An abnormality recognition and alarm system for the elderly living alone based on mobile phone behavior analysis

Abstract

The application relates to an abnormality identification and alarm system for a solitary old person based on mobile phone behavior analysis, and relates to the technical field of alarm devices, which comprises a behavior collection module used for collecting historical interaction behavior data of a mobile phone of a user; a portrait construction module used for constructing a mobile phone interaction behavior portrait of the user based on the historical interaction behavior data of the mobile phone of the user; the behavior collection module is further used for collecting current interaction behavior data and state data of the mobile phone of the user; a multi-modal perception module used for collecting multi-modal physiological behavior data of the user; an abnormality identification module used for judging whether abnormal interaction occurs based on the mobile phone interaction behavior portrait of the user, the current interaction behavior data and the state data of the mobile phone; if yes, hierarchical early warning is carried out based on the interaction behavior data, the state data and the multi-modal physiological behavior data of the mobile phone of the user, and the application has the advantages that abnormal conditions possibly occurring to the old person can be accurately identified, and alarm information can be timely sent out.

Description

Technical Field

[0001] This invention relates to the field of alarm device technology, and in particular to an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis. Background Technology

[0002] Elderly people living alone face numerous potential risks and challenges in their lives. Lacking the constant companionship and care of family members, they often cannot receive timely and effective assistance when encountering sudden illnesses, accidental falls, or other dangerous incidents, posing a serious threat to their lives. Therefore, how to monitor the living conditions of elderly people living alone in real time and issue timely alerts in case of abnormalities has become an urgent problem to be solved.

[0003] A common monitoring method involves installing fixed sensors in the elderly person's living environment, such as infrared sensors, door and window sensors, and smoke sensors. These sensors can monitor the elderly person's activity range and environmental changes such as the presence of fires. However, this monitoring method also has many shortcomings. First, the installation location of fixed sensors is relatively fixed, and they can only monitor the environmental conditions of specific areas, unable to track the elderly person's behavior and state. Video surveillance systems can capture the elderly person's activities in real time through cameras, thereby monitoring their behavior. However, this technology has serious privacy invasion issues, and the elderly person may refuse to install cameras for fear of privacy leaks. Moreover, video surveillance requires a large amount of storage space to store monitoring data and also has high requirements for network bandwidth, increasing the cost of use. In addition, video surveillance is prone to inaccurate monitoring in low light conditions or when the elderly person's view is obstructed, making it impossible to guarantee real-time and accurate identification of the elderly person's behavior.

[0004] Therefore, there is a need to provide an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis, which can accurately identify possible abnormal situations of the elderly and issue alarm information in a timely manner. Summary of the Invention

[0005] This invention provides an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis, comprising: a behavior acquisition module for acquiring the target user's historical medical records, identifying multiple similar users with similar medical records based on the user's historical medical records, identifying multiple key interaction behavior factors based on the multiple similar users with similar medical records, and collecting the user's historical interaction behavior data based on the multiple key interaction behavior factors; a profile construction module for determining the baseline value and weight of each key interaction behavior factor based on the user's historical interaction behavior data, and constructing a profile of the user's mobile phone interaction behavior; the behavior acquisition module is also used to collect the user's current interaction behavior data and status data acquired by sensors set on the mobile phone; a multimodal perception module for collecting the user's multimodal physiological behavior data; and an abnormal identification module for determining the current value of each key interaction behavior factor, the current value, baseline value, weight, and status data of each key interaction behavior factor, and determining whether abnormal interaction has occurred. If so, a graded warning is issued based on the user's mobile phone interaction behavior data, status data, and multimodal physiological behavior data.

[0006] Furthermore, the behavior collection module is specifically used to: determine multiple key interaction behavior factors based on historical interaction behavior data and historical abnormal records of multiple similar users; and collect historical interaction behavior data and current interaction behavior data of the user's mobile phone based on multiple key interaction behavior factors.

[0007] Furthermore, the behavior acquisition module is specifically used for: determining multiple key abnormal states based on historical abnormal records of multiple similar users; determining multiple interaction behavior factors; for each key abnormal state and each interaction behavior factor, calculating the difference value of the key abnormal state corresponding to the interaction behavior factor based on the historical interaction behavior data of each similar user; for each key abnormal state and any two interaction behavior factors, calculating the correlation value of the key abnormal state corresponding to the two interaction behavior factors based on the historical interaction behavior data of each similar user; calculating the key value of the interaction behavior factor based on the difference value of each interaction behavior factor corresponding to each key abnormal state and the correlation value of any two interaction behavior factors corresponding to each key abnormal state; and determining multiple key interaction behavior factors based on the key value of each interaction behavior factor.

[0008] Furthermore, the behavior acquisition module is specifically used to: for each interactive behavior factor, determine the associated interactive behavior factors based on the correlation values ​​of any two interactive behavior factors corresponding to each key abnormal state, and calculate the key value of the interactive behavior factor based on the difference values ​​of the interactive behavior factors corresponding to each key abnormal state and the difference values ​​of each associated interactive behavior factor corresponding to each key abnormal state.

[0009] Furthermore, the behavior acquisition module includes multiple accelerometers installed on the mobile phone; the status data includes acceleration data collected by the multiple accelerometers; the behavior acquisition module is also used to: set multiple accelerometers to be screened on the mobile phone; and screen multiple accelerometers from the multiple accelerometers to be screened based on multiple key abnormal states.

[0010] Furthermore, the behavior acquisition module is specifically used to: for each key abnormal state, acquire test data of multiple similar users in the key abnormal state, wherein the test data includes acceleration data collected by multiple accelerometers; for the normal state, acquire test data of multiple similar users in the normal state; and, using an improved particle swarm optimization algorithm, select multiple accelerometers from multiple accelerometers to be screened based on the test data of multiple similar users in each key abnormal state and the test data in the normal state.

[0011] Furthermore, the profile building module is specifically used to: determine the baseline value of each key interaction behavior factor based on the user's historical interaction behavior data on their mobile phone; and determine the weight of each key interaction behavior factor based on the difference value of each key abnormal state corresponding to each key interaction behavior factor.

[0012] Furthermore, the anomaly identification module is used to: determine the current value of each key interaction behavior factor based on the current interaction behavior data of the mobile phone; calculate the current difference value of each key interaction behavior factor based on the current value and the baseline value of each key interaction behavior factor; determine whether an interaction behavior anomaly has occurred based on the current difference value and weight of each key interaction behavior factor; if so, determine that an abnormal interaction has occurred; if not, determine whether a mobile phone drop or hand tremor anomaly has occurred based on the acceleration data collected by multiple acceleration sensors; if so, determine that an abnormal interaction has occurred; if not, determine that no abnormal interaction has occurred.

[0013] Furthermore, the multimodal sensing module includes a physiological monitoring bracelet and a sleep behavior monitoring device, wherein the physiological monitoring bracelet is used to collect the user's physiological data, and the sleep behavior monitoring device is used to collect the user's sleep behavior data.

[0014] Furthermore, the anomaly identification module is specifically used to: determine candidate abnormal states based on the user's mobile phone interaction behavior data and status data; determine the current abnormal state based on the user's physiological data and / or sleep behavior data and candidate abnormal states; and trigger a graded warning based on the current abnormal state.

[0015] Compared to existing technologies, the abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis provided in this specification has at least the following beneficial effects:

[0016] 1. By building user profiles based on historical mobile phone interaction data, the system can accurately depict the daily behavior patterns of the elderly. When identifying anomalies, it integrates mobile phone interaction, status, and multimodal physiological behavior data, with multi-dimensional information corroborating each other. This tiered early warning system allocates resources rationally according to the severity of the anomaly; mild anomalies alert children to pay attention, while severe anomalies prompt rapid emergency response, ensuring that elderly people living alone receive timely and appropriate responses in various abnormal situations, greatly improving the safety of their lives.

[0017] 2. By analyzing the historical interaction data and abnormal records of multiple similar users, key interaction factors are determined, fully considering the commonalities and characteristics of the behavioral habits and health conditions of elderly people living alone. The factors selected in this way are more representative and targeted, more accurately reflecting the boundaries between normal and abnormal behavior in the elderly, providing a reliable basis for subsequent precise identification of abnormalities. It innovatively introduces the medical prior knowledge of "multiple similar users with similar medical records," transforming the identification of abnormal behavior in elderly people living alone from a general model to personalized modeling. It doesn't simply record "whether they use a mobile phone," but analyzes the behavioral changes of people with similar conditions to the target elderly person before the onset of illness, automatically selecting the key interaction indicators most sensitive to the individual. This is further validated by combining physical events such as phone drops and hand tremors. Furthermore, after initial abnormality detection, it can also integrate external physiological data such as wristbands and sleep mats for cross-confirmation, achieving graded and precise alarms.

[0018] 3. First, based on the current interaction behavior data and the profile baseline value and weight, abnormal interaction behavior is judged. If no abnormality is judged, acceleration data is further combined to judge abnormalities such as phone drop or hand shaking. This multi-level and diversified judgment mechanism comprehensively considers the behavior of the elderly from different perspectives, avoids the limitations of a single judgment method, effectively improves the accuracy and comprehensiveness of abnormality identification, minimizes misjudgment and missed judgment, and safeguards the safety of elderly people living alone. Attached Figure Description

[0019] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0020] Figure 1 This is a block diagram of an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis, as shown in one embodiment of this application.

[0021] Figure 2This is a flowchart illustrating the determination of multiple key interaction behavior factors in one embodiment of this application;

[0022] Figure 3 This is a flowchart illustrating the determination of whether an abnormal interaction has occurred, as shown in one embodiment of this application. Detailed Implementation

[0023] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0024] Figure 1 This is a block diagram of an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis, as shown in one embodiment of this application. Figure 1 As shown, an abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis may include a behavior acquisition module, a profile construction module, a multimodal perception module, and an abnormal identification module.

[0025] The behavior collection module can be used to collect historical interaction behavior data of users' mobile phones.

[0026] Specifically, it includes:

[0027] Obtain the user's historical medical records;

[0028] Based on a user's historical medical records, identify multiple similar users with similar medical records;

[0029] Based on historical interaction data and historical anomaly records of multiple similar users, several key interaction behavior factors were identified.

[0030] Several key interaction behavior factors were identified, and historical and current interaction behavior data of users' mobile phones were collected.

[0031] Specifically, a user's medical history can include information such as past illnesses, treatments, and medications. For example, for the elderly, this includes information about their past chronic illnesses (such as hypertension, diabetes, and heart disease), major illnesses (such as stroke and cancer), and surgical history.

[0032] Because medical record formats and recording methods may differ across medical institutions, the acquired medical record data needs to be organized and standardized. For example, this involves standardizing disease names by converting different codes used by different hospitals into a unified International Classification of Diseases (ICD) code; and standardizing test results to ensure comparability of data measured by different equipment and at different times. Simultaneously, the textual information in the medical records is structured to extract key information, such as the time of diagnosis, treatment plan, and medication dosage, for subsequent analysis and processing.

[0033] Multiple dimensions can be considered, including disease type, disease severity, duration of illness, treatment history, and complications. For example, regarding disease type, the degree of matching between the two users' primary diseases can be calculated; regarding disease severity, a quantitative assessment can be made based on relevant indicators in the medical records (such as blood pressure, blood sugar, and tumor stage); regarding duration of illness, the similarity of the time span and disease progression stages of the users' illnesses can be considered; regarding treatment history, the consistency of the treatment methods and medications received by the users can be compared; and regarding complications, the similarity and severity of complications among the users can be considered.

[0034] Each user's medical record information is represented as a vector, with each dimension of the vector corresponding to an evaluation index. Then, the cosine value of the angle between two user vectors is calculated. The closer the cosine value is to 1, the higher the similarity of the two users' medical records.

[0035] Based on the calculated similarity scores, a similarity threshold (e.g., 0.7) is set, and users with similarities exceeding this threshold are identified as similar users. Alternatively, users can be sorted from highest to lowest similarity and selected as the top-ranked users. For example, the top 10 or 20 users with the highest similarity scores can be selected as multiple similar users whose medical records are similar to the target user, for subsequent in-depth analysis and application.

[0036] Figure 2 This is a flowchart illustrating the determination of multiple key interaction behavior factors in one embodiment of this application, such as... Figure 2 As shown, in some embodiments, the behavior acquisition module is used for:

[0037] Based on the historical anomaly records of multiple similar users, various key abnormal states are identified, such as heart disease attacks, hypertension attacks, fainting, and falls. Historical anomaly records can be used to record the abnormal states that occurred in similar users at multiple historical time points, extract the abnormal states that occur most frequently as key abnormal states, and when a user is in a key abnormal state, their mobile phone behavior may change.

[0038] Multiple interactive behavior factors should be identified. These factors should include not only screen unlock frequency (which reflects the elderly person's activity level and daily routine; for example, frequent unlocking may indicate that the elderly person is awake and active, while prolonged inactivity may suggest physical discomfort or sleep), SMS reply speed (which reflects the elderly person's timeliness in communicating with the outside world; a sudden slowdown in reply speed may suggest physical problems or an inability to reply in time during an emergency), and usage time (which understands the overall time distribution of the elderly person's mobile phone use; prolonged continuous use may affect the elderly person's rest and health), but also the frequency of use of specific applications (for example, changes in the frequency of use of health monitoring applications may reflect the elderly person's level of attention to their own health and changes in their physical condition), emergency call function usage records (which are directly related to the elderly person's emergency help behavior; abnormal usage frequency and time may indicate that the elderly person is in danger or has a sudden illness), and the frequency of mobile phone location changes (which, combined with the elderly person's daily activity range, abnormal location changes may suggest that the elderly person is lost or in an unsafe environment). These factors are all related to mobile phone use and are of great significance to the health and safety of the elderly.

[0039] For each key abnormal state and each interaction behavior factor, the difference value of the interaction behavior factor corresponding to the key abnormal state is calculated based on the historical interaction behavior data of each similar user.

[0040] For each key abnormal state and any two interaction behavior factors, based on the historical interaction behavior data of each similar user, calculate the correlation value between the two interaction behavior factors and the key abnormal state.

[0041] Based on the difference value of each interactive behavior factor corresponding to each key abnormal state and the correlation value of any two interactive behavior factors corresponding to each key abnormal state, the key value of the interactive behavior factor is calculated.

[0042] Based on the key value of each interaction behavior factor, multiple key interaction behavior factors are identified.

[0043] Specifically, for each key abnormal state and each interaction behavior factor, based on the historical interaction behavior data of each similar user, the value of the interaction behavior factor of each similar user in the key abnormal state is determined, and the difference between the value of the interaction behavior factor of the similar user in the key abnormal state and the value of the interaction behavior factor of the similar user in the normal state is calculated as the difference value of the interaction behavior factor of the similar user corresponding to the key abnormal state. The average value of the difference values ​​of the interaction behavior factors of multiple similar users corresponding to the key abnormal states is calculated as the difference value of the interaction behavior factor corresponding to the key abnormal state.

[0044] For each key abnormal state and any two interaction behavior factors, the difference between the key abnormal states corresponding to the two interaction behavior factors of each similar user is taken as two variables and substituted into the calculation formula of the correlation coefficient (e.g., Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) to obtain the correlation value between the two interaction behavior factors and the key abnormal states.

[0045] In some embodiments, the behavior acquisition module is used for:

[0046] For each interaction behavior factor, based on the correlation value of any two interaction behavior factors corresponding to each key abnormal state, determine the associated interaction behavior factors of the interaction behavior factors. Based on the difference value of each interaction behavior factor corresponding to each key abnormal state and the difference value of each associated interaction behavior factor corresponding to each key abnormal state, calculate the key value of the interaction behavior factor.

[0047] Specifically, for each interaction behavior factor, interaction behavior factors whose absolute value of association with that interaction behavior factor is greater than a threshold (e.g., 0.5) can be used as associated interaction behavior factors.

[0048] For example, the key value of the interaction behavior factor can be calculated based on the following formula:

[0049]

[0050] in, For the key value of the i-th interaction behavior factor, Let be the mean of the differences between the various key abnormal states corresponding to the i-th interaction behavior factor. Let be the mean of the association values ​​of the i-th interaction behavior factor and the m-th associated interaction behavior factor corresponding to various key abnormal states. Let be the mean of the differences between the m-th associated interactive behavior factors corresponding to the i-th interactive behavior factor and the various key abnormal states. This represents the total number of associated interactive factors for the i-th interactive factor.

[0051] First, identify the associated interaction behavior factors. Using the absolute value of the association value being greater than a threshold (e.g., 0.5) as a criterion, find other factors closely associated with the current interaction behavior factor. Next, calculate the key value. On one hand, this involves averaging the differences between various key abnormal states corresponding to the current interaction behavior factor. As a fundamental component, it reflects the degree of difference between the factor itself and the normal state under different abnormal states; the greater the difference, the more important it may be in distinguishing abnormal states. On the other hand, considering the influence of associated interaction behavior factors, the mean of the association value is used... This measures the strength of the association between the current factor and related factors; the higher the association value, the stronger the association. It is then multiplied by the mean of the differences between the related factors and various key anomalies. This demonstrates the role of correlation factors in distinguishing abnormal states. The contribution of all correlation factors in this aspect is summed and added to the base value to obtain the final key value. By comprehensively considering the relationship between the key value itself and its correlation factors and the abnormal state, the importance of interactive behavior factors in abnormal state identification can be assessed more comprehensively and accurately. This avoids the one-sidedness of judging based on a single factor, making the key value more reflective of the actual situation. This helps to accurately screen out the factors that play a key role in identifying critical abnormal states from numerous interactive behavior factors, providing a reliable basis for subsequent abnormal state analysis and early warning, and improving the accuracy and effectiveness of the system's abnormal state judgment.

[0052] Based on the key value of each interaction behavior factor, multiple interaction behavior factors can be sorted, and the top n (n is a positive integer greater than 0) interaction behavior factors can be selected as key interaction behavior factors.

[0053] First, user medical records are acquired and processed to accurately identify similar users, providing a reliable foundation for subsequent analysis. Key abnormal states are extracted from the historical anomaly records of similar users, allowing focus on situations that significantly impact the health and safety of the elderly. A comprehensive consideration of various meaningful interactive behavioral factors related to mobile phone use is undertaken, encompassing multifaceted information. By calculating the difference and correlation values ​​between interactive behavioral factors and key abnormal states, key values ​​are derived, scientifically measuring the importance of each factor in abnormal state identification. Finally, based on these key values, key interactive behavioral factors are determined, accurately identifying the most effective factors for monitoring the health and safety of the elderly, facilitating timely detection of abnormalities.

[0054] In some embodiments, the behavior acquisition module includes multiple accelerometers mounted on the mobile phone. The status data includes acceleration data acquired by the multiple accelerometers.

[0055] Specifically, since mobile phones are items that elderly people carry with them every day, integrating an accelerometer into the phone can conveniently and continuously collect acceleration data generated by the elderly person's body movements, providing basic data support for subsequent analysis of the elderly person's behavior and health status.

[0056] In some embodiments, the behavior acquisition module is used for:

[0057] Multiple accelerometer sensors to be selected are set on the mobile phone, and the location of the different accelerometer sensors to be selected is different (such as the top, bottom, side, etc. of the mobile phone);

[0058] Based on a number of key abnormal states, multiple accelerometers are selected from a pool of potential accelerometers.

[0059] In some embodiments, the behavior acquisition module is used for:

[0060] For each critical abnormal state, test data from multiple similar users under the critical abnormal state is obtained, including acceleration data collected by multiple acceleration sensors.

[0061] For the normal state, obtain test data from multiple similar users under the normal state;

[0062] By using an improved particle swarm optimization algorithm, multiple accelerometers are selected from multiple accelerometers to be screened based on test data from multiple similar users in each key abnormal state and normal state.

[0063] Specifically, screening multiple accelerometers from a pool of potential sensors based on various key abnormal states aims to identify those that can most effectively and sensitively capture the characteristics of these abnormal states. Different abnormal states (such as heart attacks, hypertension attacks, fainting, falls, etc.) can cause different hand tremor frequencies, movement patterns, and acceleration changes in the elderly. By selecting sensors that are sensitive to these abnormal states, it is possible to more accurately identify whether the elderly are experiencing these abnormalities.

[0064] For each key abnormal state, acquiring test data from multiple similar users under that state and collecting their acceleration data allows for a more accurate understanding of the acceleration characteristics that the target user might experience under similar circumstances. For example, if the target user is at risk of falling, collecting acceleration data from multiple similar users with similar medical records during falls can help summarize typical patterns of acceleration changes during falls.

[0065] Simultaneously, test data from multiple similar users under normal conditions is acquired. Comparing the normal state data with the abnormal state data allows for a clearer identification of the unique acceleration characteristics under abnormal conditions. For example, by comparing the acceleration data of an elderly person walking normally and falling, it can be found that the acceleration changes drastically in a short period of time when falling, while the acceleration changes relatively steadily when walking normally.

[0066] Based on the following process, using an improved particle swarm optimization algorithm, multiple accelerometers can be selected from multiple potential accelerometers based on test data from multiple similar users in each key abnormal state and normal state:

[0067] S11. For each accelerometer, based on the test data of similar users in key abnormal states and test data in normal states, extract the acceleration sequences of similar users in key abnormal states and normal states collected by the accelerometer, calculate the similarity between the acceleration sequences of similar users in key abnormal states and normal states collected by the accelerometer, and use it as the acceleration similarity of the accelerometer for the corresponding similar user. Calculate the mean of the acceleration similarity of the accelerometer for each similar user and use it as the mean of the acceleration similarity of the accelerometer.

[0068] S12. For any two accelerometers, calculate the similarity of the acceleration sequences of similar users in key abnormal states collected by the two accelerometers, and use it as the acceleration similarity of the two accelerometers for the key abnormal states. Calculate the average of the acceleration similarity of the two accelerometers for each key abnormal state, and use it as the average acceleration similarity of the two accelerometers.

[0069] S13. Construct a fitness function, where the fitness function is related to the mean acceleration similarity of each accelerometer included in the particle and the mean acceleration similarity of any two accelerometers. The larger the mean acceleration similarity of each accelerometer and the smaller the mean acceleration similarity of any two accelerometers, the larger the value of the fitness function.

[0070] S14. Initialize the particle swarm, where one particle in the particle swarm represents at least three accelerometers to be screened.

[0071] S15. Based on the initialized particle swarm and fitness function, perform iterative optimization to determine the optimal particle, thereby determining the selection of multiple accelerometers from multiple accelerometers to be screened.

[0072] By calculating the average acceleration similarity of each accelerometer under normal and abnormal conditions, as well as the average similarity between any two sensors under abnormal conditions, and constructing a relevant fitness function, the optimal sensor combination can be accurately located during iterative optimization. The selected sensor combination effectively reduces redundant data acquisition, avoiding resource waste and data processing burden caused by collecting too much irrelevant data. Simultaneously, these sensors can capture abnormal state characteristics from different perspectives, complementing each other and improving the sensitivity to abnormal state detection, thus enhancing detection effectiveness. In this way, both data quality and detection efficiency are ensured, providing reliable and streamlined data support for subsequent applications such as acceleration data-based health monitoring.

[0073] By constructing a fitness function related to the mean acceleration similarity of the accelerometer itself and the mean acceleration similarity between sensors, the improved algorithm comprehensively considers both individual sensor performance and group synergy. In terms of individual performance, a large mean acceleration similarity indicates that the sensor has high discrimination between abnormal and normal states, and the collected data is reliable. Regarding group synergy, a small mean acceleration similarity between sensors means that the selected sensors collect data with greater differences, providing richer and more complementary information and avoiding information redundancy. Initially, each particle represents at least three sensors to be screened. Compared to the traditional one-by-one screening method, it can simultaneously explore the possibilities of multiple sensor combinations, greatly expanding the search range. During iteration, particles continuously adjust their positions according to the fitness function, quickly converging towards the optimal solution, significantly shortening the screening time and improving screening efficiency. Furthermore, the improved algorithm has stronger robustness and adaptability. Faced with complex test data and diverse screening requirements, it can flexibly adjust the search strategy, overcome local optima traps, and find the globally optimal sensor combination, laying a solid foundation for subsequent monitoring and analysis work based on these sensors and effectively ensuring the stability and reliability of the entire system.

[0074] The profile building module can be used to build a user's mobile phone interaction behavior profile based on the user's historical interaction behavior data.

[0075] Specifically, it includes:

[0076] Based on the user's historical interaction data on their mobile phone, a baseline value is determined for each key interaction factor.

[0077] Based on the difference value of each key interaction behavior factor corresponding to each key abnormal state, the weight of each key interaction behavior factor is determined. The user's mobile phone interaction behavior profile includes the baseline value and weight of each key interaction behavior factor.

[0078] Specifically, for each key interactive behavior factor, statistical analysis methods are used to determine its baseline value. For example, for the factor of screen unlock frequency, the average number of times a user unlocks their phone each day over a period of time (such as a week) can be statistically analyzed, and this average value can be used as the baseline value for this factor. For SMS reply speed, the average time interval between user replies to SMS messages can be calculated as the baseline value. In this way, a relatively stable reference standard can be set for each interactive behavior factor, reflecting the user's typical performance in that behavior. The baseline value is a quantitative representation of the user's normal mobile phone interaction behavior, providing a basis for comparison against changes in user behavior. When the user's actual interaction behavior deviates significantly from the baseline value, it may indicate a change in the user's behavior pattern, or the possible occurrence of an abnormal situation, such as physical discomfort or an emergency.

[0079] For each key interaction behavior factor, calculate the mean of the difference values ​​for each key abnormal state corresponding to that key interaction behavior factor, and use this as the mean difference value. Calculate the sum of the mean difference values ​​for all key interaction behavior factors, and use the ratio of the mean difference value of a key interaction behavior factor to the sum of the mean difference values ​​as the weight of the key interaction behavior factor.

[0080] The behavior collection module can also be used to collect the user's current interactive behavior data and status data.

[0081] The multimodal perception module can be used to collect multimodal physiological and behavioral data from users.

[0082] Specifically, the multimodal sensing module includes a physiological monitoring bracelet and a sleep behavior monitoring device. The physiological monitoring bracelet is used to collect the user's physiological data, and the sleep behavior monitoring device is used to collect the user's sleep behavior data.

[0083] Physiological monitoring wristbands can collect various key physiological data, such as heart rate, which reflects the user's heart rate in real time, helping to determine whether heart function is normal and whether the user is in different states such as tension or exercise; blood pressure data reflects the pressure of blood vessels, and for patients with hypertension, continuous monitoring of blood pressure changes helps to adjust treatment plans in a timely manner; blood oxygen saturation data reflects the oxygen content in the user's blood, which is of great value for monitoring and assessing respiratory diseases; in addition, it may also include body temperature data, which can help determine whether the user has a fever, infection, etc. The wristband uses sensor technology, such as photoelectric sensors to measure heart rate and blood oxygen saturation, acquiring data by detecting changes in blood absorption of light; and pressure sensors to measure blood pressure, estimating blood pressure values ​​by sensing pressure at the wrist. These sensors can continuously and in real time collect data and transmit the data to the wristband's built-in storage unit or to terminal devices such as mobile phones via wireless communication technologies such as Bluetooth.

[0084] Sleep behavior monitoring devices mainly collect users' limb movements during sleep, such as turning over and kicking. Frequent limb movements may indicate restless sleep or certain health problems.

[0085] Sleep behavior monitoring devices can use pressure sensor arrays placed under the mattress or beside the bed to collect sleep behavior data by sensing changes in the pressure of the user's body on the mattress.

[0086] The anomaly detection module can be used to determine whether abnormal interaction has occurred based on the user's mobile phone interaction behavior profile, the current interaction behavior data of the mobile phone, and the status data. If so, it can provide graded warnings based on the user's mobile phone interaction behavior data, status data, and multimodal physiological behavior data.

[0087] Figure 3 This is a flowchart illustrating the determination of whether an abnormal interaction has occurred, as shown in one embodiment of this application. Figure 3 As shown, in some embodiments, the anomaly detection module is used for:

[0088] Based on the current interaction behavior data of the mobile phone, determine the current value of each key interaction behavior factor;

[0089] Calculate the current difference value for each key interaction behavior factor based on its current value and baseline value.

[0090] Based on the current difference value and weight of each key interaction behavior factor, determine whether an interaction behavior abnormality has occurred; if so, determine that an abnormal interaction has occurred.

[0091] If not, based on acceleration data collected by multiple accelerometers, determine whether the phone has been dropped or whether there has been abnormal hand shaking. If yes, determine that an abnormal interaction has occurred; if no, determine that no abnormal interaction has occurred.

[0092] Specifically, based on the weight of each key interaction behavior factor, the current difference value of each key interaction behavior factor is summed in a weighted manner to obtain a weighted difference value. When the weighted difference value is greater than a threshold (e.g., 3), it is determined that an interaction behavior abnormality has occurred.

[0093] Based on 3D acceleration data collected from multiple accelerometers, a sliding window technique is used to segment the data for processing. A fixed-length sliding window is set, and the window slides gradually over time, processing data within one window at a time. First, the mean of the acceleration data within each window is calculated to initially determine whether the phone is in a relatively stationary or stable motion state. Next, the free fall phase is determined by detecting changes in acceleration data along the gravitational direction (i.e., the Z-axis). When the Z-axis acceleration remains close to the gravitational acceleration (approximately 9.8 m / s²) for a period of time, and the acceleration on other axes is close to zero, it is marked as potentially entering the free fall phase. Then, peak detection is performed on the data after the free fall phase to look for drastic changes in acceleration in multiple directions. An acceleration peak threshold is set; when the acceleration peaks on multiple axes exceed this threshold within a short period (e.g., several hundred milliseconds), it is determined that the phone has collided with the ground or other objects, i.e., the phone has fallen.

[0094] Spectral analysis was performed on acceleration data collected from multiple accelerometers using Fast Fourier Transform (FFT). The time-domain acceleration signal was converted into a frequency-domain spectrum to better analyze the frequency characteristics of hand tremors. A frequency range and energy threshold for normal hand tremors were set, as normal tremors typically have low frequencies and relatively uniform energy distribution. For data within each window, the spectral energy distribution was calculated and compared with the set normal range. If the proportion of high-frequency components (e.g., above 5Hz) in the spectrum exceeds the preset threshold, or if the energy is concentrated in certain specific high-frequency bands, and combined with the amplitude variation of acceleration, when the amplitude variation exceeds the normal tremor amplitude range, it is determined to be an abnormal hand tremor.

[0095] In some embodiments, the anomaly detection module is used for:

[0096] Based on the user's mobile phone interaction behavior data and status data, candidate abnormal states are identified.

[0097] Based on the user's physiological data and / or sleep behavior data and candidate abnormal states, determine the current abnormal state;

[0098] Based on the current abnormal state, a tiered warning is triggered.

[0099] Specifically, a state recognition model can identify candidate abnormal states based on the user's mobile phone interaction behavior data and state data, and determine the current abnormal state based on the user's physiological data and / or sleep behavior data and candidate abnormal states. The state recognition model employs a multi-layer neural network structure. The input layer receives interaction behavior data (such as click frequency, swipe speed, operation duration, etc.) and state data (such as geographical location, movement speed, acceleration, etc.) from the user's mobile phone, as well as physiological data (heart rate, blood pressure, blood oxygen, etc.) and sleep behavior data (sleep stage, duration, limb movements, etc.). This multi-dimensional data is standardized to unify the units of measurement, eliminating differences in magnitude between different features, facilitating model training. The hidden layer contains multiple neurons that perform complex transformations and feature extraction on the input data using non-linear activation functions (such as ReLU). It can automatically learn potential patterns and associations in the data, such as discovering the potential link between abnormal operation and physical discomfort from interaction behavior data, and mining the correspondence between changes in health indicators and abnormal states from physiological data. Different hidden layers can progressively extract features from low to high levels, enhancing the model's ability to recognize complex situations. The output layer is set according to task requirements. When determining candidate abnormal states, it outputs the probability values ​​of different types of possible abnormal states, and the one with the highest probability is selected as the candidate abnormal state. When determining the current abnormal state, it combines the candidate abnormalities with physiological and sleep data to output the finally confirmed abnormal state category.

[0100] During model training, a large amount of labeled historical data is used to minimize the error between the predicted results and the true labels. The connection weights between neurons are adjusted through the backpropagation algorithm. After multiple rounds of iterative optimization, the model performs well on both the training and validation sets and has the ability to accurately identify abnormal states.

[0101] Once the status recognition model identifies an abnormal state, a tiered early warning mechanism is activated. For mild abnormalities, such as a user experiencing a slightly faster heart rate, a slight decrease in sleep quality, or some sluggishness in phone operation that doesn't affect basic daily life, the system will quickly push a notification to the user's children via the mobile app. The notification includes the type of abnormal state, the approximate time of occurrence, and some preliminary suggestions, such as reminding children to communicate with the user promptly, inquire about their health, and monitor for any subsequent changes, ensuring that children are informed and can offer support immediately.

[0102] If a severe abnormality is detected, such as a sudden heart attack, severe hypertension leading to coma, or a life-threatening fall that renders the user immobile, the system will immediately and automatically activate the community emergency response team. On one hand, it will quickly and accurately transmit the user's detailed location information and key details of the abnormal condition to emergency personnel; on the other hand, it will open emergency communication channels to facilitate contact between the emergency team and on-site personnel or relevant parties. Upon receiving the information, the community emergency response team will rapidly dispatch professional personnel and rescue equipment to the scene to conduct emergency rescue operations, maximizing the user's safety, minimizing harm caused by the accident, and buying precious time for treatment.

[0103] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A system for identifying and alarming abnormal behavior in elderly people living alone based on mobile phone behavior analysis, characterized in that, include: The behavior collection module is used to acquire the target user's historical medical records, identify multiple similar users with similar medical records based on the user's historical medical records, identify multiple key interactive behavior factors based on the multiple similar users with similar medical records, and collect the user's historical interactive behavior data on the mobile phone based on the multiple key interactive behavior factors. The interactive behavior factors are factors related to mobile phone use and are of great significance to the health and safety of the elderly. The profile building module is used to determine the baseline value and weight of each key interaction behavior factor based on the user's historical interaction behavior data on their mobile phone, and to build a profile of the user's mobile phone interaction behavior. The behavior acquisition module is also used to collect the user's current interactive behavior data and the status data obtained by the sensors set on the phone; The multimodal perception module is used to collect users' multimodal physiological and behavioral data; The anomaly detection module is used to determine the current value of each key interaction behavior factor based on the current interaction behavior data of the mobile phone. Based on the current value, baseline value, weight and status data of each key interaction behavior factor, it determines whether an abnormal interaction has occurred. If so, it performs graded warnings based on the user's mobile phone interaction behavior data, status data and multimodal physiological behavior data. The behavior acquisition module is specifically used for: Based on the historical anomaly records of multiple similar users, several key anomaly states were identified; Identify multiple interaction behavior factors; For each key abnormal state and each interaction behavior factor, based on the historical interaction behavior data of each similar user, the difference value of the interaction behavior factor corresponding to the key abnormal state is calculated. Specifically, for each key abnormal state and each interaction behavior factor, based on the historical interaction behavior data of each similar user, the value of the interaction behavior factor of each similar user in the key abnormal state is determined, and the difference between the value of the interaction behavior factor of the similar user in the key abnormal state and the value of the interaction behavior factor of the similar user in the normal state is calculated as the difference value of the interaction behavior factor corresponding to the key abnormal state of the similar user. The average of the difference values ​​of the interaction behavior factors corresponding to the key abnormal states of multiple similar users is taken as the difference value of the interaction behavior factor corresponding to the key abnormal state of the interactive behavior factor. For each key abnormal state and any two interaction behavior factors, based on the historical interaction behavior data of each similar user, the correlation value between the two interaction behavior factors and the key abnormal state is calculated. The difference between the two interaction behavior factors and the key abnormal state of each similar user is taken as two variables and substituted into the formula for calculating the correlation coefficient to obtain the correlation value between the two interaction behavior factors and the key abnormal state. Based on the difference value of each interactive behavior factor corresponding to each key abnormal state and the correlation value of any two interactive behavior factors corresponding to each key abnormal state, the key value of the interactive behavior factor is calculated. Based on the key value of each interaction behavior factor, multiple key interaction behavior factors are identified.

2. The abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis according to claim 1, characterized in that, The behavior acquisition module is also used for: Based on several key interaction behavior factors, historical and current interaction behavior data of users' mobile phones are collected.

3. The abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis according to claim 2, characterized in that, The behavior acquisition module is also used for: For each interaction behavior factor, based on the correlation value of any two interaction behavior factors corresponding to each key abnormal state, determine the associated interaction behavior factors of the interaction behavior factors. Based on the difference value of each interaction behavior factor corresponding to each key abnormal state and the difference value of each associated interaction behavior factor corresponding to each key abnormal state, calculate the key value of the interaction behavior factor.

4. The abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis according to claim 3, characterized in that, The behavior acquisition module includes multiple acceleration sensors installed on the mobile phone; The status data includes acceleration data collected by multiple acceleration sensors; The behavior acquisition module is also used for: Set up multiple accelerometer sensors to be filtered on the phone; Based on a number of key abnormal states, multiple accelerometers are selected from a pool of potential accelerometers.

5. The abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis according to claim 4, characterized in that, The behavior acquisition module is also used for: For each critical abnormal state, test data from multiple similar users under the critical abnormal state is obtained, including acceleration data collected by multiple acceleration sensors. For the normal state, obtain test data from multiple similar users under the normal state; By using an improved particle swarm optimization algorithm, multiple accelerometers are selected from multiple accelerometers to be screened based on test data from multiple similar users in each key abnormal state and normal state.

6. The abnormal identification and alarm system for elderly people living alone based on mobile phone behavior analysis according to claim 5, characterized in that, The portrait construction module is specifically used for: Based on the user's historical interaction data on their mobile phone, a baseline value is determined for each key interaction factor. The weight of each key interaction behavior factor is determined based on the difference value of each key abnormal state corresponding to each key interaction behavior factor.

7. A system for identifying and alarming abnormal behavior in elderly people living alone based on mobile phone behavior analysis according to claim 6, characterized in that, The anomaly detection module is specifically used for: Based on the current interaction behavior data of the mobile phone, determine the current value of each key interaction behavior factor; Calculate the current difference value for each key interaction behavior factor based on its current value and baseline value. Based on the current difference value and weight of each key interaction behavior factor, determine whether an interaction behavior abnormality has occurred; if so, determine that an abnormal interaction has occurred. If not, based on acceleration data collected by multiple accelerometers, determine whether the phone has been dropped or whether there has been abnormal hand shaking. If yes, determine that an abnormal interaction has occurred; if no, determine that no abnormal interaction has occurred.

8. A system for identifying and alarming abnormal behavior in elderly people living alone based on mobile phone behavior analysis according to any one of claims 1-7, characterized in that, The multimodal sensing module includes a physiological monitoring bracelet and a sleep behavior monitoring device, wherein the physiological monitoring bracelet is used to collect the user's physiological data, and the sleep behavior monitoring device is used to collect the user's sleep behavior data.

9. A system for identifying and alarming abnormal behavior in elderly people living alone based on mobile phone behavior analysis according to claim 8, characterized in that, The anomaly detection module is also used for: Based on the user's mobile phone interaction behavior data and status data, candidate abnormal states are identified. Based on the user's physiological data and / or sleep behavior data and candidate abnormal states, determine the current abnormal state; Based on the current abnormal state, a tiered warning is triggered.

Images