Alzheimer's disease old-age care platform management method and system based on big data

Abstract

The invention provides a big-data-based Alzheimer's disease old-age care platform management method and system, and relates to the technical field of old-age care platform management.The method comprises the steps that patient behavior data information is received, missing data points are deduced according to pre-stored patient behavior file historical behavior pattern data and patient behavior data information, and the missing data points are sent to a server; by evaluating the reliability degree of the event chain and comparing the event chain with a dynamically generated reliability evaluation threshold value, when a comparison result reaches an early warning standard, early warning information with a reliability quantitative index is generated, and the problem that an existing platform is poor in reliability under a non-ideal operation condition is effectively solved. The problems of reliable transmission and accurate interpretation of key information are solved, and early warning failure caused by data deviation or missing is avoided, so that the nursing quality and safety of the Alzheimer's disease patient are remarkably improved, and the optimal time for early intervention and intervention is provided for nursing personnel.

Description

Technical Field

[0001] This application relates to the field of elderly care platform management technology, and more specifically, to a big data-based Alzheimer's disease elderly care platform management method and system. Background Technology

[0002] In the field of Alzheimer's disease care, continuous monitoring and proactive intervention through management platforms have become essential—the platforms provide early warnings and personalized care by collecting and analyzing patients' health and behavioral data. However, the transmission, processing, and verification of data from collection to analysis face the dual challenges of hardware performance and data integrity: any deviation or omission may cause early warnings to fail, affecting patient safety.

[0003] The core of the platform is the continuous collection and analysis of patient data. The local data aggregation gateway was originally designed to process sensor data from a standard number of patients within its area of ​​responsibility. However, with the addition of high-frequency data collection devices in elderly care facilities (such as high-sampling-rate sleep monitoring pads and frequent location tracking anti-wandering bracelets), the data reception and computing load on the gateway CPU has increased significantly. Prolonged high load has caused the gateway CPU's heat dissipation module to degrade (e.g., aging thermal paste, dust accumulation on the fan), and the core temperature frequently approaches the throttling threshold, triggering throttling—while protecting the hardware, this slows down local data processing, resulting in a continuous slight delay in the preprocessing of real-time physiological indicators (such as sudden heart rate changes) and subtle behavioral data (such as sleep disturbance movements).

[0004] If the server is upgraded, to improve the accuracy of anomaly prediction, the analysis module will be optimized and the timing consistency and integrity checks on gateway data uploads will be tightened (e.g., stricter time limits). Previously tolerable minor data delays are now judged by the server as timeouts or incomplete. Because the server does not consider the actual operating status of the front end, it frequently discards timed-out data packets or fails to correlate delayed data with historical sequences. Ultimately, due to incomplete data and broken timelines, it cannot promptly identify subtle patient anomalies and issue warnings. Summary of the Invention

[0005] This application provides a big data-based management method and system for Alzheimer's disease elderly care platforms, aiming to solve the problems in existing Alzheimer's disease elderly care platforms. Due to hardware performance limitations and strict data integrity requirements, it is difficult to reliably transmit and accurately interpret key information during data transmission, processing, and verification. Furthermore, data loss or deviation may cause early warning failures, thereby affecting patient safety and care quality.

[0006] On the one hand, this application provides a big data-based Alzheimer's disease elderly care platform management method, including: Receive patient behavior data from the local data aggregation gateway; The system monitors whether there are missing data points in the patient behavior data information that constitute the key behavioral event chain of the patient. When missing data points are detected, the missing data points are inferred based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. Assess the reliability of the target patient's key behavioral event chain and compare it with a reliability assessment threshold to obtain the comparison results; When the comparison results reach the warning criteria, a warning message with a reliability quantification index is generated.

[0007] Optionally, the step of inferring missing data points and obtaining the target patient's key behavioral event chain based on the pre-stored historical behavioral pattern data of the patient's behavioral profile and the patient's behavioral data information includes: The weights of the patient behavior data, the missing data points, and the fuzzy signal information output by the auxiliary sensing device are obtained based on the timeliness and contextual relevance of the patient behavior data, the missing data points, and the fuzzy signal information. The weights of the patient behavior data, the missing data points, and the fuzzy signal information are adjusted based on the synergistic and conflicting relationships among the patient behavior data, the missing data points, and the fuzzy signal information. Based on the adjusted patient behavior data, the missing data points, and the weights of the fuzzy signal information, the probability of occurrence of different potential behavioral events is iteratively calculated until the probability of occurrence converges. The missing data points are then inferred to obtain the target patient's key behavioral event chain.

[0008] Optionally, the step of assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain the comparison result includes: Continuously monitor patients' real-time behavioral data; Based on the real-time behavioral data, and considering the urgency and tolerance requirements for early warning in the current time period, the situation sensitivity factor is calculated. Based on the real-time behavioral data, the trend of changes in the patient's behavioral patterns over a period of time is continuously analyzed, the degree of individualization and the speed of evolution of the behavioral patterns are quantified, and behavioral pattern variation factors are obtained. Based on the context sensitivity factor and the behavior pattern variation factor, a reliability assessment threshold is generated; Assess the reliability of the target patient's key behavioral event chain and compare it with a reliability assessment threshold to obtain the comparison results.

[0009] Optionally, the step of assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain the comparison result includes: Continuously monitor key factors in the care environment to identify the current situational state; Obtain the information of the currently executed care strategy and match it with the situational state to determine the urgency requirement for triggering the warning; Calculate the early warning urgency factor based on the matching results of the situational state and care strategy; Adjust the reliability assessment threshold according to the aforementioned early warning urgency factor; The reliability of the target patient's key behavioral event chain was compared with the adjusted reliability assessment threshold to obtain the comparison results.

[0010] Optionally, the step of generating a reliability assessment threshold based on the context sensitivity factor and the behavior pattern variation factor includes: Monitor the reliability of the original data sources that constitute the context sensitivity factor and the behavioral pattern variation factor; Calculate the reliability weight of the original data source based on its reliability. Based on the patient's stage of illness and care environment, the influence weights of the situational sensitivity factor and the behavioral pattern variation factor are adjusted to obtain the target influence weights. Based on the reliability weight of the original data source, the situation sensitivity factor and the behavior pattern variation factor are weighted and adjusted to obtain the target situation sensitivity factor and the target behavior pattern variation factor. A reliability assessment threshold is generated based on the target impact weight, the target context sensitivity factor, and the target behavior pattern variation factor.

[0011] Optionally, the step of monitoring the reliability of the original data source constituting the context sensitivity factor and the behavioral pattern variation factor includes: Receive data from multiple raw data sources; The data from each original data source is analyzed in multiple dimensions, including data completeness, data temporal consistency, and the degree of deviation between the data and adjacent data points. Obtain the historical performance baseline for each original data source, which includes the typical reliability performance of the original data source under different operating conditions; Based on the data integrity, the data temporal consistency, the degree of deviation between the data and adjacent data points, and the historical performance baseline, a reliability score is calculated for each original data source. When the reliability of an original data source changes, the reliability scores of other original data sources it depends on are adjusted according to the interdependencies between different original data sources. Adjust the reliability of each original data source based on the adjusted reliability score.

[0012] Optionally, the step of calculating the reliability weight of the original data source based on its reliability includes: Based on the reliability score of the original data source, a nonlinear function is used to convert the reliability score of the original data source into a reliability weight, wherein the parameters of the nonlinear function are preset according to the type of the original data source.

[0013] Optionally, the step of continuously monitoring the patient's real-time behavioral data includes: It receives data streams from multiple heterogeneous sensors and identifies the sensor type, sampling frequency, and data format of each data stream; Timestamp calibration is performed on each received data stream; Based on the sensor type and the data format, each data stream is converted into a unified data structure and data type; Analyze the timestamp intervals and data value changes in each data stream to remove duplicate or redundant data points; Outlier detection and noise reduction are performed on the data in each data stream, and the data is classified and labeled according to preset semantic rules to form standardized behavioral data.

[0014] Optionally, the step of timestamping each received data stream includes: Continuously monitor the clock synchronization service status of the central server; Continuously monitor the battery level and internal crystal oscillator status of each sensor; The reference weight of the central clock is obtained based on the clock synchronization service status of the central server. The reference weight of the sensor's own clock is obtained based on the sensor's battery level and internal crystal oscillator state. The correction factor for timestamp calibration is calculated based on the reference weight of the central clock and the reference weight of the sensor's own clock. Each data stream is timestamped based on the correction factor for timestamping calibration.

[0015] On the other hand, this application provides a big data-based Alzheimer's disease elderly care platform management system, which includes: The data receiving module is used to receive patient behavior data from the local data aggregation gateway; The data inference module is used to monitor whether there are missing data points in the patient behavior data information that constitute the key behavioral event chain of the patient. When missing data points are detected, the missing data points are inferred based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. The reliability assessment module is used to evaluate the reliability of the target patient's key behavioral event chain and compare it with the reliability assessment threshold to obtain the comparison results; The early warning information generation module is used to generate early warning information with reliability quantification indicators when the comparison results meet the early warning criteria.

[0016] This application relates to a big data-based Alzheimer's disease elderly care platform management method and system. By receiving patient behavior data and intelligently inferring missing data points, it effectively compensates for data processing delays caused by local data aggregation gateway processor overload and heat dissipation degradation, as well as data loss and temporal sequence breakage caused by the strict verification rules of the central server. This method infers missing data points based on pre-stored patient behavior profiles, historical behavior patterns, and patient behavior data, thereby obtaining a complete chain of key behavioral events for the target patient. This overcomes the limitation of existing technologies where incomplete data prevents comprehensive and reliable judgment. Furthermore, by evaluating the reliability of this event chain and comparing it with dynamically generated reliability assessment thresholds, when the comparison result reaches the warning standard, a warning message with quantitative reliability indicators is generated, ensuring the timeliness and accuracy of the warning. This technical solution effectively solves the problem of reliable transmission and accurate interpretation of key information under non-ideal operating conditions in existing platforms, avoiding warning failures due to data bias or missing data. This significantly improves the quality and safety of care for Alzheimer's patients, providing caregivers with the optimal opportunity for early intervention. Attached Figure Description

[0017] To illustrate this application more clearly, the accompanying drawings used in the embodiments will be briefly described below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0018] Figure 1 The illustration above is a flowchart of a big data-based Alzheimer's disease elderly care platform management method in an embodiment. Figure 2 The diagram above illustrates a schematic representation of a big data-based Alzheimer's disease elderly care platform management system in an embodiment.

[0019] Figure labeling: 100, Big Data-Based Alzheimer's Disease Elderly Care Platform Management System; 10, Data Receiving Module; 20, Data Inference Module; 30, Reliability Assessment Module; 40, Early Warning Information Generation Module. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] Traditional Alzheimer's care platforms often suffer from data gaps and temporal discrepancies during data collection, transmission, and processing due to hardware limitations of local data aggregation gateways and stringent data verification requirements of central servers. This lack of data integrity and temporal discrepancies prevent the platform from effectively identifying and proactively warning of subtle abnormal behavioral patterns in specific patients at critical points in time, directly impacting the quality of elderly care services and patient safety.

[0023] like Figure 1 The illustration shows a flowchart of a big data-based Alzheimer's disease elderly care platform method in an embodiment. This application proposes a big data-based Alzheimer's disease elderly care platform method, comprising: S10, Receive patient behavior data information from the local data aggregation gateway; Among them, patient behavior data refers to the data set collected through various sensors (such as wearable devices, environmental sensors, cameras, etc.) that reflects the patient's daily activities, physiological indicators, location information, etc.

[0024] S20, monitor whether there are missing data points constituting the key behavioral event chain of the patient in the patient behavior data information. When missing data points are detected, infer the missing data points based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. The patient's key behavioral event chain refers to a sequence of events consisting of a series of logically related patient behavioral data points that reflect a patient's specific state or trend, such as getting out of bed at night, prolonged inactivity, or unusual wandering. Data point missing refers to the phenomenon where certain key data in patient behavioral information is not fully recorded or transmitted due to various reasons (such as equipment failure, network interruption, data processing delays, etc.). A patient behavior profile is a pre-stored comprehensive database containing information such as the patient's historical behavioral patterns, physiological characteristics, and disease progression.

[0025] S30, assess the reliability of the target patient's key behavioral event chain and compare it with the reliability assessment threshold to obtain the comparison results; The reliability assessment threshold is a standard used to measure the credibility of the chain of key behavioral events of a target patient, and its value is dynamically adjusted according to the specific context and patient status.

[0026] S40: When the comparison result reaches the warning standard, a warning message with a reliability quantification index is generated.

[0027] Early warning criteria refer to the conditions that trigger the generation of early warning information, which are usually related to reliability assessment thresholds and the urgency of the early warning.

[0028] The big data-based Alzheimer's disease elderly care platform management method proposed in this application aims to improve the early warning capability for abnormal behavior in Alzheimer's patients through intelligent data processing and analysis.

[0029] Firstly, there are several ways to receive patient behavior data from a local data aggregation gateway. For example, the local data aggregation gateway can be configured to periodically upload collected patient behavior data to the management platform via a wired or wireless network interface. This data can be raw sensor data or data that has undergone preliminary cleaning and formatting. Alternatively, the local data aggregation gateway can adopt an event-driven approach, immediately pushing relevant patient behavior data to the management platform when a specific change in patient behavior is detected or preset conditions are met.

[0030] Secondly, regarding the monitoring of whether there are missing data points constituting the key behavioral event chain in patient behavior data, the following methods can be used: Continuous analysis of the received patient behavior data stream can be conducted by checking the sequence number of the data packets, the continuity of timestamps, and the completeness of data fields to determine if there are missing data points. For example, if a series of data points from a specific sensor are expected to be received within a certain time period, but the actual number of data points received is less than expected, or if timestamps jump, then it can be determined that there are missing data points.

[0031] Furthermore, when missing data points are detected, the missing data points are inferred based on pre-stored historical behavioral pattern data and patient behavior data from the patient's behavioral profile, thus obtaining the target patient's key behavioral event chain. One implementation approach is to utilize statistical methods, such as time series analysis (e.g., ARIMA models) or Markov chain models, to predict the possible values ​​of missing data points based on the patient's historical behavioral pattern data and existing patient behavior data. For example, if a patient typically engages in a certain activity within a specific time period, and data for that period is missing, the missing data points can be inferred based on their historical behavioral patterns. Another implementation approach is to employ a rule-based inference engine, combining pre-defined expert rules and the patient's behavioral profile to perform logical inference on missing data points. For example, if a patient is recorded as being in their bedroom at a certain time, but subsequent data is missing, but their historical behavioral profile shows they typically go to the bathroom at that time, it can be inferred that the missing data point may be related to going to the bathroom.

[0032] Next, regarding the assessment of the reliability of the target patient's key behavioral event chain and its comparison with a reliability assessment threshold, one approach is to calculate the reliability of the event chain based on the number of inferred missing data points, the confidence level of the inference method, and the completeness of the target patient's key behavioral event chain. For example, fewer missing data points and higher inference confidence result in a higher reliability of the event chain. The calculated reliability is then compared with a preset reliability assessment threshold to determine if it meets the warning criteria. Another approach is to combine cross-validation of multi-source information to assess reliability. For example, if the inferred event chain is consistent with data from other assistive sensing devices (such as environmental sensors, smart mattresses, etc.), the reliability score of the event chain can be improved.

[0033] Finally, when the comparison results meet the warning criteria, a warning message with a reliability quantification index is generated. One implementation is to immediately generate a warning message when the reliability of the target patient's key behavioral event chain falls below the reliability assessment threshold, and notify caregivers through various channels (such as mobile apps, SMS, and voice broadcasts). The warning message will include a reliability quantification index, such as a percentage value or a confidence interval, to intuitively inform caregivers of the warning's credibility. Another implementation is to generate warning messages of different urgency levels based on different levels of reliability quantification indexes. For example, when reliability is extremely low, a top-level emergency warning is generated; when reliability is moderate, a general warning is generated, suggesting further observation by caregivers.

[0034] The big data-based Alzheimer's disease elderly care platform management method proposed in this application firstly continuously receives patient behavior data from a local data aggregation gateway. This data is fundamental to understanding the patient's condition. However, due to various complex factors, this data is often incomplete. Therefore, the method proactively monitors this data for missing data points that constitute the patient's key behavioral event chain. Once a data gap is detected, the analysis of that event chain is not simply abandoned. Instead, the missing data points are intelligently inferred using historical behavioral pattern data from pre-stored patient behavior profiles and existing patient behavior data, thereby recovering the complete target patient's key behavioral event chain as much as possible. This inference process is one of the core innovations of this application, enabling a continuous understanding of patient behavior even under non-ideal data conditions.

[0035] After obtaining the key behavioral event chain of the target patient, its reliability is assessed. This assessment considers not only the confidence level of data inference but may also incorporate other contextual factors to ensure the accuracy of the warning. The assessment results are then compared with dynamically adjusted reliability assessment thresholds. If the comparison results meet the warning criteria—that is, the reliability of the event chain is insufficient to support routine judgment or there is a potential risk—a warning message with a quantitative reliability indicator is generated. This quantitative indicator intuitively informs caregivers of the credibility of the warning, avoiding false alarms or missed alarms, thereby improving the practicality of the warning. The entire process forms a complete chain from data reception, missing data inference, reliability assessment to warning generation, ensuring timely and accurate identification and warning of abnormal behaviors in Alzheimer's patients even with incomplete data. This effectively solves the problem of warning failure due to data missing data mentioned in the background technology, significantly improving the quality of elderly care services and patient safety.

[0036] The core innovation of this application lies in its ability to proactively and intelligently infer the missing data points—which constitute the key behavioral event chain of a patient—when the patient's behavioral data is found to be missing, instead of passively accepting the negative impact of data loss, it intelligently deduces the missing data points based on pre-stored historical behavioral pattern data from the patient's behavioral profile and the patient's behavioral data itself. This inference mechanism effectively compensates for data fragmentation caused by hardware performance limitations and strict data verification rules, enabling a continuous understanding of patient behavior even under suboptimal operating conditions.

[0037] Furthermore, this application introduces an assessment of the reliability of the target patient's key behavioral event chain and compares it with a reliability assessment threshold. This assessment mechanism ensures that even after data inference, the resulting early warning information is credible. When the comparison result meets the early warning standard, an early warning message with a quantitative reliability index is generated. This allows caregivers to intuitively understand the reliability of the early warning, thereby making more accurate judgments and interventions. The introduction of this quantitative index is generally lacking in existing technologies, and it greatly enhances the practicality and decision support value of the early warning information.

[0038] In summary, this application effectively solves the problem of early warning failure due to data gaps in existing technologies by introducing intelligent data inference and reliability quantification assessment mechanisms. It not only improves the data processing capabilities of Alzheimer's disease elderly care platforms in complex operating environments, but more importantly, it significantly enhances the accuracy of identifying abnormal patient behaviors and the timeliness and reliability of early warnings, thereby providing safer and more personalized care services for Alzheimer's patients.

[0039] In some embodiments, the step of inferring missing data points and obtaining the target patient's key behavioral event chain based on pre-stored historical behavioral pattern data from patient behavior profiles and the patient behavior data information includes: The weights of the patient behavior data, the missing data points, and the fuzzy signal information output by the auxiliary sensing device are obtained based on the timeliness and contextual relevance of the patient behavior data, the missing data points, and the fuzzy signal information. The weights of the patient behavior data, the missing data points, and the fuzzy signal information are adjusted based on the synergistic and conflicting relationships among the patient behavior data, the missing data points, and the fuzzy signal information. Based on the adjusted patient behavior data, the missing data points, and the weights of the fuzzy signal information, the probability of occurrence of different potential behavioral events is iteratively calculated until the probability of occurrence converges. The missing data points are then inferred to obtain the target patient's key behavioral event chain.

[0040] Specifically, the fuzzy signal information can be understood as indirect behavioral data from auxiliary sensing devices (e.g., environmental sensors, wearable devices, smart home devices, etc.). This data may not directly indicate specific behaviors, but it can provide contextual clues, such as ambient temperature, light intensity, sound patterns, and the status of doors and windows. Timeliness refers to the proximity of the data point or fuzzy signal to the time of the current event; generally, the closer the proximity, the higher its weight. Contextual relevance refers to the correlation between the data point or fuzzy signal and the patient's current environment or ongoing activity; the higher the relevance, the greater its weight. These weights can be calculated using preset rules, machine learning models (e.g., logistic regression, support vector machines), or expert systems to quantify the contribution of different information sources to the inference result.

[0041] The synergistic relationship refers to a relationship where different data sources or signals corroborate each other, enhancing their credibility. For example, if patient behavior data indicates that they may be active in the kitchen, and environmental sensors detect signs of activity in the kitchen area, then the two are synergistic. The conflicting relationship refers to a relationship where different data sources or signals contradict each other, reducing their credibility. For example, if patient behavior data indicates that they are resting in the bedroom, but assistive sensing devices detect vigorous activity in the living room, then the two are conflicting. The weights can be adjusted using methods such as Bayesian networks, evidence theory, or fuzzy logic to dynamically reflect the complex interactions between different information sources, thereby more accurately assessing the overall credibility of the information.

[0042] In practical applications, iterative computation is achieved by repeatedly executing a series of computational steps to gradually approach a stable or optimal solution. The probability of occurrence of different potential behavioral events refers to the likelihood of various behavioral events that a patient may be performing or about to perform, given existing data and adjusted weights. Convergence refers to the gradual decrease in the magnitude of change in the calculated probability of occurrence during the iteration process, eventually reaching a stable state, indicating that the model's inferences about missing data points have become consistent and reliable. This can be achieved through probabilistic inference techniques such as the Expectation-Maximization (EM) algorithm, Markov Chain Monte Carlo (MCMC) method, or deep learning models (e.g., Recurrent Neural Networks, RNNs) to ensure the accuracy and stability of the inference results.

[0043] The technical solution of this application introduces fuzzy signal information output by auxiliary sensing devices and comprehensively considers its timeliness and contextual relevance to assign dynamic weights to patient behavior data and missing data points, thereby overcoming the limitations of relying solely on historical behavior patterns for inference. Furthermore, these weights are adjusted by analyzing the synergistic and conflicting relationships between different information sources, enabling more intelligent identification and utilization of effective information while suppressing the influence of noise and contradictory information. Finally, iterative calculations are performed based on the adjusted weights until the probability of potential behavioral events converges, ensuring a more rigorous and accurate inference process for missing data points and significantly improving the quality and reliability of constructing the key behavioral event chain for the target patient.

[0044] Through the aforementioned technical solution, this application can more comprehensively utilize multi-source heterogeneous data, including direct behavioral data and auxiliary contextual signals, thereby significantly improving the accuracy and robustness of inference for missing data points in situations where data is missing or information is ambiguous. This technical solution, through dynamic weight adjustment and iterative probability calculation, enables more precise capture of subtle changes in patient behavior patterns, effectively reducing the risk of misjudgment and missed judgment, and ultimately generating a more reliable chain of key behavioral events for the target patient, providing a more accurate early warning foundation for the elderly care management platform for Alzheimer's patients.

[0045] In some alternative embodiments, it is assumed that an Alzheimer's patient has missing data points regarding "whether medication was taken on time" in their key behavioral event chain during a specific time period. Traditional methods might infer this from historical medication habits (e.g., taking medication at 8 a.m. daily) and recent behavioral data (e.g., the patient was near the medicine cabinet around 8 a.m.). However, such inferences may be subject to uncertainty.

[0046] First, the system receives fuzzy signal information from assistive sensing devices, such as records of smart pillbox openings, changes in the weight of medication within the pillbox, and changes in physiological indicators (such as heart rate and activity levels) recorded by wearable devices worn by the patient. Based on the timeliness (e.g., the proximity of pillbox opening time to medication administration time) and contextual relevance (e.g., the correlation between changes in physiological indicators and post-medication bodily responses) of this fuzzy signal information, it calculates patient behavior data, data point deficiencies, and initial weights for these fuzzy signal information.

[0047] Secondly, the system analyzes the synergistic and conflicting relationships between this information. For example, if the smart pillbox shows that it was opened at 8:05 AM and the medication weight decreased, while the patient's activity level stabilized after 8:15 AM, there is a synergistic relationship between these pieces of information, and they will mutually reinforce each other's weights. Conversely, if the pillbox was opened but the medication weight did not decrease, or if the patient's activity level increased abnormally after the medication time, there may be a conflicting relationship, and the weights of this information will be adjusted accordingly to reduce the impact of conflicting information.

[0048] Finally, based on the adjusted weights, the probabilities of the potential behavioral events "patient has taken medication" and "patient has not taken medication" are iteratively calculated. For example, using a Bayesian inference model, the posterior probabilities of these two events are continuously updated until the probability values ​​converge. When the probability of "patient has taken medication" converges to a high confidence value, the missing medication data points can be inferred, thus obtaining a more accurate chain of key behavioral events for the target patient. In this way, even when direct behavioral data is incomplete, the accuracy of identifying and inferring key patient behavioral events can be improved through the fusion of multi-source information and intelligent inference.

[0049] In some embodiments, the step of assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain a comparison result includes: Continuously monitor patients' real-time behavioral data; Based on the real-time behavioral data, and considering the urgency and tolerance requirements for early warning in the current time period, the situation sensitivity factor is calculated. Based on the real-time behavioral data, the trend of changes in the patient's behavioral patterns over a period of time is continuously analyzed, the degree of individualization and the speed of evolution of the behavioral patterns are quantified, and behavioral pattern variation factors are obtained. Based on the context sensitivity factor and the behavior pattern variation factor, a reliability assessment threshold is generated; Assess the reliability of the target patient's key behavioral event chain and compare it with a reliability assessment threshold to obtain the comparison results.

[0050] Specifically, various sensors deployed in the elderly care environment (such as wearable devices, environmental sensors, and cameras) continuously collect patients' physiological, activity, and environmental data to achieve real-time monitoring of patients' behavioral data. This data is aggregated and processed in real time to reflect the patient's current health status, activity patterns, and environment.

[0051] Based on the real-time behavioral data, and considering the urgency and tolerance requirements for early warning in the current time period, a situational sensitivity factor is calculated. The situational sensitivity factor can be understood as a dynamic assessment of the risk of potential events in the current situation. Its calculation takes into account both the urgency of early warning (e.g., the urgency of falls at night is higher than during the day) and tolerance (e.g., for high-risk patients, it is better to overreport false alarms than to miss alarms). For example, when a patient is at night, alone, or in a specific dangerous area (such as near a bathroom or stairs), the situational sensitivity factor is increased, indicating that the need for early warning of potential abnormal behavior is more sensitive and urgent at this time.

[0052] In practical applications, based on the real-time behavioral data, the trend of changes in the patient's behavioral patterns over a period of time is continuously analyzed to quantify the degree of individualization and the rate of evolution of the behavioral patterns, thus obtaining the behavioral pattern variance factor. The behavioral pattern variance factor aims to capture abnormal or significant changes in the patient's behavioral patterns. The degree of individualization refers to the uniqueness of the patient's behavioral pattern, while the rate of evolution reflects how quickly this pattern changes. For example, if a patient's daily routine, activity range, or habitual actions suddenly deviate significantly, and this deviation is rapid, the behavioral pattern variance factor will be increased, indicating a potential risk.

[0053] Therefore, a reliability assessment threshold is generated based on the context sensitivity factor and the behavioral pattern variation factor. This threshold is not a fixed value, but is adjusted in real time based on the two dynamic factors mentioned above. For example, the context sensitivity factor and the behavioral pattern variation factor can be combined using weighted averaging, machine learning models, or other algorithms to generate a dynamic reliability assessment threshold that reflects the current context and patient status.

[0054] Finally, the reliability of the target patient's key behavioral event chain is compared with the dynamically generated reliability assessment threshold mentioned above to obtain the comparison results.

[0055] The technical solution of this application introduces a context-sensitive factor and a behavioral pattern variation factor, enabling the reliability assessment threshold to be dynamically adjusted based on the patient's real-time context and behavioral changes. The context-sensitive factor ensures responsiveness to the urgency and tolerance requirements of warnings within the current environment and time period; for example, patients are more vigilant at night or in high-risk areas. The behavioral pattern variation factor, by continuously analyzing the individualization and evolution speed of patient behavior patterns, captures subtle or rapid changes in patient behavior, which are often early signals of potential health problems. It is precisely this dynamic combination of these two factors that allows the reliability assessment threshold to accurately reflect the true risk level currently faced by the patient, thereby avoiding the false alarms or false negatives that may arise from fixed thresholds.

[0056] Through the aforementioned technical solution, this application enables a more accurate and dynamic assessment of the reliability of key behavioral event chains in target patients. Because the reliability assessment threshold can adaptively adjust according to changes in the patient's real-time context and behavioral patterns, the accuracy and timeliness of early warning information are significantly improved, effectively reducing the risk of false alarms and missed alarms. This allows the elderly care platform to provide more personalized and contextualized care services, thereby improving the safety and quality of life for Alzheimer's patients.

[0057] In some alternative embodiments, it is assumed that an Alzheimer's patient's daily behavior pattern shows that they are typically asleep between 10 p.m. and 6 a.m. the following morning.

[0058] Specifically, at 2 AM, the patient was detected leaving the bed and moving towards the bedroom door. Due to the late hour, the urgency of the alert was high, and considering the patient's specific condition, the margin for error was low; therefore, the calculated situational sensitivity factor was increased. Simultaneously, analysis of the patient's real-time behavioral data revealed subtle differences in their movement speed and gait compared to their usual nighttime bathroom routine, causing a slight increase in the behavioral pattern variation factor. Based on these two dynamic factors, a relatively high reliability assessment threshold was generated. If the reliability of the inferred "patient's abnormal nighttime activity" key behavioral event chain fell below this dynamically adjusted threshold, an immediate warning was generated, prompting caregivers to intervene and effectively prevent potential fall risks.

[0059] For example, during the day, the patient typically spends time in the living room. However, one afternoon, the patient was observed repeatedly attempting to open the front door for an extended period. This behavior deviates significantly from the patient's usual activity pattern, thus the behavior pattern variability factor is significantly increased. Even if the situational sensitivity factor is not high at this time (e.g., during the day and in a relatively safe area), the reliability assessment threshold is dynamically adjusted to a higher level due to the significant abnormality of the behavior pattern. If the reliability of the inferred "patient attempting to leave home" key behavioral event chain falls below this threshold, an immediate alert will be issued, prompting caregivers to pay attention and take timely action to prevent the patient from getting lost.

[0060] In some embodiments, the step of assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain a comparison result includes: Continuously monitor key factors in the care environment to identify the current situational state; Obtain the information of the currently executed care strategy and match it with the situational state to determine the urgency requirement for triggering the warning; Calculate the early warning urgency factor based on the matching results of the situational state and care strategy; Adjust the reliability assessment threshold according to the aforementioned early warning urgency factor; The reliability of the target patient's key behavioral event chain was compared with the adjusted reliability assessment threshold to obtain the comparison results.

[0061] By utilizing various sensors and data sources deployed within the elderly care platform, such as environmental sensors, location sensors, and activity sensors, information about the patient's physical environment, time, spatial location, and the activities of surrounding people is collected in real time. This enables continuous monitoring of key factors in the care environment and identification of the current situational state. These key factors may include, but are not limited to, the patient's current location, room temperature, light intensity, sound level, the presence of caregivers, and the current time of day (e.g., daytime, nighttime, mealtime). Through comprehensive analysis of this multi-dimensional data, the specific situational state of the patient can be identified, such as "the patient is resting in the bedroom," "the patient is active in a public area," or "the patient is near the exit area."

[0062] Based on the patient's individualized care plan, medical orders, and general safety protocols of the elderly care facility, the currently implemented or required care strategies are obtained. This information on the currently implemented care strategies is then matched with the situational state to determine the urgency requirement for triggering an alert. For example, for an Alzheimer's patient at risk of nighttime wandering, their nighttime care strategy might include restricting their movement out of the bedroom area. When an identified situational state (e.g., "patient is near the exit area") conflicts with the current care strategy (e.g., "restricting nighttime movement out of the bedroom"), the urgency requirement for triggering an alert is determined based on this matching result. The urgency requirement can be a quantifiable indicator reflecting the degree to which a potential behavioral event is identified as abnormal and requires immediate attention in the current situation.

[0063] Based on the matching results between the aforementioned situational state and care strategy, a warning urgency factor is calculated using pre-defined rules or a machine learning model. This factor is a numerical value used to quantify the necessity and urgency of the warning in the current situation. For example, if a patient is in a high-risk area during inactive hours and the care strategy explicitly prohibits such behavior, the warning urgency factor will be high; conversely, if a patient is engaged in routine activities in a safe area during active hours, the warning urgency factor will be low.

[0064] The calculated warning urgency factor is applied to the dynamic adjustment of the reliability assessment threshold. A high warning urgency factor indicates a lower tolerance for potential abnormal behavior in the current situation. In this case, the reliability assessment threshold is lowered accordingly, making it easier to trigger warnings even if the reliability of the target patient's critical behavioral event chain is relatively low, thus ensuring a timely response. Conversely, a low warning urgency factor indicates a higher tolerance for behavior in the current situation. The reliability assessment threshold is then raised to avoid generating unnecessary warnings at non-critical times.

[0065] After the reliability assessment threshold is dynamically adjusted, the reliability of the target patient's key behavioral event chain is compared with this new, context-sensitive threshold to obtain the comparison results. The comparison results will more accurately reflect whether the behavioral event chain has met the warning criteria under the current specific care context and strategy.

[0066] The technical solution of this application continuously monitors key factors in the care environment and, in conjunction with the currently implemented care strategies, dynamically calculates an early warning urgency factor, and adjusts the reliability assessment threshold based on this factor. It is precisely because the reliability assessment threshold can adaptively adjust according to the actual situation and care needs that the judgment of abnormal patient behavior is more accurate in different scenarios. For example, in high-risk periods or areas where patients require strict care, the threshold is lowered to increase the sensitivity of the early warning; while when patients are in safe, low-risk situations, the threshold is raised to reduce unnecessary interference. This dynamic adjustment mechanism effectively solves the problems of untimely early warnings or false alarms that may be caused by traditional fixed thresholds.

[0067] Through the aforementioned technical solution, this application enables the elderly care platform management method to better align with actual care needs and environmental changes when assessing the reliability of patients' key behavioral event chains. Consequently, the generation of early warning information will be more context-aware and personalized, significantly improving the accuracy and timeliness of early warnings, effectively avoiding missed or false alarms caused by inappropriate thresholds, thereby providing more precise and efficient care services for Alzheimer's patients and enhancing the overall intelligent management level of the elderly care platform.

[0068] In some alternative embodiments, suppose an Alzheimer's patient's care strategy dictates that they should rest in their bedroom at night (e.g., from 10 p.m. to 6 a.m. the following morning) and are prohibited from leaving the bedroom area.

[0069] First, key factors in the care environment are continuously monitored. For example, at 2 a.m., location sensors show that the patient has left the bedroom and is moving towards the exit area. At this point, the situation is identified as "patient is approaching the exit area at night".

[0070] Next, the currently implemented care strategy information is obtained, namely "resting in the bedroom at night and not leaving the bedroom area." The situational state is matched with the care strategy, and a conflict is found between the two, thus determining that the urgency requirement for triggering the warning is high.

[0071] Based on this matching result, a higher early warning urgency factor is calculated.

[0072] Subsequently, based on this higher urgency factor for early warning, the reliability assessment threshold was lowered. This means that in the current high-risk situation, the tolerance for potential abnormal behavior is reduced, and early warnings are more likely to be triggered even if the reliability of the patient's key behavioral event chain is not very high.

[0073] Finally, the reliability of the inferred target patient key behavioral event chain is compared with this downgraded reliability assessment threshold. If the comparison meets the warning criteria (e.g., reliability is higher than the downgraded threshold), a warning message with quantitative reliability indicators is immediately generated to notify caregivers to take intervention measures.

[0074] In this way, the technical solution of this application can ensure that potential dangers can be identified more sensitively and timely warnings can be issued when patients are in high-risk situations, thereby effectively protecting the safety of patients.

[0075] In some embodiments, the step of generating a reliability assessment threshold based on the context sensitivity factor and the behavior pattern variation factor includes: Monitor the reliability of the original data sources that constitute the context sensitivity factor and the behavioral pattern variation factor; Calculate the reliability weight of the original data source based on its reliability. Based on the patient's stage of illness and care environment, the influence weights of the situational sensitivity factor and the behavioral pattern variation factor are adjusted to obtain the target influence weights. Based on the reliability weight of the original data source, the situation sensitivity factor and the behavior pattern variation factor are weighted and adjusted to obtain the target situation sensitivity factor and the target behavior pattern variation factor. A reliability assessment threshold is generated based on the target impact weight, the target context sensitivity factor, and the target behavior pattern variation factor.

[0076] Specifically, continuous quality and reliability assessments are performed on all input data sources used to calculate the context sensitivity factor and behavioral pattern variation factor, thereby monitoring the reliability of the original data sources that constitute these factors. These original data sources can include data from various heterogeneous sensors, medical records, environmental monitoring devices, etc. The aim is to ensure that the underlying data upon which subsequent calculations rely is reliable, thus avoiding assessment bias due to data quality issues.

[0077] The process involves calculating the reliability weight of each original data source. This can be understood as assigning a quantified weight value to each source based on its monitored reliability level. The reliability weight reflects the data source's contribution to the generation of context sensitivity factors and behavioral pattern variation factors. For example, data sources with high reliability will be assigned higher weights, and vice versa. The purpose is to allow for differentiated processing based on the quality differences of the data sources in subsequent calculations.

[0078] In practical applications, the relative importance of situational sensitivity factors and behavioral pattern variation factors in generating reliability assessment thresholds is dynamically adjusted based on the specific stage of Alzheimer's disease (e.g., early, middle, or late stage) and the current care environment (e.g., home care, professional nursing homes, hospitals, etc.). For example, for patients with more severe conditions or in high-risk care environments, it may be necessary to increase the influence weight of the situational sensitivity factor to respond more promptly to potential abnormal situations. The aim is to make the reliability assessment thresholds better adaptable to the individualized needs of patients and the actual situation.

[0079] Furthermore, the reliability weights of the original data source calculated earlier are applied to the calculation of the context sensitivity factor and the behavioral pattern variation factor, or their final values. This allows for a weighted correction of the context sensitivity factor and the behavioral pattern variation factor based on the reliability weights of the original data source, resulting in the target context sensitivity factor and the target behavioral pattern variation factor. For example, if the reliability of a data source is low, its contribution to the context sensitivity factor or the behavioral pattern variation factor will be reduced accordingly, resulting in the quality-corrected target context sensitivity factor and the target behavioral pattern variation factor. The aim is to eliminate or mitigate the errors introduced by poor quality original data sources and improve the accuracy of the factors.

[0080] Therefore, by combining the reliability-corrected situational sensitivity factor and behavioral pattern variation factor with the impact weights adjusted according to the patient's condition and care environment, a final reliability assessment threshold is calculated. This allows for the generation of a reliability assessment threshold based on the target impact weight, target situational sensitivity factor, and target behavioral pattern variation factor. This threshold will more accurately reflect the reasonable boundaries for triggering an early warning in the current situation.

[0081] The technical solution of this application effectively solves the problems of insufficient accuracy and poor adaptability in the generation of reliability assessment thresholds in basic technical solutions by introducing monitoring and weight calculation of the reliability of the original data source, and dynamically adjusting the influence weights of situation sensitivity factors and behavioral pattern variation factors according to the patient's disease stage and care environment. Because of the consideration of the reliability of the original data source, the situation sensitivity factors and behavioral pattern variation factors have already undergone quality correction before being used to generate the threshold, thus avoiding assessment bias caused by low-quality data. At the same time, by adjusting the influence weights of these factors according to the patient's individual characteristics and environment, it ensures that the generated reliability assessment thresholds are more in line with actual needs, avoiding a "one-size-fits-all" assessment standard, thereby improving the intelligence and personalization of early warning.

[0082] Through the aforementioned technical solutions, this application significantly improves the accuracy and adaptability of reliability assessment thresholds. Specifically, by quantitatively assessing and weighting the reliability of the original data source, the negative impact of data quality issues on early warning judgments can be effectively reduced, allowing situational sensitivity factors and behavioral pattern variation factors to more accurately reflect the patient's actual condition and risk. Furthermore, by dynamically adjusting the influence weights of these factors, the early warning thresholds can be personalized according to the patient's disease progression and changes in the care environment, thereby avoiding false alarms or missed alarms, improving the reliability and practical value of early warning information, and providing stronger technical support for the precise care of Alzheimer's patients.

[0083] In some alternative embodiments, it is assumed that an Alzheimer's patient wears multiple sensors, including a smartwatch for monitoring heart rate and activity levels, a mattress sensor for monitoring sleep patterns, and a smart home device for monitoring indoor environmental factors such as temperature and humidity.

[0084] First, the reliability of the original data sources that constitute the context sensitivity factor and behavioral pattern variation factor is monitored. For example, data streams from smartwatches may experience data loss or anomalies due to low battery or signal interference, mattress sensors may have inaccurate data due to location changes, and environmental sensors may have deviations due to calibration issues. The data integrity, temporal consistency, and deviation from adjacent data points of these data sources are continuously evaluated, and a reliability score is calculated for each original data source based on its historical performance baseline. For example, a smartwatch with 98% data integrity and good temporal consistency over the past week might have a high reliability score, while a mattress sensor with fluctuating data due to recent maintenance might have a slightly lower reliability score.

[0085] Secondly, based on the reliability scores of these raw data sources, a pre-defined non-linear function (e.g., the sigmoid function) is used to convert them into reliability weights. For example, a smartwatch with a high reliability score might receive a weight of 0.9, while a mattress sensor with a low reliability score might receive a weight of 0.7.

[0086] Next, the influence weights of the situation sensitivity factor and the behavior pattern variation factor are adjusted according to the patient's disease stage and care environment. For example, if a patient is in the middle stage of Alzheimer's disease and is currently in a home care environment, they may perceive small variations in behavior patterns as more indicative than sudden changes in the situation. Therefore, the influence weight of the behavior pattern variation factor is set to 0.6, and the influence weight of the situation sensitivity factor is set to 0.4, thus obtaining the target influence weights.

[0087] Then, the context sensitivity factor and behavior pattern variation factor are weighted and adjusted according to the reliability weights of the original data sources. For example, if the context sensitivity factor mainly relies on data from smartwatches and environmental sensors, its final value will be a weighted average or adjusted based on the reliability weights of these two data sources to obtain the target context sensitivity factor. Similarly, the behavior pattern variation factor will also be adjusted based on the reliability weights of the data sources it depends on (such as smartwatches and mattress sensors) to obtain the target behavior pattern variation factor.

[0088] Finally, the modified target context sensitivity factor and target behavior pattern variation factor, combined with the previously adjusted target influence weights, are used to calculate the final reliability assessment threshold through a pre-defined fusion model (e.g., a weighted summation model or a machine learning model). For example, the reliability assessment threshold = (target context sensitivity factor) / (target behavior pattern variation factor). 0.4) + (Target behavior pattern variation factor) 0.6). This generated threshold will more accurately reflect the warning triggering criteria under the current patient status and data quality.

[0089] In some embodiments, the step of monitoring the reliability of the original data source constituting the context sensitivity factor and the behavioral pattern variation factor includes: Receive data from multiple raw data sources; The data from each original data source is analyzed in multiple dimensions, including data completeness, data temporal consistency, and the degree of deviation between the data and adjacent data points. Obtain the historical performance baseline for each original data source, which includes the typical reliability performance of the original data source under different operating conditions; Based on the data integrity, the data temporal consistency, the degree of deviation between the data and adjacent data points, and the historical performance baseline, a reliability score is calculated for each original data source. When the reliability of an original data source changes, the reliability scores of other original data sources it depends on are adjusted according to the interdependencies between different original data sources. Adjust the reliability of each original data source based on the adjusted reliability score.

[0090] Specifically, receiving data from multiple raw data sources refers to collecting data related to patient behavior, environmental context, etc., from various sensors, devices, or databases. These raw data sources may include wearable devices, environmental sensors, smart home devices, etc., with the aim of gathering comprehensive information to support subsequent reliability assessments.

[0091] This involves multi-dimensional analysis of data from each original data source, including data completeness, data temporal consistency, and the degree of deviation between the data and adjacent data points. Data completeness refers to the integrity of data records, such as the presence of missing values ​​or incomplete records; data temporal consistency refers to the continuity and logic of data over time, such as whether timestamps are continuous and whether data points are generated at expected time intervals; and the degree of deviation between the data and adjacent data points measures whether there are abnormal fluctuations or outliers in the data. The aim is to comprehensively evaluate the quality and reliability of the data from multiple perspectives.

[0092] In practical applications, a historical performance baseline is obtained for each original data source. This historical performance baseline includes the typical reliability performance of the original data source under different operating conditions. The historical performance baseline refers to the historical statistical data and performance patterns of the data source in terms of data integrity, temporal consistency, and degree of deviation under normal or known abnormal conditions. Its purpose is to provide a reference standard for the reliability assessment of the current data source, so as to more accurately determine whether its current state is abnormal.

[0093] Furthermore, based on the data completeness, the data temporal consistency, the degree of deviation between the data and adjacent data points, and the historical performance baseline, a reliability score is calculated for each original data source. This reliability score is a quantitative indicator that comprehensively reflects the data source's performance in the aforementioned multi-dimensional analysis and its deviation from the historical baseline, aiming to provide a unified reliability metric for each data source.

[0094] Furthermore, when the reliability of a primary data source changes, the reliability scores of other primary data sources it depends on should be adjusted based on the interdependencies between different primary data sources. For example, if a wearable device used to monitor a patient's heart rate malfunctions and its data reliability decreases, the reliability scores of other data sources that rely on that heart rate data to determine activity intensity (such as accelerometers) should also be adjusted accordingly.

[0095] Therefore, based on the adjusted reliability score, the reliability of each original data source is adjusted, resulting in a more accurate and dynamic reliability status of the original data sources.

[0096] The technical solution of this application comprehensively, dynamically, and meticulously monitors the reliability of the original data source by introducing multi-dimensional analysis, historical performance baselines, and interdependencies between data sources. Specifically, multi-dimensional analysis can capture potential problems from multiple levels of data quality; historical performance baselines provide objective reference standards, making the judgment of current data reliability more accurate; and considering the interdependencies between data sources ensures that reliability assessment is no longer isolated but forms an interconnected whole, thus avoiding a chain of misjudgments caused by problems with a single data source. It is precisely because of the synergistic effect of these mechanisms that the reliability assessment of the original data source is closer to reality.

[0097] Through the aforementioned technical solution, the reliability of the original data source can be monitored and quantified more accurately and comprehensively. Compared to the simple monitoring methods that may exist in basic technical solutions, this application can effectively identify and quantify potential problems in the data source, such as missing data, disordered time series, or abnormal fluctuations, and further consider the complex correlations between data sources. This refined reliability assessment mechanism significantly improves the quality assurance level of the original data that constitutes the context sensitivity factor and behavioral pattern variation factor, thereby making the reliability assessment thresholds generated based on these data more accurate and robust. As a result, the final generated early warning information will have higher reliability quantification indicators, effectively reducing the risk of false alarms and false negatives, and providing a more reliable decision-making basis for the elderly care management of Alzheimer's patients.

[0098] In some alternative embodiments, suppose an elderly care platform management system needs to monitor the daily behavior of Alzheimer's patients, and its data sources include: smart bracelets worn by patients (providing heart rate and step count data), environmental sensors installed in the room (providing temperature and humidity data), and mattress sensors (providing sleep quality data).

[0099] First, data is received from these multiple raw data sources.

[0100] Next, multi-dimensional analysis is performed on the data from each data source. For example, for smart bracelet data, the analysis covers data completeness (whether there is missing heart rate data), data temporal consistency (whether the heart rate data is continuous and whether the timestamp jumps), and the degree of deviation between the data and adjacent data points (whether the heart rate value suddenly becomes abnormally high or low). For environmental sensors, the analysis covers whether the data is complete, whether the temporal sequence is consistent, and whether the temperature and humidity values ​​fluctuate within a reasonable range.

[0101] At the same time, historical performance baselines for each data source are obtained. For example, the average data integrity of smart bracelets under normal wear, and the typical temperature and humidity fluctuation range of environmental sensors in different seasons.

[0102] Then, based on these analytical results and historical baselines, a reliability score is calculated for each original data source. For example, if a smart bracelet's low battery leads to an increased data loss rate, its reliability score will decrease.

[0103] Furthermore, when the reliability score of the smart bracelet decreases, adjustments will be made based on interdependencies. For example, if the reliability of the smart bracelet's step count data decreases, the algorithm that relies on step count data to determine the patient's activity level will also be affected by the reliability score of the original data source (i.e., the smart bracelet) it depends on, thus impacting the reliability of the final activity level assessment.

[0104] Ultimately, based on the adjusted reliability score, the reliability status of each original data source will be dynamically adjusted, thereby providing more accurate input for the subsequent calculation of context sensitivity factor and behavioral pattern variation factor.

[0105] In some embodiments, the step of calculating the reliability weight of the original data source based on the reliability of the original data source includes: Based on the reliability score of the original data source, a nonlinear function is used to convert the reliability score of the original data source into a reliability weight, wherein the parameters of the nonlinear function are preset according to the type of the original data source.

[0106] Specifically, the reliability score of the original data source refers to a quantitative indicator obtained through multi-dimensional analysis of data integrity, data temporal consistency, deviation between the data and adjacent data points, and historical performance baseline. To transform this score into weights with practical influence in subsequent calculations, this application introduces a nonlinear function. This nonlinear function can be understood as a mapping relationship, with the reliability score of the original data source as input and the corresponding reliability weight as output. Compared to linear functions, nonlinear functions can better capture the complex relationship between the reliability score and the actual weight. For example, when the reliability score is low, the weight may decrease faster, while when the reliability score is high, the weight may tend to stabilize. The parameters of the nonlinear function are preset according to the type of the original data source. This means that different nonlinear function parameters can be configured for different types of original data sources, such as wearable sensors, environmental sensors, video surveillance equipment, or manually input data. For example, for medical-grade sensors with extremely high data accuracy requirements, the nonlinear function may cause the weight to decrease rapidly when the reliability score decreases slightly; while for some auxiliary environmental sensors, the nonlinear function may cause the weight to decrease relatively slowly when the reliability score decreases moderately. These parameters can be preset and optimized based on expert experience, historical data analysis, or machine learning models to ensure that each type of data source receives a reliability weight that best matches its characteristics.

[0107] The technical solution of this application solves the problems of uniformity and limitations in the traditional method of converting reliability scores to weights by introducing a nonlinear function and pre-setting its parameters according to the type of the original data source. Because different types of original data sources play significantly different roles and have different data characteristics in Alzheimer's disease elderly care platform management, it is difficult to achieve optimal weight allocation using a uniform conversion mechanism. The nonlinear function allows the reliability score to have different degrees of influence on the weights in different ranges. For example, when the reliability score is high, the weight may change slowly to maintain stability; while when the reliability score falls to a low level, the weight may drop sharply to quickly reduce the impact of unreliable data. Furthermore, by pre-setting the parameters of the nonlinear function according to the type of the original data source, the weight conversion process can fully consider the inherent characteristics of each data source and its importance in specific contexts. For example, for vital sign monitoring devices, small changes in their reliability scores may need to be amplified into significant weight adjustments, while for ambient light sensors, fluctuations in their reliability scores may only require small weight adjustments. This refined, type-aware weighting mechanism ensures that the subsequent weighted adjustments of context sensitivity factors and behavioral pattern variation factors can more accurately reflect the actual situation, thereby improving the accuracy and robustness of the entire early warning system.

[0108] The above technical solution enables a more refined and adaptive conversion from raw data source reliability scores to reliability weights. Specifically, by employing a nonlinear function, weight distortion that might occur with linear conversion can be avoided, allowing changes in reliability scores to more accurately reflect their impact on decision-making. Furthermore, by pre-setting the parameters of the nonlinear function according to the type of raw data source, the characteristics and importance of different data source types are fully reflected in the weight calculation, thus avoiding a "one-size-fits-all" weight allocation approach. Consequently, the obtained reliability weights more accurately reflect the actual reliability of each raw data source, providing a more solid and precise foundation for subsequent weighted adjustments of context sensitivity factors and behavioral pattern variation factors. Ultimately, this significantly improves the accuracy of reliability assessment of the target patient's key behavioral event chain and the reliability of early warning information.

[0109] In some alternative embodiments, it is assumed that the elderly care platform management system receives raw data from three different types of sources: the first type is high-precision wearable vital signs sensors (e.g., heart rate, blood oxygen), the second type is environmental sensors (e.g., indoor temperature, humidity), and the third type is activity monitoring devices based on image recognition.

[0110] For the first type of high-precision wearable vital signs sensor, since its data is crucial for assessing a patient's health condition, even a slight decrease in the reliability score requires a rapid reduction in its reliability weight to avoid misleading results. Therefore, a steep S-shaped nonlinear function can be preset for it, so that the weight decays rapidly when the reliability score drops slightly from the high range.

[0111] For the second type of environmental sensor, its data is usually used as auxiliary information, and its impact on the urgency of early warning is relatively small. Therefore, a relatively smooth logarithmic nonlinear function can be preset for it, so that when the reliability score fluctuates within a moderate range, the weight changes are not so drastic, thus maintaining a certain degree of fault tolerance.

[0112] For the third type of image recognition-based activity monitoring equipment, its reliability may be significantly affected by factors such as lighting and occlusion, and it plays a decisive role in certain critical behavioral event chains. A piecewise linear or exponential nonlinear function can be preset for it. When the reliability score is below a certain threshold, the weight drops sharply, while when it is above the threshold, the weight remains at a high level, ensuring that the reliability of the data is given sufficient attention at critical moments.

[0113] In this way, the reliability score of each original data source can be accurately converted into a reliability weight that best reflects its true impact through a non-linear function that matches its type, thus providing more accurate input for subsequent reliability assessment and early warning generation.

[0114] In some embodiments, the step of continuously monitoring the patient's real-time behavioral data includes: It receives data streams from multiple heterogeneous sensors and identifies the sensor type, sampling frequency, and data format of each data stream; Timestamp calibration is performed on each received data stream; Based on the sensor type and the data format, each data stream is converted into a unified data structure and data type; Analyze the timestamp intervals and data value changes in each data stream to remove duplicate or redundant data points; Outlier detection and noise reduction are performed on the data in each data stream, and the data is classified and labeled according to preset semantic rules to form standardized behavioral data.

[0115] This system receives data streams from multiple heterogeneous sensors, identifies the sensor type, sampling frequency, and data format of each stream, and aims to process data from diverse sources (such as wearable devices, environmental sensors, and video surveillance). These data may possess different physical characteristics and data representation methods. Identifying their type, sampling frequency, and data format lays the foundation for subsequent data preprocessing and integration, ensuring data compatibility and resolvability.

[0116] Timestamp calibration of each received data stream is performed to address potential clock drift or synchronization errors between different sensors. A unified time reference ensures consistency of all behavioral data across the time dimension, thereby accurately reconstructing the patient's behavioral sequence and event chain, and avoiding misjudgments due to time discrepancies.

[0117] Based on the sensor type and data format, each data stream is converted into a unified data structure and data type to eliminate structural and format differences caused by heterogeneous data sources. By mapping all data to a predefined consistency model, subsequent data processing logic can be simplified, improving the efficiency and accuracy of data analysis.

[0118] Analyzing the timestamp intervals and data value changes in each data stream, and removing duplicate or redundant data points, aims to optimize data quality and reduce computational burden. Duplicate data points may result from sensors sending the same information multiple times or retransmitting data, while redundant data points may provide the same or similar information within a short period of time. Removing these data points helps improve data processing efficiency and avoids redundant analysis of the same information.

[0119] Outlier detection and denoising are performed on each data stream, followed by classification and labeling according to predefined semantic rules to form standardized behavioral data, aiming to further improve the reliability and usability of the data. Outlier detection can identify and handle data points that deviate from normal patterns, such as data spikes caused by sensor malfunctions or environmental interference. Denoising can filter out random noise in the data, making the true behavioral signals clearer. Subsequently, the processed data is classified and labeled according to predefined semantic rules, such as interpreting the raw acceleration data as "walking," "falling," or "stationary," thereby transforming low-level sensor data into standardized behavioral data with high-level semantics, providing high-quality input for the subsequent construction and evaluation of key patient behavioral event chains.

[0120] The technical solution of this application, through the aforementioned detailed data processing flow, ensures that the raw data received from heterogeneous sensors can be effectively cleaned, integrated, and standardized. Specifically, firstly, identifying the data characteristics of heterogeneous sensors provides a basis for subsequent unified processing. Secondly, timestamp calibration ensures time synchronization between different data sources, enabling the accurate serialization of patients' real-time behavioral data. Next, unified conversion of data structures and types eliminates processing barriers caused by heterogeneity, allowing all data to be analyzed within a unified framework. Subsequently, removing duplicate or redundant data points effectively reduces the data volume, improves processing efficiency, and reduces potential false alarms. Finally, outlier detection, denoising, and semantic classification and annotation transform the raw, low-level sensor data into highly reliable, highly semantically accurate standard behavioral data. This provides high-quality, high-accuracy input for the subsequent calculation of context sensitivity factors and behavioral pattern variation factors, thereby ensuring the accuracy and effectiveness of the generated reliability assessment threshold and the final warning information.

[0121] Through the aforementioned technical solution, this application can effectively process real-time behavioral data from various heterogeneous sensors, overcoming the shortcomings of traditional methods in handling data heterogeneity, noise, and redundancy. Specifically, by refining data reception, timestamp calibration, data format unification, redundancy removal, outlier detection, denoising, and semantic classification annotation, the quality, accuracy, and reliability of real-time patient behavioral data are significantly improved. This not only lays a solid foundation for the accurate calculation of subsequent context-sensitive factors and behavioral pattern variation factors, making the generated reliability assessment threshold more closely aligned with reality, but also effectively reduces the risk of false alarms and false negatives. This improves the robustness and effectiveness of the entire Alzheimer's disease elderly care platform management method in assessing key behavioral event chains and generating early warnings, providing patients with more accurate and timely care warnings.

[0122] In some embodiments, the step of timestamping each received data stream includes: Continuously monitor the clock synchronization service status of the central server; Continuously monitor the battery level and internal crystal oscillator status of each sensor; The reference weight of the central clock is obtained based on the clock synchronization service status of the central server. The reference weight of the sensor's own clock is obtained based on the sensor's battery level and internal crystal oscillator state. The correction factor for timestamp calibration is calculated based on the reference weight of the central clock and the reference weight of the sensor's own clock. Each data stream is timestamped based on the correction factor for timestamping calibration.

[0123] Specifically, continuously monitoring the clock synchronization service status of the central server refers to acquiring and analyzing the operational status of the clock synchronization service provided by the central server in real time, such as the availability of the Network Time Protocol (NTP) service, synchronization accuracy, and the presence of abnormal fluctuations. Its purpose is to assess the reliability of the central clock as a time reference.

[0124] Continuously monitoring the battery level and internal crystal oscillator status of each sensor refers to acquiring real-time information on the power supply status (e.g., battery level) and the operational stability of its internal clock source (e.g., crystal oscillator) for each sensor. Low battery level may cause crystal oscillator instability, thus affecting the accuracy of the sensor's own clock; the crystal oscillator status directly reflects the degree of clock drift. The purpose is to evaluate the reliability of each sensor's own clock.

[0125] In practical applications, the importance of the central clock in time calibration is quantified based on the quality of the central server's clock synchronization service (e.g., high weight for stable service and high synchronization accuracy, and low weight for unstable service or poor accuracy), so as to obtain the reference weight of the central clock based on the clock synchronization service status of the central server.

[0126] Furthermore, the reliability of the sensor's own clock in time calibration is quantified based on the sensor's battery power (e.g., high weight for sufficient power and low weight for low power) and the stability of the internal crystal oscillator (e.g., high weight for stable crystal oscillator and low weight for severe crystal oscillator drift).

[0127] Therefore, taking into account both the authority of the central clock and the actual operating conditions of the sensor's own clock, a comprehensive correction value for adjusting the data stream timestamp is calculated using a weighted average or other fusion algorithm. For example, when the central clock is highly reliable but the sensor's own clock is unstable, the correction factor will be more biased towards the central clock; conversely, when the sensor's own clock is relatively stable but the central clock service fluctuates, the correction factor may give more consideration to the local stability of the sensor's own clock.

[0128] Finally, the calculated correction factor is applied to the timestamp of each received data stream to eliminate or reduce time deviations caused by clock asynchrony or drift, ensuring that the timestamps of all data streams are aligned on a unified time base, and realizing timestamp calibration for each data stream based on the correction factor calibrated by the timestamp.

[0129] The technical solution of this application dynamically acquires key reference information required for time calibration by continuously monitoring the clock synchronization service status of the central server and the battery power and internal crystal oscillator status of each sensor. It is precisely because these factors affecting clock accuracy are quantitatively evaluated and transformed into reference weights for the central clock and the sensor's own clock that a more accurate and adaptive timestamp calibration correction factor can be calculated based on these weights. In this way, the timestamp calibration strategy can be flexibly adjusted according to changes in the actual operating environment and equipment status, effectively solving the problem of inconsistent or drifting timestamps in heterogeneous sensor data streams and ensuring the timing accuracy of real-time behavioral data.

[0130] Through the above technical solution, this application can significantly improve the accuracy and robustness of timestamp calibration for each received data stream. Compared to simple timestamp calibration, this application introduces dynamic evaluation of the clock status of the central server and sensors themselves, making the time calibration process more adaptive and effectively addressing clock drift issues caused by factors such as equipment aging, insufficient power, or network fluctuations. This ensures a high degree of consistency of real-time patient behavior data across the time dimension, providing a solid data foundation for accurate inference and reliability assessment of subsequent key behavioral event chains, thereby improving the accuracy and reliability of the entire elderly care platform management method.

[0131] This application also proposes a big data-based Alzheimer's disease elderly care platform management system, such as... Figure 2 As shown, a big data-based Alzheimer's disease elderly care platform management system 100 includes: Data receiving module 10 is used to receive patient behavior data information from the local data aggregation gateway; The data inference module 20 is used to monitor whether there are missing data points in the patient behavior data information that constitute the key behavioral event chain of the patient. When a missing data point is detected, the missing data point is inferred based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. The reliability assessment module 30 is used to assess the reliability of the target patient's key behavioral event chain and compare it with the reliability assessment threshold to obtain the comparison result; The early warning information generation module 40 is used to generate early warning information with reliability quantification indicators when the comparison result reaches the early warning standard.

[0132] The big data-based Alzheimer's disease elderly care platform management system proposed in this application demonstrates significant technological progress and innovation in addressing the problems of data gaps and ineffective early warning in existing technologies. Traditional Alzheimer's disease elderly care platforms, when faced with processor throttling and data processing delays due to high load on local data aggregation gateways, and strict data verification rules on the central server, often simply discard data packets deemed timed out or incomplete. This approach directly disrupts the chain of behavioral events requiring continuous data points to form a complete judgment, preventing the platform from effectively identifying and proactively warning of subtle abnormal behavioral patterns in specific patients at critical time points.

[0133] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A big data-based management method for Alzheimer's disease elderly care platforms, characterized in that, include: Receive patient behavior data from the local data aggregation gateway; The system monitors whether there are missing data points in the patient behavior data information that constitute the key behavioral event chain of the patient. When missing data points are detected, the missing data points are inferred based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. Assess the reliability of the target patient's key behavioral event chain and compare it with a reliability assessment threshold to obtain the comparison results; When the comparison results reach the warning criteria, a warning message with a reliability quantification index is generated.

2. The management method for Alzheimer's disease elderly care platform based on big data according to claim 1, characterized in that, The step of inferring missing data points and obtaining the target patient's key behavioral event chain based on the pre-stored patient behavior profile's historical behavior pattern data and the patient behavior data information includes: The weights of the patient behavior data, the missing data points, and the fuzzy signal information output by the auxiliary sensing device are obtained based on the timeliness and contextual relevance of the patient behavior data, the missing data points, and the fuzzy signal information. The weights of the patient behavior data, the missing data points, and the fuzzy signal information are adjusted based on the synergistic and conflicting relationships among the patient behavior data, the missing data points, and the fuzzy signal information. Based on the adjusted patient behavior data, the missing data points, and the weights of the fuzzy signal information, the probability of occurrence of different potential behavioral events is iteratively calculated until the probability of occurrence converges. The missing data points are then inferred to obtain the target patient's key behavioral event chain.

3. The management method for Alzheimer's disease elderly care platform based on big data according to claim 1, characterized in that, The steps for assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain the comparison results include: Continuously monitor patients' real-time behavioral data; Based on the real-time behavioral data, and considering the urgency and tolerance requirements for early warning in the current time period, the situation sensitivity factor is calculated. Based on the real-time behavioral data, the trend of changes in the patient's behavioral patterns over a period of time is continuously analyzed, the degree of individualization and the speed of evolution of the behavioral patterns are quantified, and behavioral pattern variation factors are obtained. Based on the context sensitivity factor and the behavior pattern variation factor, a reliability assessment threshold is generated; Assess the reliability of the target patient's key behavioral event chain and compare it with a reliability assessment threshold to obtain the comparison results.

4. The management method for Alzheimer's disease elderly care platform based on big data according to claim 1, characterized in that, The steps for assessing the reliability of the target patient's key behavioral event chain and comparing it with a reliability assessment threshold to obtain the comparison results include: Continuously monitor key factors in the care environment to identify the current situational state; Obtain the information of the currently executed care strategy and match it with the situational state to determine the urgency requirement for triggering the warning; Calculate the early warning urgency factor based on the matching results of the situational state and care strategy; Adjust the reliability assessment threshold according to the aforementioned early warning urgency factor; The reliability of the target patient's key behavioral event chain was compared with the adjusted reliability assessment threshold to obtain the comparison results.

5. The management method for Alzheimer's disease elderly care platform based on big data according to claim 3, characterized in that, The step of generating a reliability assessment threshold based on the context sensitivity factor and the behavior pattern variation factor includes: Monitor the reliability of the original data sources that constitute the context sensitivity factor and the behavioral pattern variation factor; Calculate the reliability weight of the original data source based on its reliability. Based on the patient's stage of illness and care environment, the influence weights of the situational sensitivity factor and the behavioral pattern variation factor are adjusted to obtain the target influence weights. Based on the reliability weight of the original data source, the situation sensitivity factor and the behavior pattern variation factor are weighted and adjusted to obtain the target situation sensitivity factor and the target behavior pattern variation factor. A reliability assessment threshold is generated based on the target impact weight, the target context sensitivity factor, and the target behavior pattern variation factor.

6. The management method for Alzheimer's disease elderly care platform based on big data according to claim 5, characterized in that, The step of monitoring the reliability of the original data sources constituting the context sensitivity factor and the behavioral pattern variation factor includes: Receive data from multiple raw data sources; The data from each original data source is analyzed in multiple dimensions, including data completeness, data temporal consistency, and the degree of deviation between the data and adjacent data points. Obtain the historical performance baseline for each original data source, which includes the typical reliability performance of the original data source under different operating conditions; Based on the data integrity, the data temporal consistency, the degree of deviation between the data and adjacent data points, and the historical performance baseline, a reliability score is calculated for each original data source. When the reliability of an original data source changes, the reliability scores of other original data sources it depends on are adjusted according to the interdependencies between different original data sources. Adjust the reliability of each original data source based on the adjusted reliability score.

7. The management method for Alzheimer's disease elderly care platform based on big data according to claim 6, characterized in that, The step of calculating the reliability weight of the original data source based on its reliability includes: Based on the reliability score of the original data source, a nonlinear function is used to convert the reliability score of the original data source into a reliability weight, wherein the parameters of the nonlinear function are preset according to the type of the original data source.

8. The management method for Alzheimer's disease elderly care platform based on big data according to claim 3, characterized in that, The steps for continuously monitoring the patient's real-time behavioral data include: It receives data streams from multiple heterogeneous sensors and identifies the sensor type, sampling frequency, and data format of each data stream; Timestamp calibration is performed on each received data stream; Based on the sensor type and the data format, each data stream is converted into a unified data structure and data type; Analyze the timestamp intervals and data value changes in each data stream to remove duplicate or redundant data points; Outlier detection and noise reduction are performed on the data in each data stream, and the data is classified and labeled according to preset semantic rules to form standardized behavioral data.

9. The management method for Alzheimer's disease elderly care platform based on big data according to claim 8, characterized in that, The step of timestamp calibration for each received data stream includes: Continuously monitor the clock synchronization service status of the central server; Continuously monitor the battery level and internal crystal oscillator status of each sensor; The reference weight of the central clock is obtained based on the clock synchronization service status of the central server. The reference weight of the sensor's own clock is obtained based on the sensor's battery level and internal crystal oscillator state. The correction factor for timestamp calibration is calculated based on the reference weight of the central clock and the reference weight of the sensor's own clock. Each data stream is timestamped based on the correction factor for timestamping calibration.

10. A big data-based Alzheimer's disease elderly care platform management system, characterized in that, The system includes: The data receiving module is used to receive patient behavior data from the local data aggregation gateway; The data inference module is used to monitor whether there are missing data points in the patient behavior data information that constitute the key behavioral event chain of the patient. When missing data points are detected, the missing data points are inferred based on the historical behavior pattern data of the pre-stored patient behavior file and the patient behavior data information to obtain the target patient key behavioral event chain. The reliability assessment module is used to evaluate the reliability of the target patient's key behavioral event chain and compare it with the reliability assessment threshold to obtain the comparison results; The early warning information generation module is used to generate early warning information with reliability quantification indicators when the comparison results meet the early warning criteria.

Images