Multi-sensor-based intelligent monitoring method and system for preventing old people from falling down

Abstract

The invention discloses an intelligent monitoring method and system for preventing an old person from falling down based on multiple sensors, and relates to the technical field of falling monitoring, and the method comprises the following steps: merging behavior data, physiological data and environmental data into a data set; obtaining a feature combination from the related features based on an association rule, and quantifying the influence of the feature combination on the falling probability of the elderly; based on a machine learning algorithm, learning a relationship between the related features and whether the old person falls down, and predicting an old person falling probability corresponding to each sample of the new data set; optimizing the falling probability of the elderly based on the quantification result; if the optimized falling probability of the old people exceeds a certain threshold value, executing a corresponding early warning scheme; the system comprises a data acquisition module, a feature combination acquisition module, a probability prediction module, a probability optimization module and an early warning execution module. According to the invention, the fall probability corresponding to each sample can be obtained, the quantitative influence of the detected feature combination is superposed to the prediction result, and the accuracy of fall risk assessment is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fall monitoring, and in particular to a multi-sensor based intelligent fall prevention monitoring method and system for the elderly. Background Art

[0002] Multi-sensor refers to a combination of different types of sensors that are used to monitor and collect data from multiple sources simultaneously. These sensors may include, but are not limited to, accelerometers and gyroscopes. By fusing data from multiple sensors, multi-sensor systems can provide more comprehensive, accurate, and reliable information. By real-time monitoring and analysis of the elderly's behavior, physical condition, and environmental factors, possible fall risks can be identified and warned in advance, so that appropriate intervention measures can be taken to prevent falls among the elderly.

[0003] Multi-sensor-based intelligent fall prevention monitoring for the elderly combines multiple sensor technologies to more comprehensively and accurately monitor the activities and health status of the elderly to prevent falls. The core idea is to use data fusion and comprehensive analysis of multiple sensors to improve the accuracy and real-time nature of fall risk identification.

[0004] For example, Chinese patent 201910213812.7 discloses an intelligent monitoring system and method for preventing elderly people from falling based on a composite sensor, which uses a deep learning algorithm for data processing to solve the problems of weak anti-interference ability and instability. However, the above-mentioned intelligent monitoring system and method for preventing elderly people from falling based on a composite sensor still have the following shortcomings: it is relatively insufficient in the diversity and comprehensiveness of data sources and intelligent early warning and feedback mechanisms. At the same time, the existing technology usually predicts the fall of the elderly based on a single data feature, without considering the relationship and combination effect between multiple features, so it is difficult to accurately reflect the overall fall risk of the elderly.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes an intelligent monitoring method and system for preventing falls for the elderly based on multiple sensors to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a multi-sensor based intelligent monitoring method for preventing elderly people from falling is provided. The multi-sensor based intelligent monitoring method for preventing elderly people from falling comprises the following steps: S1. Collect behavioral data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavioral data, physiological data and environmental data into a data set.

[0008] S2. Extract relevant features from the data set, obtain feature combinations from the relevant features based on association rules, and quantify the impact of feature combinations on the probability of falls of the elderly.

[0009] S3. Based on the machine learning algorithm, learn the relationship between relevant features and whether the elderly fall, and predict the probability of the elderly falling corresponding to each sample in the new data set.

[0010] S4. Analyze the new data set to determine whether the feature combination appears. If so, optimize the probability of elderly people falling based on the quantitative results.

[0011] S5. Set several thresholds. If the optimized probability of elderly people falling exceeds a certain threshold, execute the corresponding early warning plan.

[0012] The extracting of relevant features from the data set, obtaining feature combinations from the relevant features based on association rules, and quantifying the influence of the feature combinations on the probability of falls of the elderly include the following steps: S21. Determine features that affect the probability of falling based on domain knowledge as relevant features; S22, prepare association rule mining and set association rule parameters; S23, mining association rules according to support and confidence, generating association rules, and determining feature combinations from the association rules; S24. Complete the quantification of the impact of feature combinations on the probability of falls of the elderly.

[0013] Furthermore, collecting the behavior data, physiological data and environmental data of the elderly based on multiple sensors and merging the behavior data, physiological data and environmental data into a data set includes the following steps: S11, synchronously collecting behavioral data, physiological data and environmental data through multiple sensors, and preprocessing the collected raw data; S12. Storing the preprocessed data in a database to obtain a data set.

[0014] Further, preparing for association rule mining and setting association rule parameters include the following steps: S221, defining the value of each relevant feature as an item, and converting the data into a format suitable for association rule analysis; S222, calculating the support and confidence of all candidate feature combinations; S223. Set the minimum support and minimum confidence.

[0015] Furthermore, mining association rules according to support and confidence, and generating association rules, and determining feature combinations from association rules include the following steps: S231, filter out candidate feature combinations with support higher than the minimum support as frequent item sets; S232. For each frequent item set, generate an association rule A→B, where A is the premise of the rule and B is the result of the rule; S233. For each association rule, calculate the confidence of the association rule; S234, retaining association rules with confidence levels higher than the minimum confidence level as strong association rules; S235, checking whether the generated strong association rules are consistent with the actual situation; S236. Use the item set in the strong association rule as a feature combination, and construct a feature combination list.

[0016] Furthermore, the quantification of the impact of feature combination on the probability of elderly people falling includes the following steps: S241, counting the number of occurrences of each feature combination in the data set, and counting the number of elderly fall events when each feature combination appears, and dividing the number of elderly fall events by the number of occurrences of the feature combination in the data set to calculate the fall probability of the feature combination; S242, determining respective weights based on the occurrence frequency of the feature combination and expert opinions, and calculating the weighted probability corresponding to each feature combination as a quantification of the impact on the probability of the elderly falling; Among them, the calculation formula for calculating the weighted probability corresponding to a certain feature combination is: ; In the formula, wp Represents the weighted probability corresponding to a certain feature combination; x Indicates the number of falls of the elderly. y Indicates the number of occurrences of a feature combination in the data set. z Represents the sum of the occurrences of all feature combinations; w 1 represents the weight of expert opinion, The weight coefficient representing the weight of the frequency of occurrence, β The weight coefficient representing the weight of expert opinion.

[0017] Furthermore, learning the relationship between relevant features and whether the elderly have fallen based on the machine learning algorithm and predicting the probability of the elderly falling corresponding to each sample in the new data set includes the following steps: S31, using gradient boosting decision tree to assign weights to each relevant feature; S32. Based on the random forest algorithm and in combination with the feature weights determined by the gradient boosting decision tree, several decision trees are constructed; S33, training the random forest model according to the acquired relevant features and corresponding labels, and optimizing the parameters of the random forest model; S34. Use cross-validation to evaluate the stability and accuracy of the random forest model and evaluate the performance of the random forest model; S35. Apply the trained random forest model to the new data set and predict the probability of the elderly falling for each sample in the new data set.

[0018] Furthermore, using the gradient boosting decision tree to assign weights to each relevant feature includes the following steps: S311, initializing the decision tree and selecting the loss function; S312, calculating the prediction residuals of all current decision trees, and training a new decision tree based on the residuals; S313, determining the weight of the new tree, and using the prediction of the new tree and the calculated weight to update the overall model; S314, counting the frequency of each relevant feature being used as a split node in all trees, and measuring the contribution of each feature split to the reduction of the loss function, and at the same time accumulating the feature importance scores in all trees to obtain the total importance score of each relevant feature; S315. Assign a weight to each relevant feature according to the total importance score.

[0019] Further, the new data set is analyzed to determine whether a feature combination appears. If so, the probability of elderly people falling is optimized based on the quantitative results, including the following steps: S41, extract relevant features from each sample of the new data set, and compare them with the feature combination to determine whether the feature combination exists in the sample; S42. If there is a feature combination, the weighted probability corresponding to the feature combination is added to the predicted probability of the elderly falling, to obtain an optimized probability of the elderly falling.

[0020] Furthermore, several thresholds are set. If the optimized probability of elderly people falling exceeds a certain threshold, the corresponding early warning plan is executed, including the following steps: S51, setting several thresholds of the probability of falling, and setting different warning levels and response measures corresponding to each threshold; S52, comparing the optimized probability of the elderly falling with a preset threshold value to determine the warning level of the probability of the elderly falling; S53. Select a corresponding warning plan based on the warning level of the probability of the elderly falling.

[0021] According to another aspect of the present invention, a multi-sensor based intelligent monitoring system for preventing falls for the elderly is provided. The multi-sensor based intelligent monitoring system for preventing falls for the elderly includes a data acquisition module, a feature combination acquisition module, a probability prediction module, a probability optimization module and an early warning execution module; wherein the data acquisition module is connected to the feature combination acquisition module, the feature combination acquisition module is connected to the probability prediction module, the probability prediction module is connected to the probability optimization module, and the probability optimization module is connected to the early warning execution module.

[0022] The data acquisition module is used to collect the behavioral data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavioral data, physiological data and environmental data into a data set.

[0023] The feature combination acquisition module is used to extract relevant features from the data set, obtain feature combinations from relevant features based on association rules, and quantify the impact of feature combinations on the probability of falls of the elderly.

[0024] The probability prediction module is used to learn the relationship between relevant features and whether the elderly have fallen based on the machine learning algorithm, and predict the probability of the elderly falling corresponding to each sample in the new data set.

[0025] The probability optimization module is used to analyze the new data set to determine whether the feature combination appears. If so, the probability of elderly people falling is optimized based on the quantitative results.

[0026] The early warning execution module is used to set several thresholds. If the optimized probability of elderly people falling exceeds a certain threshold, the corresponding early warning plan will be executed.

[0027] The beneficial effects of the present invention are: (1) The multi-sensor-based intelligent fall prevention monitoring method and system provided by the present invention collects the behavioral data, physiological data and environmental data of the elderly through the use of multiple sensors, provides the overall health status and activity status of the elderly, and helps to accurately monitor the behavioral changes of the elderly. Extracting features related to the elderly's falls, and using the trained model to predict the new data set, calculating the fall probability corresponding to each sample, can help identify high-risk elderly people in advance and take preventive measures. With the continuous increase of data and the continuous learning of the model, the prediction model can be continuously optimized to improve the prediction accuracy and stability. Based on the quantitative results, the probability of falls is optimized, and the quantitative impact of the detected feature combination is superimposed on the prediction results, which further improves the accuracy of the fall risk assessment and reduces missed reports and false alarms. Different risk thresholds are set to correspond to different warning levels. The graded warning can take different intervention measures according to the actual situation, improving the flexibility and pertinence of the warning system.

[0028] (2) The present invention uses association rules to mine the relationship between features and find out frequently occurring feature combinations, thereby revealing hidden patterns, which helps to better understand the formation mechanism of fall risk, quantify the impact of feature combinations on the probability of falls, and clarify the extent to which each feature combination increases the risk of falls for the elderly. This quantitative analysis provides a scientific basis for subsequent risk prediction and intervention measures. At the same time, when quantifying the impact of feature combinations on the probability of falls for the elderly, the respective weights are determined based on the frequency of occurrence of feature combinations and expert opinions, which can comprehensively consider the dual factors of data-driven and empirical knowledge, thereby improving the scientificity and reliability of the quantitative results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 is a flow chart of a multi-sensor based intelligent fall prevention monitoring method for the elderly according to an embodiment of the present invention; Figure 2 It is a principle block diagram of an intelligent fall prevention monitoring system for the elderly based on multiple sensors according to an embodiment of the present invention.

[0031] In the figure: 1. Data collection module; 2. Feature combination acquisition module; 3. Probability prediction module; 4. Probability optimization module; 5. Early warning execution module. DETAILED DESCRIPTION

[0032] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0033] According to an embodiment of the present invention, a multi-sensor based intelligent monitoring method and system for preventing falls for the elderly are provided.

[0034] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a multi-sensor-based intelligent monitoring method for preventing elderly people from falling is provided, and the multi-sensor-based intelligent monitoring method for preventing elderly people from falling comprises the following steps: S1. Collect behavioral data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavioral data, physiological data and environmental data into a data set.

[0035] In a further embodiment, collecting the behavior data, physiological data and environmental data of the elderly based on multiple sensors, and merging the behavior data, physiological data and environmental data into a data set includes the following steps: S11. Use multiple sensors to synchronously collect behavioral data, physiological data, and environmental data, such as walking speed, sedentary time, balance ability, temperature and humidity, etc., and pre-process the collected raw data to remove noise and abnormal data, convert data from different sensors into a unified format for subsequent storage and processing, and interpolate missing data to ensure data continuity and integrity.

[0036] S12. Store the preprocessed data in a database to obtain a data set (implement data encryption and access control to protect the elderly's personal information from unauthorized access).

[0037] S2. Extract relevant features from the data set, and obtain feature combinations from the relevant features based on association rules (find frequently occurring feature combinations). The relevant features include walking speed (fast, slow, appropriate), sedentary time (long, short, appropriate), balance ability (good, medium, poor), etc., and quantify the impact of feature combinations on the probability of falls of the elderly.

[0038] The extracting of relevant features from the data set, obtaining feature combinations from the relevant features based on association rules, and quantifying the influence of the feature combinations on the probability of elderly people falling include the following steps: S21. Based on domain knowledge, such as experts in geriatric care, rehabilitation medicine, sports science, etc., identify features that affect the probability of falls as relevant features.

[0039] S22. Prepare association rule mining and set association rule parameters.

[0040] S23. Mining association rules based on support and confidence, generating association rules, and determining feature combinations from the association rules; association rule mining algorithms, such as Apriori algorithm, FP-Growth algorithm, etc.

[0041] S24. Complete the quantification of the impact of feature combinations on the probability of falls of the elderly.

[0042] In a further embodiment, preparing for association rule mining and setting association rule parameters comprises the following steps: S221, define the value of each relevant feature as an item, such as "fast", "slow", or "appropriate" for the pace, and convert the data into a format suitable for association rule analysis, such as One-Hot encoding, that is, convert each value into a binary variable. For example, when the pace is "fast", the corresponding One-Hot encoding is [1, 0, 0], and when the pace is "slow", it is [0, 1, 0].

[0043] S222. Use an association rule mining algorithm (such as the Apriori algorithm) to generate all possible candidate feature combinations, and calculate the support and confidence of all candidate feature combinations; support (Support) indicates the frequency of a specific feature appearing in all transactions, and confidence (Confidence) indicates the frequency of another feature being included in a transaction containing a specific feature.

[0044] S223. Set the minimum support and minimum confidence. Only feature combinations with support higher than the threshold are considered meaningful, and only rules with confidence higher than the threshold are considered strong association rules. Set a reasonable minimum support threshold to ensure that the mined rules are representative. Set a reasonable minimum confidence threshold to ensure that the mined rules have high reliability.

[0045] In a further embodiment, mining association rules according to support and confidence, and generating association rules, and determining feature combinations from association rules include the following steps: S231. Filter out candidate feature combinations with support higher than the minimum support as frequent item sets; assuming that the minimum support threshold is 0.2, if the combination of feature A and feature B appears in 30% of the transactions, then the combination of A and B is a frequent item set.

[0046] S232. For each frequent item set, generate an association rule A→B, where A is the premise of the rule and B is the result of the rule.

[0047] S233. For each association rule, calculate the confidence of the association rule, that is, the frequency of occurrence of result B when premise A appears; if A and B appear in 10 transactions at the same time, and A appears in 30 transactions, then the confidence of rule A→B is 10 / 30=0.33.

[0048] S234. Association rules with confidences higher than the minimum confidence are retained as strong association rules. Assuming that the minimum confidence threshold is 0.5, if the confidence of rule A→B is 0.33, the rule will be discarded.

[0049] S235. Check whether the generated strong association rules are in line with the actual situation, that is, whether they have practical significance; through business knowledge and actual application scenarios, check whether the generated strong association rules are reasonable and have practical significance.

[0050] S236, taking the item set in the strong association rule as a feature combination and constructing a feature combination list. That is, extracting frequent item sets from the strong association rule as feature combinations. Constructing the feature combinations in all strong association rules into a feature combination list for subsequent analysis and application.

[0051] In a further embodiment, completing the quantification of the influence of the feature combination on the probability of the elderly falling includes the following steps: S241. Count the number of occurrences of each feature combination in the data set, and count the number of elderly people's fall events when each feature combination appears, and divide the number of elderly people's fall events by the number of occurrences of the feature combination in the data set to calculate the fall probability of the feature combination.

[0052] S242. Determine the respective weights based on the frequency of occurrence of feature combinations and expert opinions, invite field experts to participate, evaluate the importance of feature combinations based on practical experience, determine the weights of each feature combination through expert scoring, and calculate the weighted probability corresponding to each feature combination as a quantification of the impact on the probability of falls of the elderly.

[0053] Among them, the calculation formula for calculating the weighted probability corresponding to a certain feature combination is: ; In the formula, wp Represents the weighted probability corresponding to a certain feature combination; x Indicates the number of falls of the elderly. y Indicates the number of occurrences of a feature combination in the data set. z Represents the sum of the occurrences of all feature combinations; w 1 represents the weight of expert opinion, The weight coefficient representing the weight of the frequency of occurrence, β The weight coefficient representing the weight of expert opinion.

[0054] S3. Based on the machine learning algorithm, learn the relationship between relevant features and whether the elderly fall, and predict the probability of the elderly falling corresponding to each sample in the new data set.

[0055] In a further embodiment, learning the relationship between relevant features and whether the elderly person falls based on a machine learning algorithm and predicting the probability of the elderly person falling corresponding to each sample of the new data set includes the following steps: S31. Use the gradient boosting decision tree to assign weights to each relevant feature to reflect its impact on the probability of falling.

[0056] S32. Based on the random forest algorithm and combined with the feature weights determined by the gradient boosting decision tree, several decision trees are constructed; the random forest consists of multiple decision trees, each tree uses a different feature subset and sample subset during training, and the feature weights of the gradient boosting decision tree are combined to optimize the construction process of each decision tree.

[0057] S33. The random forest model is trained according to the obtained relevant features and corresponding labels (whether the elderly person falls or not), and the parameters of the random forest model, such as the number of trees, the maximum depth, etc., are optimized.

[0058] S34. The cross-validation method was used to evaluate the stability and accuracy of the random forest model, and the performance of the random forest model was evaluated using indicators such as accuracy, precision, recall and F1 value.

[0059] S35. Apply the trained random forest model to the new data set and predict the probability of the elderly falling for each sample in the new data set.

[0060] It should be noted that each new data set consists of multiple samples, each of which contains several related features, which are input variables for prediction. The samples in the new data set are input into the trained random forest model one by one. Each decision tree in the random forest model processes the input sample features independently. Each tree assigns a category to the sample or predicts a value through a series of feature-based decision rules (i.e., tree bifurcations). Random forests usually take the average of all tree predictions as the final prediction.

[0061] In a further embodiment, assigning weights to each relevant feature using a gradient boosted decision tree comprises the following steps: S311. Initialize the decision tree and select the loss function. The choice of loss function depends on the type of problem (for example, mean square error is used for regression and cross entropy is used for classification).

[0062] S312. Calculate the prediction residuals of all current decision trees and train a new decision tree based on the residuals; the difference is the difference between the actual value and the value predicted by the current model, and the purpose of training the new decision tree is to predict these residuals.

[0063] S313, determine the weight of the new tree (also called learning rate or step size), which is usually a value less than 1, used to control the contribution of the new tree to the final prediction, and use the prediction of the new tree and the calculated weight to update the overall model. Among them, in GBDT, the overall model refers to the weighted sum of multiple decision trees. The overall model forms a powerful predictor through the accumulation process, which integrates the prediction capabilities of all decision trees.

[0064] S314. Count the frequency with which each relevant feature is used as a split node in all trees, and measure the contribution of each feature split to the reduction of the loss function. At the same time, add up the feature importance scores in all trees to obtain the total importance score of each relevant feature.

[0065] S315: Assign weights to each relevant feature according to the total importance score. The total importance score of each feature is divided by the maximum value of the total scores of all features, and normalized, and the normalized feature importance is converted into weights, for example, using a linear ratio to assign weights.

[0066] S4. Analyze the new data set to determine whether a feature combination appears. If so, optimize the probability of the elderly falling based on the quantified result, that is, add the quantified result to the probability of the elderly falling.

[0067] In a further embodiment, analyzing the new data set to determine whether a feature combination appears, if so, optimizing the probability of elderly people falling based on the quantified results includes the following steps: S41. Extract relevant features from each sample of the new data set and compare them with the feature combination to determine whether the feature combination exists in the sample.

[0068] S42. If there is a feature combination, the weighted probability corresponding to the feature combination is added to the predicted probability of the elderly falling, to obtain an optimized probability of the elderly falling.

[0069] If a feature combination is detected in a sample in the new data set, the weighted probability of this feature combination is added to the currently predicted probability of the elderly falling. This addition of weighted probabilities can more accurately reflect the actual risk of the sample. For example, if the current model predicts that the probability of a certain elderly person falling is 20%, and a certain feature combination is found (for example, increasing the probability of falling by 1%), then the optimized probability of falling will be 21%.

[0070] S5. Set several thresholds. If the optimized probability of elderly people falling exceeds a certain threshold, execute the corresponding early warning plan.

[0071] In a further embodiment, several thresholds are set. If the optimized probability of elderly people falling exceeds a certain threshold, executing the corresponding early warning scheme includes the following steps: S51. Set several thresholds of the probability of falling, for example, low risk, medium risk and high risk thresholds, and set different warning levels and response measures for each threshold.

[0072] S52: Compare the optimized elderly person's fall probability with a preset threshold value to determine the warning level of the elderly person's fall probability.

[0073] S53. Select a corresponding warning plan based on the warning level of the probability of the elderly falling.

[0074] The early warning plan includes the following: No risk: No early warning measures are taken.

[0075] Low risk: Remind the elderly and their families to pay attention to their activities and recommend regular safety checks.

[0076] Medium risk: Notify family members and caregivers to closely monitor the elderly, which may require adjustments to the elderly’s activity plans or environment.

[0077] High risk: Immediately trigger an emergency alert and notify medical institutions or emergency services to ensure that the elderly receive timely help.

[0078] Feedback mechanism: Collect feedback information after early warning execution and evaluate the early warning effect. Adjust and optimize the early warning plan and threshold setting according to the feedback results to improve the accuracy and reliability of early warning.

[0079] In order to facilitate understanding of the above technical solutions of the present invention, the working principle or operation mode of the present invention in the actual process is described in detail below.

[0080] Obtain a data set and set the pace: fast (>1.2 m / s), slow (<0.8 m / s), appropriate (0.8-1.2 m / s). Sedentary time: long (>6 hours), short (<3 hours), appropriate (3-6 hours). Balance ability: good, medium, poor.

[0081] Get the number of occurrences of feature combinations and the corresponding number of falls.

[0082] The number of fast walking, short sitting time and good balance ability is 100, which corresponds to 1 fall.

[0083] The number of occurrences of appropriate walking speed, long sitting time, and medium balance ability is 200, and the corresponding number of falls is 10.

[0084] Slow walking speed, long sitting time, and poor balance ability occurred 150 times, corresponding to 30 falls.

[0085] The weight coefficients of expert opinion weight and frequency weight are set to 0.6 and 0.4 respectively, and the weight coefficient of expert opinion weight is set to 0.5. The weighted probability is calculated using the formula.

[0086] The predicted probability of the samples in the new data set is predicted by the random forest model, and then added to the weighted probability calculated above. For example, if the predicted probability is 0.1 and the weighted probability is 0.00333, the optimized probability is 0.10333.

[0087] Set the fall probability threshold, no risk: the fall probability is less than or equal to 0.1.

[0088] Low risk: The probability of falling is greater than 0.1 and less than or equal to 0.2.

[0089] Medium risk: The probability of falling is between 0.2 and 0.5.

[0090] High risk: The probability of falling is greater than or equal to 0.5.

[0091] After optimization, the probability of falling is 0.10333, which is low risk, and the low-risk early warning plan is implemented.

[0092] like Figure 2 As shown, according to another embodiment of the present invention, a multi-sensor based intelligent monitoring system for preventing falls for the elderly is provided, and the multi-sensor based intelligent monitoring system for preventing falls for the elderly includes a data acquisition module 1, a feature combination acquisition module 2, a probability prediction module 3, a probability optimization module 4 and an early warning execution module 5; wherein the data acquisition module 1 is connected to the feature combination acquisition module 2, the feature combination acquisition module 2 is connected to the probability prediction module 3, the probability prediction module 3 is connected to the probability optimization module 4, and the probability optimization module 4 is connected to the early warning execution module 5.

[0093] The data collection module 1 is used to collect the behavior data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavior data, physiological data and environmental data into a data set.

[0094] The feature combination acquisition module 2 is used to extract relevant features from the data set, obtain feature combinations from the relevant features based on association rules, and quantify the impact of the feature combination on the probability of the elderly falling.

[0095] The probability prediction module 3 is used to learn the relationship between relevant features and whether the elderly have fallen based on the machine learning algorithm, and predict the probability of the elderly falling corresponding to each sample of the new data set.

[0096] The probability optimization module 4 is used to analyze the new data set to determine whether the feature combination appears. If so, the probability of the elderly falling is optimized based on the quantitative results.

[0097] The early warning execution module 5 is used to set several thresholds. If the optimized probability of the elderly falling exceeds a certain threshold, the corresponding early warning plan is executed.

[0098] In summary, the multi-sensor-based intelligent monitoring method and system for preventing falls of the elderly provided by the present invention collects behavioral data, physiological data and environmental data of the elderly by using multi-sensors, provides the overall health status and activity of the elderly, and helps to accurately monitor the behavioral changes of the elderly. Extracting features related to the fall of the elderly, and using the trained model to predict the new data set, calculating the probability of fall corresponding to each sample, can help identify high-risk elderly people in advance and take preventive measures. With the continuous increase of data and the continuous learning of the model, the prediction model can be continuously optimized to improve the prediction accuracy and stability. Based on the quantitative results, the probability of fall is optimized, and the quantitative impact of the detected feature combination is superimposed on the prediction result, which further improves the accuracy of the fall risk assessment and reduces missed reports and false alarms. Different risk thresholds are set, corresponding to different warning levels, and graded warnings can take different intervention measures according to actual conditions, thereby improving the flexibility and pertinence of the warning system. The present invention uses association rules to mine the relationship between features, find out the frequently occurring feature combinations, so as to reveal hidden patterns, help to better understand the formation mechanism of the risk of fall, quantify the impact of feature combinations on the probability of fall, and clarify to what extent each feature combination increases the risk of fall of the elderly. This quantitative analysis provides a scientific basis for subsequent risk prediction and intervention measures. At the same time, when quantifying the impact of feature combinations on the probability of falls of the elderly, the respective weights are determined based on the frequency of occurrence of feature combinations and expert opinions, which can comprehensively consider the dual factors of data-driven and empirical knowledge, thereby improving the scientificity and reliability of the quantitative results.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent monitoring method for preventing elderly people from falling based on multiple sensors, characterized in that: The multi-sensor based intelligent monitoring method for preventing elderly people from falling includes the following steps: S1. Collect behavioral data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavioral data, physiological data and environmental data into a data set; S2, extract relevant features from the data set, obtain feature combinations from the relevant features based on association rules, and quantify the impact of feature combinations on the probability of falls of the elderly; S3, based on machine learning algorithms, learn the relationship between relevant features and whether the elderly fall, and predict the probability of the elderly falling corresponding to each sample in the new data set; S4, analyze the new data set to determine whether the feature combination appears. If so, optimize the probability of elderly people falling based on the quantitative results; S5. Set several thresholds. If the optimized probability of elderly people falling exceeds a certain threshold, execute the corresponding early warning plan; The extracting of relevant features from the data set, obtaining feature combinations from the relevant features based on association rules, and quantifying the influence of the feature combinations on the probability of falls of the elderly include the following steps: S21. Determine features that affect the probability of falling based on domain knowledge as relevant features; S22, prepare association rule mining and set association rule parameters; S23, mining association rules according to support and confidence, generating association rules, and determining feature combinations from the association rules; S24. Complete the quantification of the impact of feature combinations on the probability of falls of the elderly.

2. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 1 is characterized in that: The method of collecting the behavior data, physiological data and environmental data of the elderly based on multiple sensors and merging the behavior data, physiological data and environmental data into a data set includes the following steps: S11, synchronously collecting behavioral data, physiological data and environmental data through multiple sensors, and preprocessing the collected raw data; S12. Storing the preprocessed data in a database to obtain a data set.

3. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 2 is characterized in that: The preparation for association rule mining and setting association rule parameters include the following steps: S221, defining the value of each relevant feature as an item, and converting the data into a format suitable for association rule analysis; S222, calculating the support and confidence of all candidate feature combinations; S223. Set the minimum support and minimum confidence.

4. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 3 is characterized in that: The mining of association rules according to support and confidence, generating association rules, and determining feature combinations from association rules include the following steps: S231, filter out candidate feature combinations with support higher than the minimum support as frequent item sets; S232. For each frequent item set, generate an association rule A→B, where A is the premise of the rule and B is the result of the rule; S233. For each association rule, calculate the confidence of the association rule; S234, retaining association rules with confidence levels higher than the minimum confidence level as strong association rules; S235, checking whether the generated strong association rules are consistent with the actual situation; S236. Use the item set in the strong association rule as a feature combination, and construct a feature combination list.

5. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 4 is characterized in that: The quantification of the influence of the feature combination on the probability of falling of the elderly comprises the following steps: S241, counting the number of occurrences of each feature combination in the data set, and counting the number of elderly fall events when each feature combination appears, and dividing the number of elderly fall events by the number of occurrences of the feature combination in the data set to calculate the fall probability of the feature combination; S242, determining respective weights based on the occurrence frequency of the feature combination and expert opinions, and calculating the weighted probability corresponding to each feature combination as a quantification of the impact on the probability of the elderly falling; Among them, the calculation formula for calculating the weighted probability corresponding to a certain feature combination is: ； In the formula, wp Represents the weighted probability corresponding to a certain feature combination; x Indicates the number of falls of the elderly. y Indicates the number of occurrences of a feature combination in the data set. z Represents the sum of the occurrences of all feature combinations; w 1 represents the weight of expert opinion, The weight coefficient representing the weight of the frequency of occurrence, β The weight coefficient representing the weight of expert opinion.

6. The multi-sensor based intelligent fall prevention monitoring method for the elderly according to claim 5 is characterized in that: The method of learning the relationship between relevant features and whether the elderly person falls based on a machine learning algorithm and predicting the probability of the elderly person falling corresponding to each sample of the new data set includes the following steps: S31, using gradient boosting decision tree to assign weights to each relevant feature; S32. Based on the random forest algorithm and in combination with the feature weights determined by the gradient boosting decision tree, several decision trees are constructed; S33, training the random forest model according to the acquired relevant features and corresponding labels, and optimizing the parameters of the random forest model; S34. Use cross-validation to evaluate the stability and accuracy of the random forest model and evaluate the performance of the random forest model; S35. Apply the trained random forest model to the new data set and predict the probability of the elderly falling for each sample in the new data set.

7. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 6, characterized in that: The method of using the gradient boosting decision tree to assign weights to each relevant feature includes the following steps: S311, initializing the decision tree and selecting the loss function; S312, calculating the prediction residuals of all current decision trees, and training a new decision tree based on the residuals; S313, determining the weight of the new tree, and using the prediction of the new tree and the calculated weight to update the overall model; S314, counting the frequency of each relevant feature being used as a split node in all trees, and measuring the contribution of each feature split to the reduction of the loss function, and at the same time accumulating the feature importance scores in all trees to obtain the total importance score of each relevant feature; S315. Assign a weight to each relevant feature according to the total importance score.

8. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 7, characterized in that: The analyzing of the new data set to determine whether a feature combination appears, and if so, optimizing the probability of elderly people falling based on the quantified results includes the following steps: S41, extract relevant features from each sample of the new data set, and compare them with the feature combination to determine whether the feature combination exists in the sample; S42. If there is a feature combination, the weighted probability corresponding to the feature combination is added to the predicted probability of the elderly falling, to obtain an optimized probability of the elderly falling.

9. The multi-sensor based intelligent monitoring method for preventing elderly people from falling according to claim 8, characterized in that: The setting of several thresholds, if the optimized probability of elderly people falling exceeds a certain threshold, executing the corresponding early warning plan includes the following steps: S51, setting several thresholds of the probability of falling, and setting different warning levels and response measures corresponding to each threshold; S52, comparing the optimized probability of the elderly falling with a preset threshold value to determine the warning level of the probability of the elderly falling; S53. Select a corresponding warning plan based on the warning level of the probability of the elderly falling.

10. An intelligent monitoring system for preventing elderly people from falling based on multiple sensors, used to implement the intelligent monitoring method for preventing elderly people from falling based on multiple sensors as claimed in any one of claims 1 to 9, characterized in that: The multi-sensor-based intelligent fall prevention monitoring system for the elderly includes a data acquisition module, a feature combination acquisition module, a probability prediction module, a probability optimization module and an early warning execution module; Wherein, the data acquisition module is connected to the feature combination acquisition module, the feature combination acquisition module is connected to the probability prediction module, the probability prediction module is connected to the probability optimization module, and the probability optimization module is connected to the early warning execution module; The data acquisition module is used to collect the behavior data, physiological data and environmental data of the elderly based on multiple sensors, and merge the behavior data, physiological data and environmental data into a data set; The feature combination acquisition module is used to extract relevant features from the data set, obtain feature combinations from the relevant features based on association rules, and quantify the impact of the feature combination on the probability of the elderly falling; The probability prediction module is used to learn the relationship between relevant features and whether the elderly have fallen based on a machine learning algorithm, and predict the probability of the elderly falling corresponding to each sample of the new data set; The probability optimization module is used to analyze the new data set to determine whether the feature combination appears. If so, the probability of the elderly falling is optimized based on the quantified results; The early warning execution module is used to set several thresholds. If the optimized probability of the elderly falling exceeds a certain threshold, the corresponding early warning plan is executed.

Images