A method and system for generating cognitive training strategies based on time-series databases

Abstract

This invention relates to a cognitive training strategy generation system based on a time-series database, comprising at least one sensor capable of collecting different user action information, and configuring at least one server for all sensors. The server is configured to construct an information dataset by at least building a sensor device information table for describing the basic information of the sensors and a sensor data information table for the information content collected by the sensors. This invention also relates to a cognitive training strategy generation method based on a time-series database, comprising the following steps: constructing a list of predicted behaviors, a list of interests, a list of lifestyle habits, and a list of rehabilitation measures; based on the correlation analysis results between the list of predicted behaviors, the list of interests, the list of lifestyle habits, and the list of rehabilitation measures, constructing a behavior guidance list by inserting the specific rehabilitation measures from the list of rehabilitation measures into the list of predicted behaviors.

Description

[0001] The original basis for this divisional application is patent application No. 201910421470.8, filed on May 20, 2019, entitled "A method for constructing a time-series database for cognitive early warning". Technical Field

[0002] This invention relates to the field of database construction technology, and in particular to a method and system for generating cognitive training strategies based on time-series databases. Background Technology

[0003] Existing technologies, such as Yang Shuai's "A Method for Recognizing Elderly Behavior Based on Indoor Trajectory Analysis," in the "China Excellent Master's Thesis Full-Text Database Information Science and Technology Series," pp. 2-11, 26-39, disclose a method for recognizing elderly behavior based on indoor trajectory analysis. This method constructs a trajectory analysis model based on elderly people living alone, including a user behavior model, an indoor key point model, and a behavior temporal model. It proposes a behavior trajectory analysis method based on trajectory segmentation, segmenting the key dwell times contained in the collected behavior trajectory to identify the actual dwell points within the trajectory, using these actual dwell points as the location markers for the start or end of the behavior. Based on the location and time data in the behavior trajectory, it establishes location-behavior association templates, time-behavior association templates, and preceding key point association templates, respectively. It also proposes a behavior recognition method based on template matching, utilizing the different weights of the current location, time, and preceding key points in the behavior trajectory data to establish associations with daily behaviors sequentially. Furthermore, it analyzes the correlation between the probability of behavior trajectory occurrence and user behavior habits, defining behaviors not appearing in the normal behavior trajectory sequence as abnormal behaviors. The database is constructed based on key point data and start-time data experienced by the user during movement. The behavioral trajectory is represented by a series of key point sets arranged chronologically. Essentially, this technical solution is based on the modular division of time periods during the behavioral process. Based on this database, the only way to determine abnormal behavior is "the user's dwell time in this area exceeds the normal value," and it cannot handle abnormal behavior between overlapping actions.

[0004] Furthermore, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the inventors studied a large number of documents and patents when making this invention, but due to space limitations, not all details and contents were listed in detail. However, this does not mean that the present invention does not possess the features of these prior art. On the contrary, the present invention already possesses all the features of the prior art, and the applicant reserves the right to add relevant prior art to the background art. Summary of the Invention

[0005] As used in this article, the term "module" describes any hardware, software, or combination of hardware and software that can perform the functions associated with the "module".

[0006] To address the shortcomings of existing technologies, this invention provides a method for constructing a time-series database for cognitive early warning. This method involves collecting and integrating real-time user action information to form an information dataset. The construction method further includes the following steps: constructing an expert knowledge database by performing higher-level classification on the information dataset to obtain several classification data; constructing a behavior prediction classifier based on the classification data to predict behaviors that will occur at a future set time; scoring the predicted behaviors; and generating early warning information when the score drops below a set threshold.

[0007] According to a preferred embodiment, the scoring process includes at least the following steps: selecting the n most frequently triggered behaviors as standard tasks X = (X1, X2, ..., X...). n The time taken to complete the n tasks is used as the standard time T for that behavior and is represented as T = (T1, T2, ..., Tn). n Determine the time t = (t1, t2, ..., tn) required for the user to complete the standard task. n The user's task completion rate for the standard task is α = (α1, α2, ..., α3). n ), and determine the respective weight w = (w1, w2, ..., w) based on the importance of each standard task. n Establish a scoring formula for cognitive abilities. The user's cognitive ability score is determined based on the scoring formula.

[0008] According to a preferred embodiment, the construction of the information dataset includes at least the following steps: setting up a number of sensors capable of collecting different user action information in the user's activity space, and configuring at least one server for all sensors; the server is configured to complete the construction of the information dataset in a manner that at least a sensor device information table for describing the basic information of the sensors and a sensor data information table for describing the information content collected by the sensors are constructed.

[0009] According to a preferred embodiment, the construction of the expert knowledge database includes at least the following steps: classifying the information dataset based on the correlation between basic life behaviors, sensors, and sensor attributes to construct a sensor behavior table that includes at least behavior name, behavior definition, triggering item, triggering sensor, triggering time, and location, wherein: based on the sensor behavior table, the basic life behaviors can be limited to a number of sensors that can be triggered by decomposing them into specific actions, items triggered by the actions, and sensors associated with the items.

[0010] According to a preferred embodiment, when generating early warning information, a rehabilitation training strategy is generated based on the predicted behavior. Generating the rehabilitation training strategy includes at least the following steps: constructing a list of predicted behaviors, a list of interests, a list of lifestyle habits, and a list of rehabilitation measures; based on the correlation analysis results between the list of predicted behaviors, the list of interests, the list of lifestyle habits, and the list of rehabilitation measures, constructing a behavior guidance list by inserting the rehabilitation measures in the specific rehabilitation measures list into the list of predicted behaviors.

[0011] According to a preferred embodiment, generating the rehabilitation training strategy further includes the following steps: constructing several different behavioral guidance lists in a way that inserts different rehabilitation measures into each other within a set time period; and configuring a voice interactive device or display within the set time period to guide the user to complete cognitive rehabilitation training by alternately executing the behavioral guidance lists.

[0012] According to a preferred embodiment, constructing the behavior guidance list includes at least the following steps: performing a first-level correlation analysis on the rehabilitation measures list and the lifestyle habits list to filter out rehabilitation measures that do not meet the requirements, thereby generating a first rehabilitation measures list; performing a second-level correlation analysis on the first rehabilitation measures list and the interests list to maximize the satisfaction of the user's interests and hobbies, thereby generating a second rehabilitation measures list; performing a third-level correlation analysis on the second rehabilitation measures list and the predicted behavior list, and inserting the rehabilitation measures in the second rehabilitation measures list into the predicted behavior list to form a behavior guidance list.

[0013] According to a preferred embodiment, the list of rehabilitation measures consists of several behavioral measures that can inhibit the decline in the user's cognitive abilities, wherein the server can create the list of rehabilitation measures through customization or network acquisition.

[0014] According to a preferred embodiment, the behavior prediction classifier can be constructed based on a Hidden Markov Model, wherein: the behavior prediction classifier can be determined by the element value of the i-th row and j-th column of matrix A, which is determined by the i-th state q. i Transition to the j-th state q j The probability a ij =P(q) j |q i ), and define the value of the k-th row and j-th column of matrix B in state q. j The probability b of observing data k at that time j (k)=P(k|q j The Hidden Markov Model λ = (A, B, π) is constructed in a manner that allows the Hidden Markov Model to be trained using pre-collected sensor data.

[0015] According to a preferred embodiment, behavior prediction is performed using a trained Hidden Markov Model (HMM) on sensor data collected within a set time period, wherein: when using the HMM for behavior prediction, the sensor data O = (o0, o1, ..., o2) is... T-1 As the observation sequence, the behavior I = (i0, i1, ..., i...) T-1 () as a latent variable.

[0016] The beneficial technical effects of this invention are as follows: By establishing an implicit sensing environment, this invention collects users' daily life data and builds a behavioral model. This allows for long-term implicit data collection, reduces disruption to users' daily lives, obtains long-term stable data, and thereby establishes the behavioral model. Cognitive ability is assessed through long-term and effective monitoring of daily life data, reducing errors in subjective human judgment. Simultaneously, through real-time monitoring and prediction, cognitive ability is scored, providing early warnings during cognitive decline. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a preferred method for constructing a time-series database according to the present invention; and

[0018] Figure 2 This is a schematic diagram of the modular connection relationship of the preferred modules of the present invention.

[0019] List of reference numerals

[0020] 1: Sensor; 2: Server; 3: Voice Interaction Device

[0021] 4: Monitor Detailed Implementation

[0022] The following is a detailed explanation with reference to the accompanying drawings.

[0023] Example 1

[0024] like Figure 1 and Figure 2 As shown, the present invention provides a method for constructing a time-series database for cognitive early warning, comprising at least the following steps:

[0025] S1: Collect and integrate the user's real-time action information to form an information dataset.

[0026] Specifically, several sensors 1 capable of collecting different user movement information are installed in the user's activity space, and at least one server 2 is configured for each sensor 1, enabling the server 2 to store at least the real-time movement information collected by the sensors 1. For example, the server 2 can be a cloud server, and the sensors 1 can communicate with the server 2 wirelessly. The activity space refers to the user's living space. For example, when a user lives in their own house, the activity space is the living space constituted by the house. When a user is in a hospital ward, the activity space is the ward itself. Preferably, to facilitate the collection of different user movement information, different sensors can be arranged in the house, for example, as shown in Table 1.

[0027] Table 1

[0028]

[0029] Preferably, server 2 is configured to construct the information dataset by at least building a sensor device information table and a sensor data information table. The sensor device information table describes the basic information of the sensor. This basic information includes at least the sensor name, MAC address, function description, sensor type, data unit, and installation location. The sensor data information table describes the information content collected by the sensor. This information content includes at least the sensor device, data content, and upload time. For example, to collect real-time action information of a user in a home environment, such as using kitchen utensils, a flame sensor can be installed on the kitchen stove. The flame sensor can be a far-infrared flame sensor or an ultraviolet flame sensor. In this case, the MAC address serves as a unique identifier for the sensor, used to distinguish different sensors; different sensors have different MAC addresses. The sensor name is "Flame Sensor." The function description is "Determining whether gas is being used." The sensor type is the type of sensor. The data unit is the unit of the detected data signal. For example, when the sensor is a flame sensor, it can collect the wavelength emitted by the flame, and the data unit is nanometers. The installation location is the kitchen stove. The sensor device is the specific model of the sensor. The data content includes at least the gas on / off time, the gas off time, and the intensity of the infrared light emitted by the gas. Upload time refers to the time it takes for the flame sensor to upload its collected real-time action information to the server. For example, when the sensor is triggered to collect data, it uploads the collected data to the server in real time for unified storage. The server can periodically archive and process the data based on the upload time. For example, the server can archive the data at midnight every day to package and store the data collected that day. Simultaneously, the data is stored in the database in binary format. That is, for door magnetic switch sensors, flame sensors, laser sensors, water flow sensors, micro-switch sensors, pressure sensors, and smart switches, when the value of the data collected by the sensor is greater than zero, it is stored as 1; otherwise, it is stored as 0. For temperature and humidity sensors, before storage, the sensor value is divided by the maximum temperature or maximum humidity value within the sensor's measurement range, and the resulting value between 0 and 1 is stored in the database.

[0030] S2: Construct an expert knowledge database by performing higher-level classification on the information dataset to obtain several categorized data. Expert knowledge provides prior knowledge about the relationship between sensors and behavior, increasing the accuracy of behavior prediction and thus improving the accuracy of cognitive early warning. Performing higher-level classification refers to further subdividing the data to enable the association between behavior and sensors.

[0031] Specifically, an expert knowledge database is constructed by classifying basic daily behaviors, sensors, and the relationships between sensor attributes. This database contains at least a sensor behavior table. The sensor behavior table includes at least the behavior name, behavior definition, triggering item, triggering sensor, trigger time, and location. By defining basic daily behaviors, these behaviors can be broken down into specific actions, the triggering items, and the associated sensors. Limiting a behavior to a few sensors it might trigger effectively improves the accuracy of behavior prediction. For example, when a user needs to leave home, the behavior can be defined as "leaving home." Leaving home requires at least opening and closing the door, thus the behavior can be defined as opening the door, closing the door, and no other indoor behavior trigger. That is, when the server analyzes data collected by sensors and determines that the user has sequentially opened and closed the door, and there are no other indoor behavior triggers, it predicts that the user is in a state of leaving home. In this case, the triggering item is the door. The triggering sensor is the door magnetic switch sensor installed on the door. The trigger time is not limited; that is, the user leaving home at any time is considered normal. Location refers to the address where the behavior occurs; for example, leaving home can be defined as the living room. Preferably, a sensor behavior table, as shown in Table 2, containing several different behavior names can be constructed based on the correlation between different behaviors and different sensors. Constructing the sensor behavior table clarifies the many-to-many relationship between different behaviors and sensors. That is, a complete behavior requires multiple sub-behaviors to trigger different sensors. For example, the behavior of preparing breakfast, as shown in Table 2, requires the operation of a refrigerator, a gas stove, and a microwave oven, which will trigger the door magnetic switch sensor of the refrigerator, the flame sensor of the gas stove, and the door magnetic switch sensor of the microwave oven. Preferably, each row of data in the sensor behavior table can be used as a categorized data item. Preferably, the attributes include at least the sensor's installation location, the bound item, and the trigger time.

[0032] Table 2

[0033]

[0034]

[0035] S3: Construct a behavior prediction classifier based on categorical data to predict behaviors that will occur at a future set time.

[0036] Specifically, behavior prediction classifiers can be built based on Hidden Markov Models (HMMs). For example, to address the implicit temporal sequence in sensor data and behavior, HMMs can be used to predict behavior that will occur at a future time based on sensor data. HMMs introduce a latent variable and assume that changes in the state of the data are caused by and only by the latent variable at the previous time point, but this latent variable cannot be directly observed by the observer. In an HMM, the behavior at the time point to be predicted is solved from the transition matrix between the latent variables, the initial state probability matrix, and the emission matrix of the behavior corresponding to the latent variable. When using HMMs for behavior prediction, behavior is treated as a latent variable of the sensor; sensor data is observable, while behavior data is hidden.

[0037] Preferably, behavior prediction is performed using a trained Hidden Markov Model (HMM) on sensor data collected within a set time period. For example, the HMM can be trained using pre-collected sensor data, and then used to predict behavior on sensor data collected each day, with a 24-hour cycle. Specifically, the HMM training and prediction process includes at least the following steps:

[0038] A1: Define the value of the element in the i-th row and j-th column of matrix A as determined by the i-th state q. i Transition to the j-th state q j probability ai j = P(q) j |q i Define the value of matrix B in the k-th row and j-th column when the state is q. j When, the probability b of observing data k j (k)=P(k|q j At this point, the Hidden Markov Model λ = (A, B, π) is complete.

[0039] A2: In the constructed Hidden Markov Model (HMM), both A and B are unknown parameters, requiring the Viterbi algorithm to solve the HMM. Specifically, the Viterbi algorithm aims to maximize the probability maxP(I / O) of the latent variable sequence given the observed data. The Viterbi algorithm is defined as follows: The formula defined based on the Viterbi algorithm can be used to derive the recursive formula. ,

[0040] Among them, the definition Therefore, the Viterbi algorithm process is as follows:

[0041] Input: Model λ = (A, B, π), Observational data sequence O = (o0, o1, ..., o T-1 ).

[0042] Output: Latent variable sequence I = (i0, i1, ..., i... T-1 ).

[0043] Initialize local state:

[0044] δ0=π i b i (o0), i = 0, 1, ..., N-1

[0045] Ψ0(i)=0, i=0, 1,...,N-1

[0046] Perform dynamic programming to recursively calculate the local states at times t = 1, 2, ..., T-1:

[0047]

[0048]

[0049] Calculate the maximum Ψ at time T T (i), at this time Ψ T (i) is the most likely latent variable state at time t = T.

[0050]

[0051] Using Ψ(i) backtracking, we obtain I = (i0, i1, ..., i...). T-1 ):

[0052] i t =Ψ t+1 (i t+1 )

[0053] When using a Hidden Markov Model for prediction, the sensor data O = (o0, o1, ..., o2) is used. T-1 Viewing the behavior I = (i0, i1, ..., i...) as an observation sequence, we can define the behavior as I = (i0, i1, ..., i...). T-1 This can be considered a latent variable. The behavior at each time step can be obtained using the Viterbi algorithm.

[0054] Preferably, for ease of understanding, the specific meanings of the parameters involved in the above process are defined. T represents the data sequence length. A represents the transition matrix between states (latent variables). B represents the emission matrix. π represents the initial state probability matrix. N represents the number of values ​​for the latent variables. Q = {q0, q1, ..., q...} N-1} represents the set of possible values ​​for the latent variable. I = (i0, ..., i1 , ..., i T-1 O represents a sequence of latent variables. O = (o0, o1, ..., o2) T-1 ) represents the observed data sequence.

[0055] S4: The predicted behaviors are scored, and an alert is generated when the score drops below a set threshold. A decline in cognitive ability manifests in daily life as behavioral errors, such as skipping or incorrectly executing important steps in an activity, rendering the activity meaningless. Examples include forgetting to turn off the gas, leaving the refrigerator door open for an extended period, or spending too much time completing a simple task. Cognitive ability is standardized using scores to quantify the quality of activity completion. Based on daily scoring information, an alert is generated when a user's cognitive ability score drops below, for example, a 10% threshold.

[0056] Specifically, the scoring process for behavior includes at least the following steps:

[0057] B1: Select the most n behaviors that are triggered daily as the standard task X = (X1, X2, ..., X...). n ).

[0058] B2: The time it takes for an average person to complete these n tasks is obtained from the test and taken as the standard time T for this behavior, denoted as T = (T1, T2, ..., T...). n ).

[0059] B3: Determine the time t = (t1, t2, ..., t3) required for the user to complete the standard task. n The user's task completion rate for the standard task is α = (α1, α2, ..., α3). n ), and determine the respective weight w = (w1, w2, ..., w) based on the importance of each standard task. n ).

[0060] Specifically, the task completion rate of a standard task can be determined by the ratio of the number of sub-tasks actually completed to the total number of sub-tasks. For example, the dishwashing behavior shown in Table 2 includes 5 sub-tasks: turning on the water valve, turning off the water valve, turning the water valve on and off several times, opening the cupboard, and closing the cupboard. When the user completes 3 of these sub-tasks, the task completion rate for the dishwashing behavior is (3 / 5)*100% = 60%.

[0061] Preferably, the importance of standard tasks is categorized by the frequency of their occurrence. For example, the importance of standard tasks can be divided into three levels. The weight of the first level is set at 0.5, the weight of the second level at 0.3, and the weight of the third level at 0.2. The frequency of different behaviors in the database is statistically calculated. Behaviors occurring more than 3 times per day are classified as Level 1. Behaviors occurring 1 to 3 times per day are classified as Level 2. Behaviors occurring less than 1 time per day are classified as Level 3.

[0062] B4: Establish a scoring formula for cognitive abilities The cognitive ability-based scoring formula can calculate a user's cognitive ability score.

[0063] Example 2

[0064] This embodiment is a further improvement on embodiment 1, and repeated content will not be described again.

[0065] Preferably, when generating early warning information, a rehabilitation training strategy is generated based on the predicted behavior. Generating the rehabilitation training strategy includes at least the following steps:

[0066] C1: Construct a list of predictive behaviors, a list of interests, a list of lifestyle habits, and a list of rehabilitation measures.

[0067] Specifically, the list of predicted behaviors refers to a list of specific behaviors that are highly probable to occur at each future moment, as predicted by a Hidden Markov Model. This list can be constructed in a temporal order, meaning the predicted behaviors are arranged chronologically.

[0068] Preferably, the interest list refers to a specific list of behaviors composed of the user's interests. For example, a user may have various hobbies, including but not limited to listening to music, dancing, listening to opera, and singing opera. The user or their caregiver can input the user's interests into the server through an external input terminal. For example, server 2 can be configured with an external input terminal, such as a keyboard or voice input device, allowing the user or their children or other caregivers to input the user's interests into the server to build the interest list.

[0069] Preferably, the lifestyle habit list refers to the specific behaviors performed by the user within a set time period, obtained through analysis of data collected by sensors. Specifically, all sensor data collected within 24 hours can be analyzed to determine the specific behaviors performed by the user in different time periods. For example, for an elderly user, who wakes up at 7:00 AM, their first action is to use the toilet, followed by washing up. Thus, the user's lifestyle habit list can be obtained by associating time with specific behaviors. Partial content of the lifestyle habit list can be shown in Table 3. Preferably, the user's lifestyle habit list can be adjusted through regular updates. For example, in summer and winter, users' wake-up times may change due to climate changes, thus requiring regular updates to the lifestyle habit list.

[0070] Table 3

[0071] Serial Number time Specific actions 1 7：00～7：10 get up 2 7：10～7：30 Go to the toilet 3 7：30～7：50 Washing 4 7：50～8：30 Preparing breakfast 5 8：30～9：00 Eat breakfast

[0072] Preferably, the rehabilitation measure list consists of several behavioral measures that can inhibit the decline in the user's cognitive abilities. The server can create the rehabilitation measure list through customization or network access. Specifically, the server can establish an access relationship with, for example, a hospital expert system, thereby obtaining behavioral measures to inhibit the decline in cognitive abilities from the hospital expert system. Generally, hospital expert systems have expert knowledge databases that store effective relief measures for different diseases. For example, for Alzheimer's disease, the expert knowledge database can record measures that can effectively improve or prevent Alzheimer's disease, such as finger exercises, improving diet quality, and improving mood. After establishing an access relationship with the hospital expert system, the server can directly add the recorded measures to the rehabilitation measure list. Preferably, the server can also customize the rehabilitation measure list through external input. For example, after a professional diagnosis by a doctor, the doctor will formulate specific behavioral measures for the user based on their actual situation, and the user or their caregiver can then input the behavioral measures given by the doctor into the rehabilitation measure list through external input.

[0073] C2: Based on the correlation analysis results between the predicted behavior list, interest list, lifestyle habit list and rehabilitation measure list, construct a behavior guidance list by randomly inserting the rehabilitation measures in the specific rehabilitation measure list into the predicted behavior list.

[0074] Specifically, the server is configured to perform a first-level correlation analysis between the list of rehabilitation measures and the list of lifestyle habits to filter out unsuitable rehabilitation measures, thereby generating a first list of rehabilitation measures. The server can determine the user's lifestyle characteristics based on the list of lifestyle habits. These characteristics are primarily used to identify behaviors the user cannot perform. For example, the list of lifestyle habits might record that the user has a habit of praying at night. By combining this with sensors such as voice sensors or image sensors placed in the room, the content of the prayers can be determined, thus analyzing whether the user has a religious belief. The server can then obtain information about the taboos of that religion through external input or network connection, thereby deleting rehabilitation measures related to these taboos from the list of rehabilitation measures. The server can be equipped with correlation analysis algorithms such as grey relational analysis, FP-Growth, or Apriori to perform the first-level correlation analysis between the list of rehabilitation measures and the list of lifestyle habits.

[0075] Preferably, the server is also configured to perform a second-level association analysis on the first list of rehabilitation measures and the list of interests to maximize the satisfaction of the user's interests. Rehabilitation measures in the first list may completely conflict with the user's interests. For example, rehabilitation measures may include exercising such as running or square dancing to increase physical activity. Users may be unable or unwilling to perform exercise-related rehabilitation measures due to obesity or mobility issues. In this case, the server can delete rehabilitation measures that conflict with the interest list from the first list to form the second list of rehabilitation measures. The server can also use association analysis algorithms such as grey relational analysis, FP-Growth algorithm, or Apriori algorithm to perform the second-level association analysis between the first list of rehabilitation measures and the interest list.

[0076] Preferably, the server is also configured to perform a third-level association analysis on the second rehabilitation measure list and the predicted behavior list, and insert the rehabilitation measures from the second rehabilitation measure list into the predicted behavior list to form a behavior guidance list. The rehabilitation measures need to determine whether there is an execution conflict based on the occurrence time, location, and behavioral attributes of the specific behaviors in the predicted behavior list. Specifically, behavioral attributes can be divided into three categories: limb restriction, attention focus, and no restriction. Limb restriction indicates that the specific behavior requires the user's limbs to complete. For example, going to the toilet requires temporarily occupying the user's hands to unzip their pants and continuously occupying their legs. Another example is preparing breakfast, which requires prolonged and frequent use of the user's hands and occasional use of their legs to move around and change positions to retrieve or place items. Attention focus indicates that the specific behavior requires the user to concentrate highly and not be excessively disturbed by external factors. For example, when praying, the user should try to avoid external noise interference. No restriction indicates that the user is in a relaxed and leisurely state, and can adjust their state at any time to perform any task they need to complete. For example, when watching TV, a user can simply turn it on with the remote and then lie down or sit on the sofa for the entire time. In this relaxed state, their attention doesn't need to be highly focused, and their limbs are unused, allowing them to handle other tasks at any time. Preferably, the server can also use association analysis algorithms such as grey relational analysis, FP-Growth, or Apriori to perform a third-level association analysis between the second list of rehabilitation measures and the list of predicted behaviors. For example, regarding the specific behavior of going to the toilet after waking up at 7:00 AM, the time of this behavior is 7:00 AM, and the location is the bathroom. The behavior attribute is limb restriction. Regarding the rehabilitation measure of playing music or opera, since the location is the user's private space, the association analysis algorithm can determine that the location does not conflict with playing music or opera. However, the time of occurrence is relatively early, and the user's neighbors may still be asleep. Therefore, the association analysis algorithm can determine that there is some conflict between the time of occurrence and playing music or opera, but this conflict can be mitigated by lowering the volume. Rehabilitation measures involving playing music or opera do not require the user's attention or physical activity, and correlation analysis algorithms can determine that there is no conflict between them and behavioral attributes. Ultimately, the server can control a player in the bathroom to play music or opera at a volume similar to normal human speech (i.e., 20 to 50 decibels).

[0077] C3: Within a set time period, construct several different lists of behavioral guidelines by inserting different rehabilitation measures into each other.

[0078] Specifically, a behavioral guidance list can be created for each day of the week, ensuring that each day's list is different. The differences in the behavioral guidance lists are reflected in the different rehabilitation measures implemented within the same time period. For example, at 7:00 AM on the first day, a rehabilitation training measure of playing music could be introduced. On the morning of the second day, a rehabilitation training measure of playing the morning news could be introduced.

[0079] C4: Within a set time period, configure a voice interaction device 3 or a display 4 to guide the user to complete cognitive rehabilitation training by following an alternating list of behavioral guidelines.

[0080] Specifically, several voice interaction devices or displays can be installed in the user's living space. For example, voice interaction devices and displays can be installed in the living room, bedroom, and bathroom. The voice interaction devices can play voice prompts to the user and receive voice input signals from the user, allowing the server to interpret the user's feedback. Both the voice interaction devices and displays are connected to the server, enabling the server to control each one. The voice interaction devices can guide the user through rehabilitation measures via voice prompts, or the displays can provide intuitive instructions on how to perform rehabilitation measures.

[0081] Preferably, a one-week time cycle can be used to create seven different behavior guidance lists. Alternating the behavior lists means that the first behavior guidance list is implemented on the first day, the second behavior guidance list on the second day, and so on, with the seventh behavior guidance list implemented on the seventh day. This ensures that the user's rehabilitation training content is different every day of the week, which can prevent the user from becoming bored and thus effectively improve the rehabilitation training effect.

[0082] It should be noted that the specific embodiments described above are exemplary, and those skilled in the art can devise various solutions inspired by the disclosure of this invention. These solutions all fall within the scope of this invention and its protection. Those skilled in the art should understand that this specification and its accompanying drawings are illustrative and not intended to limit the scope of the claims. The scope of protection of this invention is defined by the claims and their equivalents.

Claims

1. A timing database based cognitive training strategy generation system, characterized by, This includes setting up several sensors (1) in the user's activity space that can collect information about different user movements, and configuring at least one server (2) for all sensors (1), wherein, The server (2) is configured to construct an information dataset by at least building a sensor device information table for describing the basic information of the sensor and a sensor data information table for describing the information content collected by the sensor. Furthermore, the server (2) is capable of periodically archiving and processing the data based on the upload time. Based on the correlation between basic life behaviors, sensors, and sensor attributes, the information dataset is classified to construct a sensor behavior table that includes at least the behavior name, behavior definition, triggering item, triggering sensor, triggering time, and location. Based on the sensor behavior table, basic life behaviors can be limited to a number of sensors that they can trigger by decomposing them into specific actions, the items triggered by the actions, and the sensors associated with the items. The server (2) constructs a behavior prediction classifier based on the classification data to predict the behavior that will occur at a future set time. The server (2) is also configured to construct a list of predicted behaviors, a list of interests, a list of lifestyle habits, and a list of rehabilitation measures; based on the correlation analysis results between the list of predicted behaviors, the list of interests, the list of lifestyle habits, and the list of rehabilitation measures, a list of behavioral guidance measures is constructed by inserting the rehabilitation measures in the list of specific rehabilitation measures into the list of predicted behaviors. The rehabilitation measures need to determine whether there is an execution conflict based on the time, place, and attribute of the specific behaviors in the predicted behavior list.

2. The timing database based cognitive training strategy generation system of claim 1, wherein, The server (2) can establish a list of rehabilitation measures through customization or network acquisition. When the sensor is triggered to perform data collection, the collected data is uploaded to the server (2) in real time for unified storage.

3. The timing database based cognitive training strategy generation system of claim 1, wherein, The server (2) can be configured with an external input terminal with a keyboard or voice input device. The user or the user's caregiver can input the user's interests into the server through the external input terminal to build an interest list.

4. The timing database based cognitive training strategy generation system of claim 1, wherein, The server (2) is able to determine the user's lifestyle characteristics based on a list of lifestyle habits.

5. A method of using the timing database based cognitive training strategy generation system of claim 1, wherein, The process includes the following steps: constructing a list of predicted behaviors, a list of interests, a list of lifestyle habits, and a list of rehabilitation measures; based on the correlation analysis results between the list of predicted behaviors, the list of interests, the list of lifestyle habits, and the list of rehabilitation measures, constructing a list of behavioral guidance by inserting the specific rehabilitation measures from the list of rehabilitation measures into the list of predicted behaviors.

6. The method of claim 5, wherein, It also includes the following steps: Within a set timeframe, several distinct behavioral guidance lists are constructed by inserting different rehabilitation measures into each other. Within the set time period, a voice interaction device (3) or a display (4) is configured to guide the user to complete cognitive rehabilitation training by alternating the execution of a behavioral guidance list.

7. The method of claim 6, wherein, The establishment of the behavioral guidance list includes the following steps: performing a first-level correlation analysis on the rehabilitation measures list and the lifestyle habits list to screen out rehabilitation measures that do not meet the requirements, thereby generating a first rehabilitation measures list.

8. The method of claim 7, wherein, The establishment of the behavioral guidance list also includes the following steps: performing a second-level correlation analysis on the first rehabilitation measures list and the interest list to maximize the satisfaction of the user's interests and hobbies, thereby generating a second rehabilitation measures list.

9. The method of claim 8, wherein, The establishment of the behavior guidance list also includes the following steps: performing a third-level association analysis on the second rehabilitation measures list and the predicted behavior list, and inserting the rehabilitation measures in the second rehabilitation measures list into the predicted behavior list to form the behavior guidance list.

Images