Method and system for monitoring the life state of the elderly based on multi-dimensional analysis of cooking behavior

Abstract

The application discloses a kind of old person life state monitoring method and system based on cooking behavior multidimensional analysis, it is related to health monitoring technical field, including the following steps: S1, acquisition equipment state data, and execute data cleaning and pretreatment;S2, from the data after cleaning, the working interval of each equipment is extracted, after time sequencing, equipment event is combined based on adaptive time window, when the total length of candidate event after combination exceeds preset threshold, it is split according to equipment activity gap.The old person life state monitoring method and system proposed in the application, by fine perception and analysis of old person daily cooking behavior, multi-dimensional, accurate monitoring of the life state of the elderly is realized, effectively overcome the deficiencies of traditional monitoring methods such as wearable devices needing regular charging, low acceptance of the elderly, and video monitoring invading privacy, data difficult to quantitative analysis, through non-contact data acquisition mode, the privacy of the elderly is protected.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically to a method and system for monitoring the living conditions of the elderly based on multidimensional analysis of cooking behavior. Background Technology

[0002] With the continued acceleration of population aging, home health monitoring and living status assessment for empty-nest and single-living elderly individuals have become core research directions in the field of smart elderly care. Cooking behavior is a core component of the daily home activities of the elderly. Its frequency, duration, and operation patterns can directly reflect the elderly's dietary and nutritional status, the stability of their daily rhythms, and their daily self-care ability. It is an important non-invasive monitoring indicator for predicting health risks and cognitive function changes in the elderly. Therefore, home status monitoring technology based on cooking behavior has significant application value and research significance.

[0003] Current technologies for monitoring the home life of the elderly have developed into several mature approaches, with each solution achieving corresponding monitoring functions in its respective application scenarios. However, certain limitations remain. Wearable devices can collect human activity data in real time, but they require regular charging and suffer from low compliance among the elderly. Video surveillance solutions can visually present the elderly's activity status, but they have application limitations related to privacy protection, and the data is difficult to analyze quantitatively and predict long-term trends in a standardized manner. While existing smart home devices can collect data on individual devices, they mostly operate in independent modes, resulting in relatively singular data dimensions. They lack in-depth correlation and fusion analysis of data from multiple devices, failing to establish a systematic correlation model between device operation data and health dimensions such as the elderly's dietary habits, nutritional intake, and changes in self-care ability, making it difficult to achieve early identification and accurate warning of health risks. To address this, we propose a method and system for monitoring the life status of the elderly based on multidimensional analysis of cooking behavior. Summary of the Invention

[0004] To address the aforementioned technical challenges, this paper provides a method and system for monitoring the living conditions of the elderly based on multidimensional analysis of cooking behavior. This technical solution overcomes the limitations of existing monitoring solutions, such as the cumbersome charging of wearable devices and insufficient compliance among the elderly; the privacy limitations of video surveillance and the difficulty in standardizing and quantifying data analysis; and the isolated data from individual smart home devices, which lack multidimensional and in-depth correlation analysis, making it impossible to establish a systematic model with the health dimensions of the elderly and achieve early risk warning.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for monitoring the living status of older adults based on multidimensional analysis of cooking behavior includes the following steps: S1. Collect equipment status data and perform data cleaning and preprocessing; S2. Extract the working range of each device from the cleaned data, sort by time and merge the device events based on the adaptive time window. When the total duration of the merged candidate events exceeds the preset threshold, split them according to the device activity interval. Combine the meal time period dynamically identified by DBSCAN clustering to label the merged events by meal type and output the cooking event record. S3. Calculate the nutritional intake index, dietary regularity score, and self-care ability index based on cooking event records. The nutritional intake index is calculated and normalized based on the weighted score of the equipment working time and the preset equipment-cooking method mapping weight. The dietary regularity score is calculated by mapping the standard deviation of the start time of the three meals through an exponential decay function. The self-care ability index is calculated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. S4. The nutritional intake index, dietary regularity score, and self-care ability index are weighted and integrated to generate a comprehensive status score. Based on the risk level of each module and the comprehensive score, the status monitoring results and early warning signals are output.

[0006] Preferably, in step S1, collecting device status data and performing data cleaning and preprocessing specifically involves: Complete the installation and parameter configuration of the kitchen sensing equipment according to the preset deployment rules. The infrared sensor is deployed at the entrance of the kitchen, the thermometer and hygrometer are deployed directly above the stove and the viewing angle covers the entire stove surface, the smart sockets are plugged into the power supply terminals of all kitchen electrical equipment, the refrigerator door magnet is deployed at the opening and closing of the refrigerator door, and the water sensor is deployed at the end of the kitchen water pipe. Configure the acquisition trigger threshold and output specifications for each device. The temperature trigger threshold of the thermometer and hygrometer is set to the ambient temperature plus the preset ambient temperature increment threshold. The power change trigger threshold of the smart socket is set to the preset power change threshold and the duration reaches the preset trigger duration. All device output data carries the device's unique identifier, timestamp, and status value. The gateway receives output data from all devices and performs noise reduction processing on the power data collected by the smart socket using a sliding median filter with a preset window size. Perform duration validation on all device power-on events and remove power-on events with a duration less than a preset short-term event threshold; Perform integrity checks on power data missing due to network packet loss, and use linear interpolation to complete power data with missing duration less than the preset effective completion duration threshold.

[0007] Preferably, in step S2, extracting the working range of each device specifically involves: Select the valid device data after noise reduction, elimination and completion processing as the sole input data source; For each type of device, a fixed standby power threshold is preset. A continuous interval scan is performed on the power sequence of a single device to extract a continuous interval where the power value is greater than the sum of the device's standby power and the preset working power increment threshold and the duration is greater than or equal to the preset effective working duration threshold. This interval is marked as the device's working interval. Record the unique device identifier, start time, and end time of each device's work interval to generate standardized single device events; All single-device events are sorted in ascending order by the start time of the work interval to form an ordered sequence of device events.

[0008] Preferably, in step S2, the merging and splitting of device events based on the adaptive time window specifically involves: Set the preset initial merge time window for event merging; Traverse the ordered sequence of device events, compare the start time interval between the current event and the previous event in turn. If the time interval is less than or equal to the current merging window and the unique device identifiers corresponding to the two events are different, merge the two events into the same candidate cooking event. Perform a total duration check on the merged candidate cooking events. If the total duration of the candidate cooking events exceeds the preset threshold for excessively long events, scan the equipment activity gaps within the candidate cooking events, extract the longest gap whose duration exceeds the preset threshold for splitting gaps as the splitting point, and split the candidate cooking events into multiple independent cooking events. The system collects cooking event data from users within the corresponding historical period according to a preset update cycle, calculates the duration distribution characteristics of user cooking events using the quantile method, and dynamically adjusts the value of the merging window to ensure that the total duration of the merged cooking events does not exceed the preset target event duration.

[0009] Preferably, in step S2, dynamically identifying the three meal times and completing the meal labeling specifically involves: Collect the start time data of all cooking events within the user's preset historical period, and draw a time distribution histogram with a preset time interval width; The DBSCAN clustering algorithm is set with a preset neighborhood radius and a preset minimum number of samples. Clustering is performed on the time data within the time distribution histogram to identify three core time clusters. The cluster center times of the three core time clusters are marked as the base times for the user's personalized breakfast, lunch, and dinner periods, respectively, and the cluster center times are updated by sliding at a preset update frequency. For newly registered users, the initial settings include preset breakfast, lunch, and dinner times. After the new user has been registered for a preset adaptation period, the settings will switch to personalized time slot configurations. Match the start time of an independent cooking event with the user's current mealtimes, add a corresponding meal category tag to the successfully matched cooking events, add a supplementary meal tag to the unmatched cooking events, and generate a complete cooking event record.

[0010] Preferably, in step S3, the calculation of the nutrient intake index specifically involves: A preset equipment-cooking method mapping table matches the corresponding cooking method category and preset weight coefficient for each type of equipment; Extract the working time of all devices within a single cooking event, accumulate the multiple working times of the same device according to device type, multiply the device weight coefficient by the corresponding accumulated working time, and divide by the preset time unit conversion coefficient to obtain the nutritional score of a single cooking event. Extract the nutritional scores of all cooking events of the user on the same day and sum them up to get the total nutritional score of the day. Select the average daily nutritional score of the user within the preset historical period excluding abnormal days as the baseline average score. Divide the total nutritional score of the day by the baseline average score and then multiply it by the preset normalized benchmark score. Take the smaller value between the calculated result and the normalized benchmark score to obtain the normalized nutritional intake index of the day. Perform valid meal verification on single cooking events. Events whose cooking time is less than the preset simple meal time threshold, involve the preset single device threshold, and are microwave ovens are marked as simple meals and are not counted as valid meals. Count the number of valid meals for the day. The number of unique equipment types involved in a single cooking event is counted and marked as the single-event cooking operation complexity. The average cooking operation complexity of all cooking events on the day is calculated to obtain the average cooking operation complexity of the day. Linear regression calculation is performed on the average daily cooking operation complexity within the user's preset trend statistical period to obtain the slope of the cooking complexity change. Linear regression calculation is also performed on the number of effective meals per day within the user's preset trend statistical period to obtain the slope of the meal change. Based on the daily nutrient intake index, the number of effective meals, and the slope of change in cooking complexity, the risk level of malnutrition is determined and divided into three levels: low risk, medium risk, and high risk.

[0011] Preferably, in step S3, calculating the dietary regularity score specifically involves: Extract the start time of cooking events corresponding to the user's three meals a day, convert the start time into a standardized value corresponding to the time series of that day, and exclude missing meals from the calculation of that day; Calculate the mean and standard deviation of the start times for breakfast, lunch, and dinner within the user's preset statistical period; Set preset scale parameters and use an exponential decay function with the natural constant as the base to map the standard deviation of the start time of a single meal to a single meal stability score within a preset score range; The missing rate of each meal within the user's preset statistical period is calculated. Meals with a missing rate less than or equal to the preset effective meal missing threshold are included in the comprehensive calculation. The average of the single meal stability scores of all meals included in the calculation is calculated to obtain the comprehensive dietary regularity score. Calculate the lag first-order autocorrelation coefficient of the user's daily three meal start time series, and mark the cases where the autocorrelation coefficient is less than the preset rhythm disorder threshold as rhythm disorder; Perform consecutive day statistics on cooking events occurring during user-preset nighttime non-activity periods, and mark cooking events occurring in that period as serious rhythm abnormalities when the consecutive days reach a preset abnormal duration threshold. The system performs statistical analysis on the changes in the user's comprehensive dietary regularity score over a preset long period. If the score decline exceeds the preset decline threshold or the increase in the standard deviation of the start time of a single meal exceeds the preset standard deviation growth threshold, a cognitive decline risk warning is triggered.

[0012] Preferably, in step S3, the calculation of the self-care ability index specifically involves: Cooking event data within the baseline establishment period after the user enables monitoring and presets the adaptation period are selected. The daily effective meal count, daily total cooking time, daily average cooking operation complexity, daily nutritional intake index, and daily comprehensive dietary regularity score are extracted to form a daily indicator vector. Calculate the mean and standard deviation of each indicator within the indicator vector to construct a user behavior baseline. The behavior baseline is updated according to a preset baseline update cycle. The periodic mean of each indicator for the user within the corresponding period is calculated according to the preset monitoring period. The periodic mean is compared with the behavioral baseline mean of the corresponding indicator. The absolute value of the difference between the periodic mean and the baseline mean is calculated and then divided by the baseline standard deviation of the corresponding indicator to obtain the deviation of each indicator. The deviation of each indicator is statistically analyzed for consecutive periods. Indicators whose deviation for consecutive preset periods exceeds a preset significant deviation threshold are marked as significant decline indicators. Each indicator is assigned a corresponding weight, and an indicator function is used to assign values ​​to the significant decline indicators. The comprehensive self-care ability index is obtained by subtracting the preset decline deduction coefficient from the preset baseline score and multiplying the sum of all indicator weights and corresponding indicator function values. The target value for the user's total daily cooking time is set as the baseline mean, and the allowable deviation is a multiple of the preset baseline standard deviation. The positive and negative cumulative sums of the total daily cooking time are calculated using a cumulative sum control chart. Cases where the cumulative sum exceeds the preset anomaly judgment threshold are marked as sudden anomalies. A missed meal alarm is marked when the number of valid meals for a user's consecutive preset number of days is 0; a forgotten alarm is marked when the continuous working time of a single device exceeds the preset safe time threshold; and a time surge alarm is marked when the duration of a single cooking event exceeds the user's historical preset percentile and the number of devices involved is less than the preset number of devices. Events where a user has no device activity for a preset duration during regular activity periods and simultaneously experiences missed meals are marked as suspected emergency events.

[0013] Preferably, in step S4, the weighted fusion to generate a comprehensive status score and output the monitoring results and early warning signals specifically involves: Preset corresponding weight coefficients for the nutrient intake index, the comprehensive dietary regularity score, and the comprehensive self-care ability index; The user's overall status score is obtained by performing a weighted summation of the indices of the three dimensions and their corresponding weights. The system verifies the risk level across three dimensions. If any dimension triggers a high-risk level, the user's overall status is directly downgraded to the corresponding risk level. Configure tiered push rules to push the previous day's meal overview and corresponding dimension scores according to the preset daily push time, and push the indicator trend charts and risk warnings for the corresponding period according to the preset weekly push cycle. Based on the comprehensive status score and risk level, the corresponding push channels are matched. Non-emergency risk events are pushed through the application, while emergency risk events are pushed through SMS and telephone voice. The corresponding status monitoring results and early warning signals are output simultaneously.

[0014] An elderly life status monitoring system based on multidimensional analysis of cooking behavior includes: The module includes a data acquisition and preprocessing module, a cooking event recognition module, a multi-dimensional quantitative evaluation module, and a comprehensive decision output module. The data acquisition and preprocessing module is used to deploy kitchen sensing devices, collect device status data, perform sliding median filtering on power data to remove short-term events, perform linear interpolation to complete missing data, and output a standardized device event sequence after cleaning. The cooking event recognition module is electrically connected to the data acquisition and preprocessing module. The cooking event recognition module is used to extract the working interval of each device from the device event sequence, merge and split the device events based on the adaptive time window, and label the merged events by meal type in combination with the meal time period dynamically identified by the clustering algorithm to generate cooking event records. The multi-dimensional quantitative assessment module is electrically connected to the cooking event recognition module. The multi-dimensional quantitative assessment module is used to calculate the nutritional intake index, dietary regularity score and self-care ability index based on the cooking event records. The nutritional intake index is based on the weighted normalization of the equipment-cooking mapping weight and working time. The dietary regularity score is generated based on the standard deviation of the start time of the three meals through exponential decay mapping. The self-care ability index is generated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. The comprehensive decision output module is electrically connected to the multi-dimensional quantitative assessment module. The comprehensive decision output module is used to weight and integrate the nutrient intake index, dietary regularity score and self-care ability index to generate a comprehensive status score, and push monitoring results and early warning signals according to the risk level.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a method and system for monitoring the living conditions of the elderly. By precisely sensing and analyzing the daily cooking behavior of the elderly, it achieves multi-dimensional and accurate monitoring of their living conditions. This effectively overcomes the shortcomings of traditional monitoring methods, such as wearable devices requiring regular charging and low acceptance among the elderly, as well as the privacy invasion and difficulty in quantifying and analyzing data from video surveillance. Through non-contact and seamless data collection, it protects the privacy of the elderly while improving the convenience and continuity of monitoring. The system can deeply explore subtle features in cooking behavior and comprehensively assess the living conditions of the elderly from multiple dimensions, such as nutritional intake, dietary regularity, and self-care ability, providing a scientific basis for health early warning. The system also has dynamic ability assessment and early warning functions, which can promptly detect signs of gradual decline in the living abilities of the elderly and warn of potential health risks, providing a safer and more considerate home-based elderly care environment. Attached Figure Description

[0016] Figure 1 The flowchart of the elderly living condition monitoring method of the present invention Figure 1 ; Figure 2 The flowchart of the elderly living condition monitoring method of the present invention Figure 2 ; Figure 3 This is a block diagram of the elderly living condition monitoring system of the present invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 and 2As shown, the method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior includes the following steps: S1. Collect equipment status data and perform data cleaning and preprocessing; In step S1, the process of collecting device status data and performing data cleaning and preprocessing specifically includes: Install and configure the kitchen sensing equipment according to the preset deployment rules. The infrared sensor is deployed at the kitchen entrance, 1.2 to 1.4 meters above the ground, horizontally facing the core operating area inside the kitchen. The detection angle should be no less than 90 degrees horizontally and 60 degrees vertically, covering the entire width of the kitchen entrance. The installation tolerance is ±15 degrees for horizontal angle deviation and ±0.1 meters for vertical height deviation. The thermometer and hygrometer are deployed 1.5 meters directly above the stove, with a horizontal offset of no more than 0.3 meters from the center of the stove. The detection angle should be no less than 120 degrees horizontally and 90 degrees vertically, covering the entire stove surface. The installation tolerance is ±0.2 meters for vertical height deviation. The horizontal offset shall not exceed ±0.5 meters. The smart sockets shall be plugged into the power supply terminals of all kitchen electrical appliances. When plugging in, ensure that the power supply circuit of the device passes through the metering module of the smart socket completely without bypass wiring. The installation tolerance is that the contact resistance of the power supply circuit shall not exceed 0.1 ohms. The refrigerator door magnet shall be deployed at the opening and closing of the refrigerator door. The installation gap between the door magnet body and the magnet shall not exceed 15 mm. When the door is closed, the magnet and the body shall be completely aligned. The installation tolerance is that the alignment deviation shall not exceed ±5 mm. The water sensor shall be deployed at the end of the kitchen water pipe and installed downstream of the cold and hot water pipes, covering all water points in the kitchen. The installation tolerance is that the pipe connection is sealed without leakage and the trigger flow threshold shall not exceed 0.5 liters / minute. Configure the acquisition trigger thresholds and output specifications for each device. The temperature trigger threshold for the thermometer and hygrometer is set to the ambient temperature plus a preset ambient temperature increment threshold. The typical value for the ambient temperature increment threshold is 15℃. The setting is based on the minimum stable increase in ambient temperature above the stove surface compared to the indoor base ambient temperature after the stove is ignited. The power change trigger threshold for the smart socket is set to a preset power change threshold and a duration that reaches a preset trigger duration. The typical value for the power change threshold is 20W, and the typical value for the trigger duration is 5 seconds. The setting is based on the minimum power change amplitude and the shortest stable working duration when commonly used kitchen electrical equipment is started normally. All device output data carries a unique device identifier, timestamp, and status value. The unique device identifier adopts a 16-bit fixed-length encoding rule. The first 2 bits are the device type code, the middle 6 bits are the gateway unique code, and the last 8 bits are the device serial number, ensuring that each device has a globally unique identity within the system. The gateway receives output data from all devices and performs noise reduction processing on the power data collected by the smart socket using a sliding median filter with a preset window size. The typical window size of the sliding median filter is 5 sampling points. The principle for determining the window size is to match the typical sampling frequency of 1Hz of the smart socket. The 5 sampling points correspond to a 5-second time window. This can effectively filter out instantaneous power anomalies caused by grid spikes and electromagnetic interference while fully preserving the power step changes caused by normal device start-up and shutdown. The adjustable range of the window size is 3 to 7 sampling points, and the adjustment constraint is that the time length corresponding to the window does not exceed 10 seconds. The system performs duration validation on all device activation events, eliminating activation events with a duration less than a preset short-term event threshold. The typical value for the short-term event threshold is 30 seconds, set based on the shortest working time of kitchen electrical equipment under normal cooking scenarios. This effectively filters invalid events caused by non-cooking operations such as accidental user touch of device switches or momentary power-on self-tests. The threshold can be adjusted from 20 to 60 seconds, with the adjustment constraint being that it does not exceed 50% of the minimum effective working time of the device. For power data missing due to network packet loss, integrity checks are performed. For power data with a missing duration less than a preset effective completion duration threshold, linear interpolation is used to complete the data. The typical effective completion duration threshold is 5 minutes. The applicable boundary conditions for linear interpolation are that there are at least 3 consecutive valid power data points before and after the missing period, and there are no device on / off event trigger records during the missing period. The judgment logic for data integrity checks is as follows: first, the percentage of valid data for a single device within a single day is checked. If the percentage of valid data is not less than 95%, the data is considered complete. If the percentage of valid data is less than 95% but more than 80%, linear interpolation is performed on the missing segments that meet the applicable boundary conditions. If the percentage of valid data is less than 80%, the data for that day is considered invalid and is not included in the subsequent analysis process.

[0019] S2. Extract the working range of each device from the cleaned data, sort by time and merge the device events based on the adaptive time window. When the total duration of the merged candidate events exceeds the preset threshold, split them according to the device activity interval. Combine the meal time period dynamically identified by DBSCAN clustering to label the merged events by meal type and output the cooking event record. In step S2, the extraction of the working range of each device specifically involves: Select the valid device data after noise reduction, elimination and completion processing as the sole input data source; A fixed standby power threshold is preset for each type of device. A continuous interval scan is performed on the power sequence of a single device to extract continuous intervals where the power value is greater than the sum of the device's standby power and the preset working power increment threshold and the duration is greater than or equal to the preset effective working time threshold. These intervals are marked as the device's working intervals. Typical values ​​for the fixed standby power thresholds for each type of device are as follows: rice cooker 2W, microwave oven 1W, oven 3W, range hood 1.5W, electric kettle 1W, and electric slow cooker 2W. The physical meaning of the standby power threshold is the maximum no-load power of the device in the power-on standby state. The typical value for the working power increment threshold is 10W, which means the minimum power increase required for the device to enter the working state from the standby state. The typical value for the effective working time threshold is 30 seconds, which means the shortest continuous working time required for the device to complete a single effective cooking operation. Record the unique device identifier, start time, and end time of each device's work interval to generate standardized single device events; All single-device events are sorted in ascending order by the start time of the work interval to form an ordered sequence of device events.

[0020] In step S2, the merging and splitting of device events based on the adaptive time window specifically involves: Set a preset initial merging time window for event merging. The typical value of the initial merging time window is 30 minutes. The method for determining the initial merging window is based on the statistical results of the maximum time interval between different device start-up operations in a single cooking behavior of a healthy elderly person. It can cover the entire time interval of food preparation, cooking, and activation of auxiliary equipment in a single cooking process. Traverse the ordered sequence of device events, compare the start time interval between the current event and the previous event in turn. If the time interval is less than or equal to the current merging window and the unique device identifiers corresponding to the two events are different, merge the two events into the same candidate cooking event. The total duration of the merged candidate cooking events is checked. If the total duration of the candidate cooking events exceeds the preset threshold for excessively long events, the device activity gaps within the candidate cooking events are scanned. The longest gap with a duration exceeding the preset threshold for splitting gaps is extracted as the splitting point, and the candidate cooking events are split into multiple independent cooking events. The typical value of the threshold for excessively long events is 3 hours, which is set based on the maximum statistical upper limit of the maximum duration of a single complete cooking behavior of a healthy elderly person. The typical value of the threshold for splitting gaps is 45 minutes. The calculation logic of the threshold for splitting gaps is based on the distribution of the time interval between two adjacent device operations in the user's historical cooking events. The time interval corresponding to the 95th percentile is taken as the dynamic adjustment value of the threshold for splitting gaps. The basic threshold is not less than 30 minutes and not more than 60 minutes. Data on cooking events from the user's historical period is collected according to a preset update cycle. The duration distribution characteristics of the user's cooking events are calculated using the quantile method. The value of the merging window is dynamically adjusted so that the total duration of the merged cooking events of a preset proportion does not exceed the preset target event duration. The typical value of the update cycle is 1 month, and the value of the historical period is consistent with the update cycle. The typical value of the preset proportion is 90%, and the typical value of the target event duration is 2 hours. The setting is based on the upper limit of the average duration of a single cooking behavior of healthy elderly people. The specific implementation steps of the quantile method are as follows: First, the start time interval data of all adjacent device events in the historical period are collected. The interval data is sorted in ascending order, and the time interval value corresponding to the 90th quantile is calculated. This value is used as the dynamic adjustment value of the merging window. The adjustment constraint is that the minimum value of the merging window is not less than 15 minutes, and the maximum value is not more than 45 minutes.

[0021] In step S2, dynamically identifying meal times and labeling meal types specifically involves: Collect the start time data of all cooking events within the user's preset historical period, and draw a time distribution histogram with a preset time interval width. The typical value of the historical period is 30 days, and the typical value of the time interval width is 15 minutes. The principle for determining the interval width is to match the smallest fluctuation unit of the start time of the elderly's cooking behavior, while ensuring that the 30 days of historical data can form a sufficient sample density on the 24-hour time axis of a single day. The adjustable range of the interval width is 10 minutes to 30 minutes, and the adjustment constraint is that the total number of intervals on the time axis of a single day is not less than 48 and not more than 144. The DBSCAN clustering algorithm is configured with a preset neighborhood radius and a preset minimum number of samples. Clustering is performed on the time data within the time distribution histogram to identify three core time clusters. The typical value for the neighborhood radius is 45 minutes, and the typical value for the minimum number of samples is 5. The parameters are selected as follows: the neighborhood radius matches the normal fluctuation range of the start time of single meal cooking for the elderly; the minimum number of samples ensures that the clusters formed by the clustering are statistically significant; the adjustable range of the neighborhood radius is 30 to 60 minutes, and the adjustable range of the minimum number of samples is 3 to 7. The adjustment constraint is to stably identify three core time clusters corresponding to three meals. The cluster center times of the three core time clusters are marked as the base times for the user's personalized breakfast, lunch and dinner periods, respectively. The cluster center times are updated by sliding at a preset frequency. The typical update frequency is daily. The specific algorithm for the sliding update of the cluster center times is as follows: take the start time of all cooking events for the corresponding meal in the previous 30 days, convert it into minutes for the day, calculate the arithmetic mean, and use this mean as the updated cluster center time. At the same time, based on the cluster center time, extend it by 45 minutes before and after as the effective time range for the corresponding meal. For newly registered users, there are initial preset breakfast, lunch, and dinner times. After the new user's registration period reaches a preset adaptation period, the time slots are switched to personalized settings. The preset values ​​for the initial breakfast time are 5:00-9:00, the initial lunch time is 11:00-14:00, and the initial dinner time is 17:00-20:00. The typical length of the adaptation period is 7 days. The setting is based on the fact that the cooking behavior data of new users over 7 days can form a statistically significant time distribution, supporting the clustering and identification of personalized meal times. Match the start time of an independent cooking event with the user's current mealtimes, add a corresponding meal category tag to the successfully matched cooking events, add a supplementary meal tag to the unmatched cooking events, and generate a complete cooking event record.

[0022] S3. Calculate the nutritional intake index, dietary regularity score, and self-care ability index based on cooking event records. The nutritional intake index is calculated and normalized based on the weighted score of the equipment working time and the preset equipment-cooking method mapping weight. The dietary regularity score is calculated by mapping the standard deviation of the start time of the three meals through an exponential decay function. The self-care ability index is calculated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. In step S3, the calculation of the nutrient intake index is specifically as follows: A preset device-cooking method mapping table matches each type of device with a corresponding cooking method category and a preset weight coefficient. The specific contents of the device-cooking mapping table are as follows: device type rice cooker corresponds to food category staple food, weight coefficient 1; device type stove corresponds to food category stir-fry / stew, weight coefficient 1.5; device type microwave oven corresponds to food category heating / simple meals, weight coefficient 0.5; device type oven / air fryer corresponds to food category baking / roasting, weight coefficient 2; device type range hood corresponds to food category auxiliary cooking, weight coefficient 0.2. The weight coefficient is assigned based on the cooking operation complexity, food processing depth, and nutritional diversity contribution of different devices. The higher the processing complexity and the more diverse the types of food involved, the higher the corresponding weight coefficient. Extract the working time of all devices within a single cooking event, accumulate the multiple working times of the same device according to device type, multiply the device weight coefficient by the corresponding accumulated working time, and then divide by the preset time unit conversion coefficient to obtain the nutritional score for the single cooking event. The corresponding calculation formula is as follows: in, Nutritional value per cooking event. The weighting coefficients for the corresponding devices. The cumulative working time of the corresponding equipment during this cooking event is expressed in minutes. The conversion factor for the time unit is defined as 10 minutes, corresponding to the denominator 10 in the formula. The basis for setting this is to quantify the contribution of the equipment working time to the nutritional points using 10 minutes as the basic unit, so as to avoid excessive accumulation of nutritional points due to excessive time. The nutritional scores of all cooking events performed by the user on that day are extracted and accumulated to obtain the total nutritional score for that day. The average daily nutritional score within the user's preset historical period, excluding abnormal days, is selected as the baseline average score. The total nutritional score for that day is divided by the baseline average score and then multiplied by a preset normalized benchmark score. The smaller value between the calculated result and the normalized benchmark score is taken to obtain the normalized daily nutritional intake index. The corresponding calculation formula is as follows: in, The daily nutrition intake index is selected based on a normalized baseline score of 100, corresponding to a standardized scoring range of 0-100. This facilitates the integration of multi-dimensional indicators and risk level determination. The typical historical period is 30 days. The criteria for determining an abnormal day is that the total nutrition score of the day deviates from the historical daily average nutrition score by more than 3 standard deviations, which can eliminate interference from occasional abnormal scenarios such as going out or being sick. For single cooking events, a valid meal check is performed. Events whose cooking time is less than the preset simple meal time threshold, involve the preset single device threshold, and are microwave ovens are marked as simple meals and are not counted as valid meals. The number of valid meals for the day is counted. The typical value of the simple meal time threshold is 10 minutes, and the single device threshold is 1. The criteria for judging simple meals are that they must meet three conditions at the same time: the total event time is less than 10 minutes, the event involves only 1 electrical device, and the device is a microwave oven. The basis for setting this is that such events usually correspond to simple operations such as reheating leftovers and do not contribute to the nutritional intake of a complete meal. Not counting them as valid meals can avoid misjudgment of meal integrity. The number of unique equipment types involved in a single cooking event is counted and marked as the single-event cooking operation complexity. The average cooking operation complexity of all cooking events on the day is calculated to obtain the average cooking operation complexity of the day. Linear regression calculation is performed on the average daily cooking operation complexity within the user-preset trend statistical period to obtain the slope of cooking complexity change. Linear regression calculation is also performed on the number of effective meals per day within the user-preset trend statistical period to obtain the slope of meal change. The typical value of the trend statistical period is 7 days. Based on the daily nutrient intake index, the number of effective meals, and the slope of change in cooking complexity, a malnutrition risk level is determined, categorized into three levels: low risk, medium risk, and high risk. Specifically, a low-risk level requires a nutrient intake index greater than or equal to 80 and a daily effective meal count greater than or equal to 2; a medium-risk level requires a nutrient intake index between 50 and 80, or a daily effective meal count less than or equal to 1 for three consecutive days; and a high-risk level requires a nutrient intake index less than 50 and a daily effective meal count of 0 for two consecutive days, or a cooking complexity slope less than -0.1 and a continuous decrease for five consecutive days. The highest risk level is used for risk level determination; if multiple risk level conditions are met, the highest risk level is applied.

[0023] In S3, the calculation of the dietary regularity score is specifically as follows: Extract the start time of cooking events corresponding to the user's three meals a day, convert the start time into a standardized value corresponding to the time series of the day. The standardized value is the number of minutes from 0 to 1440 minutes of the day. Missing meals on the day are not included in the calculation range of the day. The mean and standard deviation of the start time for breakfast, lunch and dinner within the user's preset statistical period are calculated respectively. The typical statistical period is 7 days, which can adapt to the short-term fluctuation characteristics of the elderly's daily routine, and at the same time has a sufficient statistical sample size. By setting preset scaling parameters and employing an exponential decay function with a base of the natural constant, the standard deviation of the start time of a single meal is mapped to a single meal stability score within a preset score range. The corresponding calculation formula is as follows: in, To score the stability of a single meal, This is a scale parameter, typically with a value of 0.02. The standard deviation of the start time of the corresponding meal is expressed in minutes. The method for determining the scale parameter is as follows: when the standard deviation is 50 minutes, the single meal stability score is approximately 36.8 points. To match the correspondence between the fluctuation of the elderly’s daily routine and the stability of their rhythm, the standard deviation value from 0 to infinity can be mapped to the standardized score range of 0-100. The adjustable range of the scale parameter is 0.01 to 0.03. The adjustment constraint is that the score corresponding to a standard deviation of 30 minutes is not lower than 70 points, and the score corresponding to a standard deviation of 60 minutes is not higher than 40 points. The missing rate of each meal within the user's preset statistical period is calculated. Meals with a missing rate less than or equal to the preset valid meal missing threshold are included in the comprehensive calculation. The average of the single-meal stability scores of all meals included in the calculation is then used to calculate the comprehensive dietary regularity score. The corresponding calculation formula is as follows: in, To assess the overall regularity of eating habits, To determine the number of meals to be included in the calculation, the typical threshold for missing effective meals is 30%. This threshold is set based on the fact that when the missing meal rate exceeds 30%, the time data for that meal is not statistically significant, and including it in the calculation will lead to a bias in the regularity score. The lagged first-order autocorrelation coefficient of the user's daily three meal start time series is calculated. If the autocorrelation coefficient is less than the preset rhythm disorder threshold, it is marked as rhythm disorder. The typical value of the rhythm disorder threshold is 0.5. The basis for setting it is that when the autocorrelation coefficient is less than 0.5, it indicates that there is no significant correlation between the three meal times of two adjacent days, and the user's circadian rhythm shows obvious disordered fluctuations. The system performs consecutive days statistics on cooking events that occur during the user's preset non-activity hours at night. If the number of consecutive days reaches the preset threshold for abnormal duration, the cooking events in that period are marked as a serious circadian rhythm abnormality. The preset value for the non-activity hours at night is 22:00-4:00 the next day. The typical value for the threshold for abnormal duration is 3 days. The basis for setting this value is that if cooking behavior occurs for 3 consecutive days during the non-activity hours at night, it indicates that the user's day and night sleep is seriously reversed, which is a high-risk circadian rhythm abnormality. The system performs statistical analysis on the changes in the user's comprehensive dietary regularity score over a preset long period. If the score decline exceeds the preset decline threshold or the increase in the standard deviation of the start time of a single meal exceeds the preset standard deviation growth threshold within a unit period, a cognitive decline risk warning is triggered. The typical long period is 4 weeks, the typical decline threshold is 15%, and the typical standard deviation growth threshold is 10 minutes per week. The settings are based on existing research showing that early signs of cognitive decline include progressive disruption of eating and sleeping rhythms. A continuous decline of more than 15% in the regularity score over 4 consecutive weeks or a sustained increase in the standard deviation can effectively identify the early risk of cognitive decline.

[0024] In S3, the self-care ability index is calculated as follows: Cooking event data within the baseline establishment period after the user enables monitoring and the preset adaptation period are selected. The number of effective meals per day, total cooking time per day, average cooking operation complexity per day, daily nutritional intake index, and daily comprehensive dietary regularity score are extracted to form a daily indicator vector. The typical value for the adaptation period is 1 week, and the typical value for the baseline establishment period is 2 weeks. That is, the data from the 2nd to 4th week after the user registers is used as the baseline establishment data source. The basis for this setting is that the user can complete the adaptation to the system in 1 week, and the stable data in the following 2 weeks can reflect the user's normal life behavior characteristics and have statistical significance for establishing a behavioral baseline. The mean and standard deviation of each indicator in the indicator vector are calculated separately to construct a user behavior baseline. The behavior baseline is updated according to a preset baseline update cycle. The typical value of the baseline update cycle is one quarter. The update constraint is that the baseline update can only be performed when the user's indicator data is stable for four consecutive weeks without significant decline or abnormal fluctuations, and after confirmation by the user's family. This avoids the risk of missed detection caused by baseline updates after user behavior decline. The periodic mean of each indicator for the user within the corresponding period is calculated according to the preset monitoring period. The periodic mean is compared with the behavioral baseline mean of the corresponding indicator. The absolute value of the difference between the periodic mean and the baseline mean is calculated and then divided by the baseline standard deviation of the corresponding indicator to obtain the deviation of each indicator. The typical value of the monitoring period is 1 week. The deviation of each indicator is statistically analyzed for consecutive periods. Indicators with a deviation greater than a preset significant deviation threshold for consecutive periods are marked as significant decline indicators. The typical value for the preset consecutive periods is 3 weeks, and the typical value for the significant deviation threshold is 2. The basis for setting this value is that a deviation greater than 2 indicates that the indicator value deviates from the baseline mean by more than 2 standard deviations, which is a statistically significant abnormality. If this abnormality occurs for 3 consecutive weeks, it can be determined that the behavioral ability of this dimension has gradually declined. Each indicator is assigned a corresponding weight. Indicator functions are used to assign values ​​to indicators of significant decline. The comprehensive self-care ability index is obtained by subtracting the preset decline deduction coefficient from the preset baseline score, multiplying it by the sum of all indicator weights and their corresponding indicator function values, and then calculating the index using the following formula: in, To comprehensively assess self-care ability, Let be the weight coefficient for the i-th indicator. By default, equal weighting is used, meaning each indicator has a weight coefficient of 0.2. The weight coefficients are allocated based on the equal contribution of each indicator to the elderly person's daily living abilities. These weight coefficients can be personalized according to the user's physical condition and care needs, with the constraint that the sum of all weight coefficients is 1. Let be the deviation of the i-th indicator. This is an indicative function. It takes a value of 1 when the condition in parentheses is met and a value of 0 when it is not met. The baseline score is 100, and the decline deduction coefficient is 10. The setting is based on the deduction of 10 points corresponding to each significant decline indicator, which can intuitively reflect the degree of decline in the user's self-care ability. The target value for the user's total daily cooking time is set as the baseline mean, with an allowable deviation of a preset multiple of the baseline standard deviation. A cumulative sum control chart is used to calculate the positive and negative cumulative sums of the daily total cooking time. Cases where the cumulative sum exceeds a preset anomaly threshold are marked as sudden anomalies. The corresponding calculation formula is as follows: in, Let be the positive cumulative sum on day t. For the negative cumulative sum of day t, Let be the total cooking time on day t. The target value is the baseline mean of the total cooking time. To allow for offset, the typical value is 0.5 times the baseline standard deviation. The setting is based on the ability to effectively capture offsets greater than 0.5 times the process mean, balancing detection sensitivity and false alarm rate. The typical value for the anomaly judgment threshold h is 5 times the baseline standard deviation. When the positive cumulative sum or negative cumulative sum exceeds h, it is judged as a sudden anomaly. A positive anomaly corresponds to a sudden increase in cooking time, and a negative anomaly corresponds to a sudden decrease in cooking time. A missed meal alarm is triggered when the number of valid meals for a user is 0 for a consecutive preset number of days. A forgotten alarm is triggered when the continuous working time of a single device exceeds the preset safe time threshold. A time surge alarm is triggered when the duration of a single cooking event exceeds the user's historical preset quantile and the number of devices involved is less than the preset device number threshold. The consecutive days threshold for missed meal alarms is 2 days. The basis for this is that two consecutive days without valid meals indicate that the user may not be able to complete cooking independently, posing a high health risk. The typical value for the safe time threshold is 2 hours. The basis for this is the longest continuous working time of commonly used kitchen equipment during normal cooking. Exceeding this time is likely because the user forgot to turn off the equipment, posing a safety hazard such as fire. The typical value for the historical preset quantile is the 95th quantile. The device number threshold is 2. The basis for this is that when the cooking time of a single event exceeds 95% of the historical time and only one device is used, it indicates that the user may have signs of declining self-care ability, such as difficulty in operation and slow movements. An event where a user has no activity from any device for a preset duration during a regular activity period and simultaneously misses a meal is marked as a suspected emergency event. The regular activity period is defined as the core activity interval corresponding to the user's personalized three meal times, usually from 6:00 to 20:00 per day. The typical value for the preset duration is 3 hours. The trigger condition for a suspected emergency event is that there is no activity record of any kitchen equipment for 3 consecutive hours during the regular activity period, and simultaneously there is a situation where there are 0 valid meals for more than 1 consecutive day.

[0025] S4. The nutritional intake index, dietary regularity score, and self-care ability index are weighted and integrated to generate a comprehensive status score. Based on the risk level of each module and the comprehensive score, the status monitoring results and early warning signals are output.

[0026] In step S4, the weighted fusion is used to generate a comprehensive status score and output the monitoring results and early warning signals, specifically as follows: Preset corresponding weight coefficients for the nutrient intake index, the comprehensive dietary regularity score, and the comprehensive self-care ability index; The user's overall status score is obtained by weighted summation of the indices from the three dimensions and their corresponding weights. The corresponding calculation formula is as follows: in, The user's overall status is scored, and the weight coefficients of each dimension are allocated as follows: the self-care ability index corresponds to the core living abilities and emergency safety risks of the elderly, with the highest weight of 0.4; the nutrition intake index and the diet regularity score correspond to the health status and early decline risks of the elderly, with a weight of 0.3 each. The weight coefficients can be adjusted in a personalized manner according to the user's care needs, with the adjustment constraint that the sum of all weight coefficients is 1. The risk level of the three dimensions is checked. When any dimension triggers a high risk level, the user's overall status is directly downgraded to the corresponding risk level. The judgment rule for the overall status downgrade triggered by the high risk level is that, regardless of the calculation result of the overall status score, as long as any of the three dimensions triggers a high risk level, the overall status is directly judged as high risk, ensuring priority response to high risk events. Configure tiered push rules to push the previous day's meal overview and corresponding dimension scores according to the preset daily push time, and push the indicator trend charts and risk warnings for the corresponding period according to the preset weekly push cycle. The preset daily push time is 8:00 a.m. every day, and the preset weekly push cycle is every Monday, pushing the statistical data of the previous week. Based on the comprehensive status score and risk level, the corresponding push channels are matched. Non-emergency risk events are pushed through the application, while emergency risk events are pushed through SMS and telephone voice. The corresponding status monitoring results and warning signals are output simultaneously. The channel matching strategy for the tiered push is as follows: daily and weekly reports are pushed only through the mobile application of the user's family members; medium-risk alerts are pushed through the application and SMS reminders are sent simultaneously; high-risk non-emergency events are pushed through both the application and SMS; emergency risk events include device forgetting alarms and suspected emergency events, which are first pushed through the application and SMS. If there is no confirmation response within 15 minutes, a telephone voice push is automatically triggered.

[0027] Reference Figure 3 As shown, the elderly living status monitoring system based on multidimensional analysis of cooking behavior includes: The module includes a data acquisition and preprocessing module, a cooking event recognition module, a multi-dimensional quantitative evaluation module, and a comprehensive decision output module. The data acquisition and preprocessing module is used to deploy kitchen sensing devices, collect device status data, perform sliding median filtering on power data to remove short-term events, perform linear interpolation to complete missing data, and output a standardized device event sequence after cleaning. The cooking event recognition module is electrically connected to the data acquisition and preprocessing module. The cooking event recognition module is used to extract the working interval of each device from the device event sequence, merge and split the device events based on the adaptive time window, and label the merged events by meal type in combination with the meal time period dynamically identified by the clustering algorithm to generate cooking event records. The multi-dimensional quantitative assessment module is electrically connected to the cooking event recognition module. The multi-dimensional quantitative assessment module is used to calculate the nutritional intake index, dietary regularity score and self-care ability index based on the cooking event records. The nutritional intake index is based on the weighted normalization of the equipment-cooking mapping weight and working time. The dietary regularity score is generated based on the standard deviation of the start time of the three meals through exponential decay mapping. The self-care ability index is generated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. The comprehensive decision output module is electrically connected to the multi-dimensional quantitative assessment module. The comprehensive decision output module is used to weight and integrate the nutrient intake index, dietary regularity score and self-care ability index to generate a comprehensive status score, and push monitoring results and early warning signals according to the risk level.

[0028] The data interaction between the various modules of the system adopts the MQTT IoT communication protocol, supports TLS 1.3 encrypted transmission, and the heartbeat cycle of data interaction between modules is 30 seconds to ensure real-time monitoring of the module connection status.

[0029] The data storage and backup mechanism adopts a combination of local gateway caching and cloud server distributed storage. The local gateway caches the raw collected data of the most recent 7 days, while the cloud server stores the full historical data. The cloud data adopts multi-copy redundant storage, performs full backups every day, and retains the backup data for no less than 6 months.

[0030] User privacy protection measures include data anonymization, separate storage of all user personal identity information from device-collected data, device data only carries the device's unique identifier and is not directly associated with users' sensitive personal information, data access is subject to hierarchical permission control, and all data access operations are logged.

[0031] The system's self-check and anomaly recovery mechanism is as follows: each module performs a status self-check once per hour; real-time alarms are issued for issues such as device offline, data interruption, and module communication anomalies; local data caching caused by network interruption is automatically resumed after network recovery; and automatic restart recovery is performed for module malfunctions.

[0032] In actual deployment, all preset parameters can be personalized according to the user's behavioral characteristics. The adjustment method is to perform adaptive optimization on the parameters based on the statistical distribution characteristics of the user's historical behavior data, or to be manually adjusted by authorized users through the application. The constraint condition for parameter adjustment is that the adjustment of all parameters must not exceed the adjustable range of the corresponding indicator, and must not change the core analysis logic and risk judgment principle of the system.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for monitoring the life state of the elderly based on multi-dimensional analysis of cooking behavior, characterized in that, Includes the following steps: S1. Collect equipment status data and perform data cleaning and preprocessing; S2. Extract the working range of each device from the cleaned data, sort by time and merge the device events based on the adaptive time window. When the total duration of the merged candidate events exceeds the preset threshold, split them according to the device activity interval. Combine the meal time period dynamically identified by DBSCAN clustering to label the merged events by meal type and output the cooking event record. S3. Calculate the nutritional intake index, dietary regularity score, and self-care ability index based on cooking event records. The nutritional intake index is calculated and normalized based on the weighted score of the equipment working time and the preset equipment-cooking method mapping weight. The dietary regularity score is calculated by mapping the standard deviation of the start time of the three meals through an exponential decay function. The self-care ability index is calculated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. S4. The nutritional intake index, dietary regularity score, and self-care ability index are weighted and integrated to generate a comprehensive status score. Based on the risk level of each module and the comprehensive score, the status monitoring results and early warning signals are output.

2. The method for monitoring the life state of the elderly based on multi-dimensional analysis of cooking behavior according to claim 1, characterized in that, In step S1, the process of collecting device status data and performing data cleaning and preprocessing specifically includes: Complete the installation and parameter configuration of the kitchen sensing equipment according to the preset deployment rules. The infrared sensor is deployed at the entrance of the kitchen, the thermometer and hygrometer are deployed directly above the stove and the viewing angle covers the entire stove surface, the smart sockets are plugged into the power supply terminals of all kitchen electrical equipment, the refrigerator door magnet is deployed at the opening and closing of the refrigerator door, and the water sensor is deployed at the end of the kitchen water pipe. Configure the acquisition trigger threshold and output specifications for each device. The temperature trigger threshold of the thermometer and hygrometer is set to the ambient temperature plus the preset ambient temperature increment threshold. The power change trigger threshold of the smart socket is set to the preset power change threshold and the duration reaches the preset trigger duration. All device output data carries the device's unique identifier, timestamp, and status value. The gateway receives output data from all devices, and the power data collected by the smart socket is denoised using a sliding median filter with a preset window size. Perform duration validation on all device power-on events and remove power-on events with a duration less than a preset short-term event threshold; Perform integrity checks on power data missing due to network packet loss, and use linear interpolation to complete power data with missing duration less than the preset effective completion duration threshold.

3. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 2, characterized in that, In step S2, the extraction of the working range of each device specifically involves: Select the valid device data after noise reduction, elimination and completion processing as the sole input data source; For each type of device, a fixed standby power threshold is preset. A continuous interval scan is performed on the power sequence of a single device to extract a continuous interval where the power value is greater than the sum of the device's standby power and the preset working power increment threshold and the duration is greater than or equal to the preset effective working duration threshold. This interval is marked as the device's working interval. Record the unique device identifier, start time, and end time of each device's work interval to generate standardized single device events; All single-device events are sorted in ascending order by the start time of the work interval to form an ordered sequence of device events.

4. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 3, characterized in that, In step S2, the merging and splitting of device events based on the adaptive time window specifically involves: Set the preset initial merge time window for event merging; Traverse the ordered sequence of device events, compare the start time interval between the current event and the previous event in turn. If the time interval is less than or equal to the current merging window and the unique device identifiers corresponding to the two events are different, merge the two events into the same candidate cooking event. Perform a total duration check on the merged candidate cooking events. If the total duration of the candidate cooking events exceeds the preset threshold for excessively long events, scan the equipment activity gaps within the candidate cooking events, extract the longest gap whose duration exceeds the preset threshold for splitting gaps as the splitting point, and split the candidate cooking events into multiple independent cooking events. The system collects cooking event data from users within the corresponding historical period according to a preset update cycle, calculates the duration distribution characteristics of user cooking events using the quantile method, and dynamically adjusts the value of the merging window to ensure that the total duration of the merged cooking events does not exceed the preset target event duration.

5. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 4, characterized in that, In step S2, dynamically identifying meal times and labeling meal types specifically involves: Collect the start time data of all cooking events within the user's preset historical period, and draw a time distribution histogram with a preset time interval width; The DBSCAN clustering algorithm is set with a preset neighborhood radius and a preset minimum number of samples. Clustering is performed on the time data within the time distribution histogram to identify three core time clusters. The cluster center times of the three core time clusters are marked as the base times for the user's personalized breakfast, lunch, and dinner periods, respectively, and the cluster center times are updated by sliding at a preset update frequency. For newly registered users, the initial settings include preset breakfast, lunch, and dinner times. After the new user has been registered for a preset adaptation period, the settings will switch to personalized time slot configurations. Match the start time of an independent cooking event with the user's current mealtimes, add a corresponding meal category tag to the successfully matched cooking events, add a supplementary meal tag to the unmatched cooking events, and generate a complete cooking event record.

6. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 5, characterized in that, In step S3, the calculation of the nutrient intake index is specifically as follows: A preset equipment-cooking method mapping table matches the corresponding cooking method category and preset weight coefficient for each type of equipment; Extract the working time of all devices within a single cooking event, accumulate the multiple working times of the same device according to device type, multiply the device weight coefficient by the corresponding accumulated working time, and divide by the preset time unit conversion coefficient to obtain the nutritional score of a single cooking event. Extract the nutritional scores of all cooking events of the user on the same day and sum them up to get the total nutritional score of the day. Select the average daily nutritional score of the user within the preset historical period excluding abnormal days as the baseline average score. Divide the total nutritional score of the day by the baseline average score and then multiply it by the preset normalized benchmark score. Take the smaller value between the calculated result and the normalized benchmark score to obtain the normalized nutritional intake index of the day. Perform valid meal verification on single cooking events. Events whose cooking time is less than the preset simple meal time threshold, involve the preset single device threshold, and are microwave ovens are marked as simple meals and are not counted as valid meals. Count the number of valid meals for the day. The number of unique equipment types involved in a single cooking event is counted and marked as the single-event cooking operation complexity. The average cooking operation complexity of all cooking events on the day is calculated to obtain the average cooking operation complexity of the day. Linear regression calculation is performed on the average daily cooking operation complexity within the user's preset trend statistical period to obtain the slope of the cooking complexity change. Linear regression calculation is also performed on the number of effective meals per day within the user's preset trend statistical period to obtain the slope of the meal change. Based on the daily nutrient intake index, the number of effective meals, and the slope of change in cooking complexity, the risk level of malnutrition is determined and divided into three levels: low risk, medium risk, and high risk.

7. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 6, characterized in that, In S3, the calculation of the dietary regularity score is specifically as follows: Extract the start time of cooking events corresponding to the user's three meals a day, convert the start time into a standardized value corresponding to the time series of that day, and exclude missing meals from the calculation of that day; Calculate the mean and standard deviation of the start times for breakfast, lunch, and dinner within the user's preset statistical period; Set preset scale parameters and use an exponential decay function with the natural constant as the base to map the standard deviation of the start time of a single meal to a single meal stability score within a preset score range; The missing rate of each meal within the user's preset statistical period is calculated. Meals with a missing rate less than or equal to the preset effective meal missing threshold are included in the comprehensive calculation. The average of the single meal stability scores of all meals included in the calculation is calculated to obtain the comprehensive dietary regularity score. Calculate the lag first-order autocorrelation coefficient of the user's daily three meal start time series, and mark the cases where the autocorrelation coefficient is less than the preset rhythm disorder threshold as rhythm disorder; Perform consecutive day statistics on cooking events occurring during user-preset nighttime non-activity periods, and mark cooking events occurring in that period as serious rhythm abnormalities when the consecutive days reach a preset abnormal duration threshold. The system performs statistical analysis on the changes in the user's comprehensive dietary regularity score over a preset long period. If the score decline exceeds the preset decline threshold or the increase in the standard deviation of the start time of a single meal exceeds the preset standard deviation growth threshold, a cognitive decline risk warning is triggered.

8. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 7, characterized in that, In S3, the self-care ability index is calculated as follows: Cooking event data within the baseline establishment period after the user enables monitoring and presets the adaptation period are selected. The daily effective meal count, daily total cooking time, daily average cooking operation complexity, daily nutritional intake index, and daily comprehensive dietary regularity score are extracted to form a daily indicator vector. Calculate the mean and standard deviation of each indicator within the indicator vector to construct a user behavior baseline. The behavior baseline is updated according to a preset baseline update cycle. The periodic mean of each indicator for the user within the corresponding period is calculated according to the preset monitoring period. The periodic mean is compared with the behavioral baseline mean of the corresponding indicator. The absolute value of the difference between the periodic mean and the baseline mean is calculated and then divided by the baseline standard deviation of the corresponding indicator to obtain the deviation of each indicator. The deviation of each indicator is statistically analyzed for consecutive periods. Indicators whose deviation for consecutive preset periods exceeds a preset significant deviation threshold are marked as significant decline indicators. Each indicator is assigned a corresponding weight, and an indicator function is used to assign values ​​to the significant decline indicators. The comprehensive self-care ability index is obtained by subtracting the preset decline deduction coefficient from the preset baseline score and multiplying the sum of all indicator weights and corresponding indicator function values. The target value for the user's total daily cooking time is set as the baseline mean, and the allowable deviation is a multiple of the preset baseline standard deviation. The positive and negative cumulative sums of the total daily cooking time are calculated using a cumulative sum control chart. Cases where the cumulative sum exceeds the preset anomaly judgment threshold are marked as sudden anomalies. A missed meal alarm is marked when the number of valid meals for a user's consecutive preset number of days is 0; a forgotten alarm is marked when the continuous working time of a single device exceeds the preset safe time threshold; and a time surge alarm is marked when the duration of a single cooking event exceeds the user's historical preset percentile and the number of devices involved is less than the preset number of devices. Events where a user has no device activity for a preset duration during regular activity periods and simultaneously experiences missed meals are marked as suspected emergency events.

9. The method for monitoring the living status of the elderly based on multidimensional analysis of cooking behavior according to claim 8, characterized in that, In step S4, the weighted fusion is used to generate a comprehensive status score and output the monitoring results and early warning signals, specifically as follows: Preset corresponding weight coefficients for the nutrient intake index, the comprehensive dietary regularity score, and the comprehensive self-care ability index; The user's overall status score is obtained by performing a weighted summation of the indices of the three dimensions and their corresponding weights. The system verifies the risk level across three dimensions. If any dimension triggers a high-risk level, the user's overall status is directly downgraded to the corresponding risk level. Configure tiered push rules to push the previous day's meal overview and corresponding dimension scores according to the preset daily push time, and push the indicator trend charts and risk warnings for the corresponding period according to the preset weekly push cycle. Based on the comprehensive status score and risk level, the corresponding push channels are matched. Non-emergency risk events are pushed through the application, while emergency risk events are pushed through SMS and telephone voice. The corresponding status monitoring results and early warning signals are output simultaneously.

10. A monitoring system for the living conditions of the elderly based on multidimensional analysis of cooking behavior, characterized in that, include: The module includes a data acquisition and preprocessing module, a cooking event recognition module, a multi-dimensional quantitative evaluation module, and a comprehensive decision output module. The data acquisition and preprocessing module is used to deploy kitchen sensing devices, collect device status data, perform sliding median filtering on power data to remove short-term events, perform linear interpolation to complete missing data, and output a standardized device event sequence after cleaning. The cooking event recognition module is electrically connected to the data acquisition and preprocessing module. The cooking event recognition module is used to extract the working interval of each device from the device event sequence, merge and split the device events based on the adaptive time window, and label the merged events by meal type in combination with the meal time period dynamically identified by the clustering algorithm to generate cooking event records. The multi-dimensional quantitative assessment module is electrically connected to the cooking event recognition module. The multi-dimensional quantitative assessment module is used to calculate the nutritional intake index, dietary regularity score and self-care ability index based on the cooking event records. The nutritional intake index is based on the weighted normalization of the equipment-cooking mapping weight and working time. The dietary regularity score is generated based on the standard deviation of the start time of the three meals through exponential decay mapping. The self-care ability index is generated based on the deviation of the current indicator from the historical baseline and the cumulative sum control chart detection results. The comprehensive decision output module is electrically connected to the multi-dimensional quantitative assessment module. The comprehensive decision output module is used to weight and integrate the nutrient intake index, dietary regularity score and self-care ability index to generate a comprehensive status score, and push monitoring results and early warning signals according to the risk level.

Images