Early intelligent early warning method and device for health risk of left-behind old people

Abstract

The invention discloses a left-behind old people health risk early-stage intelligent early warning method and device, and the method comprises the steps: collecting daily life data of the left-behind old people every day in a multi-dimensional manner, and obtaining original time sequence data; slicing and standardizing the original time series data, and extracting daily multi-dimensional features of the left-behind old people; based on the historical data of the left-behind old people in a certain time period and a pre-constructed prediction model, predicting to obtain predicted multi-dimensional features of the left-behind old people on the current day; calculating the deviation between the actually measured multi-dimensional features and the predicted multi-dimensional features of the left-behind old people on the current day; performing risk judgment according to the deviation and a pre-constructed risk judgment model to obtain a judgment result; and performing early warning according to the judgment result. By using the scheme of the invention, early recognition and intervention of the health risk of the left-behind old people can be realized.

Description

Technical Field

[0001] This invention relates to the field of health monitoring, specifically to a method and device for early intelligent warning of health risks among elderly people left behind in rural areas. Background Technology

[0002] Currently, with the migration of laborers, the number of hollowed-out rural areas and towns is increasing, leaving behind a growing number of elderly people. Concerning the health of these elderly has become a common concern for migrant workers. However, due to factors such as poor local medical conditions, limited economic resources, weak medical awareness, and reluctance to undergo frequent checkups, proactive monitoring of the health of these elderly individuals is difficult. With technological advancements, in addition to proactive health checkups, monitoring the health of these elderly individuals can also be achieved through comprehensive assessments and predictions based on data generated from their daily lives, thus identifying potential health problems. However, existing solutions often suffer from the following shortcomings: 1) Early warning systems for elderly people left behind in rural areas can only focus on some extreme situations, such as "risk of death," and lack the ability to intervene in early and moderate health problems.

[0003] 2) The data sources collected are relatively limited, with only data on water and electricity consumption and data detected by wearable devices used as the data sources for evaluation.

[0004] 3) The method is simple and uses fixed thresholds and rules for judgment. This method cannot learn individual differences in behavioral patterns and the false alarm rate will be very high. For example, when an elderly person visits a neighbor's house during the day, the water and electricity in the house will be "consistently 0".

[0005] 4) The warning features are relatively crude, using only absolute or instantaneous values, without exploring deeper features such as time sequence patterns, frequency, and regularity. Summary of the Invention

[0006] This invention provides an early intelligent warning method and device for health risks of left-behind elderly people, so as to realize the early identification and intervention of health risks of left-behind elderly people.

[0007] Therefore, the present invention provides the following technical solution: An early intelligent warning method for health risks among left-behind elderly, the method comprising: Collect daily life data of left-behind elderly people from multiple dimensions to obtain raw time-series data; The original time-series data is sliced ​​and standardized to extract multi-dimensional features of the elderly left behind each day; Based on the historical data of the elderly left behind within a certain period of time and the pre-built prediction model, the predicted multi-dimensional characteristics of the elderly left behind on that day are obtained. Calculate the deviation between the measured multidimensional features and the predicted multidimensional features of the elderly left behind on the day; Risk assessment is performed based on the aforementioned deviation and the pre-built risk assessment model to obtain the assessment result; An early warning will be issued based on the judgment result.

[0008] Optionally, the daily life data includes: daily life usage data, communication behavior data, body monitoring data, and indoor activity data.

[0009] Optionally, the prediction model is constructed in the following manner: Obtain the multi-dimensional features of the elderly left behind within a certain time period, and generate a training sample set in a supervised learning format; The training samples in the training sample set are input into the LSTM network for training to obtain the prediction model.

[0010] Optionally, the risk assessment model is constructed in the following manner: Obtain a large number of measured multidimensional features of the left-behind elderly people every day and the corresponding predicted multidimensional features based on the prediction model; Calculate the deviation between the measured multidimensional features and the predicted multidimensional features, determine the risk type label corresponding to the deviation, and generate an inference training sample set; The risk judgment model is obtained by training using the XGBoost algorithm and the inference training sample set.

[0011] Optionally, the step of issuing an early warning based on the judgment result includes: mapping the judgment result to a specific early warning level and content according to a preset mapping rule.

[0012] Optionally, the method further includes: establishing the mapping rules using the Drools rule engine framework.

[0013] Optionally, the method further includes: The warning level and content will be pushed to relevant personnel; Receive feedback information from the relevant personnel; Based on the feedback information, risk type labels are generated, and new inference training samples are generated. The risk judgment model is optimized and updated using the accumulated new inference training samples.

[0014] An early intelligent warning device for health risks among left-behind elderly, the device comprising: The data acquisition module is used to collect daily life data of left-behind elderly people from multiple dimensions to obtain raw time-series data; The data processing module is used to slice and standardize the raw time-series data to extract multi-dimensional features of the elderly left behind each day. The prediction module is used to predict the multi-dimensional features of the elderly left behind on a given day based on historical data of the elderly left behind within a certain period of time and a pre-built prediction model. The deviation determination module is used to calculate the deviation between the measured multidimensional features and the predicted multidimensional features of the elderly left behind on the same day. The risk assessment module is used to assess risks based on the deviation and a pre-built risk assessment model, and obtain the assessment result. The early warning module is used to issue early warnings based on the judgment results.

[0015] Optionally, the device further includes: a prediction model building module, used to build a prediction model based on historical data of the left-behind elderly over a certain period of time; The prediction model construction module includes: A multi-dimensional feature acquisition unit is used to acquire the multi-dimensional features of the elderly left behind within a certain time period and generate a training sample set in a supervised learning format. The training unit is used to input training samples from the training sample set into the LSTM network for training to obtain the prediction model.

[0016] Optionally, the device further includes: a risk assessment model construction module, used to construct the risk assessment model; The risk assessment model construction module includes: The data collection unit is used to acquire a large number of measured multi-dimensional features of the left-behind elderly people every day and the corresponding predicted multi-dimensional features based on the prediction model. The sample generation unit is used to calculate the deviation between the measured multi-dimensional features and the predicted multi-dimensional features, determine the risk type label corresponding to the deviation, and generate an inference training sample set. The inference unit is used to train the risk judgment model using the XGBoost algorithm and the inference training sample set.

[0017] Optionally, the early warning module is specifically used to map the judgment result to a specific early warning level and content according to a preset mapping rule; The device further includes: a push module, a receiving module, and a recording module; The push module is used to push the warning level and content to relevant personnel; The receiving module is used to receive feedback information from the relevant personnel; The recording module is used to generate risk type labels based on the feedback information and generate new inference training samples; The reasoning unit is also used to optimize and update the risk judgment model using the accumulated new reasoning training samples.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the steps of the distributed model collaborative training method in the scenario of heterogeneous computing power.

[0019] This invention provides an intelligent early warning method and device for health risks among left-behind elderly. It collects daily life data from left-behind elderly individuals from multiple dimensions to obtain raw time-series data. The raw time-series data is then sliced ​​and standardized to obtain multi-dimensional characteristics of each elderly person's daily life. The deviation between the measured and predicted multi-dimensional characteristics for each day is determined. Risk assessment is based on this deviation, and early warnings are issued according to the assessment results. This invention uses AI to learn the unique lifestyle behavior data of left-behind elderly individuals within a specific period, automatically acquiring regularity and frequency characteristics, and automatically establishing warning values ​​for each behavior. This not only allows for monitoring of extreme situations but also enables the analysis of data over a certain time period to discover the evolutionary trends of health problems, achieving early identification and intervention of health risks for left-behind elderly individuals. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an early intelligent warning method for health risks of left-behind elderly provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the optimization and updating process of the risk assessment model through early warning and manual intervention feedback in an embodiment of the present invention. Figure 3 This is a schematic diagram of a structure of an early intelligent warning device for health risks of left-behind elderly provided in an embodiment of the present invention; Figure 4 This is another structural schematic diagram of the intelligent early warning device for health risks of left-behind elderly provided in the embodiments of the present invention. Detailed Implementation

[0022] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] To address the problems existing in current health monitoring programs for left-behind elderly, this invention provides an early intelligent warning method and device for health risks of left-behind elderly. By collecting daily life data of left-behind elderly from multiple dimensions and combining AI technology, it enables real-time monitoring of health problems of left-behind elderly, and early identification and intervention of health risks.

[0025] The use of the daily life data of left-behind elderly collected in this invention complies with the provisions of the Civil Code and the Personal Information Protection Law. It is implemented reasonably within the scope of the consent of the individual or his / her guardian, and reasonably processes information that the individual has disclosed on his / her own or that has been legally disclosed, except where the individual expressly refuses or it infringes on his / her significant interests. It is an action that is reasonably implemented to safeguard the public interest or the legitimate rights and interests of the natural person.

[0026] like Figure 1 The diagram shown is a flowchart of an early intelligent warning method for health risks among left-behind elderly provided by an embodiment of the present invention, which includes the following steps: Step 101: Collect daily life data of left-behind elderly people from multiple dimensions to obtain raw time-series data.

[0027] Specifically, daily life data of left-behind elderly can be collected seamlessly, continuously, and in multiple dimensions. This daily life data may include, but is not limited to: daily living data, communication behavior data, physical monitoring data, and indoor activity data. Daily life data can include electricity, water, and gas consumption data. For example, this data can be obtained by querying the mobile apps of the power grid, water utilities, and gas companies.

[0028] Communication behavior data can include data such as call time, call recipient, and call duration. For example, communication behavior data can be obtained by sharing the elderly's mobile phone call records with their children who are left behind in rural areas.

[0029] Body monitoring data can include: heart rate, blood pressure, steps taken, stress data, blood oxygen data, etc. for the elderly. For example, data can be obtained by sharing the smartwatches worn by the elderly with their children who are left behind in rural areas.

[0030] Indoor activity trajectory data can include human movement data in areas such as bedrooms, bathrooms, living rooms, and kitchens. For example, this data can be obtained by acquiring networked infrared devices installed in the home.

[0031] The data of the above-mentioned different dimensions can be represented in a set format. For example, in a non-limiting embodiment, the data format shown in Table 1 below is used.

[0032] Table 1

[0033] Step 102: Slice and standardize the original time-series data to extract multi-dimensional features of elderly people left behind each day.

[0034] The purpose of standardization is to transform low-frequency, messy raw data streams into high-frequency, regular, high-level features that can characterize behavioral patterns.

[0035] In one non-limiting embodiment, the open-source Pandas framework can be used to slice and standardize the time series data, and the tsfresh feature calculation library can be used to extract features from the processed time series data to obtain the multi-dimensional features of left-behind elderly people every day.

[0036] For example, the raw data collected from Mr. Zhang on the 10th and 11th are shown in Table 2 below.

[0037] Table 2

[0038] Feature calculations were performed on the data in Table 2 above, generating the following feature vectors: Day10_Vector = [0.25, 0.85, 3, 125, 900, 1, 1]; Day11_Vector = [1.33, 0.10, 1, 18, 0, 0, 0].

[0039] Step 103: Based on the historical data of left-behind elderly over a certain period of time and the pre-built prediction model, predict the multi-dimensional features of left-behind elderly on the current day.

[0040] In this embodiment of the invention, future data is predicted based on historical data. For example, an LSTM (Long Short-Term Memory) network can be selected to train the prediction model. Data from consecutive past time periods is used to predict the data range for future time periods to determine whether future measured data is normal.

[0041] The training process of the prediction model is as follows: obtain the multi-dimensional features of the elderly left behind within a certain period of time, and generate a training sample set in a supervised learning format; input the training samples in the training sample set into the LSTM network for training to obtain the prediction model.

[0042] The following example further illustrates the training process of the prediction model.

[0043] The prediction model takes a sequence of feature vectors from the past N consecutive days (e.g., 30 days) as input and outputs the predicted range of that feature value for day N+1 (tomorrow) (e.g., [100, 150]). The training process is as follows: (1) Data preparation: First, obtain Mr. Zhang's daily activity frequency data for the past 30 days, forming a time series: [120, 118, 135, 110, ..., 125], with a total of 30 data points; Then, the data is converted into a supervised learning format, which uses the values ​​from the past few days to predict the values ​​for the next day.

[0044] For example, to predict the next day using the past 7 days, the following sample pairs are generated: X_train_sample1: [120, 118, 135, 110, ...] (data from day 1 to day 7) -> y_train_sample1: 105 (actual value on day 8); X_train_sample2: [118, 135, 110, ..., 105] (data from day 2 to day 8) -> y_train_sample2: 115 (actual value on day 9); ...; This process is repeated to generate 23 training samples.

[0045] (2) Model training and prediction: Training: Input the above samples into the LSTM network for training. The internal gating mechanism of LSTM (forget gate, input gate, output gate) will learn the long-term regularity and pattern in the time series (e.g., more activity during the week and less activity on weekends; Mr. Zhang's activity level fluctuates between 100 and 150).

[0046] Prediction: After the model is trained, take the real data of the last 7 days [108, 115, 122, 130, 119, 125, 118] (days 24 to 30) and input it into the prediction model. The prediction model will output a predicted value for day 31, for example, predicted value = 125.

[0047] Meanwhile, the prediction model calculates the prediction error (root mean square error RMSE) of the training data. Assuming RMSE = 10, the normal prediction range for day 31 can be set as [125 - 2×10, 125 + 2×10] = [105, 145]. This range represents the normal fluctuation range of Mr. Zhang's activity frequency tomorrow, as predicted by the prediction model.

[0048] For example, using the prediction model obtained from the above training and Mr. Zhang's data from the previous 30 days, the normal range of his daytime activity frequency on Day 11 is predicted to be [105, 145].

[0049] The measured number of Mr. Zhang's daytime activity frequency on Day 11 was 18. Since 18 << 105 (far below the lower limit of the normal range), this characteristic was determined to be significantly abnormal.

[0050] It should be noted that in some embodiments, an early warning can also be issued when a significant anomaly is detected in a feature item, and this embodiment of the present invention does not limit this.

[0051] Step 104: Calculate the deviation between the measured multidimensional characteristics and the predicted multidimensional characteristics of the elderly left behind on the day.

[0052] Specifically, for each feature, the deviation between the measured multidimensional features for each day and the predicted multidimensional features for that day can be calculated.

[0053] Step 105: Perform risk assessment based on the deviation and the pre-built risk assessment model to obtain the assessment result.

[0054] The supervised classification algorithm XGBoost (eXtremeGradientBoosting) is used to classify the deviation vector between the daily feature vector and the baseline prediction value, and to determine which risk the daily feature vector belongs to.

[0055] The construction process of the risk assessment model is as follows: (1) Obtain a large number of measured multidimensional features of the left-behind elderly people every day and the corresponding predicted multidimensional features based on the prediction model; (2) Calculate the deviation between the measured multidimensional features and the predicted multidimensional features, determine the risk type label corresponding to the deviation, and generate an inference training sample set; (3) The risk judgment model is obtained by training the XGBoost algorithm and the inference training sample set.

[0056] The following example further illustrates the training process of the risk assessment model.

[0057] 1. Training data preparation: Obtain sample features and labels.

[0058] The sample feature is a deviation vector, which is the deviation vector between each measured feature and the predicted feature. For example, the deviation vector of Grandpa Zhang on a certain day is [number of activities: -87, call duration: -900, kitchen water usage: -2, ...]; The calculation method is: (measured value - predicted value) / predicted value.

[0059] The corresponding label is the risk type. The label can be manually marked based on historical experience. For example, after a village doctor verifies the situation at home, they can select "Cause: Difficulty moving after a fall" in the app, and the system can map it to the label "Health deterioration / disability risk".

[0060] 2. Model Training: XGBoost is a powerful gradient boosting tree model that builds multiple decision trees, each learning how to correctly classify the residuals of the previous tree. During training, the model learns complex feature combination patterns. For example: Rule 1: If the deviation in the number of activities is less than -50 and the deviation in the call duration is less than -300, then the sample has a high probability of being at risk of health deterioration / disability.

[0061] Rule 2: If the activity entropy deviation > 0.5 and the kitchen water usage deviation < -1, then the sample may belong to the risk category of cognitive impairment. Through continuous training, the model integrates the results of all trees to form a powerful classifier, namely the risk assessment model.

[0062] Using the risk assessment model obtained from the above training, the reasoning process for assessing Mr. Zhang's daily health status is as follows: First, calculate the deviation vector for all features of Mr. Zhang (character number 11). For example: Activity frequency deviation = 18 - 105 = -87 (actual value - lower limit of prediction); Call duration deviation = 0 - 300 = -300 (assuming the lower limit of prediction is 300s); Kitchen water usage deviation = 1 - 2 = -1 (assuming the lower limit of prediction is 2 times); ...; The final deviation vector V = [-87, -300, -1, ...].

[0063] The aforementioned deviation vector V is input into the pre-trained risk judgment model, which then traverses all decision trees within it to make judgments.

[0064] Since the patterns in vector V (sudden decrease in activity, communication interruption, reduced cooking) are very similar to the samples in the "health deterioration / disability" class in the training data, the model outputs the following probability distribution: P(health deterioration / risk of disability) = 0.80; P (risk of social isolation) = 0.15; P(other) = 0.05; The type with the highest probability, "health deterioration / disability risk", is selected as the final classification result, and this result and its probability value are sent to the early warning decision engine.

[0065] Step 106: Issue an early warning based on the judgment result.

[0066] In some embodiments, the judgment result can be mapped to specific warning levels and content according to preset mapping rules. For example, the mapping rules can be established using the Drools rule engine framework.

[0067] The early warning decision engine receives the anomaly score and risk type output by the risk assessment model, and maps them to specific early warning levels and content based on factors such as the persistence of the anomaly (e.g., occurring for several consecutive days).

[0068] For example, the execution logic of the Drools rule engine is as follows: {IF Risk type == “Potential health deterioration / disability risk” AND Abnormal score > threshold 1 AND Yesterday’s activity was already at a low level (considering persistence); THEN Generate a [High Risk - Emergency Alert], with the following warning information: {Level: High Risk; Type: Potential Health Event / Disability Risk; Details: "Mr. Zhang's activity level today is extremely abnormal (only 15% of his usual level), and he has not made any external communications. Blood pressure monitoring has been interrupted. Please visit him immediately to verify."} Recipients: Village doctor, emergency contact person.

[0069] Furthermore, the warning level and content can be pushed to relevant personnel, feedback information can be received from relevant personnel, and this feedback information and corresponding status data can be recorded as historical data and stored in the database. When new data accumulates to a certain amount or the data is used periodically, the risk assessment model can be optimized and updated to improve the performance of the risk assessment model.

[0070] like Figure 2The diagram shown is a flowchart illustrating the optimization and update process of the risk assessment model based on early warning and manual intervention feedback in an embodiment of the present invention, including the following steps: Step 201: Push the warning level and content to relevant personnel.

[0071] Step 202: Receive feedback information from the relevant personnel.

[0072] For example, after receiving the alert, the village doctor or a relative or child immediately goes to Mr. Zhang's home and finds him bedridden due to difficulty moving around after a fall the previous day. The village doctor or relative or child processes the alert on a mobile app and fills in the feedback information: "Verification result: true; specific reason: difficulty moving around after a fall; handling measures: sent to the hospital for examination, and his children have been contacted." Step 203: Generate risk type labels based on the feedback information and generate new inference training samples.

[0073] The corresponding risk type label "Health deterioration_Fall" is generated based on the above feedback information. Then, Day11_Vector (the data collected on the 11th and the deviation vector of each feature) and its corresponding label "Health deterioration_Fall" are stored in the database as new inference training samples.

[0074] Step 204: Optimize and update the risk judgment model using the accumulated new inference training samples.

[0075] For example, an optimization update may be performed after a certain number of new inference training samples have been accumulated, or an optimization update may be performed at regular intervals. This embodiment of the invention does not limit the scope of the optimization update.

[0076] Optimization and updates can improve the performance of the risk assessment model, making its judgments more accurate. For example, when encountering the "sudden decrease in activity + no communication" pattern in the future, the system may combine time factors (such as occurring in winter) to provide a more accurate judgment or a new risk label.

[0077] Accordingly, embodiments of the present invention also provide an early intelligent warning device for health risks among left-behind elderly, such as... Figure 3 The diagram shown is a structural schematic of the device.

[0078] The early warning device 300 includes the following modules: The data acquisition module 301 is used to collect daily life data of left-behind elderly people from multiple dimensions to obtain raw time-series data. The daily life data may include, but is not limited to, any one or more of the following: life usage data, communication behavior data, body monitoring data, indoor activity data, etc. Different types of data can be acquired in different ways. For details, please refer to the description in the previous embodiments of the present invention, which will not be repeated here.

[0079] Data processing module 302 is used to slice and standardize the original time-series data to extract multi-dimensional features of the elderly left behind each day; The prediction module 303 is used to predict the multi-dimensional features of the left-behind elderly on the current day based on the historical data of the left-behind elderly within a certain period of time and the pre-built prediction model. The deviation determination module 304 is used to calculate the deviation between the measured multidimensional features and the predicted multidimensional features of the left-behind elderly on the day. Risk assessment module 305 is used to assess risk based on the deviation and a pre-built risk assessment model, and obtain a assessment result. The early warning module 306 is used to issue an early warning based on the judgment result.

[0080] The prediction model can be constructed by a corresponding prediction model building module based on historical data of left-behind elderly over a certain period of time. One non-restricted structure of the prediction model building module may include a multi-dimensional feature acquisition unit and a training unit. The multi-dimensional feature acquisition unit is used to acquire multi-dimensional features of left-behind elderly over a certain period of time and generate a training sample set in a supervised learning format; the training unit is used to input the training samples from the training sample set into an LSTM network for training to obtain the prediction model.

[0081] The risk assessment model can be constructed by a corresponding risk assessment model construction module based on the deviation between measured data and predicted data obtained using a prediction model. A non-limiting structure of the risk assessment model construction module may include a data collection unit, a sample generation unit, and an inference unit. The data collection unit is used to acquire daily measured multidimensional features of a large number of elderly people left behind in rural areas and corresponding predicted multidimensional features based on the prediction model. The sample generation unit is used to calculate the deviation between the measured multi-dimensional features and the predicted multi-dimensional features, and to determine the risk type label corresponding to the deviation, thereby generating an inference training sample set. The inference unit is used to train the risk assessment model using the XGBoost algorithm and the inference training sample set.

[0082] like Figure 4 The diagram shown is another structural schematic of the intelligent early warning device for health risks of left-behind elderly provided in an embodiment of the present invention.

[0083] In this embodiment, the early warning module 306 is specifically used to map the judgment result to a specific early warning level and content according to a preset mapping rule.

[0084] In addition, with Figure 3 The difference in the illustrated embodiment is that, in Figure 4 In the illustrated embodiment, the early warning device 300 further includes: a push module 307, a receiving module 308, and a recording module 309. Wherein: The push module 307 is used to push the warning level and content generated by the warning module 306 to relevant personnel; The receiving module 308 is used to receive feedback information from the relevant personnel; The recording module 309 is used to generate risk type labels based on the feedback information and generate new inference training samples.

[0085] Correspondingly, the inference unit in the risk judgment model construction module can also use the accumulated new inference training samples to optimize and update the risk judgment model, so as to improve the performance of the risk judgment model and make its judgment results more accurate.

[0086] This invention provides an intelligent early warning method and device for health risks among left-behind elderly. It collects daily life data from left-behind elderly individuals from multiple dimensions to obtain raw time-series data. The raw time-series data is then sliced ​​and standardized to obtain multi-dimensional characteristics of each elderly person's daily life. The deviation between the measured and predicted multi-dimensional characteristics for each day is determined. Risk assessment is based on this deviation, and early warnings are issued according to the assessment results. This invention uses AI to learn the unique lifestyle behavior data of left-behind elderly individuals within a specific period, automatically acquiring regularity and frequency characteristics, and automatically establishing warning values ​​for each behavior. This not only allows for monitoring of extreme situations but also enables the analysis of data over a certain time period to discover the evolutionary trends of health problems, achieving early identification and intervention of health risks for left-behind elderly individuals.

[0087] Compared with existing solutions, this solution offers a superior approach, primarily in the following aspects: 1. Early warning objectives: Shift from single-extreme early warning (such as death warning) to multi-dimensional and hierarchical "early identification of health risks", covering multiple aspects such as living ability, cognitive function, and psychosocial aspects, and realize early intervention for the health of left-behind elderly.

[0088] 2. Data Insights: Upgrading from using simple statistical values ​​(sum, average) of data to mining its pattern characteristics (regularity, entropy, time series patterns), achieving a deeper understanding of behavior.

[0089] 3. Technical Approach: Abandoning the "fixed threshold" rule (such as the mechanism that a value above 100 is considered abnormal), we innovatively adopt an individual dynamic behavior baseline modeling method based on unsupervised learning, enabling the system to have the ability to perceive individual differences and an extremely low false alarm rate.

[0090] 4. Closed-loop optimization capability: By introducing a multi-dimensional risk classifier based on machine learning and a human-machine collaborative feedback optimization closed loop, the system can not only quantify anomalies, but also characterize risk types and has the ability to continuously evolve.

[0091] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0092] The present invention also provides a storage medium, which is a computer-readable storage medium storing a computer program thereon, the computer program being executable when it runs. Figure 1 or Figure 2 The method shown may include some or all of the steps. The storage medium may include read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc. The storage medium may also include non-volatile memory or non-transitory memory, etc.

[0093] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data provider to another website, computer, server, or data provider via wired or wireless means.

[0094] The embodiments of the present invention have been described in detail above. Specific implementation methods have been used to illustrate the present invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and apparatus of the present invention, and are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention, and the content of this specification should not be construed as a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An early intelligent warning method for health risks of an elderly left at home, characterized in that, The method comprises: Multi-dimensional collection of daily life data of the left-behind elderly people to obtain original time series data; Slicing and standardization processing of the original time series data to extract multi-dimensional features of the left-behind elderly people every day; Prediction of the predicted multi-dimensional features of the left-behind elderly people on the current day based on historical data of the left-behind elderly people within a certain time period and a prediction model constructed in advance; Calculation of the deviation between the measured multi-dimensional features and the predicted multi-dimensional features of the left-behind elderly people on the current day; Risk judgment based on the deviation and a risk judgment model constructed in advance to obtain a judgment result; Early warning based on the judgment result; The risk judgment model is constructed in the following manner: A large number of measured multi-dimensional features of the left-behind elderly people every day and corresponding predicted multi-dimensional features obtained based on the prediction model are obtained; The deviation between the measured multi-dimensional features and the predicted multi-dimensional features is calculated, and a risk type label corresponding to the deviation is determined to generate an inference training sample set; The risk judgment model is obtained by training using an XGBoost algorithm and the inference training sample set.

2. The method for early intelligent warning of health risks of left-behind elderly people according to claim 1, characterized in that, The daily life data comprises: life use data, communication behavior data, body monitoring data, and indoor activity data. 3.The method of claim 1, wherein the method further comprises: The prediction model is constructed in the following manner: The multi-dimensional features of the left-behind elderly people within a certain time period are obtained to generate a training sample set in a supervised learning format; The training samples in the training sample set are input into an LSTM network for training to obtain the prediction model.

4. The method for early intelligent warning of health risks of left-behind elderly people according to claim 3, characterized in that, The early warning based on the judgment result comprises: Mapping the judgment result to specific early warning levels and contents according to a preset mapping rule.

5. The method for early intelligent warning of health risks of the left-behind elderly according to claim 4, characterized in that, The method further comprises: Establishing the mapping rule using a Drools rule engine framework.

6. The method for early intelligent warning of health risks of the left-behind elderly according to claim 5, characterized in that, The method further comprises: Pushing the early warning levels and contents to relevant personnel; Receiving feedback information from the relevant personnel; Generating a risk type label based on the feedback information and generating new inference training samples; Optimizing and updating the risk judgment model using the accumulated new inference training samples.

7. An early intelligent warning device for health risks of an elderly person left at home, characterized by, The device comprises: A data collection module for multi-dimensional collection of daily life data of the left-behind elderly people to obtain original time series data; A data processing module for slicing and standardization processing of the original time series data to extract multi-dimensional features of the left-behind elderly people every day; A prediction module for predicting the predicted multi-dimensional features of the left-behind elderly people on the current day based on historical data of the left-behind elderly people within a certain time period and a prediction model constructed in advance; A deviation determination module for calculating the deviation between the measured multi-dimensional features and the predicted multi-dimensional features of the left-behind elderly people on the current day; A risk judgment module for risk judgment based on the deviation and a risk judgment model constructed in advance to obtain a judgment result; An early warning module for early warning based on the judgment result; The device further comprises a risk judgment model construction module for constructing the risk judgment model; The risk judgment model construction module comprises: a data collection unit configured to obtain a plurality of measured multi-dimensional features of the left-behind elderly person each day and corresponding predicted multi-dimensional features obtained based on the prediction model; a sample generation unit configured to calculate deviations between the measured multi-dimensional features and the predicted multi-dimensional features, determine risk type labels corresponding to the deviations, and generate an inference training sample set; an inference unit configured to train the risk judgment model by using an XGBoost algorithm and the inference training sample set. 8.The health risk early warning device for left-behind elderly people according to claim 7, characterized in that, The device further comprises a prediction model construction module configured to construct a prediction model based on historical data of the left-behind elderly person within a certain time period. The prediction model construction module comprises: a multi-dimensional feature acquisition unit configured to obtain the multi-dimensional features of the left-behind elderly person within a certain time period and generate a training sample set in a supervised learning format; a training unit configured to input training samples in the training sample set into an LSTM network for training to obtain the prediction model.

9. The left-behind elderly health risk early intelligent warning device according to claim 8, wherein the warning module is specifically configured to map the judgment result to specific warning levels and contents according to a preset mapping rule; The device further comprises a push module, a receiving module and a recording module; the push module is configured to push the warning levels and contents to relevant personnel; the receiving module is configured to receive feedback information of the relevant personnel; the recording module is configured to generate risk type labels according to the feedback information and generate new inference training samples; the inference unit is further configured to optimize and update the risk judgment model by using the accumulated new inference training samples.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, performs the steps of the left-behind elderly health risk early intelligent warning method of any one of claims 1 to 6.

Images