, allowing them to check on Mr. Zhang's water safety and health status at any time.

[0170] During system initialization, Mr. Zhang's basic information was collected: age 72, male, weight 68 kg, height 170 cm, suffering from hypertension and diabetes, and taking antihypertensive and hypoglycemic medications daily. Based on this information, the system calculated Mr. Zhang's recommended daily water intake to be 2100 ml, taking into account the need for hypertensive patients to appropriately increase water intake to promote metabolism, as well as the dry autumn climate.

[0171] The first 30 days of system operation were a learning period to establish a baseline for Mr. Zhang's personalized water usage behavior. Data collection revealed that Mr. Zhang usually used water for the first time between 6:30 and 7:00 in the morning, drank water about 6 to 8 times during the day, with each drink being between 250 and 300 milliliters, and occasionally used water once at night around 2:00 in the morning, with a total daily water consumption of about 80 liters.

[0172] Table 1 shows an example of some of Mr. Zhang's water usage data for a certain week collected by the system: Table 1: Example of Mr. Zhang's partial water usage data for a certain week;

[0173] On November 6th, the system detected that Mr. Zhang only used water for the first time at 9:15 AM, about two hours later than usual, triggering a mild abnormality of prolonged water inactivity. The system issued a gentle reminder via voice broadcast on the water purifier but did not immediately notify his children. The system continued monitoring and found that Mr. Zhang used water normally after the reminder, and the human body sensor detected normal indoor activity that day, determining that Mr. Zhang might simply have been sleeping in occasionally, and automatically removed the abnormality flag. It was later learned that Mr. Zhang had spoken to a friend late the previous night and slept in a little longer the next day; the system's judgment was accurate and avoided a false alarm.

[0174] One afternoon in mid-November, the system detected continuous water usage exceeding 50 minutes, far exceeding Mr. Zhang's usual 30-minute time for washing vegetables and cooking, triggering an abnormal prolonged water usage warning. The system immediately sent a push notification to Mr. Zhang's phone: "Prolonged continuous water usage detected. Please check if you forgot to turn off the tap." After seeing the notification, Mr. Zhang checked and found that he had indeed forgotten to completely turn off the tap after washing items, causing a continuous drip. He immediately turned it off. Mr. Zhang then clicked the "Resolved" button in the app, and the system lifted the warning. This warning helped Mr. Zhang avoid wasting water and potential property damage.

[0175] One day in early December, the system received a gas leak alarm from Mr. Zhang's gas alarm. The system immediately activated a high-alert warning, sending an emergency alert to Mr. Zhang and his children simultaneously: "Gas leak detected. Please open windows for ventilation immediately and evacuate. The water supply solenoid valve has been automatically shut off." It also automatically shut off the water purifier's solenoid valve to prevent sparks from the electrical switch from igniting the leaking gas. Upon receiving the alarm, Mr. Zhang's son immediately called his father to confirm the situation and instructed him to open windows for ventilation and contact the gas company. This coordinated response was timely and effective, preventing a potential safety accident.

[0176] Regarding drinking water health management, the system continuously monitors Mr. Zhang's actual water intake and provides reminders. Table 2 shows the weekly water intake achievement data: Table 2: Weekly water intake achievement data;

[0177] Mr. Zhang had not drunk enough water for two consecutive days, December 11th and 12th. On the evening of the 12th, the system generated a reminder about insufficient water intake: "You have not drunk enough water in the past two days. Adequate water intake helps control blood pressure and blood sugar metabolism. Please pay attention to increasing your water intake." After seeing the reminder, Mr. Zhang realized that he had indeed been busy with other things and neglected to drink water for the past two days. He then made an effort to improve his water intake, and his water intake compliance rate rebounded.

[0178] The system generates a weekly report for Mr. Zhang every Sunday, summarizing his water safety and drinking health status for the week. The report shows "Safety status was good this week, with an average drinking water compliance rate of 89%. It is recommended to strengthen drinking water intake during the afternoon." Mr. Zhang's children view the weekly report through a mobile application and are relieved to know that their father is in good overall condition. The monthly report at the end of each month provides a more comprehensive analysis, showing that Mr. Zhang's water usage patterns are stable, his drinking habits are gradually improving, and his overall health score is 85 points, which is at a good level among his peers.

[0179] After three months of use, the system successfully helped Mr. Zhang develop more regular drinking habits, increasing his compliance rate from 75% initially to over 90%. The system issued two moderate alerts, both of which were genuine abnormalities, demonstrating 100% accuracy with no false alarms. Mr. Zhang and his children are very satisfied with the system's intelligent monitoring functions, believing that it provides tangible safety and health management value without adding to their psychological burden.

[0180] As can be seen from the above application examples, the intelligent water purification system of the present invention has successfully achieved unobtrusive safety monitoring and scientific drinking water management for elderly people living alone, effectively improving the home safety and health level of the elderly, and providing an effective tool for children to provide remote care, which has significant practical value and social significance.

[0181] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An intelligent water purification system integrating safety monitoring and healthy drinking water management, characterized in that, include: The data acquisition and preprocessing module is used to acquire raw data from built-in sensors and external safety devices and preprocess it to obtain a standard multi-source heterogeneous dataset. The water use behavior modeling module identifies water use events and analyzes their temporal distribution patterns based on standard multi-source heterogeneous datasets, and builds a personalized water use behavior baseline model. The abnormal behavior identification module, based on a personalized water use behavior baseline model, identifies abnormal behaviors and integrates standard multi-source heterogeneous datasets for anomaly confirmation, and obtains abnormal behavior identification results and severity assessment reports. The healthy drinking water management module, based on a personalized water behavior baseline model, calculates the recommended daily water intake and formulates a drinking water plan, conducts achievement assessment and trend analysis, and obtains a personalized drinking water solution. The tiered early warning execution module formulates and executes tiered early warning strategies based on the abnormal behavior identification results and severity assessment reports, and obtains the early warning execution results. The visualization analysis module, based on the results of abnormal behavior identification, personalized drinking water plans, and early warning execution results, provides visual presentation and trend analysis to generate a whole-house safety trend report; The adaptive optimization module performs adaptive optimization and upgrades of the system based on standard multi-source heterogeneous datasets and early warning execution results, resulting in an adaptive optimization scheme for the system.

2. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The data acquisition and preprocessing module includes: Establish a unified data acquisition interface to encapsulate data from built-in sensors and external security devices into a unified data frame structure; The main controller is time-calibrated using the Network Time Protocol, and multi-source data is arranged by timestamp to form a time-series data stream; Establish multi-dimensional data quality assessment indicators, including completeness indicators, continuity indicators, reasonableness indicators, and consistency indicators, and use a weighted summation method to calculate the comprehensive quality score as the credibility weight; The health status of each sensor is evaluated using a sliding window statistical method, and dynamic weights are assigned to the sensors according to the health status grading standard.

3. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The water use behavior modeling module includes: A threshold segmentation algorithm is used to identify independent water use events. When the water flow rate rises from below the water flow rate threshold to above the water flow rate threshold, it is determined that the water use event has started. When the water flow rate drops from above the water flow rate threshold to below the water flow rate threshold and continues for a preset time, it is determined that the water use event has ended. Feature parameters of each water use event are extracted. The sliding window method is used to calculate statistical characteristics at multiple time scales, with the window sliding forward daily to update the statistical indicators. Extract the time of first water use in the morning, the frequency of drinking water during the day, the water use at night, and the interval between water uses to construct a feature vector of water use behavior.

4. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 3, characterized in that, The water use behavior modeling module also includes: Calculate the statistical distribution parameters for each dimension of the water use behavior feature vector. For the normal distribution feature, calculate the mean and standard deviation. For the skewed distribution feature, calculate the median, quartiles and percentiles. Establish confidence intervals for features to define the boundaries of normal behavior; The correlation coefficient is used to calculate the degree of linear correlation between feature pairs. When the absolute value of the correlation coefficient is greater than the preset correlation threshold, it is determined to be a related feature. A conditional distribution model is established for the related features. A sliding window strategy is used to periodically recalculate the parameters of the conditional distribution model. When updating the parameters, outlier data is excluded and normal data is used to update the model.

5. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The abnormal behavior identification module includes: The standardized deviation measure method is used to calculate the difference between the current water use data and the corresponding feature in the personalized water use behavior baseline model, and divide the difference by the standard deviation to obtain the standardized deviation. The degree of deviation is determined based on the standardized deviation. Define abnormal modes such as prolonged inactivity, prolonged continuous water use, abnormal water use time patterns, and insufficient drinking water. Set corresponding judgment conditions and abnormal thresholds for each abnormal mode. An anomaly is confirmed by using a multi-sensor data fusion method. Data from human body sensors, door and window sensors, temperature sensors, pressure sensors, and humidity sensors are queried to assist in the judgment. The overall confidence level of the anomaly is calculated. When the confidence level exceeds a preset threshold, the anomaly is confirmed and an early warning is triggered.

6. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 5, characterized in that, The abnormal behavior identification module also includes: Define the assessment dimensions for the severity of anomalies, including the degree of deviation, duration, and historical frequency. A scoring mechanism is used to comprehensively determine the severity. Each dimension is scored based on its quantitative value, and the weighted sum is used to obtain the comprehensive score. Based on the comprehensive score, anomalies are divided into mild, moderate, and severe anomalies. Mild anomalies correspond to Level 1 alerts, which are local notifications on the device. Moderate anomalies correspond to Level 2 alerts, which send push messages to the user's mobile device. Severe anomalies correspond to Level 3 alerts, which trigger the highest level alarm and notify all emergency contacts.

7. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The healthy drinking water management module includes: Establish user health records, calculate basic drinking water requirements using a weight-related water intake calculation method, and introduce age correction and disease correction factors; The system dynamically adjusts based on real-time environmental factors, collects outdoor temperature and indoor humidity data, and introduces temperature correction coefficients, humidity correction coefficients, and seasonal correction coefficients. The basic requirement is then multiplied by each correction coefficient to obtain the recommended total daily water intake. Develop personalized time-segmented water drinking plans, collect historical water consumption and frequency data for each time period to calculate the time period allocation ratio, set reminder points to check users' water consumption and trigger reminders; The system calculates the achievement rate by comparing the actual total water consumption with the recommended water consumption on a daily basis and establishes a trend analysis over several consecutive days. When the achievement rate is consistently lower than the preset threshold for more than a preset number of days, a serious water shortage warning report is generated.

8. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The hierarchical early warning execution module includes: Build a structured early warning strategy knowledge base, where each rule includes triggering conditions, execution actions, notification targets, and escalation mechanisms; For minor anomalies, the water purifier plays a prompt voice through its built-in voice module and sets a delayed confirmation mechanism. After the reminder, it continues to monitor water usage behavior. If the user uses water within the first preset time, the anomaly mark will be automatically removed. For moderate anomalies, a push notification is sent to the user's mobile terminal. The push message includes a quick response button and a user response timeout judgment mechanism is set. If the user does not respond within a second preset time, it will automatically be upgraded to a severe warning. For severe anomalies, a multi-channel synchronous alarm mechanism is activated to send alarm notifications to users and all emergency contacts, and to perform device linkage control according to the anomaly type.

9. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 8, characterized in that, The hierarchical early warning execution module also includes: Collect complete data on historical early warning events; train an optimized early warning model using supervised learning methods, divide the historical early warning data into training and testing sets, construct feature vectors, and use labels to represent the authenticity of the early warning data; A classification algorithm is used to learn the correspondence between feature combinations and real anomalies. The output of the early warning optimization model is the probability that an anomaly is real under given feature conditions. The early warning judgment threshold is adjusted according to the prediction results of the early warning optimization model. Regularly retrain the early warning optimization model using the latest accumulated early warning data, update the parameters, convert the optimized early warning model parameters into adjustments to the early warning strategy rules, and push them to user devices.

10. The intelligent water purification system integrating safety monitoring and healthy drinking water management according to claim 1, characterized in that, The adaptive optimization module includes: Establish a sensor performance monitoring index system to monitor the stability of the output signal and response time of the water flow sensor; Establish a solenoid valve performance monitoring system to record the number of actuations and cumulative operating time. The filter life prediction algorithm is used to record the total flow of water, monitor the trend of water quality changes, calculate the filtration efficiency, use a weighted scoring method to comprehensively evaluate the life status, estimate the remaining days of use based on the average daily water consumption and the current status, and implement a phased reminder strategy. The system operation is modeled as a reinforcement learning problem, defining the state space, action space and reward function, and the decision model is trained using reinforcement learning algorithms; Establish a remote diagnostic and maintenance channel. When a device fault is detected, a fault report is automatically generated and uploaded to the cloud platform. The cloud platform's intelligent diagnostic system matches the fault cause and solution based on the fault knowledge base.

Images